Most all you hear about "running an AI model" involves a GPU of some sort, but not every AI task is suited to that hardware. Smaller models or single-user latency-sensitive operations can benefit from running on the CPU instead, as it avoids the overhead of shuffling data to and from the GPU. There are also many situations where there is no GPU available to begin with, or it's a meek integrated affair with limited capabilities. Intel and AMD have recently released the full specification for the ACE CPU extensions that make it easier and more power-efficient to run the aforementioned AI tasks on x86 processors.

ACE comes in by offering a technical standard that leverages the existing AVX10 registers but adds silicon dedicated to matrix multiplication. This brings multiple benefits, but the key advantages are better power efficiency, easier development and optimization, and leveraging AVX's 512-bit inputs. The latter makes for easy integration with existing designs by eschewing the need for ACE-specific inputs.

Matrix multiplication is the cornerstone of AI workloads: take a table of numbers, and run a multiplication-addition loop over the whole thing. This has always been possible with most any CPU, though at limited speed. Even today, running these loops uses a lot of power, even when leveraging x86's AVX10 multiply-accumulate instructions — something that's technically a hack, as AVX wasn't designed with 2D matrix operations multiplication in mind.

For the same number of input vectors, ACE can perform 16x as many operations, compared to AVX10. Note this doesn't necessarily mean a 16x speedup, as that will depend on each individual implementation, but it's reasonable to expect that Intel and AMD will dedicate more silicon to this task in future designs to improve performance. Plus, as each ACE instruction performs more work than its equivalent AVX10 loop, there's less CPU instruction overhead and potentially better RAM bandwidth usage right off the bat.

The benefits go far beyond just using fewer instructions for the same thing. ACE is intended to be implementation-agnostic, meaning that ML frameworks and their underlying libraries (PyTorch, TensorFlow) can just write one code path instead of having multiple variations depending on the underlying hardware and its degree of AVX support.

ACE native supports most every data type used in ML operations (including but not limited to INT8, INT32, FP8, FP16, FP32, BF16), but it also can use Open Compute Project's MX block-scaled formats natively, something that AVX10 does not provide. Developers will also be able to move some NPU-specific workloads back to CPU when they need something done now and fast. In those situations, not having to deal with the fact that each NPU is different is a huge boon, too, as ACE offers a consistent target across x86 hardware.

Probably the most important processor in PC history was introduced on this day 48 years ago. June 8, 1978, marked the birth of the x86 architecture with the arrival of the 16-bit Intel 8086 CPU. This lineage continues to the majority of PCs today, almost half a century later. Ironically, this chip and its x86 architecture, the result of 18 months of R&D, was meant only as a stopgap because Intel’s complex, clean-sheet 32-bit iAPX 432 project was delayed.

The Intel 8086 was designed by a team of four engineers and 12 layout people led by Stephen P. Morse. Reports indicate that the impetus behind this project was to provide a practical, timely alternative to upcoming 16-bit Motorola and Zilog CPU designs. The fabled 8086 processor was only meant as a stopgap, as Intel had bitten off a bit more than it could chew with the iAPX 432 project, begun a year prior. As a side note, the 432 finally shipped in 1981 and was deemed too expensive, too complex, and fatally too slow when it arrived.

The 8086, the founding CPU in the x86 lineage, was marketed as Intel’s first 16-bit processor. It benefited from a degree of backwards compatibility with prior Intel 8-bit designs like the 8008, 8080, and 8085. Notable advancements over its predecessors included microcode for multiply and divide assembly language instructions.

Looking closer at the hardware tech specs, the Intel 8086 had around 20,000 transistors (29,277 including ROM and PLA) and was manufactured using Intel’s HMOS (High performance MOS) manufacturing process, originally developed for manufacturing fast static RAM products. The resulting 40-pin chip measured 33mm², and the minimum feature size was 3.2μm. Over its lifetime, it was released in clock speeds ranging from 5 to 10 MHz.

While the Intel 8086 founded the x86 architecture, the subsequent 8088 design (1979) would become the beating heart of the first IBM PC (1981) and that particular storied lineage.

Direct 8086 successors like the 80286, 80386, and 80486 would spearhead the Wintel alliance and establish the PC compatible as the default choice for productivity, home computing, and computer gaming enthusiasts until being sidelined by the Pentium CPU (also x86) from the mid-90s onwards.

Intel released its Core i7-8086K 40th anniversary chip in 2018. What next?

To celebrate 40 years since the original 8086, Intel launched the Core i7-8086K 40^th anniversary chip in 2018. It looks like a pretty safe bet that we should get another tribute in 2028, marking the half-century anniversary. What will we get for the 50^th? Something that embodies the fun and enthusiastic side of PCs, we hope.

Also, it will be interesting to see if Arm processors begin to impinge upon the dominant x86 designs from the likes of Intel and AMD in the Windows PC market in the next couple of years. We’ve had Windows-on-Arm efforts from Qualcomm and Mediatek try to usurp x86 with muted success.

At the recent Computex 2026 there was a lot of buzz about Nvidia’s RTX Spark Superchip, a powerful new Arm platform designed to transform Windows 11 into an agentic AI operating system. Looking back two years from now, will Nvidia and its partners have started to turn the tide against x86?

China's National Supercomputing Center in Shenzhen announced the Lingshen supercomputer project at a conference on April 24, targeting sustained performance above 2 ExaFLOPS using only CPUs and no foreign-made components.

The system would pack 47,000 processors into 92 compute cabinets, making it the first exascale machine designed to reach that performance tier without GPU accelerators. Lu Yutong, director of the Shenzhen supercomputing center and the system's chief designer, presented the technical details at the event.

Every other exascale system in operation relies heavily on GPUs or accelerator hardware. The U.S. Department of Energy's El Capitan, currently the world's fastest supercomputer, runs on 44,544 AMD MI300A APUs that tightly couple CPU and GPU silicon on a single package. Lingsheng’s CPU-only architecture would represent a fundamentally different approach to reaching exascale throughput.

The project is split into a pilot verification phase and a full production system, with the pilot using 100 Huawei Kunpeng servers built on Arm-based Taishan cores, totaling 12,800 cores. The production system scales to 1,580 blade servers using x86 CPUs with 101,120 cores and a theoretical peak above 10 petaflops. 16 four-way servers could add another 2,048 cores, and four eight-way servers could contribute a further 1,280, based on a machine translation of Chinese-language press materials.

The full system would include 36 network cabinets supporting a million-port interconnect, 650PB of planned storage across 428 nodes, and 67 liquid-cooled storage cabinets with 10 TB/s of bandwidth.

Lingsheng's claimed sustained performance of 2+ ExaFLOPS would, if achieved, exceed El Capitan's measured Linpack score of 1.809 ExaFLOPS. While El Capitan's theoretical peak is 2.79 ExaFLOPS, real-world Linpack results are always lower. Obviously, no Linpack or equivalent benchmark data exists for Lingsheng because the system hasn’t been built yet.

China’s claims are at best dubious. On a literal reading of the announcement, China says it might be able to achieve 2+ ExaFLOPS at some point in the future. But El Capitan is already theoretically capable of 2.79 ExaFLOPS, so it’s difficult to see how China’s project is ever going to "cast a new benchmark for global supercomputing," when it’s unlikely to be running even five years from now.

Then China says there’ll be no reliance on outside vendors for Lingsheng while also claiming that the production system will use x86 CPUs. China’s domestic x86 options are limited to Zhaoxin, a joint venture between VIA Technologies and the Shanghai municipal government, and Hygon, which originally licensed AMD's Zen architecture but lost access to updated designs following U.S. export restrictions.

Neither company has demonstrated processors competitive with current-gen parts from Intel or AMD, and the center and its partners didn’t name any suppliers or operational timeline for the completed system.

AMD appears to be fundamentally redesigning its CPU architecture for the upcoming Zen 6 chips, rather than making it a subset of Zen 5. As a part of those improvements, thanks to X user InstLatX64, we now know that AMD has finally adopted Intel's FRED instructions for its new silicon, along with fresh matrix multiplication and bit reversal instructions.

FRED is a wholesale replacement of the 80286-era Interrupt Descriptor Table (IDT) mechanism that's well over 40 years old by now. The IDT is currently the standard way to handle system events, like a network packet being delivered or mouse input, and passing its data to a driver or application.

AMD's adoption of FRED is the result of both companies creating the x86 Ecosystem Advisory Group to coordinate work on the new instruction set. Last October, one year after the alliance's inception, AMD agreed to implement FRED on its upcoming chips. At this moment, no production chips from either company support FRED, though it's a reasonable expectation that Intel Nova and Panther Lake, and AMD Zen 6, will be the first lineups to do so.

Things looked a little shaky on the interoperability front for a little while, as AMD had its Supervisor Entry Extensions (SEE). While Intel's FRED is a full and complete replacement of IDT, AMD SEE can best be described as a viable workaround that requires as few changes as possible to legacy software.

Linus Torvalds himself clearly outlined his opinion in a forum post, hoping that both vendors would implement both ideas, but praising Intel's approach as a more complete clean-room solution that gets rid of legacy cruft entirely.

Regarding software support, the Linux kernel has had provisional support for FRED since version 6.9. Once again, it's reasonable enough to expect that whichever version of desktop or server Windows comes next will also have the feature enabled. Note that only low-level software like operating systems and drivers can use the feature; it's not a concern for actual applications.

As to what FRED actually is, we can offer an oversimplification. A system event (say, getting an incoming network packet) is called an interrupt and the data and/or code transfer for handling involve a ring transition. This scenario happens constantly in any operating system, across every part, from storage I/O (a write being complete) to mouse input (say, a simple click).

Until now, programmers had to rely on IDT, for which the accurate technical adjectives are hacky and janky. IDT enabled only a half-assed transition from kernel code to application code, meaning programmers had to perform many other operations manually, be careful about edge cases, think about multiple ring levels, and work around potential race conditions (where two system events happen simultaneously and step over each other).

FRED improves on all fronts by providing one-shot instructions to guarantee a clean transition from kernel to application and back, with a consistent stack (the information about the event and where to continue execution). The main FRED instructions are atomic, meaning they execute all at once (or not at all), so programmers don't have to worry about an inconsistent state when they need to handle an interrupt. There's far less mental overhead, too, as there is much less to think about. Even the old ring levels are gone, being reduced to 0 (kernel) and 3 (user).

All told, this means that developers can just call on FRED and it'll do all the necessary work for them in one sitting, and no longer have to code their way around a lot of corner cases and theoretical issues. This should enable more stable kernels, system drivers, bootloaders, and other low-level software.

At least in theory, FRED is also bound to improve overall system performance, as it consumes fewer CPU cycles, resulting in lower event latency. These can add up under high load scenarios, like handling large network transfers, and perhaps even high-refresh-rate gaming and audio work — though potential uplifts will vary for each type of workload.

FRED will definitely be a performance boon for virtualization scenarios that have particularly involved event handling passing through multiple layers. The x86 architecture has long been criticized for holding on to legacy features for far too long, so it's good news that FRED is arriving soon.

Microsoft just announced that you can now install the Xbox app on all Arm-based Windows 11 devices. The Windows Experience Blog said that more than 85% of the Game Pass catalog is compatible with Windows 11 devices that run on Arm-based processors, and that it’s continuing to expand compatibility with the remaining titles. This news comes as Nvidia is reportedly debuting Arm-based N1X gaming laptops this quarter, challenging the established x86 architecture that still dominates the gaming PC industry and offering an alternative to Qualcomm’s Snapdragon X SoCs.

“Players can now download and enjoy a wide range of titles from the Xbox PC app game catalog on Arm-based Windows 11 PCs,” the company said in its blog post. “Today, more than 85% of the Game Pass catalog is compatible with these PCs, and we’re actively working with partners to expand support even further. For titles not yet supported running locally, Xbox Cloud Gaming remains an easy way to jump in while we work with partners to broaden compatibility.”

The company achieved this with the continued development of Prism, an emulator designed to run x86/x64 apps on Arm hardware. The addition of AVX and AVX2 support has expanded its compatibility with several modern games, while Epic Anti-Cheat support means that you can continue playing popular multiplayer titles on Arm-based Windows 11 PCs.

This is going to be excellent news for gamers, as it will allow them to have more hardware options when it comes to gaming. Even though many feel that x86 processors still deliver more raw horsepower, the efficiency of Arm-based chips is certainly attractive to those who prefer gaming on the go. Handheld console maker Ayaneo has been making Arm-based devices for several years now with the Snapdragon G3x Gen 2-powered Pocket S, but the lack of PC titles for Arm SoCs meant that it’s designed for Android gaming. This announcement would likely push handheld gaming console manufacturers to develop Arm-powered options, giving us the option to play PC games on battery for much longer than what we usually get from x86-powered devices.

Arm’s threat to x86’s dominance has got long-time rivals Intel and AMD cooperating, with the two companies forming the x86 Ecosystem Advisory Group in 2024. By the time of its first anniversary, we’ve already seen AMD and Intel agreeing to support some technologies, like the Advanced Matrix Extension, AVX10, Flexible Return and Event Delivery, and x86 Memory Tagging, for upcoming and future processors.

The Xbox app’s support for gaming on Arm-based processors will pose a threat to x86, especially now that a lot of games are compatible with the system. Even Steam is experimenting with Arm compatibility, with the client app adding native support for Apple silicon in 2025. But while this might make Intel and AMD nervous about their duopoly, especially in the gaming PC space, this is just going to be good news for gamers and consumers, as it drives competition between different ISAs and could even be the engine that would lead to more innovations in processors.

Intel’s 80386 turned 40 this week. Introduced in October 1985, the third-generation x86 processor — better known as the i386 — was the first 32-bit chip in Intel’s PC line, the origin point for the IA-32 instruction set, and the architectural turning point that transformed personal computing.

The i386 shipped with 275,000 transistors and ran at up to 16 MHz at launch. Internally, it brought 32-bit general-purpose registers, a flat memory model, and support for up to 4GB of address space, but the bigger change was to the system architecture. Protected mode, virtual 8086 mode, and hardware paging laid the groundwork for real multitasking and virtual memory on x86. Microsoft’s early i386 development kits included demos showing multiple DOS sessions running in parallel, each in its own paged VM. That became a core feature of Windows 3.0 in 1990, under the name “386 Enhanced Mode.”

Compaq was the first company to ship a machine based on Intel’s new CPU, following a rejection by IBM. The Deskpro i386, launched in September 1986, beat IBM to market by nearly a year. This was by design; the company had worked directly with Intel on the chip and had early access to silicon. Prices started at $6,499, and the system became a turning point for the PC industry, whereby Intel made the chip, but Compaq dictated the pace.

Linux, too, was built on an i386. Linus Torvalds’ first kernel release targeted 386-AT hardware explicitly, and his early development notes reference nothing older. The 386’s protected mode and paging features made it possible to build a real Unix-like system without writing elaborate workarounds. Torvalds dropped 386 support from the Linux kernel in 2012, more than 20 years after it launched.

Intel followed the i386 with the i486 in 1989, but the architecture stuck around for decades in embedded systems. The company didn’t end production of the chip until 2007. IA-32 — the i386’s instruction set — remained the backbone of Windows and most Linux distributions well into the 2010s.

Ultimately, the i386 wasn’t just a faster 286. It made x86 viable as a protected, multitasking software platform, something it had never been before. 40 years later, its core design still shows up in emulators, VMs, and legacy boot environments. The original chip might now be long gone, but the architecture it defined isn’t going anywhere.

AMD and Intel are celebrating one year since the formation of the x86 Ecosystem Advisory Group, an alliance designed to coordinate the evolution of the x86 instruction set architecture (ISA) and ensure that new features are supported by both leading CPU designers. In the first year, AMD and Intel have managed to ratify four new features that are set to be supported by the upcoming processors from these companies, including long-awaited memory tagging.

The new cross-vendor capabilities agreed upon by AMD and Intel are ACE (Advanced Matrix Extension) and AVX10 to enhance the performance of matrix multiplication and vector operations, as well as FRED (Flexible Return and Event Delivery) and ChkTag (x86 Memory Tagging) to reduce latency between software and hardware, as well as to detect errors like buffer overflows or use-after free bugs.

Intel's Granite Rapids processors already support AVX10.1 and AMX, whereas Sapphire Rapids were first to support AMX instructions. With the ratification by the x86 EDA, AVX10 and AMX will be supported by AMD's next-generation processors, though we can only wonder whether this will happen with Zen 6 or already with Zen 7. Other capabilities are less well-known.

Intel introduced FRED publicly in 2023, and by now, the capability is well-documented in developer documentation. The technology is described as a replacement for traditional x86 interrupt and exception mechanisms, so ultimately it is designed to simplify context switches, reduce latency, improve performance, and security when working with operating systems that support it.

FRED speeds up how the CPU switches between user mode (ring 3) and kernel mode (ring 0) with a hardware-defined entry and exit path. While this does not sound too impressive, replacing the old x86 mechanism (which uses the Interrupt Descriptor Table and IRET) is a big deal. At present, every time an application interacts with the OS (which happens millions of times per second), the CPU must switch between user mode and kernel mode, which introduces fairly high latencies with today's machines. Since the traditional IDT and IRET mechanisms are software-managed, while FRED provides a hardware-defined and verified entry and return path, replacing the former with the latter also improves reliability and security, in addition to performance

Up until today, AMD's stance on FRED was unclear, but now that the feature is recognized by the x86 EAG as a cross-vendor capability, it will be added to AMD's platforms over time.

Perhaps the most interesting addition to the list of cross-vendor x86 EAG features is the ChkTag (x86 Memory Tagging) capability, which has not been widely discussed before. The feature is added to catch memory safety errors — problems like buffer overflows, use-after-free, and out-of-bounds memory access — directly in hardware. Memory tagging is rapidly becoming a standard feature in modern CPUs as it is valuable (can catch a variety of bugs in hardware) and easy to implement, which is why modern processors from Apple and Ampere now support Arm's MTE technology.

It is hard to say when AMD and Intel plan to implement ChkTag (x86 Memory Tagging) in their processors. The announcement by the x86 Ecosystem Advisory Group signals both are committed to supporting this feature, but there is no obligation to implement it within a certain timeframe. Meanwhile, hardware changes of this depth typically require building them into the CPU microarchitecture itself, so expect support of FRED and ChkTag to come several years down the road.

Inkbox Software has released a recreation of a beloved isometric arcade game that runs directly from a PC’s Unified Extensible Firmware Interface (UEFI). Moreover, they coded their version of Zaxxon (1982) in x86-64 Assembly language, and have generously released it as open source under the GPLv3 license. There have been prior UEFI games and apps, but they are often incomplete or limited demos, and run within the UEFI as payloads, rather than booting directly to the game.

In the above video, Inkbox explains the reason they decided to begin this classic arcade gaming UEFI coding quest. In brief, our assembly coding hero says that they were “so tired of operating systems holding me back,” and that they wanted their hardware to run a classic game as it was meant to be played. That means “Written in assembly, without an operating system,” asserts Inkbox. “This is total freedom from big tech. Reject OS, Return to metal.”

Of course, there are significant hurdles to creating a handsome and responsive Zaxxon port for a PC UEFI. You can learn how Inkbox navigated this difficult path in their extensive and in-depth, yet engrossing video.

Parallels released Parallels Desktop 20.2.0, which lets you use x86 emulation on Apple Silicon Macs. The Parallels Desktop Blog says you can run x86_64 virtual machines on these devices. This allows you to install x86 versions of Windows 10, Windows 11, Windows Server 2019/2022, and some Linux distributions with UEFI BIOS on your Apple Silicon Mac laptop or desktop. You can also create new virtual machines with Windows 10 21H2 and Windows Server 2022.

Although Parallels brought Windows 11 on Apple-silicon Macs in 2021, you were limited to its Arm version because of Apple's processor architecture. This meant developers and users could not natively run Windows apps on their Apple-silicon Macs. This update finally answers that issue and allows for the installation of x86_64 Linux distros without resorting to Rosetta.

However, the company calls it a technology preview, not a full feature. That’s because x86 support on Parallels Desktop 20.2.0 for Apple-silicon Macs has a lot of limitations, the biggest one being its sluggish response. It says, “Performance is slow—really slow. Windows boot time is about 2-7 minutes, depending on your hardware. Windows operating system responsiveness is also low.” So, if you want to play with it right now, you must be patient.

While the feature supports installing 32-bit applications, it doesn’t support 32-bit operating systems, so you cannot install 32-bit Linux distros or the 32-bit version of Windows 10. Parallels also list several limitations, such as not supporting USBs, not recognizing devices you plug into the computer, and not making a sound. Aside from that, the virtual machine must be booted using Apple hypervisor; you can’t use Parallels hypervisor to launch it.

Since x86_64 support on Apple Silicon Mac is still in its Beta phase, the feature is hidden from the user interface, preventing average users from using it. If you want to experiment with it, right-click on the target virtual machine (which says Not Compatible underneath it) and click on Start using Emulator. Alternatively, you can run it using Parallel’s Command-Line Interface utility.

Qualcomm is assembling a team of developers to design server processors for the data center. Evidence of this comes via a vacancy posted on the company's own website (noticed by Longhorn). The company is looking for a server system-on-chip (SoC) security architect, and the details of the listing outline that the company is "developing reference platforms based on Qualcomm's Snapdragon SoC, delivering a comprehensive solution that includes hardware, software, reference designs, user guides, SDKs, and more."

The job listing confirms the existence of a 'Qualcomm Data Center' team that is designing a 'high-performance, energy-efficient server solution for data center applications.' While we are, of course, speculating, Qualcomm may plan to use the high-performance energy-efficient cores designed by its Nuvia team for data center applications. This same team developed the company's Snapdragon X processors for laptops, but it originally designed server processors before it was brought into the Qualcomm stable. 

In fact, Qualcomm used to develop and sell Arm-based data center CPUs back in the day without much success, which is presumably why it stopped. However, Amazon’s Graviton processors have proven that Arm-based solutions can thrive in the data center, and the market is wide open for more innovative entrants. Apparently, Amazon's success made Qualcomm rethink and return to the drawing board with data center CPUs. 

"We are seeking experienced SoC Security Architects to join our team," the job listing reads. "If you possess a deep understanding of hardware security architecture and are passionate about architecting and designing complex SoCs at advanced process nodes, we would be pleased to hear from you. This critical role involves architecting the next-generation security system and hardware infrastructure by collaborating with other platform architects to optimize overall Power, Performance, Area (PPA) efficiency and security assurance while ensuring compliance with industry standards."

Interestingly, when acquiring Nuvia, Qualcomm specifically noted that it planned to use Nuvia's Phoenix cores in its processors for client PC and not data centers, despite the fact that Nuvia originally designed its cores with the data center in mind. Nowadays, the cores dubbed 'Oryon' power Qualcomm's Snapdragon X processors for client PCs.

But now that Qualcomm has apparently changed its mind about data center products and is assembling a team of SoC developers (the core team is already there), it is reasonable to expect the company to develop a data center solution within the next couple of years. While we do not know for sure, the first devices to use Qualcomm's CPUs would likely be the company's platforms for 5G and then eventually 6G base stations, a space where the company controls the hardware and software stack. As for Qualcomm's development of chips for data centers currently dominated by x86, only time will tell if that is the company's ultimate intention. 

Intel's market capitalization dropped from around $198 billion in early 2024 to around $87 billion today, which makes the company an interesting acquisition target. Broadcom is certainly one of the companies that is known for making large acquisitions, and given its current focus on custom datacenter processors for AI, it could potentially be interested in taking over. However, its chief executive played down such a possibility citing a lack of Interest from Intel. 

"[Broadcom has] not been asked [to take over Intel]," said Hock Tan, chief executive of Broadcom, in an interview with the Financial Times. "I can only make a deal if it is actionable. Actionability means someone comes and asks me. Ever since Qualcomm, I learned one thing: no hostile offers." 

Back in 2018, Broadcom's $142 billion attempt to acquire rival chipmaker Qualcomm through a hostile takeover was stopped by then-President Donald Trump in a rare and extraordinary move. 

Earlier this year, we reported that Qualcomm was potentially interested in acquiring Intel, although it ultimately turned out that other reports said the company lost interest in buying Intel with all of its product divisions and manufacturing operations. 

Just like a merger with Qualcomm, an Intel merger with Broadcom could align with industry megatrends. Broadcom enjoys growing demand for AI processors it designs for clients like Google as well as connectivity solutions in data centers. By contrast, Intel's processors power the vast majority of servers as well as client PCs. Broadcom lacks presence in PCs and does not supply server CPUs. 

Intel's strength is, of course, its production scale and dominance in client computers. However, Intel is currently struggling to keep its process technologies competitive with TSMC and its processor products competitive with rival AMD. As a result, the appeal of the company is not particularly high at this point, especially because Broadcom has to integrate VMware, a company it acquired for $61 billion in 2022. 

Additionally, Broadcom is busy with its bespoke processor business for major high-tech companies, including ByteDance, Google, Meta, and reportedly Apple and OpenAI, which are seeking alternatives to Nvidia's processors. 

Tan described the ongoing investment frenzy in Silicon Valley, where tech giants are rapidly planning AI infrastructure projects for the next three to five years. These firms are constructing massive data centers, such as xAI's Memphis facility with 100,000 Nvidia GPUs and Colossus with up to one million Nvidia GPUs, to handle the computing needs of their ambitious AI programs. 

Tan has predicted that AI projects for Broadcom's clients could require clusters with up to one million processors as demand for computing power to train and run advanced AI systems continues to grow. He emphasized that the scaling principle—more data and processing yielding smarter AI—is driving these investments, despite uncertainty about when or if artificial general intelligence (AGI) will be achieved.

In the first sign of changes to come after the formation of a new industry group, Intel has confirmed to Tom's Hardware that it is no longer working on the x86S specification. The decision comes after Intel announced the formation of the x86 Ecosystem Advisory Group, which brings together Intel, AMD, Google, and numerous other industry stalwarts to define the future of the x86 instruction set.

Intel originally announced its intentions to de-bloat the x86 instruction set by developing a simplified 64-bit mode-only x86S version, publishing a draft specification in May 2023, and then updating it to a 1.2 revision in June of this year. Now, the company says it has officially ended that initiative.

"We remain deeply committed to the x86 architecture, as demonstrated by the creation of the x86 Ecosystem Advisory Group in collaboration with AMD and other industry leaders. This initiative reinforces our dedication to securing a strong future for x86, building on decades of software compatibility. While we have pivoted away from the x86S initiative, our focus remains on driving innovation and collaboration within the x86 ecosystem." - Intel spokesperson to Tom's Hardware.

Intel's decision to cease work on x86S makes sense. Any significant changes to the specification will require cooperation with AMD to ensure full interoperability and backward compatibility. Such is the nature of the 46-year-old x86—Intel and AMD are the only two primary x86 architecture licensees that build new processors in high volumes, creating a duopoly that requires standardization. 

It's also best if large software vendors and other ecosystem players add their full support, which is the crux of why the companies created the x86 Ecosystem Advisory Group. It's certainly best to have a unified, holistic ecosystem as x86 faces new challenges from Arm in both the consumer and data center markets. 

When Intel and AMD announced the group in an extremely rare joint announcement earlier this year, it was unclear how existing initiatives would be impacted. However, it certainly wouldn't make sense for Intel to forge ahead alone on x86S now that it can better ensure interoperability with any changes to the standard. Intel is also undertaking the largest layoff in the history of the company and a restructuring, which might have also played a role. That's not to mention the company is on the search for a new CEO.  

For now, it's unclear if any of the modifications that Intel made with its x86S initiative will come to fruition through the new x86 group. If they do, it will not be a unilateral decision on Intel's part — there will surely be plenty of discussion among the interested parties before any far-ranging changes are made. Naturally, that can be an issue with standards bodies and other groups, as getting a large group of stakeholders to agree on the finer points can slow progress and limit agility (we've seen that play out a thousand times).

Intel also has other future-looking x86 initiatives underway, such as its Flexible Return and Event Delivery (FRED) and the new AVX10 spec. Intel has confirmed that those initiatives are still underway.

Edit 12/20/2024 4:20am PT: Added Intel response that FRED and AVX10 initiatives are still underway.

When Zhaoxin, a joint venture between Via Technologies and Shanghai Municipal Government, introduced its KaiXian KX-7000 series of x86 processors in late 2023, it implied that the new CPUs will offer tangible performance improvements over its predecessors as they were based on a brand-new microarchitecture. While the new chips are probably faster than their predecessors, they are still years behind AMD and Intel CPUs.

PC Watch this week put Zhaoxin's eight-core KaiXian KX-7000/8 processor (Century Avenue, 8C/8T, 4MB L2, 32MB L3, two DDR4-3200/DDR5-4800 channels, 24 PCIe 4.0 lanes) into a test against AMD's six-core Ryzen 5 5600G (Zen 3, 6C/12T, 3.90 GHz – 4.40 GHz, 3 MB L2, 16 MB L3, two DDR4-3200 channels, 16 PCIe 3.0 lanes, 65W, released in 2021) and Intel's quad-core Core i3-8100 (Coffee Lake, 4C/4T, 3.60 GHz, 6 MB cache, two DDR4-2400 channels, 16 PCIe 3.0 lanes, 65W, released in Q4 2017). While these processors belong to different years and were even positioned differently when released, they operate at the same clock speed, which gives an idea about potential of the latest Zhaoxin microarchitecture (allegedly developed by Centaur Technology). 

PC Watch tested Zhaoxin's KaiXian KX-7000/8 in a variety of benchmarks, some artificial, others are more real-world applications.

When it comes to single-thread CPU performance via CPU-Z benchmark, the KX-7000/8 scored 335.9, significantly lower than the Intel Core i3-8100 (422.2) and AMD Ryzen 5 5600G (615.3). Since the KX-7000/8 has eight cores, it achieved a score of 2517.2 in multi-thread CPU-Z benchmark, outperforming the Core i3-8100 (1618.4) but falling short of the Ryzen 5 5600G (4790.5).  

To a large degree, such behavior defines the capabilities of the KX-7000/8: it can beat rivals in workloads that take advantage of eight cores, but it can never beat them in applications that require superior single-thread performance. 

In Cinebench R23, the KX-7000/8 scored 584 points in single-thread benchmark, which is significantly lower than both the Intel Core i3-8100 (953) and AMD Ryzen 5 5600G (1,435). In the multi-core Cinebench R23 test, the KaiXian KX-7000/8 achieved 3,595, roughly on par with the Core i3-8100 (3,585) but far behind the Ryzen 5 5600G (10,387).  

When it comes to PCMark 10 Express, the Zhaoxin KX-7000/8 exhibits lower performance across all categories compared to the AMD Ryzen 5 5600G and Intel Core i3-8100. The Chinese CPU struggles particularly with productivity, spreadsheet, and app scores, which pretty much indicates limited multitasking capability in heavy workload scenarios. While it may be sufficient for very basic tasks, it is outclassed by both competitors in all aspects. 

Just like its competitors, the KaiXian KX-7000/8 has a built-in graphics core, though its performance is so low that it does not really make sense to discuss it, but to focus on performance of the CPU in gaming applications when it is equipped with an external Radeon RX 6400 graphics card.

In 3DMark, the Zhaoxin KX-7000/8 demonstrates competitive performance with the Core i3-8100 in two scenarios (e.g., Time Spy and Night Raid) and outperforms it in one case (Night Raid), but generally falls behind of the Ryzen 5 5600G in all tests. But this is rather an exception than a rule. 

The KX-7000/8 is significantly behind both of its rivals in the Dragon Quest X benchmark even when it is equipped with an external GPU, which highlights its low single-thread performance. The same applies to the Final Fantasy XIV benchmark: the new CPU is barely a good choice for gaming. 

Where Zhaoxin's KaiXian KX-7000/8 can actually shine are sequential read and sequential write SSD speeds as unlike its rival, it has PCIe 4.0 lanes and therefore it can work with PCIe 4.0 x4 drives at full speed. 

While it is hard to draw conclusions based on one review and a limited number of benchmarks, it looks like Zhaoxin's latest microarchitecture is still years behind those of AMD and Intel in terms of single-thread performance. It is also behind rivals in terms of power consumption and efficiency. Nonetheless, now that Chinese government agencies and state-controlled companies are told not to buy PCs with foreign processors inside, they will buy Zhaoxin-based machines with not so many questions asked. 

Despite Qualcomm's ambitions and high hopes to control over 50% of the Windows market in the next five years, recent statistics show that its first-gen X Elite offerings are failing to gain consumer traction. TechRadar reports that of all PCs sold in Q3 24, Qualcomm's chips powered 720,000 units - capturing just 0.8% of the entire market. In other words, one out of every 125 devices sold in the last quarter featured an X Elite processor, which pales in contrast to Intel, AMD, and even Apple on the Arm front.

Undoubtedly, AI PCs are on the rise, primarily because virtually every new processor comes equipped with a Copilot PC+ compliant NPU. Per the report, 13.3 million AI PCs were shipped in Q3 this year - a 20% increase compared to last quarter. For the uninitiated, this stat refers to PCs featuring an NPU-equipped processor, like Intel's Meteor Lake and Lunar Lake offerings, AMD's Phoenix, Hawk Point, and Strix Point series, and Qualcomm's X Elite SoCs.

Despite a 180% surge in X Elite sales from Q2 24 to Q3, Qualcomm's chips power less than 1.5% of Windows PCs. This lackluster reception is primarily due to unoptimized software or the lack thereof. For example, Microsoft's tardy release of an official Windows 11 ISO for Arm devices left early X Elite adopters in the cold for months. Additionally, Linux support isn't as robust as existing x86 alternatives; longer battery life isn't sufficient or incentivizing enough to drive adoption.

Qualcomm is preparing to launch budget X Elite laptops next year to somewhat alleviate this situation. On that note, the fruits of Qualcomm's acquisition of Nuvia are materializing—both in terms of high-stakes litigation and powerful mobile SoCs, namely, the Snapdragon 8 Elite designed using the Oryon 2 architecture. Rumors suggest that the X Elite's successor will jump directly to Oryon 3, aiming for a 2025 reveal/launch.

In any case, compatibility will always be an issue with these Arm SoCs (in Windows) - be it any chip maker. Apple's sheer market share and dominance compelled developers to write and optimize software for its silicon. Collaborating with Microsoft is vital for Qualcomm in this regard, however, it'd still have to compete with other x86 players in the Windows space; Intel and AMD.

SoftBank Group plans to build the industry's first Nvidia Blackwell-based AI supercomputer, according to an announcement at Nvidia's AI Summit in Tokyo this week. This announcement confirms SoftBank's AI ambitions, which have been rumored for some time. Interestingly, the supercomputer will use x86 CPU-based DGX B200 servers, not Arm-based processors. 

SoftBank's s telecom arm will build an AI supercomputer based on Nvidia's DGX B200 servers, featuring x86 processors and eight B200 GPU modules per server. SoftBank plans to adopt the Grace Blackwell platform, which combines Arm-based Grace CPUs and Blackwell AI GPUs, eventually (it's unclear when).  

Neither Nvidia nor SoftBank disclosed how many B200 processors the Japanese company will purchase — merely that it will get "a lot of them." Elon Musk's companies Tesla and xAI are deploying tens of thousands of Nvidia processors to train their latest AI models, so SoftBank will need to be somewhere in that ballpark if it wants to compete. But getting enough Nvidia GPUs will be hard, as virtually all global high-tech companies currently need them. 

Beyond supercomputers, SoftBank's telecom unit plans to integrate Nvidia's AI technology into Japan's mobile networks to create what Huang described as an 'AI grid' across the nation. A significant feature of this rollout will be the deployment of AI radio access networks (AI-RANs), which will use less energy than traditional RAN hardware. Such RANs are said to be a better fit for innovative services, such as autonomous vehicles and remote robotics, and will repurpose the communications network as a nationwide AI infrastructure. 

SoftBank's telecom division will be trialing the AI-enhanced network in collaboration with Fujitsu and IBM subsidiary Red Hat. 

Nvidia and SoftBank claim the upcoming supercomputer will be the first Blackwell-based AI supercomputer, but this may not be technically true. Microsoft is already testing Blackwell-based machines, and is expected to be the first company to deploy them. But perhaps SoftBank will be the first company to deploy DGX B200-based servers. 

It will take some time before Microsoft optimizes its software for Nvidia Blackwell, so don't expect the company to roll out a wide range of services powered by Nvidia's latest GPUs overnight. 

After a delay involving an Nvidia B100/B200 design flaw, which was discovered and promptly fixed earlier this year, the final revision of B100/B200 entered mass production in late October. As a result, the first batch of these processors isn't expected to arrive until late Jan. 2025. Nvidia revealed that, in order to meet the high demand for Blackwell GPUs from leading cloud providers such as AWS, Google, and Microsoft, it will need to start shipping some of the early low-yield Blackwell processors in 2024; it's not clear how many of these GPUs will be delivered to datacenters throughout the year. 

Earlier today on Twitter, AMD engineer Phil Park identified a curious nugget of PC architectural history from, of all places, a year-old Quora answer posted by former Intel engineer Robert Colwell. The nugget indicates that Intel could have beaten AMD to the x86-64 punch if the former wasn't dead-set on the x64-only Itanium line of CPUs. The historical context of this story and the comments within are fascinating, especially considering today's desktop CPU architectures, where "64-bit" is immediately understood to mean x86-64 — because who wants a computer that can't natively run most applications made for a PC?

To Intel's credit, it isn't that Intel Itanium processors completely did away with x86 software and architecture. However, they were intended to in the long run, especially as a measure of definitively striking down AMD in the marketplace. Unfortunately, this meant that the pure 64-bit architecture of Intel Itanium did not allow 32-bit (x86) applications to run natively, and the emulation solutions performed poorly. As a result, Itanium landed with a thud in the market despite being among the first to the 64-bit punch. Backward compatibility matters a great deal for big software and hardware purchases of any kind, but especially in the PC market. This fact was especially true in the server and enterprise markets where Intel targeted the Itanium CPUs.

As Intel's former chief x86 architect, Colwell knew this to be true, and thus designed the earliest internal versions of Pentium 4 CPUs to be x86-64 instead of mere x86 chips. Unfortunately, this design path was axed by the higher-ups at Intel, who clearly feared it would eat into or harm Itanium (much in the same way that AMD's debut x86-64 chips would end up doing anyway). In the face of repeated firing threats, the knowledgeable Colwell left the gates for 64-bit in Pentium 4, but fused off the functionality to make the inevitable return to x86 easier for Intel when the time came.

However, the story of Intel Itanium doesn't quite end at the launch of AMD64/x86-64 in 2003 despite the obvious chilling effect. Itanium still kept support and was iterated on until February 2017, and shipments didn't completely halt until July 2021. For some fringe scenarios, Itanium was very performant and even ideal.

But as Colwell knew, even back then, backward compatibility is king, especially on the PC, and especially in the enterprise. And don't even get me started on what consumers are willing to do for some good backward compatibility.

In a rare public-facing display of cooperation between two of the industry's fiercest rivals, Intel and AMD jointly announced the formation of a new x86 advisory group to ensure a unified x86 instruction set architecture (ISA) moving forward, an important consideration given both the potential addition of new features and the various x86 simplification efforts already underway.

The companies announced the new x86 ecosystem advisory group at the 2024 OCP Summit. It already has several notable industry software and hardware stalwarts as participants, including Google, Broadcom, Dell, HP, Lenovo, Microsoft, Meta, Oracle, and Red Hat, with more to come in the future.  

The 46-year-old x86 is the most prevalent ISA used for general computing for PCs and data centers, and Intel and AMD are the only two primary x86 architecture licensees that build new processors in high volumes, creating a duopoly. Cooperation between the two, with the input of a bevy of customers and end users, will help to build a more unified approach that reduces or even eliminates custom ISA implementations that can be problematic for the duopoly's hardware and software customers. That's becoming even more important as the x86 ecosystem faces intense pressure from Arm in both the consumer and data center markets, not to mention the continuing rise of RISC-V.

The new group intends to standardize at least some of the new additions and alterations to the x86 ISA, which includes several simplification efforts already underway. The alterations and areas of collaboration aren't yet defined, but there are plenty of clear candidates that could be up for discussion. 

For instance, AMD has its Supervisor Entry Extensions, which are designed to clear up some of the older cruft in the ISA, while Intel has its Flexible Return and Event Delivery (FRED) code, which has similar goals. Intel has even begun work on X86S, a simplified 64-bit mode-only implementation designed to strip out even more of the legacy cruft. 

While cleaning up is an obvious area of potential collaboration, the x86 ISA is also constantly moving forward with new additions, and here, alignment between Intel and AMD could become even more important. For instance, Intel has recently introduced AMX, a matrix math extension that dramatically boosts performance in AI inference workloads, and AVX10. Both would benefit from more uniform implementations among the vendors. However, there will certainly be even newer and as-yet-unforeseen additions, particularly regarding various extensions that support AI operations. 

Before the announcement, we spoke with AMD's Forrest Norrod (EVP and GM of its Data Center Solutions BU), and Intel's Justin Hotard (EVP and GM of the Data Center and AI Group).

"AMD and Intel are excited to come together in this," said Norrod, "and we think it's one of the most significant shifts in the x86 architecture in decades. As Justin said, x86 is the de facto standard. It's a strong ecosystem, but it's one that Intel and AMD have co-developed in a way, but at arm's length. And that has caused some inefficiencies and some drift in portions of the ISA over time."

"When we step back and look at it, we see that it is a de facto open ecosystem. Open ecosystems benefit from having consortiums with stakeholders, all with a voice in driving the ecosystem forward," said Norrod. 

However, we shouldn't expect AMD and Intel to back off their competitive instincts. "We're going to remain fierce competitors. You know, Justin and I are friends first, but when we show up to work each day, we are fiercely trying to compete on behalf of our companies and ensure that our customers have compelling choices from each one of our companies. But we can compete even while we're driving industry standards together, and there's a rich history of Intel and AMD doing that," said Norrod. 

"I think the other question you might ask is, why now? It's because we're seeing a real shift in the demand for compute. We're at an inflection point. We think it's the right time to enable this new, consistent architecture as a source of innovation for our ecosystem," Hotard added. 

"We're looking at new innovations in the architecture, and going forward, we implement them in standard ways, and we think that provides easier adoption for the entire ecosystem, whether they're hardware vendors or software developers," said Hotard. "There have been many different feature requests that we've had for x86; some have been adopted by one of us, and some have been adopted by neither of us. I think the agreement that we will have a consistent approach is foundational, so we have good predictability."

Better collaboration and standardization will also benefit security-related features. Intel and AMD already collaborate on multiple axes behind the scenes, and standards such as PCIe, ACPI, and USB, among others, result from intense collaborative efforts across the industry. Naturally, coming to some collaborative decisions about the future of x86 could become contentious between the two rivals, but Norrod cites the presence of other companies and customers in the group as a guiding force that will help keep the companies headed in the right direction. 

Via Technologies is also another x86 licensee, but the company's status in the group is unknown for now. Intel purchased much of the company's assets, though Via does retain the right to make x86 processors. I asked Intel and AMD about the matter. Representatives said Via is free to join the group, but they will also follow up with more details.

AMD and Intel intend to chart a new course together for x86 to ensure compatibility and consistency. However, given the long design cycles for modern processors, we shouldn't expect to see the impact of the new x86 advisory group for at least a year, if not longer. For now, neither company is willing to commit to specifics, but we're sure that will change in the coming months. 

Intel chief executive Pat Gelsinger revealed the company’s next steps this week in the form of three main goals that might turn around the struggling technology company. But tech enthusiasts worry: Will it be enough?

In a recent meeting, Intel leadership discussed how to pursue their long-term strategy. The Board, comprised of independent directors, emphasized the importance of Intel’s efficiency, profitability, and market competitiveness. In the end, they agreed upon three core objectives: enhancing Intel’s foundry operations, optimizing costs to hit the $10 billion savings target, and refocusing on its core x86 technology while driving forward its AI initiatives. 

Notable announcements include a multi-billion-dollar partnership with Amazon Web Services (AWS) and the decision to put fab and packaging facility projects in Germany and Poland on hold for the next two years while continuing to build capacity in the U.S.

Foundry business remains key to Intel's future success

Intel's internal manufacturing remains crucial to the company's long-term success. However, the company plans to make Intel Foundry an independent subsidiary, completing the separation of profit and financial reporting from Intel Products started earlier this year.

This new structure will offer more apparent distinctions for customers and suppliers while allowing Intel to explore external funding and optimize capital strategy for the Foundry and Products units. The leadership team remains unchanged, reporting to the CEO, with an independent board providing oversight to ensure transparency and accountability.

A more focused Intel Foundry will enhance collaboration with Intel Products, while Intel's design and manufacturing capabilities will remain a competitive advantage.

Intel says that its Foundry unit has seen increased customer interest, with a significant uptick in demand for its advanced packaging technologies. Intel has tripled its deal pipeline since the start of the year, and the collaboration with AWS (more on this in a while) is a testament to this growth.

Germany and Poland facilities are on hold; U.S. projects to continue

Intel is also enhancing its capital efficiency within the foundry business. After a period of significant investment in manufacturing, the company is shifting toward a more balanced development pace.

Intel remains committed to its manufacturing projects in the U.S., so the fabs and packaging facilities in Arizona, Oregon, New Mexico, and Ohio will be built according to Intel's plans. It may not be particularly surprising that Intel received billions of dollars from the U.S. government to expand its production capacity in America.

Given the current demand outlook, Intel will pause its fab and packaging facility projects in Germany and Poland for two years as part of the overall business optimization. Intel recently expanded the capacity of its fab near Leixlip in Ireland and expected this facility to remain its lead European hub. The company did not reveal whether demand expectations concern foundry or internal products, but it is obvious that Intel limits the growth of its production capacity in general.

Also, Intel will build the shell for its advanced packaging facility in Malaysia but will not equip it before demand picks up.

Amazon Web Services to use custom Intel Xeon processors

One significant development is Intel's expanded partnership with Amazon Web Services (AWS). Under this multi-year agreement, Intel will produce custom AI fabric chips for AWS using its Intel 18A technology and a custom version of the Xeon 6 CPU based on Intel 3, as planned. In the future, Intel will produce AWS designs using Intel 18A, Intel 18AP, and Intel 14A process technologies. AWS has become Intel Foundry's long-term client, which is a big deal.

The manufacturing and packaging contribute to Intel's ongoing collaboration with AWS, which already uses Intel Xeon Scalable processors (many of which are the so-called off-roadmap semi-custom processors). Intel calls this collaboration a 'better together' strategy, integrating foundry services, infrastructure, and x86 products.

Streamline product portfolio

Intel’s x86 instruction set architecture (ISA) and expertise in its enhancement have been its core strengths for decades, so Intel is certainly not giving up on x86.

Instead, Intel is streamlining its portfolio, including developing competitive x86 offerings across its client, edge, and data center lineups. Intel’s collaboration with AWS exemplifies its commitment to providing customized x86 solutions for customers to better compete against Arm and RISC-V.

The company also sees opportunities to simplify its portfolio to increase efficiency and innovation. To that end, it integrates its Edge and Automotive businesses into its Client Computing Group (CCG), allowing it to expand its leadership in AI PCs and vertical edge solutions. Networking and telco will remain the focus of Intel’s Network and Edge (NEX) group, while its Integrated Photonics Solutions division will merge into the Data Center and AI (DCAI) group.

Building a leaner Intel

These changes are vital to creating a leaner, more streamlined, and efficient Intel by making internal changes and reducing the company's workforce by 15,000.

With efforts underway to cut around 15,000 jobs by year-end, the company has already reached more than half of this target through voluntary retirements and separation packages. In addition, Intel plans to reduce or exit two-thirds of its global real estate by the end of the year as part of its cost-saving initiatives. To bolster its financial position, Intel is also preparing to sell part of its stake in Altera, generating funds ahead of Altera's anticipated IPO.

If you're looking for the best CPUs for gaming or one of the best budget CPUs, there are only two choices: AMD and Intel. That fact has spawned an almost religious following for both camps, and the resulting AMD vs Intel flamewars make it tricky to get unbiased advice about the best choice for your next processor. But in many cases, the answer is actually very clear: AMD's X3D chips dominate gaming performance, from the Ryzen 5 7600X3D up to the Ryzen 9 9950X3D2. For application performance, AMD still tops the charts with flagships, but Intel's recent Arrow Lake Refresh chips provide nearly the same level of performance under $300 based on the data in our CPU benchmark hierarchy.

This article covers the never-ending argument of AMD vs Intel desktop CPUs (we're not covering laptop or server chips). We judge the chips on nine criteria based on what you plan to do with your PC: pricing, specifications and features, gaming performance, productivity performance, driver support, power consumption, overclocking, chipset and socket support, and security, giving us a clear view of the state of the competition. Throughout each section, we'll also discuss the process nodes and architectures influencing the moving goalposts. However, each brand has its strengths and weaknesses, so which CPU brand you should buy depends mostly on what blend of features, price, and performance is important to you.

The latest 'Arrow Lake Core Ultra 200S series builds on Intel's hybrid architecture formula but leverages a new design philosophy that constrains gaming performance. However, the chips do provide strong performance in single- and multi-thread productivity workloads, along with improved energy efficiency. New refreshed chips like the Core Ultra 7 270K Plus shore up the gaming front and provide a big boost in productivity performance, as well.

AMD's answer is its Zen 5 Ryzen 9000 series. These chips debuted with a fizzle on the gaming front, but subsequent firmware and operating system updates have improved the overall performance significantly, making these chips the bang-for-the-buck leader for most users looking for a balanced system.

AMD's powerful gaming-optimized Ryzen 9000 "X3D" models utilize a breakthrough 3D chip-stacking tech to take the lead in gaming performance, as you can see in our Ryzen 9 9850X3D review that finds the 9850X3D to be the fastest gaming CPU on the planet, bar none. AMD has also leveraged this X3D tech with its older architectures to make incredibly value-centric gaming chips, like the Ryzen 5 5600X3D. The gains in gaming performance are phenomenal, so much so that the competition against Intel chips isn't even close, giving AMD a walk-in touchdown for gaming PCs.

AMD vs Intel: Which CPU is Best?

Comparing Intel vs AMD in 9 categories in 2026, AMD and Intel are in lockstep with a win in three categories each, with the brands coming in tied in two categories. Despite long-standing as the value alternative, AMD’s original Zen microarchitecture has kicked off a slow rise to the top. The Steam hardware survey now shows AMD only slightly behind Intel in market share, and from the Passmark database, AMD outpaced Intel in desktop market share for the first time last year.

Although the earlier years of Zen were met with an audacious Intel, that story quickly changed as AMD started to gain ground, particularly around the release of Zen 3. Intel was finally able to move onto its 10nm node after years of manufacturing challenges, as well as introduce a hybrid architecture in a desktop x86 CPU. Alder Lake, Raptor Lake, and Raptor Lake Refresh all maintain competitive positions against their Zen 3, Zen 4, and even Zen 5 counterparts, but Intel’s market dominance – and the price it demanded from it – has disappeared in desktops.

Fierce competition is good for everyone, but Intel didn’t maintain its trajectory, instead opting for a radically different architecture with Arrow Lake that ditched Hyper-Threading. In AMD’s camp, it doubled-down on the wildly successful 3D V-Cache, introducing a second generation of the technology that places additional cache below the compute die for more thermal headroom, and as a result, higher clock speeds.

Although Arrow Lake brings much-needed efficiency improvements to Intel’s desktop line-up, as well as great overclocking headroom on CPU cores and the memory subsystem, AMD dominates gaming performance with its X3D lineup, sometimes by as much as 30%. At the same time, AMD is able to match or exceed Intel’s productivity performance, despite how impressive the Arrow Lake architecture is from a technical perspective.

Broadly, AMD earns our recommendation for most people. It clearly leads in gaming performance with X3D CPUs, and despite stagnant specifications, AMD has continued to show architectural advantage in comparative performance scenarios. Intel has gained by some ground with the 270K Plus and Core Ultra 5 250K Plus, which are both excellent processors are affordable prices. For now, however, AMD remains our go-to recommendation.

AMD vs Intel CPU Pricing and Value 2025

AMD has long stood as the value king, and although you can get a lot for your money with chips like the Ryzen 7 9800X3D, Intel has slid into the value position as AMD continues to rise in market share. Looking at like-for-like CPUs in the AMD vs Intel battle, the two are in lockstep on pricing within a margin of $20 to $30 in most cases, outside of sales.

For the current generation lineup, Intel’s Arrow Lake chips are slightly cheaper at the Core Ultra 9 and Core Ultra 7 levels, though only by $20. Factoring in sales, Arrow Lake ends up much cheaper. At the time of writing, for instance, the Core Ultra 9 285K is available for $500 while the Ryzen 9 9950X will run you $570.

The prices above are based on the average selling price of the main lineups from AMD and Intel over the past 60 days. Notably, these aren’t list prices, which aren’t reflective of what you can expect to spend, especially this late into a generation. As an example, the Core Ultra 7 265K launched at $395 while the Ryzen 7 9700X launched for $359. Now, however, Intel comes out slightly cheaper.

AMD leads in overall value for lower-end offerings like the Ryzen 5 9600X, but Intel isn't far behind with the recent Core Ultra 5 250K Plus. In gaming performance, you can see Intel is delivering better average frame rates, as well, though only with its newer Arrow Lake Refresh chips.

AMD and Intel flip positions when looking at last-gen options, with only the Core i9-14900K coming in slightly less than the Ryzen 9 7950X. Intel has clearly recognized the value proposition of the Core i5-14600K and Core i7-14700K, with both now selling for around the same price as their Arrow Lake counterparts.

Winner: Intel. Intel and AMD are more competitive on pricing than they’ve been in years. Intel comes out slightly ahead on value with lower-end SKUs, but value demands context. The Ryzen 7 9800X3D might be a worse value than the Core i5-14600K, but if you have $480 to spend on a CPU and want the best gaming performance, AMD’s X3D chip is right for you.

However, Intel shook things up massively with the 270K Plus and 250K Plus. Now, Intel is offering competitive gaming performance around $200 to $300 while pushing flagship-level application performance. For now, at least, Intel is in the driver's seat when it comes to overall value through Arrow Lake Refresh.

AMD vs Intel CPU Gaming Performance 2025

In the AMD vs Intel CPU gaming performance battle, AMD's X3D processors hold the lead in all critical price bands. Below, we have a wide selection of collective gaming performance measurements for the existing chips in the different price bands. You can see a much more holistic view in our CPU Benchmarks Hierarchy.

As you can see, AMD's standard 'Granite Ridge' Ryzen 9000 models take the lead over Intel's current-gen Core Ultra 200S 'Arrow Lake' processors, and the gaming optimized Ryzen 9000X3D chips, which cost more than the standard models, take an overwhelming lead, often by 30% or more.

Intel's current-gen Core Ultra 9 285K is the new flagship, but its unique tile-based (chiplet) design has a negative impact on gaming performance, so it isn't as fast as the prior-gen models in gaming. That leaves the Core i9-14900K as the fastest Intel gaming chip, and it often sells at a steep discount. The recent Core Ultra 7 270K Plus, however, is just marginally behind the Core i9-14900K, and it's available for around $350.

AMD's standard Ryzen 9000 models, like the flagship Ryzen 9 9950X, are faster in gaming than Intel's Core Ultra 200S series, but Intel's previous-gen 14900K holds the edge. However, AMD's X3D models, like the Ryzen 7 9800X3D and the Ryzen 9 9950X3D, are the fastest gaming chips on the planet, beating Intel's current-gen chips by 30+%. These chips also deliver the full performance of the standard models in productivity workloads, eliminating the tradeoffs associated with the X3D models that we've seen in the past. The X3D models do come at a premium, but they justify their price tag with the fastest gaming performance money can buy.

Older X3D chips leverage AMD’s first-gen 3D V-Cache, which stacks the cache on top of the compute die rather than below, which is the case with the latest Zen 5 X3D offerings. That design limits thermal headroom, and in turn, clock speeds, so expect a dip in productivity performance if you aren’t using one of AMD’s latest X3D offerings.

Kicking your resolution up to 1440p and beyond typically pushes the bottleneck back to the GPU, so you won't gain as much from your CPU's gaming prowess. However, a bit of extra CPU gaming performance could pay off if you plan on updating your graphics card (see our best graphics cards) with a newer generation while keeping the rest of your system intact. We expect most builds in the mid-range to come with lesser GPUs, which generally serve as an equalizer in terms of CPU performance.

In terms of integrated graphics performance, there's no beating AMD. The company's current-gen Phoenix Point APUs offer the best performance with the 8700G and 8600G. We’ve seen even more capable APUs out of AMD with its mobile Strix Point offerings like the Ryzen AI Max 395, but Strix Point hasn’t made its way to desktop yet.

Winner: AMD. AMD's standard Ryzen 9000 models deliver solid performance in gaming, beating Intel's current-gen models, but they trail the previous-gen Intel chips. However, the Ryzen 9000X3D models take the unequivocal lead in gaming by massive margins, giving AMD an easy win in the gaming department. The previous-gen Raptor Lake Refresh models are a distant second in gaming performance, and they aren't as performant as the Ryzen 9000 models in productivity applications, making them a tough choice as an alternative.

AMD vs Intel Productivity and Content Creation Performance 2025

In the non-gaming performance battle of AMD vs Intel CPUs, Intel's Arrow Lake chips have made great strides against AMD's finest, but they fail to deliver a tangible enough lead over AMD's core-heavy models to make productivity performance a make-or-break buying decision. AMD leads in overall multi-threaded performance with its Ryzen 9 9950X3D and 9950X by a slim margin, but Intel keeps a solid grip on the lead in single-threaded applications. The new Arrow Lake Refresh CPUs don't radically change the performance picture overall, but they come in much less than AMD's Zen 5 competition.

Arrow Lake marks Intel's continuation of using a mixture of two types of cores for the mainstream desktop PC. The big performance cores (P-cores) are best for latency-sensitive work, giving Intel the uncontested lead in single-threaded applications. The efficiency cores (E-cores) step in to add some additional heft in threaded and background applications, which pays big dividends in heavily-threaded content creation and productivity applications.

In contrast, AMD continues to leverage its tried-and-true P-core only design, which also supports AVX-512, a big consideration for those who leverage powerful productivity applications that employ the high-performance instructions. Zen 5 brought a 512-bit data path for AVX instructions – Zen 4 used two 256-bit paths – which provides a massive performance advantage in applications that leverage AVX-512, even compared to Zen 4. Intel’s latest CPUs don’t support AVX-512 instructions.

Solid performance in single-threaded work equates to faster performance in all manner of workloads, particularly day-to-day applications that rely on snappy responsiveness from the processor. The Core 7 270K Plus has taken the uncontested lead in single-threaded performance across the full suite of our benchmarks, and it comes in at a mid-range price point. And on the multithreaded front, Intel is able to match the Ryzen 9 9950X despite the Core Ultra 7 270K Plus coming in at half the price.

Winner: Intel. Although AMD offers compelling productivity performance with its most expensive offerings, the Core Ultra 7 270K Plus and Core Ultra 5 250K Plus punch far above their weight class in productivity performance, easily delivering flagship performance at entry-level prices.

AMD vs Intel Processor Specifications and Features 2025

Modern Intel and AMD CPUs sport much different architectures, so comparing specifications directly is a fool’s errand in most cases. AMD has a traditional approach with a homogeneous core architecture that sports simultaneous multi-threading. Intel moved onto a heterogeneous architecture with Alder Lake, which it has maintained since, and it removed Hyper-Threading with Arrow Lake.

This departure in architecture means you can draw inaccurate conclusions from specs alone. For instance, Intel’s flagship Core Ultra 9 285K only has 24 threads compared to AMD’s 32 on the Ryzen 9 9950X. Despite that, the two chips offer competitive multi-threaded performance, as you can see in our performance results in the above section.

The cores and clocks aren’t as important as they once were, at least when comparing a Ryzen 9 to a Core Ultra 9, or a Ryzen 5 to a Core Ultra 5. Instead, there’s been a stark shift in focus toward cache in the face of 3D V-Cache. That’s the marquee feature AMD has been running on for the past few years, offering chart-dominating performance in games that Intel has yet to answer. Previously, 3D V-Cache came with a slight compromise to productivity performance due to clock speeds on eight-core parts and CCX-to-CCX latency on 12- and 16-core parts. AMD has rectified that issue with second-gen 3D V-Cache, however.

Apples-to-apples, Intel and AMD offer similar cache amounts when excluding X3D parts, though Intel more evenly distributes the cache between L2 and L3, while AMD focuses the lion’s share of cache in the last level. Intel is rumored to pack bLLC into its upcoming Nova Lake generation to answer 3D V-Cache. Intel currently deploys this style of cache packaging in its Clearwater Forest server chips, but it has yet to show up in a consumer CPU.

Winner: AMD. Intel has highlighted several features over the past several generations, from Thread Director to Intel APO and even the new iBOT feature, but these features have mainly served to bring Intel’s unique architectural design up to par. Meanwhile, AMD has pushed ahead with a big win in gaming due to 3D V-Cache, which only improved in its second generation. Add on top of that AVX-512 support, as well as a true 512-bit data path in Zen 5, and Team Red easily wins this category.

AMD vs Intel CPU Power Consumption and Heat 2025

When comparing AMD vs Intel CPU power and heat, the former's 4nm TSMC process node makes a big difference. Power consumption comes as a byproduct of design choices, like lithography and architecture. However, higher power consumption often correlates to more heat generation, so you'll need beefier coolers to offset the heat output of greedier chips.

Overall, Intel has reduced its power consumption from meme-worthy to an acceptable level, but it still consumes more power than Ryzen. Still, in aggregate, AMD's 4nm chips either consume less power or provide much better power-to-performance efficiency.

As you can see above, you'll get more work done per watt of energy consumed, which is a win-win, and AMD's cooling requirements aren't nearly as overbearing. Arrow Lake brings Intel more in-line with AMD, both when looking at peak power consumption and productivity efficiency. Even then, however, Intel’s latest offerings end up slightly ahead of AMD in overall power consumption, and slightly behind in over efficiency.

Winner: AMD. In judging AMD vs Intel CPU performance per watt, it's impossible to overstate the importance of having the densest process node paired with an efficient microarchitecture, and TSMC's 4nm and AMD's Zen 5 are the winning combination. The latest Ryzen processors consume less power than Intel on a power-vs-performance basis, even in the face of Arrow Lake.

Intel vs AMD CPU Overclocking 2025

There's no debate when you compare Intel vs AMD CPU overclocking. Intel offers the most overclocking headroom, meaning you can gain more performance over the baseline speed with Intel chips than you can with AMD's Ryzen processors. To learn more about overclocking, head to our How to Overclock a CPU guide.

Dynamic boosting algorithms and increasingly multi-threaded applications have offset some of the previous benefits of overclocking, at least for core clock speeds. With AMD’s current CPUs, and up to Raptor Lake Refresh for Intel, a modest all-core overclock generally won’t provide a performance benefit, and it could even reduce your performance, though only in lightly-threaded applications. The boosting algorithms can usually achieve higher speeds for these applications, with all-core overclocks running into a thermal wall before providing much of a boost.

However, for Intel, that changed with Arrow Lake. It ships with much higher thermal headroom than the last several Intel generations, shortening the gap between single-core boosts and all-core speeds. Intel now also provides more fine-grain control over clock speeds, with increments as low as 16MHz. Arrow Lake scales particularly well with high DDR5 speeds, offering a small performance edge if you’re able to secure a fast kit of memory.

AMD has taken a different approach. Rather than offering more granular control over overclocking, it has continued to invest in Precision Boost Overdrive (PBO) as a one-click overclocking solution. In addition, AMD offers Curve Optimizer and Shaper, which allow you to easily undervolt your chip with a voltage offset across the frequency range.

Intel has an edge in overclocking, but it comes with a caveat. In order to overclock an Intel chip, you need to purchase an unlocked K-series processor as well as a Z-series motherboard. H- and B-series motherboards only support memory overclocking. Meanwhile, AMD supports overclocking on all of its modern CPUs, short of the Ryzen 7 5800X3D and other Zen 3 chips sporting 3D V-Cache. AMD also supports CPU overclocking on B- and X-series chipsets, saving you some money.

Winner: Intel. Although CPU overclocking isn’t as prominent as it used to be, Team Blue holds the edge in this AMD vs Intel battle. You have more granular control over core overclocking, as well as additional fabric and memory overclocking features that can net higher performance on Arrow Lake. AMD has focused on making overclocking more accessible, but that also leaves enthusiasts with less room to play.

AMD vs Intel CPU Drivers, Software, and Firmware 2025

Over the past couple of years, both AMD and Intel have seen catastrophic firmware issues, which caused CPUs to fail, or in the worst cases, literally burn themselves up. In the most recent generation, we also saw hamstrung performance due to software issues from both AMD and Intel, though both have also released a string of microcode and driver updates to address these problems.

On the driver front, Intel and AMD are both stable. In the early days of Zen, we saw lopsided support for Intel’s latest architecture, and AMD’s smaller position in the market meant issues were more common and fixes took longer to arrive. That isn’t the case today. Broadly, Intel and AMD’s chipset drivers are stable, and issues that do arise are usually resolved through an update within a few days.

Firmware is a different story, and that’s been an unfortunate area of focus for AMD and Intel over the last couple of years. Intel was at the center of a years-long issue with instability on Raptor Lake and Raptor Lake Refresh chips, and in particular, the Core i9-13900K and Core i9-14900K. Reports of instability in games on these CPUs date back at least as far as early 2023, but Intel didn’t acknowledge them until mid-2024.

Intel released a microcode update in late 2024 that solved most instability problems, but as recently as September 2025, it’s continued to work on the issue.

In AMD’s camp, high-end Ryzen CPUs have burned themselves up in the socket, ruining both the chip and motherboard – twice, though for different reasons. In the first go, the Ryzen 7 7800X3D was operating at elevated voltages. An AGESA update shortly after reports started circulating capped the voltage at 1.3V, and there haven’t been any reports since.

With the release of the Ryzen 7 9800X3D, a far larger number of reports started circulating, once again showing the burning on the CPU’s contact pads and in the motherboard socket. This issue was much messier than the original Ryzen 7 7800X3D, though the vast majority of reports came from the Ryzen 7 9800X3D paired with an ASRock motherboard. In this case, AMD shifted blame to ASRock. ASRock has reportedly solved the problem through BIOS updates, but new reports are still trickling out at a reduced cadence.

Winner: Tie. It’s hard to declare a winner in this category given the two major controversies Intel and AMD have faced over the past few years, both of which resulted in CPUs being ruined while supposedly operating within warrantied specifications.

Although it’s important to highlight these problems and the responses to them from AMD and Intel, they aren’t widespread issues. Neither triggered a proper recall.

AMD vs Intel Chipset and Socket Support 2025

An increasingly important aspect of choosing between Intel and AMD is the longevity of the chipset and socket of your motherboard. Although new generations bring new chipsets, AMD has set a new standard for socket support with AM4, which it appears to be carrying forward with AM5. Intel has responded in kind, supporting 12th-, 13th-, and 14th-gen chips on the same LGA1700 socket. However, it moved to the LGA1851 socket with the release of Arrow Lake, and it will move to the LGA1954 socket with the release of Nova Lake CPUs in late 2026.

AMD’s current commitment is to support the AM5 socket through 2027, but that timeframe could extend. With the release of Zen 4 and introduction of AM5, AMD originally only committed to support through 2025 before extending the timeframe with the release of Zen 5. With AMD’s current release cadence, that means we should see at least three generations, and possibly four, on the AM5 socket. If AMD once again extends the window, we’re likely looking at five generations of support.

It’s certainly possible that AMD will extend the timeframe. The AM4 socket set a precedent of longevity. It was originally introduced in 2016, and AMD has released new AM4 chips as recently as 2025; though, these chips are mostly rebranded versions of old silicon, or specific variants for different international markets.

For chipsets, AMD currently supports Zen 4 and Zen 5 CPUs across all chipsets with the AM5 socket. With AM4, AMD eventually brought support for Zen, Zen+, Zen 2, and Zen 3, though support varies based on manufacturer for older chipsets.

Intel has a more confusing lineup of chipsets and support, which alone spoils the reasoning for AMD winning this section. After three generations on the same socket, Intel moved to a new socket with the release of Arrow Lake. That socket (LGA1851) has seen Arrow Lake Refresh chips, and we don't expect it to see any more releases.

Intel also forces customers to its flagship Z-series chipset for CPU overclocking support; a line in the sand that it’s continued to maintain despite the fact that AMD supports CPU and memory overclocking on both B-series and X-series chipsets. Thankfully, the last several generations of Intel chipsets have opened up memory overclocking on B-series chipsets.

Winner: AMD. The AM4 socket brought about a new standard for drop-in CPU replacements, which AMD is bringing forward with AM5. Intel has made strides to support multiple generations on the same socket, but it has yet to demonstrate the ability to deliver support over several years like AMD has. Additionally, Intel continues to charge a premium for CPU overclocking by restricting the feature to flagship Z-series chipsets.

AMD vs Intel CPU Security 2025

The last few years have found security researchers poking and prodding at the speculative execution engine that's one of the key performance-boosting features behind all modern chips. The resulting research has spawned an almost never-ending onslaught of new vulnerabilities that threaten the safety of your system and private data. Unfortunately, these types of vulnerabilities are incredibly dangerous because they are undetectable—these tactics steal data by using the processor exactly as it was designed; thus, they are undetectable by any known anti-virus program.

The rash of fixes required to plug these holes also continues to grow, and many of them resulted in reduced performance initially. AMD and Intel have recognized the cost of such updates during the Spectre and Meltdown era, and more recent patches come with little to no impact on performance. It’s possible that could change with a future vulnerability, but we haven’t seen chart-breaking changes in performance from security updates in the past couple of generations.

Intel still has more known vulnerabilities, but in the years following Spectre, it’s become clear that both AMD and Intel are vulnerable to this family of attacks. At the time of writing, Intel recently patched over 30 security bugs, and AMD confirmed a critical security vulnerability in its random number generator on Zen 5 CPUs, which it plans to address with an AGESA update.

Winner: Tie. Similar to firmware, it’s hard to say anyone is a winner here. Intel claims it surfaces and addresses vulnerabilities more proactively than AMD does, but both brands chase the constantly evolving world of threats and offer driver and/or firmware updates in response. In addition, both maintain a bug bounty program to incentivize security researchers to find and surface exploits.

FAQ: Frequently Asked Questions

• MORE: Best CPUs for Gaming
• MORE: CPU Benchmark Hierarchy
• MORE: Zen 4 Ryzen 7000 All We Know
• MORE: How to Overclock a CPU
• MORE: How to check CPU Temperature
• MORE: All CPUs Content

While Apple has slowly but surely been pushing to bring more games to Macs running Apple Silicon, it's still a long way off from the massive library of games you can play on a Windows PC.

If you don't mind using a bit of elbow grease and dealing with some potential technical hiccups, you can get plenty of PC games running on your Mac through emulation. Apple has a translation layer designed to test Windows games on Macs to port them to native Mac versions. While this is really meant for Apple's Game Porting Toolkit  with a developer account, there are both open source projects and commercial products that let you play around with it. 

You could pay for Parallels to run Windows, or a paid app like Crossover that translates Windows application commands to macOS and back. There are also open source projects like Whisky, which don't have a full developer team behind it and might not get as much support, but hey, it's free.

Here's, we're going to show you how to set up Whisky to play Windows games on an Apple Silicon Mac. But before we do: one note. While some games will work great under emulation, others won't. Some won't work at all (for example, when I tried Sonic Frontiers, which requires AVX instructions, it wouldn't launch. That should be fixed to some degree with macOS Sequoia). If you only own a Mac, you're best off looking for Mac-native games for the best experience. If you also own a PC or a console, playing games there will be smoother.

But hey, if you own games on Steam or Epic Games and really want to play them on your Mac despite that they might not be flawless, here's how to do it:

1. Grab Whisky from getwhisky.app. Click "download latest." 

2. Install Whisky and place it in your applications folder. (Whisky is open source, and you can see the code on GitHub, if you wish.)

3. Open Whisky. Upon the first run, it will make sure you have both Rosetta and WhiskyWine — the necessary translation layer installed.

4. Click Next to install any dependencies you may be missing.

5. Click "Create Bottle." A bottle is effectively a Windows file system, which will be a folder on your Mac. It will appear in the left rail.

6. Name your bottle and choose your Windows version, then click "Create." I'll be using this bottle for a Steam library. Whisky recommends a single bottle per game, though I've been using one for my Steam account.

7. Grab the Steam executable for Windows if you plan to run Steam games. You could also do this with Epic Games or other launchers.

8. Press "Run…" to pick SteamSetup.exe from your downloads folder (or wherever else you put it) in the Finder. Pick the file and click "Open".

Whisky runs the software — in this case, the Steam installer. Run it as normal to install Steam in your bottle. Sometimes, it takes a minute for these windows to pop up. 

9. From here, log in to Steam, and go to your library to pick a game to install. Here, I'm using SoulCalibur VI.

From there, I can launch the game from my Steam library. Alternatively, I can go through "Run…" in Whisky to choose the game, or pin the game to the bottle menu.

It's possible that some games won't work well. Others may not work at all (especially games with anti-cheat).

Bottle Configuration

From your bottle, the Bottle Configuration menu has a few handy features, including allowing you to switch between different versions of Windows, different options for scaling, and the ability to turn on Apple's Metal HUD, which will show your frame rates, RAM usage, and more. The latter is great to see how well a game is performing in emulation.

If you didn't know it by now, Windows 11 24H2 has already been deployed to the public. The only caveat is that it has only been deployed to Copilot+ PCs (namely Snapdragon X Elite laptops and 2-in-1s). Microsoft has confirmed that update 24H2 will not be making its way to x86 PCs anytime in the immediate future, with the trillion-dollar company confirming it will be arriving "later this year."

Later in the year could mean anything, but Microsoft's wording suggests we won't see Windows 11 24H2 anytime soon (as in weeks). Realistically, we are probably looking at an August 2024 release date at the soonest — realistically, it might not even launch until later in the Q4 timeframe.

This all but confirms that Microsoft put most of its 24H2 development resources into the Windows 11 on ARM variant of 24H2, to meet the deadline of its Copilot+ PC debut. Windows 11 24H2 is the only version of Windows that supports this new category of PC right now, along with the bevy of AI-assisted features 24H2 also comes with locked exclusively to these systems. Such as the ill-fated Recall feature.

If you really want to get your hands on the x86 flavor of 24H2, Microsoft has switched the Canary and Dev insider preview channels to the new version. But be warned, these two channels are the most experimental tiers in Microsoft's line-up, so expect bugs and crashes. The beta and release preview channels are still based on the outgoing version of Windows 11, 23H2.

The good news is that — besides any of the "killer" AI features implemented or being implemented into 24H2, there aren't any game-changing additions outside of AI making it into 24H2. The vast majority of these changes are quality of life improvements, such as Wi-Fi 7 support, 7-zip and TAR support in File Explorer, support for the "sudo" command in the command line, and a re-designed Copilot experience that puts the AI assistant into a regular app that can be pinned to the taskbar. Microsoft is also adding checkpoint cumulative updates that will improve the download speed and install duration of Windows updates.

Phone Link is also getting a major update that sees it live as part of the start menu, but we aren't sure if that will be coming in the 24H2 update specifically.

Regardless, the good news is that most people won't be dying to try out Microsoft's latest patch, so it's not really going to impact the Windows 11 user base much (if at all) if Windows 11 24H2 is delayed by several extra months for x86 users. Remember that many of the AI-focused features 24H2 comes with are locked to the Copilot+ PC ecosystem.

There's a vast difference between hardware and software developers, which opens up pitfalls for those trying to coordinate the two teams. Arm and x86 researchers encountered it years ago -- and Linus Torvalds, the creator of Linux, fears RISC-V development may fall into the same chasm again.

“Even when you do hardware design in a more open manner, hardware people are different enough from software people [that] there’s a fairly big gulf between the Verilog and even the kernel, much less higher up the stack where you are working in what [is] so far away from the hardware that you really have no idea how the hardware works,” he said (video embedded below).

“So, it’s really hard to kind of work across this very wide gulf of things and I suspect the hardware designers, some of them have some overlap, but they will learn by doing mistakes — all the same mistakes that have been done before.”

RISC-V is an open-standard ISA for processors that is slowly gaining traction, especially in China, where some tech companies are using it to bypass America’s sanctions on the country. Companies like DeepComputing and Framework have started developing, building, and selling consumer laptops powered by these new processors.

But even though RISC-V is slowly being built up, it’s still not at the performance level that it could compete against current generation x86 and Arm processors. It would still take several years or decades of development to play AAA games on a RISC-V chip. But even though Arm, which also uses a reduced instruction set computer (RISC) architecture, has already undergone intensive development, Linus fears RISC-V will still make the same mistakes.

“They’ll have all the same issues we have on the Arm side and that x86 had before them,” he says. “It will take a few generations for them to say, ‘Oh, we didn’t think about that,’ because they have new people involved.”

But even if RISC-V development is still expected to make many mistakes, he also said it will be much easier to develop the hardware now. Linus says, “It took a few decades to really get to the point where Arm and x86 are competing on fairly equal ground because there was al this software that was fairly PC-centric and that has passed. That will make it easier for new architectures like RISC-V to then come in.”

Do you think real-time ray tracing requires a modern GPU? It still does for practical applications, but YouTuber nanochess has a functioning ray tracing demo with ray-traced reflections working on machines as old as an original Intel Pentium Pro (1995), and in only  *483 bytes at that. Below, we've embedded the original video of this demo, titled Boot Sector Ray Tracer.

*Previously 506. Changed during the writing of this article!

Boot Sector Ray Tracer is a simple demo showcasing a camera coming across two reflective orbs floating in the air in the middle of a large, empty space surrounded by a colorful sky. The colors of the sky reflect off of both orbs realistically, as do the reflections of the orbs themselves, resulting in a hall-of-mirrors effect if one looks into the correct part of the reflection during the animation. It's also real-time animation, so the final speed will depend on your hardware.

The main ray tracing effect being shown is ray traced reflections. Non-ray traced graphics, or rasterized graphics usually, can only display reflections based on information already available onscreen. These limitations have workarounds like rendering additional character models in video games, but there's no room for "workarounds" in 483 bytes!

The full demo is available for free on GitHub and is recommended to those who have the old hardware lying around or don't mind setting up a virtual machine for something like this. The YouTube animation runs on a Macbook Air through a VirtualBox'd Windows XP, and seems to run perfectly smoothly despite the virtual machine ahead.

Other projects by Nanochess include an Intel 8080 emulator, various Chess programs, and even various homebrew games for retro consoles and PCs (Intellivision, MSX, Atari, etc).

Additionally, Boot Sector Ray Tracer is inspired by a similar 8-bit "RayTrace Movie" demo released for Atari, though that runs at frames-per-hour. The visual touching up that this (relatively) more modern Boot Sector RT demo makes it look more akin to something like the classic Amiga ball demo instead of the pure-pixel vaporwave aesthetic of the original Atari demo.

Google this week became yet another major hyperscaler that has developed its own Arm-based processor for datacenters. The company claims its Axion CPU is faster than competing x86 and Arm-based offerings and plans to use it to run workloads that rely on general-purpose computing at its datacenters. Some of Google's services already run on Axion and the company intends to increase usage of its custom silicon to take advantage of additional performance at lower power. 

Google asserts that its Axion processors deliver up to 50% greater performance and up to 60% higher energy efficiency compared to contemporary x86-based processors, though it does not disclose with which CPUs it compares its devices. Additionally, the company claims that Axion offers a 30% performance advantage over rival Arm-based CPUs designed for datacenters, but again it does not disclose the exact rival.

Google's Axion processor is based on the Neoverse V2 cores (featuring the Arm v9 instruction set architecture), but Google does not disclose how many general-purpose cores the CPU integrates, though a picture of the processor indicates that the unit uses standard dual-rank DDR5 modules. 

One of the advantages that Google Axion-based servers have is that they come outfitted with the company's custom Titanium-badged controllers that offload networking, security, and I/O storage processing from the host CPU, which has a positive effect on the performance available to applications. Using purpose-built accelerators to offload certain workloads from the host CPU is one of the ways to increase server performance without significantly increasing its power budget and Google certainly takes advantage of its Titanium platform.

 "Google's announcement of the new Axion CPU marks a significant milestone in delivering custom silicon that is optimized for Google's infrastructure, and built on our high-performance Arm Neoverse V2 platform," said Rene Haas, CEO of Arm. "Decades of ecosystem investment, combined with Google's ongoing innovation and open-source software contributions ensure the best experience for the workloads that matter most to customers running on Arm everywhere." 

Google has already initiated deployment of Google services, including BigTable, Spanner, BigQuery, Blobstore, Pub/Sub, Google Earth Engine, and the YouTube Ads platform, on existing Arm-based servers and plans to expand and scale these services on Axion processors in the near future. 

Google was among the first hyperscalers to develop its own custom silicon for its datacenters. The company has deployed several generations of its tensor processing units (TPUs) for its AI services and has special video transcoding units (VCUs) for transcoding YouTube videos. With its own custom CPU, Google makes another step away from x86 processors from AMD and Intel and while we do not expect x86 to get eliminated from datacenters any time soon, it is evident that Arm is slowly eating AMD's and Intel's pie.

The latest version of Box64, a popular x86_64 emulator for Linux running on architectures like ARM/AArch64 and RISC-V, introduces significant performance improvements, making possible gaming on RISC-V-based platforms such as the Vision Five 2 board, reports Phoronix. 

The most important addition to the Box64 version 0.2.4 is that Dynarec now works on RISC-V. This facilitates faster operations of x86_64 Linux software on RISC-V 64-bit system and makes simplistic games, such as Stardew Valley, playable on Vision Five 2-based devices. Additionally, the new version introduces several fixes for Steam, enhanced multi-threading capabilities, and broader improvements for emulation across various CPU architectures.

While titles like Stardew Valley can hardly attract avid gamers, the Vision Five 2 is not exactly designed to run games (even though it has an OpenGL ES 3.2, Vulkan 1.2-capable integrated GPU), so running a game is already a kind of a breakthrough for this product. The motherboard is aimed at software developers and is based in the quad-core StarFive JH7110 SoC with SiFive U74 RV64GC cores running at 1.50 GHz and Imagination's  BXE-4-32 GPU.

The Box64 version 0.2.4 has refined its compatibility with several modifications, which includes better handling of ELF files, added wrapped libraries and functions, expanded opcode functionalities, and preliminary WoW64 support for 32-bit operations in Wine. 

In conjunction with Box64's release, Box86, focused on x86 32-bit emulation, launched its 0.3.2 version. This version sees improvements in system call processes, an optimized ELF loader, a decrease in memory usage, and stability improvements.

High-tech giants Amazon, Apple, Intel, Nvidia, and Samsung are looking forward to becoming anchor investors in Arm's upcoming initial public offering, according to reports from Bloomberg and Nikkei, citing sources with knowledge of the matter. This IPO might amass up to $10 billion and elevate Arm's valuation to a staggering $60 to $70 billion, according to Bloomberg's estimates.

Arm, which is owned by SoftBank Group, is preparing for what will potentially be the year's largest tech IPO, and has engaged with major customers for months regarding their participation. The company has held talks with key global chip designers about becoming anchor investors, and companies like Amazon, Apple, Intel, Samsung, and Nvidia are expected to invest in Arm upon its market listing (note that none of these companies have formally confirmed intention to buy an Arm stake, however). Arm's strategy involves offering these processor developers medium- to long-term shareholding with the intention of ensuring stock price stability during the listing.

Arm's global impact on chip designs is undeniable. In the smartphone sector, Arm-based SoCs command close to a 100% market share. They also command a sizeable share in the consumer electronics industry, and, thanks to the efforts of Apple and Qualcomm, Arm's technology is gaining traction in PCs as well. Throughout the history of Arm, over 250 billion chips based on its architectures have been produced.

Apple, Samsung, Nvidia, and Qualcomm produce some of the world's best known Arm-based SoCs, which is why Apple (who already has a stake in Arm), Samsung, and Nvidia have reason to invest in the chip IP designer. But Amazon seems to have a particularly good reason to invest in Arm: apparently, Amazon Web Services dominates the Arm server market, with over half of all Arm-based server CPUs globally deployed in AWS datacenters, according to a Bernstein Research report cited by The Register.

The Graviton system-on-chip family, introduced in 2018 and optimized for AWS powered around 20% of AWS CPU instances by mid-2022 — which is a significant portion. These SoCs, optimized for AWS's requirements, have more cores per socket than competing x86 offerings, reduced power consumption, and significant cost benefits — partly because Amazon does not have to pay premium to AMD or Intel and to a large degree because of lower total cost of ownership and power consumption. The report asserts that in some cases AWS can achieve 20% to 70% lower costs at the same performance compared to competing x86 CPUs, though the details are unclear. 

Given the tangible advantages that AWS has with its Arm-based SoCs, it is evident that the company has invested massively in datacenter software ecosystem for Arm processors. Therefore, the company is likely more than interested in further development of the Arm instruction set architecture (ISA) in general and datacenter-oriented Neoverse CPU cores in particular. 

SoftBank, which acquired Arm in 2016 for $32 billion, tried to sell the company to Nvidia for about $40 billion in 2020 – 2022, but regulatory issues prevented the sale. In the aftermath, SoftBank chose to steer Arm towards an IPO. If the IPO reaches its peak valuation, it would rival the tech sector's largest IPOs — such as Alibaba in 2014 and Meta Platforms (formerly Facebook) in 2012.

Intrepid Steam Deck modder Balázs Triszka has shared news of an impressive feat of handheld device enhancement. In a recent Tweet/X by Triszka (h/t Overclock3D), we see that they managed to boost the Valve-made x86 PC gaming portable to 32 GB of RAM. That’s double the RAM quota this SteamOS handheld is factory configured with.

Upgrading the RAM on a Steam Deck is definitely not for everyone. Unlike your desktop, and many laptops, the RAM in these devices is soldered directly to the system PCB, like on a graphics card. These RAM chips will be attached with a plethora of tiny solder balls (ball grid array, BGA) to the mainboard of the Steam Deck. You'll need a one of the best hot air rework stations to make this possible.

Sometimes memory module replacement, for repair or expansion, is hindered by glue under the BGA chip. However, Triszka confirmed that Valve hasn’t done this. To double the memory quota of the Steam Deck, the modder simply switched the existing RAM chips for similar ones with double the capacity / density.

Triszka shows the flux-strewn aftermath of his de/soldering of the system’s RAM chips. A bath of isopropyl alcohol will soon have this mod looking factory fresh. This high-temperature molten metal surgery was just the beginning of the demonstrably effective mod. Very often hardware needs a firmware modification to enable this kind of upgrade, and this is true of the Steam Deck memory doubling mod. However, the satisfying conclusion is that this mod works, and you can see evidence of the system reporting an available 32 GB of RAM, below.

The Steam Deck is one of the most popular of the new wave of PC gaming handhelds, which is understandable given that it invigorated the segment with a uniquely compelling software / hardware combo in 2022. One of the good things about a leading and popular device is that it attracts the most developer attention.

We have already documented the arrival of a multitude of upgrades and mods like: an enhanced screen replacement, liquid cooling, a transparent shell, and we have tested lots of SSD upgrades. But be warned, RAM upgrades like Triszka are far from trivial, with soldering equipment and skills required in abundance. Modders also need a strong disposition and steady hands.

Due to increasing difficulty in obtaining high-performance x86 servers, Chinese datacenter companies are ramping up the adoption of Arm-based system-on-chips. Currently, 40% of Arm-powered servers are used in China, according to investment banking firm Bernstein, as reported by @EricJhonsa.

The vast majority of servers are running on x86 processors from AMD and Intel. Still, Arm-based SoCs are gaining market share now that multiple companies, including AWS, Ampere, Google, Fujitsu, Microsoft, and Nvidia have developed and/or adopted Arm-powered SoCs. But while U.S.-based hyperscalers are slowly adopting Arm-based processors in their datacenters, Chinese companies are more inclined to use them because they may not always be able to procure Intel's Xeon or AMD's EPYC.

Multiple companies in China develop Arm-based SoCs for client and datacenter processors, including Alibaba, Huawei, and Phytium. 

Both Huawei and Phytium are blacklisted by the U.S. government and cannot have access to leading-edge process technologies used by TSMC. As a result, they cannot offer processors that challenge market leaders in terms of performance. While Alibaba's T-Head can use the latest TSMC innovations, just like other Chinese companies, it cannot license Arm's Neoverse V-series CPU cores for high-performance computing due to export control rules of the U.S. and U.K. as well as the Wassenaar arrangement on export controls for dual-use goods and technologies and conventional arms.

While Arm's cores make a lot of sense for Chinese companies, many chip designers, including Huawei, are looking towards RISC-V. RISC-V is a rapidly developing open-source instruction set architecture (ISA) that is not restricted and can enable highly-custom general-purpose cores designed for very specific workloads. 

But for now, the Armv8 and the Armv9 instruction set architectures have several advantages over RISC-V. The Arm ISA is supported by a vast number of influential companies like AWS, Google, Nvidia, Microsoft, Qualcomm, and Samsung, and these companies know how to ensure that the software ecosystem supports their CPUs, so adoption of Arm will continue to expand in the datacenter space for years to come.

The world of single board computers (SBCs) is dominated by Arm processors, but there are some x86 based boards to be found. AAEON's series of UP boards look very much like a Raspberry Pi, but its latest range, sport Intel N-series (Alder Lake-N) CPUs.

Claiming to be "The world’s smallest developer board featuring the Intel Processor N-series platform (formerly Alder Lake-N)" the UP 7000 is indeed a small SBC. measuring just 85 x 56mm, the exact same size as the Raspberry Pi 4. While they may share a similar size, their port locations means that you won't be dropping the UP 7000 in any of the best Raspberry Pi cases. The HDMI port placement means we sacrifice a USB port. We can see the logic here, these boards are most likely destined for behind the scenes roles, not the desktop.

The Intel CPUs at the heart of the board range from an Intel N50 (dual-core, 3.4GHz, 6W TDP) to an Intel N97 (quad-core, 3.6GHz, 12W TDP) which on paper are quite competitive to the Arm chip found on the Raspberry Pi. The bonus here is that being an x86 CPU, we are free to install whatever operating system we see fit.

CPUs generate heat, and being a past owner of a first generation UP board we know that they can run hot. But the UP 7000 comes with a passive cooler and a 12V fan header means we can add our own active cooling.

Storage options are limited to 32GB eMMC for the N50 and N100 and 64GB eMMC for the N97. We can't spot a micro SD card slot, so if you want a little more storage then you will need to grab an external USB 3 drive.

Connectivity is limited to just Ethernet. No Wi-Fi or Bluetooth for the UP 7000. The older UP 4000 did have a range of carrier boards which provided extra features (Wi-Fi, Bluetooth and M.2 storage), but the UP 7000 product page is devoid of these accessories. For now?

The 40 pin GPIO is a mystery. Is it compatible with the best Raspberry Pi HATs? Electrically it is possible, but we cannot confirm that. More boards are providing electrical compatibility, but often fall short with software support. If you need GPIO more than raw processing power, then a Raspberry Pi is the safer bet. If you need raw power and GPIO then UP 7000 could offer a solution, but tread carefully.

So where are these boards destined for? The official blurb states "applications such as digital signage, medical imaging" and sure these boards would do a great job. The Raspberry Pi may struggle with medical imaging, but it has proven to be a great base for digital signage and AI based projects. As this is an x86 board, could we use it as a desktop? Short answer is yes. We could even play some older games on the UP 7000, but we would have to manage our expectations.

Prices and general availability are currently unknown. If you need something similar then right now the Intel Celeron N5105 powered LattePanda 3 Delta is a low-cost SBC with built-in Arduino compatible board. If you need a little more power, then the LattePanda Sigma provides an Intel Core-i5 1340P and 16GB of LPDDR5.

Intel posted its new APX (Advanced Performance Extensions) today and also disclosed the new AVX10 [at the bottom of this page] that will bring unified support for AVX-512 capabilities to both P-Cores and E-Cores for the first time. This evolution of the AVX instruction set will help Intel sidestep the severe issues it encountered with its new x86 hybrid architecture found in the Alder and Raptor Lake processors.

However, the new AVX10 ISA won't be supported with Intel's current-gen CPUs — it's slated to arrive in future chips. Intel says that AVX10 will be its vector ISA of choice moving into the future for both consumer and server processors.

Intel AVX10 (Advanced Instruction Extensions 10)

At its most basic level, AVX10 will allow Intel's chips that have both E-cores and P-cores to still support AVX-512, though 512-bit instructions can only run on P-cores. Meanwhile, converged 256-bit AVX10 instructions can run on either the p-cores or e-cores, thus allowing the full chip to still have support for AVX-512 capabilities.

As such, Intel won't have to disable support for 512-bit vectors as it did when it disabled AVX-512 for both Alder Lake and Raptor Lake. 

Diving deeper, the AVX10 (Advanced Instruction Extensions 10) ISA is a superset of AVX-512 and comes with all of the features of the AVX-512 ISA for processors with both 256-bit and 512-bit vector register sizes.

The converged AVX10 ISA will include "AVX-512 vector instructions with an AVX512VL feature flag, a maximum vector register length of 256 bits, as well as eight 32-bit mask registers and new versions of 256-bit instructions supporting embedded rounding," and this version will run on both p-cores and e-cores.

However, the e-cores will be limited to the converged AVX10's maximum 256-bit vector length, while P-cores can use 512-bit vectors. This feels akin to Arm's support for variable vector widths with SVE. 

Intel says that existing applications will provide the same level of performance with AVX10 as they did with AVX-512, at least at the same vector lengths. Intel also claims:

• Intel AVX2-compiled applications, re-compiled to Intel AVX10, should realize performance gains without the need for additional software tuning. 
• Intel AVX2 applications sensitive to vector register pressure will gain the most performance due to the 16 additional vector registers and new instructions. 
• Highly-threaded vectorizable applications are likely to achieve higher aggregate throughput when running on E-core-based Intel Xeon processors or on Intel products with performance hybrid architecture.

Intel will support AVX10 version 1 (AVX10.1) beginning with its sixth-gen Xeon "Granite Rapids" chips, but that generation will only support 512-bit vector instructions, and not the new converged 256-bit vector instructions. Instead, this first gen will serve as the transition chip from AVX-512 to AVX10.

Chips arriving after Granite Rapids will support AVX10.2, which adds support for the converged 256-bit vector lengths and other new features, like new AI data types and conversions, data movement optimizations, and standards support. All future Xeon processors will continue fully supporting all AVX-512 instructions to ensure that legacy apps function normally.

To address developer feedback (obviously negative), Intel also plans to significantly simplify its AVX10 enumeration methods compared to AVX-512. Intel also plans to ensure that each move to a new AVX10 revision has enough new instructions and capabilities to merit a change, thus reducing version and enumeration bloat. 

Intel will freeze the AVX-512 ISA when AVX10 debuts, and all future use of AVX-512 instructions will occur through the AVX10 ISA. Meanwhile, the new AMX will be unimpacted. 

Intel APX (Advanced Performance Extensions)

Intel also announced the new APX (Advanced Performance Extensions) today (not to be confused with the old-school iAPX 432).

 Intel claims APX-compiled code contains 10% fewer loads and 20% fewer stores than the same code compiled for an Intel 64 baseline. Intel also says that register accesses are both faster and consume significantly less dynamic power than complex load and store operations. Interestingly, the new APX finds a new use for the 128B area that was left unused when Intel abandoned MPX back in 2019, and repurposes it for XSAVE.

Here are APX's top-level features:

• 16 additional general-purpose registers (GPRs) R16–R31, also referred to as Extended GPRs (EGPRs) in this document
• Three-operand instruction formats with a new data destination (NDD) register for many integer instructions
• Conditional ISA improvements: New conditional load, store and compare instructions, combined with an option for the compiler to suppress the status flags writes of common instructions 
• Optimized register state save/restore operations 
• A new 64-bit absolute direct jump instruction

Intel claims it has implemented APX in such a way that it will not impact the silicon area or power consumption of the CPU core. You can read much more about APX here, and Intel has a list of resources for both APX and AVX10 at the bottom of the linked page.

APX and AVX10 come on the heels of Intel's recent announcement that it is investigating slimming down the Intel 64 architecture to a simplified version of x86 named x86S.

Hardware sleuth Ken Shirriff took to his blog to explore the feature set and possible operations one could bring out of the world's oldest x86 CPU - the Intel 8086. By following the 8086's microcode operations, Shirriff found some unexpected design "choices": from proprietary microcode meant to trap IP thieves through questionable (but necessary) design decisions that allowed 8086 processors to run unknown instructions (with equally little-known results), the detective work shines some light on what is likely the most important CPU release in the last fifty or so years.

The world's first implementation of the x86 instruction set on Intel's 8086 was about as you'd expect from a CPU released back in the late seventies. While today's leading-edge chips can carry hundreds of billions of transistors (the basic building blocks for more complex circuits), Intel's first commercial x86 CPU, the 5-10 MHz Intel 8086, had to make do with a mere 29 thousand of them.

More transistors generally equal more performance and especially more functionality (increased support for more varied and complex instructions and hardware-based acceleration are the poster child here). Then, like now, a balance between performance, power efficiency, and transistor counts/die area enforced hard limitations on how transistor allocations were handled - and ultimately, what features were deemed "necessary" and "disposable."

According to Shirriff's finding, one particular quirk of Intel's 8086 is that the CPU didn't possess any checks and balances for running microcode - the closest-to-the-metal layer between software and hardware that's responsible for dividing complex instructions into a series of simpler, step-by-step ones that the CPU can execute. But with transistors being limited to 29,000, Intel opted not to include a hardware-level lock on what instructions could and couldn't be run. Usually, instructions work on a whitelist basis - the CPU allows for the execution of microcodes that it already knows can be processed within its hardware design.

But since Intel's 8086 was so tight for transistor space, the company actually didn't bake in the microcode operation whitelist that the CPU could process, meaning that in the presence of "illegal" (read: non-supported) microcode ops, Intel's 8086 tried its best to resolve the microcode operation anyway - without specifying the result of the "illegally requested" operation.

All in all, the Intel 8086 supported 521 instructions, held in its Microcode ROM chip (Read-Only Memory). Some of those 512 instructions were duplicates (as fall-back mechanisms and other reasons), but others were never made public: certain microcode operations were left undocumented - whether for planned functionality that was never implemented or for other reasons.

Yet one interesting microcode, as verified by Shirriff, aided Intel in defending its IP from would-be IP thieves. Populating an instruction list from the available 512 in the 8086's ROM, Shirring was able to discern the opcodes with known functions (in white); the ones in orange, yellow, and green were left blank for the 8086, but were populated on Intel's next processors, the 80186, 80286, and 80386 respectively; and finally, the purple outlier: an opcode that was implemented in Intel's 8086 and subsequent processors, but that was never openly documented.

Shirriff's take on the undocumented purple flag - which was only brought to light in Intel's documents as late as 2017, despite living within the company's x86 design since the 8086 - is that it served as a honeypot of sorts for anyone looking to copy Intel's technology. If a company had cut corners in developing their own CPU solutions and had instead pilfered Intel's microcode IP, then their CPUs would carry the same SALC (Set AL to Carry) operation when fed the relevant bits of machine code - it's equivalent to a barcode identification that'd allow Intel to more effectively prosecute any IP stealers.

Unfortunately, the "copyright trap" microcode didn't serve Intel very much, not even in the early days of the 8086 when the competition was more varied: Intel hoped that its SALC instruction had been implemented in NEC's own (and better) take on Intel's x86 processors in the form of the NEC V20 and NEC V30 processors. But it wasn't: a federal ruling from 1989 asserted that NEC hadn't infringed Intel's copyright (and the SALC instruction, therefore, wasn't there). But other things happen when a copyright infringement suit is launched, things that have little to do with the merits of the copyright itself. Intel's decision to sue and block NEC and the sale of its V20 and V30 processors is part of the reason why there aren't more players in the cut-throat-competitive scene of x86 chip design.

The US has blacklisted both Huawei and Phytium, a China-based CPU maker, but business still continues apace. Huawei last week said that it had teamed up with the state-owned China Electronics Corporation (CEC) to build a unified hardware-software ecosystem over its own Kunpeng processors and the US-sanctioned Tianjin Phytium Technology's Phytium CPUs, reports HuaweiCentral. The joint ecosystem will be called Penteng and will enable CEC to use both Kunpeng and Phytium chips in its devices without making major changes. In addition, the companies plan to invite new partners to the project.

Contemporary Kunpeng and Phytium processors use an Armv8 instruction set architecture and are compatible with Chinese cryptography algorithms, so it should not be too tricky to ensure that both CPUs run the same programs just fine. On the hardware side of matters, things will get more challenging since modern chips use different process technologies and different packaging. Therefore, it remains to be seen how these differences will be addressed by engineers working on the Penteng project. 

Both Huawei's chip design arm HiSilicon and Tianjin Phytium Technology are blacklisted by the U.S. government and cannot access leading-edge production capacities. As a result, both companies have to find a way to make competitive chips domestically. Therefore, it is probable that the companies will unify the form factors and packages of their next-generation CPUs. While the partnership includes a focus on Phytium, purportedly a subsidiary of CEC, Phytium has not yet formally announced its participation in the Penteng project.

This collaboration not only incorporates the refinement of software as well as the unification of some hardware parts but also plans to enlarge the ecosystem by inviting more partners through a certification program. This joint work provides the opportunity to engineer a wide array of products that can accommodate all modern workloads, including cloud data centers, edge computing, and client PCs. Moreover, Huawei and CEC have committed to address emerging business opportunities with Penteng.

Building up an ecosystem encompassing hardware and software developed in China seems like quite a significant move as it will make the Chinese IT industry generally more competitive. Programs for x86 platforms running Windows or Linux work on appropriate PCs from all vendors. By contrast, this is not always the case with Chinese programs and PCs from different suppliers.

Oracle this week said that it had ported its Oracle Database 19c Enterprise Edition, the current long-term support release of Oracle Database, to Ampere's Altra processors that use the Arm instruction set architecture (ISA). The move marks a milestone both for Ampere and Arm ISA as Oracle is one of the most widely used enterprise software suites.

Separately, the company said that eventually it might ditch x86-based instances running Database on processors from AMD and Intel from its data centers in favor of instances enabled by Ampere CPUs. Oracle hopes that by tailoring its Database software for Ampere's single-thread CPUs, it will tangibly increase performance efficiency of its data centers. Furthermore, Ampere, in which Oracle is a lead investor, could also implement tweaks into CPUs to better run Oracle's Database.

"It is a major commitment to move to a new supplier. We've moved to a new architecture and we've moved to new supplier," Larry Ellison, Oracle's founder, said at an event hosted by Ampere, reports Reuters. "We think that this is the future. The old Intel x86 architecture, after many decades in the market, is reaching its limit."

Oracle Database 19c Enterprise Edition is now certified for work on Ampere Altra-based servers for both on-premises deployments and in the cloud by subscribing to Oracle Database Service using OCI Ampere A1 compute instances enabled by Ampere Altra on Oracle Cloud Infrastructure (OCI). It will continue to be offered on AMD and Intel-powered instances probably for years to come. But with Arm-based Ampere servers, Oracle hopes to offer "highly economical price points."

Oracle's OCI Ampere A1 can be used in a flexible VM format ranging from 1 to 57 CPU cores, each with 8GB of memory (maxing out at 456GB), 1 Gbps of network bandwidth for each CPU core (up to a total of 40 Gbps per VM).

Oracle's Database enterprise database management software is used by large businesses, banks, government agencies, retailers, and manufacturers for running online transaction processing (OLTP), data warehousing (DW) and mixed (OLTP & DW) workloads. The software has been in development since 1979 and supports a wide variety of hardware and software platforms, including IBM's Mainframe and Power-bases systems, x86 CPUs from AMD and Intel, Sun's SPARC processors, Intel's IA64 (Itanium) chips, and now Arm-based Ampere Altra SoCs.

"Today's announcement highlights the broad architectural shift across the market to Ampere processors that meet the demands of both modern cloud and on-premises environments," said Jeff Wittich, Chief Product Officer of Ampere. "With the Ampere Altra family of processors, customers of the world's most popular database — Oracle Database — now have a high-performance, energy efficient architecture built with sustainability in mind for organizations of all sizes."

While running Database on energy-efficient cloud-native Ampere's Altra CPUs promises to make a lot of economic sense for Oracle, OCI still needs to offer high-performance computing (HPC), and Dense-IO instances for those who need maximum performance and ultra-fast local storage. So, this week the company announced OCI Compute E5 HPC and OCI Compute E5 Dense-IO instanced based on AMD's 4th Generation EPYC processors with their vast core counts and rich I/O capabilities.

Intel is kicking its internal foundry model into high gear. Today, on a call with investors, David Zinsner, executive vice president and chief financial officer, and Jason Grebe, corporate vice president and general manager of corporate planning, detailed how Intel's manufacturing group, including Internal Foundry Services, will be broken out into its own profit and loss (P&L) statement.

Zisner and Grebe suggested this new IDM 2.0 model will put Internal Foundry Services along with manufacturing and technology development into a reportable unit, who will sell to both external fabless customers and internal business units that will choose to use either Intel's foundries or external competitors. (Intel already has some x86 chips being made at TSMC like the upcoming Meteor Lake processors.)

Intel stated that its IDM 1.0 strategy worked for a long time, but the capital required to create new leading edge nodes has increased, and there has been a lag behind rivals in a competitive business environment. Intel does claim, however, that its Intel 18A will come in 2025 and regain process leadership.

The company states that it's still the goal for the Internal Foundry Model to produce the second-largest external foundry in the industry. It claims that putting this group in its own P&L will also lead to $8 billion to $10 billion in "cost reduction opportunities." including ramp rates and test and sort times based on a "market-based" price. It is likely, however, to start with a negative operating margin.

Intel also suggests that keeping the foundry in its own company will provide benefits to external customers. Intel says it will have five internal products on 18A, and suggests that by creating multiple products on a node internally before releasing to external customers, it "de-risks" the node for those outside partners, with more wafer starts and the kinks removed from the process.

Grebe reiterated that there will be a "compete segregation" of customer data and intellectual property, along with "world class" service and support and an affirmation that there is supply available for customers. Intel will have to compete with the likes of TSMC and other fabs both with external customers and inside Intel to ensure its success. 

When asked why Intel isn't simply splitting into a fabless design business and a separate foundry business, Zisner said that "we think there is a ton of benefit for having both a product business and a manufacturing business combined." Those benefits, he suggested, include better process technologies and products due to internal teams collaborating, and using internal teams as "customer zero" to increase volumes on new nodes. Zisner suggested he didn't see any requirement to split the business in two. That being said, it seems clear that it is a possibility in the future, where Intel could spin its foundry out.

In recent weeks, Intel has recently announced plans for huge investments, including an assembly plant in Poland a $25 billion plant in Israel. These are on top of billions of dollars on new fabs in Oregon and Arizona, as well as a site in Magdeburg, Germany.

Intel stock was down approximately 4.9% on the news.

Razer today announced a new version of its Razer Blade 14 laptop, and it’s quite an intriguing ultraportable. Whereas the current generation is powered by a Ryzen 6900HX processor (8 cores, 16 threads) and your choice of up to a GeForce RTX 3080 Ti GPU, the new 2023 edition of the Razer Blade 14 brings the latest mobile hardware from AMD and Nvidia.

To be more specific, Razer opted for AMD’s Ryzen 7940HS “Phoenix” processor, which AMD bills as the first x86 processor with a dedicated, onboard AI engine. This processor also features an 8-core/16-thread configuration but with a base clock of 4 GHz and a turbo clock of 5 GHz. Razer also upgraded the discrete graphics across-the-board to Nvidia Ada Lovelace architecture. Base configurations get a GeForce RTX 4060 (8GB, 2370MHz boost, 140-watt), but you can step up to the optional RTX 4070 (8GB, 2175MHz boost, 140-watt) if you need more performance.

All systems have a 1TB SSD (upgradable to 4TB), while 16GB of DDR5-5600 memory is standard. If you opt for the RTX 4070, you can upgrade to 32GB of DDR5-5600. The maximum installable memory on the Razer Blade 14 is 64GB using the two onboard SO-DIMM slots.

All that hardware pushes pixels to a 14-inch, QHD+ (2560 x 1600) 16:10 display with a 240Hz refresh rate and AMD FreeSync Premium support. In addition to its anti-glare finish to cut down on reflections, Razer says that it’s spec’d for 500 nits typical brightness and covers 100 percent of the DCI-P3 color space.

Fueling this mobile powerhouse is a 68.1 WHr lithium-ion polymer battery, which powers the Blade 14 for up to 10 hours. On the connectivity front, you’ll find the requisite HDMI 2.1 port, two USB 3.2 Gen 2 (Type-A) ports, two USB 4 (Type-C) ports (up to 100-watt Power Delivery supported) and a 3.5mm combo jack. For your wireless needs, Razer opted for Wi-Fi 6E and Bluetooth 5.2 (Qualcomm WCN6856).

1080p webcams are becoming more commonplace on laptops, and Razer is no exception with the Blade 14. However, the company goes one step further by including a privacy shutter for those concerned about nefarious parties spying on you without permission. For your audio needs, you’ll find dual speakers with THX Spatial Audio support. Other features include a Razer Chroma RGB keyboard and a Windows Precision Trackpad.

The Blade 14 is constructed of anodized aluminum that is resistant to your greasy fingerprints and is available in Matte Black or Mercury White (exclusive to the 32GB/RTX 4070 SKU). The laptop measures 12.23 x 8.97 x 0.70 inches and weighs just 4.05 pounds.

The base Blade 14 SKU with 16GB of RAM and an RTX 4060 costs $2,399. Upgrading to the RTX 4070 takes the price to $2,699. Finally, the flagship SKU with the RTX 4070 and 32GB of RAM costs $2,799. The Blade 14 is up for preorder today and will begin shipping on June 20th.

The event has concluded, and you can see our overview with the live blog below, However, here are links to our deeper coverage of each topic:

AMD Expands MI300 With GPU-Only Model, Eight-GPU Platform with 1.5TB of HBM3

AMD EPYC Genoa-X Weilds 1.3 GB of L3 Cache, 96 Cores

AMD Details EPYC Bergamo CPUs With 128 Zen 4C Cores, Available Now

AMD Intros Ryzen 7000 Pro Mobile and Desktop Chips, AI Comes to Pro Series

AMD is holding its Data Center and AI Technology Premiere today, June 13, 2022, at 10 am PT here in San Francisco -- which is now. We're here to cover the event live and bring you the news as it happens as AMD CEO Lisa Su takes to the stage to reveal AMD's new AI-focused silicon.

AMD has already said that it will reveal its EPYC Bergamo chips at the event. These chips come with up to 128 cores, an innovation that's enabled by the company's new 'Zen 4c' efficiency cores. These new cores are optimized for density through several techniques, yet unlike Intel's competing efficiency cores, retain support for the chips' full feature set.

AMD is also expected to announce its Instinct MI300 accelerators. This data center APU blends a total of 13 chiplets, many of them 3D-stacked, to create a chip with twenty-four Zen 4 CPU cores fused with a CDNA 3 graphics engine and 8 stacks of HBM3. Overall the chip weighs in with 146 billion transistors, making it the largest chip AMD has pressed into production. This chip is designed to compete with Nvidia's Grace Hopper. 

Other expected announcements include the debut of the company's Genoa-X processors, which use 3D-Stacked L3 cache to boost performance in technical workloads, much like the existing Milan-X processors. We also expect news about the company's first teleco-optimized chips, Sienna, and perhaps an update on the company's next-gen Zen 5 'Turin' data center chips.

We're now seated and ready for the show to begin in less than ten minutes. 

AMD CEO Lisa Su has come on stage to introduce the company's new products, noting that she will introduce a range of new products including CPUs and GPUs. 

Lisa Su is outlining AMD's progress with its EPYC processors, particularly in the cloud with instances available worldwide. 

Lisa Su touts that AMD EPYC Genoa offers 1.8x the performance of Intel's competing processors in cloud workloads, and 1.9X faster in enterprise workloads. 

The vast majority of AI runs on CPUs, and AMD says it has a commanding lead in performance over competing Xeon 8490H, offering 1.9X more performance. Su also touted a 1.9X efficiency advantage.

Here we can see AMD's AI benchmarks relative to Intel's Sapphire Rapids Xeon.

Dave Brown, the VP of AWS's EC2, came on stage to talk about the cost savings and performance advantages of using AMD's instances in its cloud. He provided several examples of customers that benefited from the AMD instances, with workloads spanning from HPC to standard general-purpose workloads. 

Amazon announced that it is building new instances with AWS Nitro and the fourth-generation EPYC Genoa processors. The EC2 M7a instances are available in preview today, offering 50% more performance than M6a instances. AWS says they offer the highest performance of the AWS x86 offerings. 

AMD will also use the EC2 M7a instances for its own internal workloads as well, including for chip-designing EDA software.  

AMD also announced that Oracle with have Genoa E5 instances available in July.

Lisa Su has now transitioned to talking about cloud-native processors, explaining that they are throughput-oriented and require the highest end density and efficiency. Bergamo is the entry for this market, and uses up to 128 cores per socket with a consistent x86 ISA support. The chip has 83 billion transistors and offers the highest vCPU density available. 

The Zen 4c core offers higher density than standard Zen 4 cores, yet maintains 100% software compatibility. AMD optimized the cache hierarchy, among other trimmings, for a savings of 35% on the die area. The CCD core chiplet is the only change. 

Here is the die breakdown. 

The core is 35% smaller than standard Zen 4 cores. 

Here is a diagram of the chip package. 

Bergamo is shipping now to AMD's cloud customers. AMD also shared the following performance benchmarks. 

A Meta representative joined Lisa Su on the stage to talk about the company's use of AMD's EPYC processors for its infrastructure. Meta is also open-sourcing its AMD-powered server designs. 

Meta says that it has learned that it can rely upon AMD for both chip supply and a strong roadmap that it delivers on schedule. Meta plans to use Bergamo, which offers 2.5X more performance than the previous-gen Milan chips, for its infrastructure. Meta will also use Bergamo for its storage platforms. 

Dan McNamara, AMD's SVP and GM of the Server Business Unit, has come to the stage to introduce two new products. Genoa-X will add more than 1 GB of L3 cache with 96 cores. 

Gen0a-X is available now. Four SKUs, 16 to 96 cores. SP5 socket compatibility, so it will work with existing EPYC platforms. 

McNamara showed performance benchmarks of Genoa-X against Intel's 80 core Xeon. 

Here we can see a comparison of Genoa-X against an Intel Xeon with the same number of cores. 

A Microsoft representative joined McNamara on the stage to show Azure HPC performance benchmarks. In just four years, Azure has seen a 4X improvement in performance with the EPYC processors. 

Azure announced the general availability of its new HBv4 and HX-series instances with Genoa-X, and new HBv3 instances. Azure also provided benchmarks to show the performance gains, which top out at 5.7X gains 

AMD's Sienna is optimized for Telco and Edge workloads but comes to market in the second half of the year. 

AMD's Forrest Norrod, MD's executive vice president and general manager of the Data Center Solutions Business Group, has come to the stage to share information about how the data center is evolving. 

Citadel Securities has joined Norrod on the stage to talk about their shift in workloads to AMD's processors, powering a 35% increase in performance. They use over a million concurrent AMD cores. 

Citadel also uses AMD's Xilinx FPGAs for its work in financial markets with its high frequency trading platform. It also uses AMD's low-latency solarflare networking.

AMD purchased Pensando to acquire DPU technology. Norrod explained how AMD is using these devices to reduce networking overhead in the data center. 

AMD's P4 DPU offloads networking overhead and improves server manageability. 

AMD's Pensando SmartNICs are an integral part of the new data center architectures. 

The next step? Integrating P4 DPU offload into the network switch itself, thus providing services at the rack level. This comes as the Smart Switch they've developed with Aruba Networks. 

Lisa Su has come back to the stage to talk about AMD's broad AI silicon portfolio, including the Instinct MI300

Lisa Su outlined the massive market opportunity for the AI market driven by large language models (LLMs), causing the TAM to grow to around $150 billion. 

AMD Instinct GPUs are already powering many of the world's fastest supercomputers. 

AMD President Victor Peng came to the stage to talk about the company's efforts around developing the software ecosystem. That's an important facet, as Nvidia's CUDA software has proven to be a moat. AMD plans to use an 'Open, Proven, and Ready' philosophy for its AI software ecosystem development, which Peng is in charge of. 

Peng showed some of AMD's latest hardware efforts. 

AMD's ROCm is a complete set of libraries and tools for its optimized AI software stack. Unlike the proprietary CUDA, this is an open platform.

AMD is continually optimizing the ROCm suite. 

PyTorch is one of the most popular AI frameworks in the industry, and they've joined Peng on the stage to talk about their collaboration with ROCm. The new PyTorch 2.0 is nearly twice as fast as the previous version. AMD is one of the founding members of the PyTorch Foundation. 

Here are details of PyTorch 2.0. 

AMD is shifting to talking about AI models, with Hugging Face joining Peng on the stage. AMD and Hugging Face announced a new partnership, optimizing their models for AMD CPUs, GPUs, and other AI hardware. 

Lisa Su has returned to the stage, and now we expect to learn about the biggest announcement of the show: The Instinct MI300. This is for training larger models, like LLMs behind the current AI revolution. 

SU is talking about the Instinct roadmap, and how the company previewed the MI300 with the CDNA 3 GPU architecture paired with 24 Zen 4 CPU cores, tied to 128GB of HBM3. This gives 8x more performance and 5x higher efficiency than the MI250. 

146 billion transistors across 13 chiplets. 

There will be a GPU-only MI300, the MI300X. This chip is optimized for LLMs. this delivers 192GB of HBM3, 5.2 TB/s of bandwidth, and 896 GB/s of Infinity Fabric Bandwidth. 

And here's new chip. 153 billion transistors all in one package with 12 5nm chiplets. 

MI300X offers 2.4X HBM density than the Nvidia H100 and 1.6X HBM bandwidth than the H100, meaning that AMD can run larger models than Nvidia's chips. 

Lisa Su conducted a demo of the MI300X running a Hugging Face AI model. The LLM wrote a poem about San Francisco, where the event is taking place. This is the first time a model this large has been run on a single GPU. A single MI300X can run a model up to 80 billion parameters. 

This allows fewer GPUs for large language models, thus delivering cost savings. 

SU also announced the AMD Instinct Platform, which has 8 MI300X in an industry-standard OCP design, offering a total of 1.5TB of HBM3 memory. 

MI300A, the CPU+GPU model, is sampling now. The MI300X and 8-GPU Instinct Platform will sample in the third quarter, and launch in the fourth quarter. 

Lisa Su wrapped up the presentation. Here's a few more wrap up slides. Stay tuned for our ongoing coverage over the coming hours. 

The event has concluded, and you can see our overview with the live blog below, However, here are links to our deeper coverage of each topic:

AMD Expands MI300 With GPU-Only Model, Eight-GPU Platform with 1.5TB of HBM3

AMD EPYC Genoa-X Weilds 1.3 GB of L3 Cache, 96 Cores

AMD Details EPYC Bergamo CPUs With 128 Zen 4C Cores, Available Now 

AMD Intros Ryzen 7000 Pro Mobile and Desktop Chips, AI Comes to Pro Series

Intel has shared a slew of new benchmarks of its fourth-gen Xeon Scalable Sapphire Rapids CPUs going head-to-head with AMD's fourth-gen EPYC Genoa processors, claiming up to 7 times more performance in AI workloads when comparing two 32-core chips. Intel also touts higher performance under certain conditions, like when Sapphire Rapids' in-built accelerators are brought into play, in a spate of standard general-purpose workloads. Intel's 56-core Xeon Max, the first x86 data center CPU with HBM memory, also takes on AMD's 96-core flagship in several HPC workloads, matching or exceeding AMD's bulkier chip.

Intel's performance comparisons come well after the company's launch of its Sapphire Rapids Xeons back in January of this year, but the company says its benchmark comparisons were delayed due to difficulties procuring AMD's competing EPYC Genoa chips, which launched in November of last year. The benchmarks come a day before AMD's AI and Data Center event that we're flying out to cover, so we'll attempt to get AMD's feedback about Intel's benchmarks while we're at the event.

With a few shipping OEM systems powered by AMD's Genoa in hand, Intel has conducted a wide range of benchmarks in multiple types of workloads spanning AI, HPC, and general-purpose workloads, to present its view of the competitive landscape. However, as with all vendor-provided benchmarks, these should be approached with caution. Intel claims it enabled all rational optimizations for both its and AMD's silicon for these tests, but be aware that the comparisons can be a bit lopsided, which we'll call out where we see it. The price of the chips used for comparison are also lopsided, too. We've included Intel's full test notes for the tested configurations in the relevant image albums below. With that, let's take a closer look at Intel's results. 

AI Workloads: Intel Sapphire Rapids Xeon vs AMD EPYC Genoa

For nearly every large organization, the question is no longer "if" or "when" they should deploy AI-driven applications into their deployments — the question is where and how. Yes, AI training remains the land of GPUs and various flavors of custom silicon, and we can expect Large Language Models (LLMs) to continue to rely upon those types of accelerators for the foreseeable future, but the majority of AI inference workloads still tend to run on CPUs. Given the quickening pace of AI infusion in the data center, the CPUs' performance in various types of inference will only become more important in the years to come.

Intel has had its eyes on accelerating AI workloads since the debut of its DL (Deep Learning) Boost suite with its second-gen Cascade Lake Xeon Scalable chips in 2019, which it claimed made them the first CPUs specifically optimized for AI workloads. Those chips came with support for new VNNI (Vector Neural Network Instructions) that optimized instructions for the smaller data types prized in AI applications. One of Intel's bedrock principles behind its AI strategy has been to use AVX-512 to vastly improve Xeon's performance and power efficiency in AI workloads by using VNNI and BF16. Intel's focus on AI acceleration features, including software optimizations, has expanded over the years to now include purpose-built AI acceleration engines on its Sapphire Rapids chips — if you're willing to pay the extra fee.

But a more important development lurks in the Sapphire Rapids silicon — Intel has now progressed to its new Advanced Matrix Extensions (AMX) x86 instructions, which deliver tremendous performance uplift in AI workloads by using a new set of two-dimensional registers called tiles. The Tile Matrix Multiply Unit (TMUL) that powers AMX is native to the Sapphire Rapids chips — you don't have to pay extra to use it like you do the dedicated AI accelerator engine — and leverages BF16 and INT8 to perform matrix multiply operations that can vastly enhance AI performance.

The benchmarks above leverage Intel's AMX, and not the optional in-built AI accelerator engine. Intel claims a 7X advantage over EPYC Genoa in ResNet34, a 34-layer object detection CNN model, using INT8 instructions at a batch size of 1 to measure latency — in this case, with an SLA of sub-100ms. Intel also claims a ~5.5X advantage in this same workload with a batched test. This model is trained in PyTorch but converted to the ONNX format.

Intel claims a ~3.3X advantage over AMD in ResNet50 (INT8 BS1) image classification with a sub-15ms SLA, and a 3X advantage in DLRM, a Deep Learning Recommendation Model, with PyTorch BF16 and INT8 in a batched workload.

We also see a ~5.5X advantage in BertLarge natural language processing with BF16, but that is versus Genoa with FP32, so it isn't an apples-to-apples test. Intel notes that BF16 datatypes were not supported with AMD's ZenDNN (Zen Deep Neural Network) library with TensorFlow at the time of testing, which leads to a data type mismatch in the BertLarge test. The remainder of the benchmarks used the same data types for both the Intel and AMD systems, but the test notes at the end of the above image album show some core-count-per-instance variations between the two tested configs -- we've followed up with Intel for more detail [EDIT: Intel responded that they swept across the various ratios to find the sweet spot of performance for both types of chips].

Overall, Intel claims that its AMX acceleration provides huge speedups in performance with industry-standard frameworks, but it's also important to call out the efficiency claims. The second yellow bar for each benchmark quantifies Intel's performance-per-watt claims, an incredibly important metric in today's power-constrained data centers — particularly with rising power costs in some climes, like the EU. Intel claims AMX delivers massive efficiency advantages when comparing two chips with similar core counts, which is surprising in light of Genoa's more advanced process node that tends to give it an efficiency advantage. Yes, dedicated silicon, as we see for AVX-512 and AMX, can be costly in terms of die area, and, thus, overall cost, but the advantages are huge if the applications can leverage the accelerators appropriately. 

Intel's 32-core Xeon Platinum 8462Y+ chip squares off with AMD's 32-core EPYC Genoa 9354, but be aware that while these are iso-core-count comparisons, Intel lists the 8462Y+ for $5,945 while AMD lists the 9354 for $3,420, so the Intel chip costs 74% more. That said, the list pricing from both vendors is usually not reflective of what customers (particularly Tier 1 customers) actually pay, so take the pricing as a fuzzy guideline.

AMD does tend to offer higher core count chips at any given price point in the product stack and has a higher peak core count of 96 compared to Intel's 56 cores. Intel stuck with a 32-core vs 32-core comparison here, with per-core software licensing fees being the company's rationale for why these remain comparable. In fairness, software licensing, and other server BOM costs, like DDR5 memory and GPUs/accelerators, do have an outsized impact on solution-level pricing.

Given the size of Intel’s AI performance advantage with the 32-core parts, it’s reasonably safe to assume that it can compete with higher core count EPYC chips in these AI workloads, even if saddled with a reasonable amount of subpar workload/power scaling as they move to their own higher core count parts. Much of this performance can be chalked up to Intel's efforts on software enablement.

Intel doesn’t have any benchmark comparisons with LLMs (of the smaller variety), largely due to the raw and rapidly-changing nature of the LLM landscape. However, the company does say it is seeing impressive results with bandwidth-hungry LLMs on its HBM-equipped Xeon Max models (more below), which could be interesting given GPU shortages. We’re told that LLM benchmarks with Xeon Max will come in the future.

General Workloads: Intel Sapphire Rapids Xeon vs AMD EPYC Genoa

Here we see a standoff between the same two chips in a broader spate of general workloads. You'll note that Intel shows itself lagging Genoa in SPECint and SPECfp by large margins in efficiency and smaller margins in performance, but the company has recently begun de-emphasizing those industry-standard SPEC benchmarks because it feels the aging benchmark no longer represents real-world customer workloads -- that's actually not an uncommon viewpoint in the industry. Intel's take on SPEC is evidenced by the slideware they shared during the Sapphire Rapids briefings (second slide in above album). Intel says it is working on helping define the next generation of the venerable SPEC to make it more representative. For SPEC, Intel used the ICC compiler for Sapphire Rapids and AOCC for EPYC. The remainder of the benchmarks use 'mostly' the GCC compiler for both types of chips.

Intel also trails in the stream triad memory benchmark, which isn't surprising given that Sapphire Rapids has eight memory channels while Genoa has 12 (due to its higher channel count, AMD has 50% more memory capacity at its disposal in all of these benchmarks). That disparity also gives AMD an advantage in several other memory-bound workloads, like HPCG and Ansyst Fluent/Mechanical (not shown). It's clear that Intel is avoiding memory-bound workloads in these comparisons, though we do see those types of workloads in the HPC section where the company employs Xeon Max for comparison.

Intel largely claims slight performance gains in these workloads but notes that some of the larger gains come from employing its on-chip accelerator engines that require an extra fee to unlock. Intel used these engines in several of the workloads, marked in bronze on the horizontal axis, like SQL HammerDB, Gromacs, LAMMPs, NAMD, and others, to highlight the advantages of using the engines in tandem with applications that are tuned to exploit the advantages.

We haven't seen many benchmarks with these engines fully employed via software support, so it's clear that Intel is moving forward on the enablement front, and to good effect — at least according to its own benchmarks. Just be mindful that these accelerators, which speed AI, encryption/decryption, compression, data movement, and analytics, aren't available consistently across the Sapphire Rapids product stack and also cost extra.

There's also the matter of DPUs, too. These discrete accelerators can perform many of the same functions, such as encryption/decryption, compression, and data movement, at higher performance levels while also offloading the CPU, managing network traffic, and providing a separate control plane. Advanced DPU-augmented data center architectures aren't as widespread as the data center architectures that we consider 'traditional,' but that means the relative value of Intel's on-die accelerator engines will vary by deployment.

HPC Workloads : Intel Xeon Max vs AMD EPYC Genoa

Here we see Intel's 56-core Xeon Max, the first x86 data center chip with on-package HBM to be pushed into full production, face off with AMD's 96-core EPYC Genoa flagship. Intel claims the onboard memory gives its smaller-but-not-less-expensive chip the advantage over the EPYC processor in a range of HPC-centric workloads, but without on-board accelerators enabled (oddly, only the data streaming accelerators are available on the Xeon Max models). 

Naturally, these benchmarks won't land without plenty of dispute - there's a whole slew of architecture-specific optimizations possible for AMD's chips that could extract more performance, particularly in heavily-optimized HPC environments - but the key takeaway here is that Intel claims its HBM-equipped CPUs can offer compelling advantages in HPC that rival AMD's core-heavy flagships.

It is worth noting that both power scaling and chip fabrics can become a constraining factor with higher core-count chips in these types of workloads. As such, it's conceivable that some of these workloads could shift to a more favorable outcome for AMD with a slightly less prodigious model, like its own 64- or 56-core Genoa’s. Also, Intel's chips here only use the HBM2e memory -- there is no DDR5 attached. As such, AMD's Genoa has a tremendous memory capacity advantage (1.5TB of DDR5 capacity vs 128GB of HBM2e capacity for Xeon Max).

AMD has taken somewhat of a different path of augmenting its chips with extra memory by 3D-stacking the L3 cache for its Milan-X processors, and we expect the company to announce its newer Genoa-X chips tomorrow at its event tomorrow. However, AMD gears Milan-X for certain technical workloads, as opposed to positioning it for the broader HPC market as Intel does for Xeon Max.

TCO: Intel Sapphire Rapids Xeon vs AMD EPYC Genoa

Every vendor has its own take on how to calculate the Total Cost of Ownership (TCO), but they're all typically questionable -- overall TCO will vary so widely by deployment that it's truly hard to derive solid metrics that are indicative of broader trends.

Intel's take is that using the SPEC benchmark for deriving TCO values isn't indicative of real-world use cases, which we touched on briefly in the general workloads section above. As such, Intel provides a range of examples here for different types of deployments, and the associated savings of using Intel chips over AMD's EPYC. Some of these comparisons, like the QAT backup, aren't much better than using SPEC, so take them with a big grain of salt and be sure to check the configuration slides we've included in the album.

Intel also points to an accelerating pace of deployment, with 200 designs of 450+ Sapphire Rapids design wins already shipping. Intel also points out that it already has cloud instances in general availability at Google Cloud (notably paired with its Mount Evans IPU/DPU), and preview instances already available at several other cloud providers, like AWS. In contrast, AMD has yet to have any publicly available Genoa instances in the US, preview or otherwise, at cloud providers (we are aware of public instances available in China). We'll inquire about the status of AMD's cloud efforts at its event tomorrow and update as necessary. 

Finally, Intel reiterates that its recently-revised data center roadmap remains on track, which you can read about here. We have a few questions still pending with Intel about the above benchmark test configurations, and will add more info as it becomes available.

The always-increasing performance demands of cloud datacenters requires CPU developers to rethink their designs in a bid to deliver maximum performance per socket while confronting cost constraints set by slowing Moore's Law. AMD's EPYC 'Begamo' is the industry's first x86 cloud-native CPU that is based on the specially-tailored Zen 4c microarchitecture that maintains essentially the same feature set with the Zen 4 microarchitecture while halving core size requirements, reports SemiAnalysis. 

AMD's EPYC 'Bergamo' processor packs 128 cores and sits in the same Socket SP5 as the 96-core EPYC 'Genoa' CPU and has a similar 12-channel DDR5-4800 memory subsystem as well as uses the same I/O die (codenamed Floyd), meaning that it also has 128 PCIe Gen5 lanes and other peculiarities of SP5 products. Being a cloud-native system-on-chip (SoC) — and, to some degree, a response to emerging Arm-based datacenter-grade SoCs from Ampere, Amazon, Google, and Microsoft — Bergamo's design was shaped by multiple factors, including efficiency, power usage, die size, and low total cost of ownership (TCO) rather than by the aim to deliver the maximum per-core performance. 

On the microarchitecture level, Zen 4c retains the same design as Zen 4, including identical features and instructions-per-clock-performance, but they are configured and implemented in a drastically different way, SemiAnalysis claims.  When it comes to Zen 4c 'Dionysus' cores, they are about 35.4% smaller compared to Zen 4 'Persephone' cores, according to SemiAnalysis. To achieve this, AMD had to implement a number of design tricks. The analysts believe: 

• It reduced boost clock targets from 3.70 GHz to 3.10 GHz. This made timing closure simpler and decreased the need for extra buffer cells to meet relaxed timing constraints. Today's designs are often constrained by routing density and congestion, so lowering  frequency allows for tighter packing of signal pathways, enhancing the density of standard cells.
• It lowered the number of physical partitions of a die and packed logic closer together, which made debugging and introducing fixes harder but reduced die size.
• It used denser 6T dual-port SRAM cells for Zen 4c as opposed to 8T dual-port SRAM circuits for Zen 4 to reduce SRAM area. As a result, while Zen 4 and Zen 4c cores have similar L1 and L2 cache sizes, the area used by caches in case of Zen 4c is lower, but these caches also are not as fast as those inside Zen 4. 
• Finally, it removed through-silicon vias (TSVs) arrays for 3D V-Cache, to further save silicon. 

These were not the only methods of die area reduction used by AMD. According to SemiAnalysis, AMD's Bergamo is based on eight Vindhya core complex dies (CCDs) that pack 16 Zen 4c cores (up from eight Zen 4 cores per CCD) — which is justified as cores got smaller, but which also impacts clock speed potential. Each CCD also features two eight-core core complexes (CCX) and 32MB of L3 cache, or 16MB per CCX. By contrast, each Zen 4 CCX has 32MB of L2, which greatly increases its size compared to Zen 4c CCX. 

Overall, we could say that AMD's Zen 4c and Bergamo make design trajectory shift as the company needed to fit 128 Zen 4-class cores into the same 360W – 400W power envelope as Genoa. Reduced frequency targets, usage of denser SRAM cells, and cutting L3 per CCX in half certainly enabled AMD to increase its core count, but how that impacted per-core performance is something that we will still have to find out.

SemiAnalysis says that AMD is preparing to launch two Bergamo processors later this month: the 128-core EPYC 9754 and its slightly cut-down sibling, the 112-core EPYC 9734. Given that operators of exascale datacenters tend to have specific requirements for their deployments, we can only wonder how many custom and semi-custom Bergamo offerings AMD will eventuall produce, but for now two models are set to be introduced already next week.

"You are going to hear about this next week with Bergamo, which is a cloud-native optimized device with high-density and very good performance-per-watt in energy efficiency for cloud-native computing," said Dan McNamara, AMD's server business chief, at the Bank of America 2023 Global Technology Conference (via SeekingAlpha).

The $648 (review configuration) LattePanda Sigma is a beast of a SBC (single board computer). Packaged deep under a dominating heatsink and fan is a 13th gen Intel I5 CPU, 16GB of DDR5 RAM and an Intel Iris XE GPU (that is good enough for light gaming).

Almost a year ago we reviewed the LattePanda 3 Delta, which was an Intel Celeron N5105 based system and onboard Arduino co-processor. We felt that it was the all-in-one desktop for makers. Providing a full x86 CPU and our choice of operating system, Arduino support, and it could play a decent game from the early 2010s. Fast forward to 2023 and the Sigma’s specifications blow the Delta out of the water. We get more RAM, better CPU and GPU and a plethora of upgrade options that use off the shelf components. Heck, the CPU and GPU upgrades mean we can play some games from 2022!

We put the LattePanda Sigma on the bench and tested how it performed as a maker desktop, and we also did some light gaming.

LattePanda Sigma Versus Delta Specifications

Using LattePanda Sigma

The LattePanda Sigma is powered by a 13th generation Intel Raptor Lake CPU, the i5-1340P to be specific. This CPU has 12 cores in total: four performance cores (max turbo 4.6 GHz) and eight efficiency cores (max turbo 3.4 GHz). There are 16 threads available (via the e-cores). This mix of cores provides plenty of punch when we need it, and a good level of general performance. The more powerful CPU means that the Sigma is larger than the 3 Delta, which features an Intel Celeron N5105. It also means that the Sigma is way larger than a Raspberry Pi 4, approximately 2.5 times the area. 

As the Sigma is an x86 based SBC, we are free to use Windows or a Linux distribution of our choice. Our review unit came with a 500GB WD Black SN770 NVMe SSD with an unregistered version of Windows 11 pre-installed. LattePanda supplies a range of custom operating system images, but you can just install your own choice and tweak it to work.

Boot time for the LattePanda Sigma is fast. At 20.92 seconds the Sigma is a close second to the Raspberry Pi Compute Module 4 with eMMC which clocked in at 20 seconds. Raspberry Pi OS is a lighter OS than Windows 11, so the difference is remarkable given Windows 11’s relative “bloat”. 

A fairer comparison would be Windows 11 on an NVMe drive with the LattePanda 3 Delta. Coming in at 70.13 seconds, the Delta was much slower than the Sigma. This is to be expected given that the Delta has a PCIe 3.0 interface, versus the PCIE 4.0 of the Sigma. We also compared the Raspberry Pi 4 (micro SD) boot time of 30 seconds, and Khadas VIM4’s 36.38 seconds boot for reference.

We noticed that the PCIe 4.0 interface used for the main drive was able to keep up with our WD Black SN850X, with a constant read speed of just under 7GB/s when using Ubuntu 22.04.

We have an AMI BIOS which affords us a great deal of configuration. We can tweak almost every aspect of the Sigma, converting it into a low power desktop, or we can alter the CPU settings to grant us a little more computational power. 

Storage options with the Sigma are condensed when compared to the Delta. Gone is the eMMC. Instead we have three M.2 PCIe slots. The main slot is for PCIe 4.0 NVMe, second is PCIe 3.0. The final slot is M.2 B Key and can be used with a SATA III drive or PCIe 3.0 x1.

All of these options are impressive for what is really a turbo-charged SBC. But whereas we had to manage our expectations with the Delta, the Sigma can be your desktop PC and the brains behind a robotics or machine learning project. The Arduino element is there to control the motors, sensors and other components that make the project move, but the Raptor Lake CPU is there to give your projects the brains.

Gaming is more than possible. The Delta saw a hard limit of mid 2010s era games, but with the Sigma we have an Intel Iris Xe GPU with 80 Execution Units running at up to 1.45 GHz. That is a considerable amount of power for an “SBC” and it means that we can push gaming a little farther. 

We tested Stray, a 2022 indie game which sees the player investigating a mystery in a futuristic world. Oh and you are a cat. To get a constant 60 fps you’ll need to drop the resolution to 720p and everything else to low. If you can survive 30 fps, then 1080p with medium settings is possible, but your framerate will take a dip in congested areas. We noticed a large drop to 8 fps when the game loaded in a new area. 

This prompted us to enter the BIOS and turn off energy saving features. This gave us a little more GPU power and our framerates hovered around 40 fps. As we said, 1080p 30 fps is the sweet spot for Stray. 

We also tested Warhammer 40,000 Boltgun, a game that has just been released. This “boomer shooter” looks like the original Doom, but on modern hardware. It ran well at 1080p, most of the time way above 60fps. There were some dips, but if the fps counter wasn’t there, we would be hard pressed to notice. 

How about something a little older? I went through my Steam library and pulled out the original Call of Duty 4: Modern Warfare and this is where I hit a small issue. Matching my monitor’s resolution of 2560 x 1440 was a breeze, as was using high quality textures. The issue was moving the player to the firing range caused the game to hang, every time! 

The solution seemed to be running the game in safe mode, performing the player orientation and then turning everything back to my preferred settings. I was soon on a container ship in the middle of the ocean with Captain Price and co. If you want the best gaming performance, then you need to think outside of the box. 

The USB-C connectors offer Thunderbolt 4, so with an external enclosure and one of the best GPUs you can unlock plenty of potential. Another means is to use one of the M.2 slots and an Occulink connection with an external GPU. This would provide similar functionality to Thunderbolt 4. YouTuber ETA Prime has connected an Intel ARC A750 to the LattePanda Sigma using Occulink and has seen remarkable results.

Design of LattePanda Sigma

There is no denying that the LattePanda Sigma is a beast of an SBC. It is more like a Mini PC than a true SBC. Dominating the top of the Sigma is a large heatsink and fan. It’s so large, in fact, that ports and pins are cut out of it for easier access. The bulk does not mean that the Sigma is undesirable. All of the ports are located on the long sides, and on the short sides are pinouts for SATA, USB, fans and even a front panel connection. Flipping the Sigma over and we are greeted to a large heatsink.

Removing the four screws and we find the pre-installed 500GB NVMe drive in the PCIe 4.0 slot. We also find the Intel Wi-Fi / Bluetooth card which we included in our review unit. The underside of the Sigma is well designed, with plenty of space for adding drives, and direct contact between the drives and the heatsink.

On the same heatsink are four screw holes, which do not appear to conform to a VESA mount. These screw holes can be used to mount the Sigma to a fixed point, but you’ll need to work out your own fixing solution, perhaps something using the best 3D printers?

Cooling LattePanda Sigma

The Intel i5-1340P Raptor Lake CPU uses Intel 7 lithography and is designed with low power / mobile devices in mind. As the Intel i5-1340P is a much more powerful processor than the N5105 of the Delta, we can expect to see a higher temperature.

We tested the Sigma and Delta using Windows 11, y-cruncher and a PinePower 65W USB C power outlet with realtime power metrics. The Delta hit a max temperature of 80 degrees Celsius. The Sigma, well the performance cores hit 100°C and the efficiency cores hit a high of 88°C. Luckily the included beefy heatsink and fan are whisper quiet, unlike other SBC cooling fans we have tested.

The idle temperature for the Sigma was 25°C compared to the Delta’s 41°C. We tested the idle temps by leaving the board idling for 10 minutes. So the take away from this test is that the Sigma can really heat up when under load.

According to its datasheet, the Intel i5-1340P chip runs between 28 and 64W, but in our tests, we noted that at idle, the Sigma pulled 6W. Under stress, via y-cruncher we pushed the LattePanda Sigma to pull 66W! How does that compare to the Delta? At idle, the delta pulled 4.56W, under stress 19.76W. Again this all boils down to the i5-1340P CPU.

Let's look at the power consumption. Under stress, the LattePanda Sigma consumes 10 times more power than the Raspberry Pi 4 and Khadas VIM4 (60W versus 6W). If you are building a project with a power budget, for example robotics or off-grid data collection, then perhaps the Raspberry Pi or Khadas VIM4 is more applicable.

If your power budget needs to be even lower, perhaps a microcontroller such as the Raspberry Pi Pico W is a better option. That said, the extra power consumption can be factored into your build and if you really need that much processing power in a portable form factor, the Sigma packs a real punch.

LattePanda Sigma GPIO

The GPIO of the LattePanda Sigma is different to that of the Delta. We get less GPIO pins than the Delta, but what we get is pretty much an Arduino Uno in a different form factor. The GPIO is for the onboard Arudino Leonardo compatible coprocessor. It seems that an Arduino compatible co-processor is quite useful for Intel based SBCs. The LattePanda 3 Delta had its own (and a breakout for RS232 and audio) as does Seeed’s Odyssey.

The double row of headers has a pinout printed on its side. Reading the pinout was simple and clear, in fact we prefer it to the Delta which saw its heatsink get in the way.

Controlling the Arduino GPIO is possible using the pre-installed Arduino IDE, an older version of the IDE (1.8.19) which works, but would’ve liked to have seen v2.0 onwards. In fact, we installed the latest Arduino IDE and set the board type to Leonardo, but the IDE would not upload to the board. We attempted to install the board, but alas it didn’t work. Not a massive loss, but the latest Arduino IDE is leaps and bounds over its predecessors. 

Using the pre-installed IDE, we tested the Arduino GPIO with the humble LED, blinking the LED every half second. With that success under our belt, we connected a one meter length of WS2812B NeoPixels and installed the appropriate libraries. Moments later, colors erupted from the chain of RGB LEDs.

This GPIO is a better compromise than the Delta, and for Raspberry Pi alternatives, it is pretty good. Other Raspberry Pi alternatives have attempted to ape (hardware and software) the Raspberry Pi format, and they have varying levels of success. By using a simple pinout form factor, while retaining compatibility with Arduino code and components, the LattePanda Sigma provides the GPIO access that we need. Sure the pinout isn’t Arduino shield compatible, but we can use a few wires to bridge the gap.

LattePanda Sigma Linux Performance

If Windows isn’t your thing, we hear you. Linux is generally a better fit for SBCs and so we installed the latest LTS (Long Term Support) release of Ubuntu 22.04.2 then fully updated the OS. Our installation was contained on a 1TB WD_Black SN850X  NVMe SSD so that our previous Windows installations remained untouched. 

We’d love to say that “Ubuntu just worked”, and while there were no installation issues and it detected our hardware correctly, there was an issue that existed between 22.04 and 23.04. The issue was that the live USB failed to fully boot, complaining of “unable to find a medium containing a live file system.”

A quick Google search revealed that this has been a problem for other computers. The solution? Weirdly it was to remove the live USB during the boot process (just as the Ubuntu logo flashes on the screen) for two seconds, then put it back in. Yes, that worked. I installed the latest Arduino IDE and hit the same Arduino IDE issue as Windows. So I followed the LattePanda 3 Delta guide to configure the LattePanda Leonardo board, but this only works with the older Arduino IDE. Suffice to say I was able to configure the board with Arduino 1.8.19 and complete a series of GPIO tests. 

Ubuntu 22.04.2 is a slightly lighter OS than Windows 11. It felt no different than Windows 11 for its responsiveness,largely down to the powerful Intel CPU and the generous 16GB of DDR5 RAM. We could happily sit down and use Linux on the Sigma for our general work and to write code. We’d leave the gaming side of things to Windows as even after installing the latest drivers, configuring Steam to use Proton and then downloading shaders, both Stray and Warhammer: Boltgun failed to launch.

LattePanda Sigma versus LattePanda 3 Delta versus Raspberry Pi 4

The Raspberry Pi 4 is, like it or not, the benchmark from which other SBCs are measured. Sure it's not a desktop replacement and the hardware package is now five years old (pandemics and global supply chain issues artificially extended the shelf life of the Pi 4), but it is a good all-rounder from which we can measure.

The LattePanda 3 Delta is much more powerful than the Pi 4; we proved that in our review. But the Sigma screams past them both in terms of raw performance, as it should given the 13th gen Intel CPU and 16GB of DDR5. But this performance comes at a price, around $648 for our review configuration. This is just over double the price of the LattePanda 3 Delta and for $648 we can buy eight Raspberry Pi 4 8GB at non-scalper prices.

What about the Jetson Orin Nano? The $499 Jetson Orin Nano is a close contender with an Nvidia GPU featuring 32 Tensor Cores, a six-core Arm A78AE v8.2 64-bit CPU and 8GB of LPDDR5. But we have to reserve our judgment until our review is complete. We had to postpone the review due to software issues at the time of its release.

Bottom Line

If you need lots of computation power and a GPIO then the LattePanda Sigma is a contender for your cash. You will need to justify the spend as $648 is a lot of cash for what is essentially a mini PC with an embedded Arduino. That said, the package is sublime and clearly well thought out.

The $648 price tag is initially hard to swallow but again this isn’t just a replacement for the Raspberry Pi, it is a desktop computer in a slightly larger SBC form factor. You could happily use this machine for day-to-day work, while using less electricity than on a traditional desktop. Or you can embed it into a robotics / autonomous / AI build and harness all of the CPU cores to perform complex operations while on the move.

Arduino compatibility is still the icing on the cake for the LattePanda machines. It is easy to use and works with the many thousands of libraries available in the Arduino ecosystem. Sure, we can’t directly use Arduino shields, but unless you have bought into that system, you aren’t going to miss them.

The LattePanda Sigma is a revolution compared to the Raspberry Pi 4. Compared to the LattePanda 3 Delta, it is an evolution, and one that will entice makers craving more power for their builds. The Sigma is the new maker’s desktop, for makers that need the extra horsepower.

The Chinese Powerstar P3-01105 CPU has popped up in the Geekbench v5 online result browser. Importantly, the benchmark's system information section appears to confirm that this 4C / 8T chip is quite certainly Intel-produced, as this Socket 1200 LGA part has an A0653 (GenuineIntel) CPUID, and purportedly uses the Intel Comet Lake architecture.

Earlier this month we reported on the newly launched first generation Powerstar P3-01105 CPU. China's PowerLeader framed the chip as a home-grown product using the "storm core" architecture, while remaining x86 compatible. However, there was quite a weight of evidence already pointing to the 'Chinese' chip being a rebranded Intel Core i3-10105(F) Comet Lake CPU with 4C / 8T.

Earlier evidence of the Powerstar / Intel Comet Lake similarities included the following:

• Physical lugged heatspreader design, and other physical characteristics
• Physical substrate design is identical, as far as we can see
• Silk screen print format on the IHS is the same
• The PowerLeader processor name slightly jumbles Intel's: compare "10105" and "01105"
• They are both marked as capable of a "3.70GHZ" base clock
• The QR code on the upper right of the Powerstar P3-01105 PCB is said to match Intel's.

The new evidence from Geekbench is probably more than enough for most people to be certain about the provenance of the Powerstar P3-01105 CPU. However, we still must retain some doubt, as pranksters fiddle with things like Geekbench system info reports just for fun.

PowerLeader has the ambitious target of selling 1.5 million units of its Powerstar P3-01105 CPU. We think that the CPU being a 'Genuine Intel' part can only help in this goal, as the "storm cores" touted at launch had no history, and even optimists would expect some glitches in an actual 'first gen' product.

PC enthusiasts and DIYers in the west might now be thinking about sourcing Powerstar P3-01105 CPUs for budget builds. However, PowerLeader's launch suggests the chips will only be supplied within full systems. We shall have to see about that over the coming weeks / months.

As we have now seen the Powerstar surface in Geekbench, we can probably just wait a few weeks before some enterprising Chinese YouTuber (or Bilibili-er, or Weibo-er) shares some more in-depth information, tests, and benchmarks of the P3-01105 CPU.

In some ways PowerLeader's rebadging of something as complex and American as an Intel CPU, contrasts with the recent announcement that Micron memory products have been banned from organizations connected to China's critical information infrastructure.

Hackboard today launched the Hackboard 2, a compact and affordable Wintel SBC. Key features of this 4.7 x 3.1 inches (120 x 80 mm) board are its Intel Celeron N4020 dual-core CPU, its good selection of ports and connectivity, the licensed Windows 11 Pro OS, and its $199 price tag.

The Hackboard makers tout it as an ideal maker board for general x86 hobbyist projects, IoT projects, for work from home folk, for edge and embedded solutions, and more. In a press release, Mike Callow, Hackboard's co-founder and CEO said that the Hackboard 2 is also designed "to address the increasing worldwide challenge of the Digital Divide and the needs of education." Those are big asks for a 2C/2T 14nm Celeron-powered SBC.

Hackboard is working to put together some kits built around the Hackboard 2, featuring extras such as a case, a wireless keyboard / touchboard, a webcam, and a portable monitor. There are also optional 4G and 5G cellular modems to expand the connectivity options already on offer.

Regular readers may remember the Hackboard 2 from its crowdfunding stage, way back in Dec 2020 / Jan 2021. However, this SBC is now available from a wide range of resellers, including RS Components, OKdo, Micro Center, Mouser, CrowdSupply, and Vilros.

Perhaps of some importance to the chance of success of the Hackboard 2 is the change in pricing. During the crowdfunding stage, there was an appealing $99 option, which is now gone. Now the cheapest model post-launch model is $175 with Debian Linux, or $199 with a Windows 11 Pro license. With the new price, the above Raspberry Pi 4 comparison might not be as fair as it used to be. And there are full Windows systems with newer hardware available for less money. Having said that, a supply / demand imbalance has affected the Pi in recent months and caused significant price inflation -- when stock is available. So for those looking for a project board who are inclined toward x86, the Hackboard 2 offers up an interesting alternative.

On Saturday, Chinese computer manufacturer PowerLeader launched a 'new' processor and compact desktop PC. During the press conference, PowerLeader talked about its first generation Powerstar P3-01105 CPU, featuring the "storm core" architecture, described as "extremely high performance," x86 compatible, and offering great support for Windows. And so it should, as it looks very much like a rebranded Intel Core i3-10105(F) Comet Lake CPU with 4C / 8T.

In Chinese tech news media coverage, such as at ITHome, there was no hint given about a partnership with Intel on this rebrand. Instead, we have a quote from a PowerLeader execs talking about how the first gen Powerstar chips were "specially designed for daily desktop users, and are suitable for government, education, energy, industry, finance, medical care, games, and retail, etc." 

As well as the CPU, a new compact tower PC dubbed the PC PT620P, featuring the Powerstar P3-01105 CPU, was unveiled. A statement about this PC's capabilities provided a small hint that the Powerstar P3-01105 wasn't entirely domestic. According to the machine translation of a PowerLeader exec's statement, the new CPU "has extremely high performance, which is several times higher than that of the domestic CPU." There, it was admitted that this isn't a domestic CPU.

One of the last official blurb worth reporting is that PowerLeader says the CPU will be launched shortly, targeting annual CPU sales of 1.5 million units.

It's an Intel Chip, Sherlock

In response to harukaze5719's Tweet about this Powerstar P3-01105 CPU, another Twitterer pointed out that what we are looking at is almost certainly an Intel Core i3-10105(F) Comet Lake CPU. The following clues convincingly point to this:

• Physical lugged heatspreader design, and other physical characteristics
• Physical substrate design is identical, as far as we can see
• Silk screen print format on the IHS is the same
• The PowerLeader processor name slightly jumbles Intel's: compare "10105" and "01105"
• They are both marked as capable of a "3.70GHZ" base clock
• Last but not least, the QR code on the upper right of the Powerstar P3-01105 PCB is said to match Intel's.

There seems to be a weight of evidence pointing to a rebrand, but we await further announcements and third-party tests to ensure it. So please stay tuned for when the truth comes out.

So, why would PowerLeader rebrand an Intel Core i3-10105? All we can find about the underlying reason for this sleight-of-hand with branding is that Chinese tech firms can benefit from attractive subsidies by developing and launching "domestic products." If that is true, it also explains how Innosilicon touted its Fantasy graphics cards as "domestic desktop GPUs," but later, we saw UK-based ImgTec confirm that they used the PowerVR architecture.

Intel researchers are working on novel solutions to cool next-gen chips up to 2000W. The iconic x86 chipmaker already threw its weight behind immersion cooling a few years back. However, the march of Moore’s Law and increasing chip densities means Intel is now busy looking to adapt/augment its best cooling technology with “new materials and structures.”

Cooling is a very serious business, especially in the data center. Improvements to cooling efficiency can significantly impact the bottom line of a data center operator. Powerful processors will chew through the lion’s share of watts, but research suggests cooling accounts for up to 40% of a facility’s power consumption. Better cooling could also allow chips to run faster.

Tantalizingly, Intel says that it will be working closely with innovative cooling technology companies on solutions that “seem squarely in the realm of science fiction.” It doesn’t name names, but the clues are big enough for us to have an educated guess about two of its major collaborators.

Coral-like Structures, Tiny Liquid Jets

One of the new cooling solutions is said to be based on technology “like 3D vapor chambers embedded in coral-shaped heat sinks.” This sounds like it could be a collaboration with Belgium’s Diabatix, which specializes in heatsinks and cold plates created by a generative design software platform. The results look highly organic, like coral growth, and the structures make for ultra-low thermal resistance.

Intel also mentions that a partner has technology that features “tiny jets, adjusted by artificial intelligence, that shoot cool water over hot spots in the chip to remove heat.” Again, Intel doesn’t name any names, but the technology sounds very similar to that championed by MIT spinoff JetCool. This Massachusetts-based firm also fits the brief of providing integrated die-cooling technology for “exceptional” performance.

Elsewhere in its blog about potential new cooling technologies, Intel mentions it is looking closely at cooling technologies utilizing 3D vapor champers, advanced materials, and boiling enhancement coatings. This work is “existentially important for our future,” said Tejas Shah, lead thermal architect for Intel’s Super Compute Platforms group.

Cooling isn’t all about efficiency and saving energy, though. Intel asserts that new cooling technology has the potential to enable processors to run at lower temperatures and thus deliver “a 5% to 7% increase in performance for the same power.”

The AMD Ryzen 7 7800X3D vs Intel Core i9-13900K vs Intel Core i7-13700K showdown is a matchup of one upper mid-range processor against two nimble and competent competitors for the title of the best CPU for gaming. 

The $448 Ryzen 7 7800X3D is now the fastest gaming CPU in the world. The 7800X3D recently debuted to critical acclaim, arriving with eight cores and 16 threads and AMD's Second-Gen 3D V-Cache that deliver ever-so-slightly more gaming performance than the flagship $699 Ryzen 9 7950X3D. The 3D V-Cache tech fuses a 3D-stacked cache chiplet atop the processor to deliver market-leading gaming performance at the expense of performance in productivity applications, but while it delivers explosive uplift in some games, not all titles benefit. 

In contrast, the $570 Core i9-13900K wields 24 cores and 32 threads with Intel's 13th-Gen Raptor Lake x86 hybrid architecture. The Intel Core i9-13900K is Intel's fastest gaming processor, and while it isn't quite as fast as AMD's X3D chips in gaming, it serves up a more predictable blend of performance in both gaming and productivity apps.

Intel also has its Core i7-13700K, which offers nearly the same level of performance as the 13900K in gaming, but at a much more amenable $417 price point that is more competitive with the Ryzen 7 7800X3D. While the 13700K isn't as fast in productivity work as the 13900K, it also offers far more performance than the 7800X3D in the types of apps. 

This makes for a complex array of choices if you're looking for a flagship gaming experience with your next rig, and that's before we even factor in pricing. Below we've taken the Ryzen 7 7800X3D vs Core i9-13900K vs Core i7-13700K rivalry through a six-round faceoff to see which chips take the crown in our gaming and application benchmarks, along with other key criteria like power consumption and pricing. Let's see how they stack up. 

Features and Specifications: AMD Ryzen 7 7800X3D vs Intel Core i9-13900K and Core i7-13700K

AMD’s $449 Ryzen 7 7800X3D comes packing eight Zen 4 cores with the company’s second-gen 3D V-Cache tech that unlocks 96MB of total L3 cache. This combo delivers the fastest gaming performance available on the market. The 7800X3D has the same underlying design as the Ryzen 7 7700X with a central I/O die and one eight-core compute chiplet that provides a total of 8 cores and 16 threads.

Unlike the standard Ryzen 7000 chips, the 7800X3D's compute chiplet has an additional 3D-stacked L3 cache. However, this tech results in a lower boost clock of 5.0 GHz than the standard single compute chiplet models, which can reach up to 5.4 GHz. Despite the lower frequency, the cache-equipped chiplet provides the best gaming performance. In contrast, AMD's 7950X3D and 7900X3D come with two compute chiplets, requiring a new driver to steer threads for different types of workloads to the correct chiplet. In contrast, the 7800X3D is a plug-and-play processor that doesn't require special software.

The Ryzen 7 7800X3D has all the accommodations we expect from a modern chip, including DDR5, PCIe 5.0, and the latest USB standards. The chip also has an integrated graphics engine, but it isn't suitable for gaming. You can see our iGPU testing here.

The 7800X3D is incredibly power efficient, but AMD assigned a base TPD of 120W and a max 162W PPT, 15/20W higher than the 105W/142W rating for the standard 7700X. As you'll see in the power testing section, the higher TDP rating doesn't make much sense. Regardless, you'll still need to plan for a 240mm to 280mm AIO (or better) for the best performance.

The Ryzen 7 7800X3D requires an AM5 motherboard, so DDR5 memory is the only option. Unfortunately, DDR5 continues to carry a premium over DDR4 memory. Granted, that pricing pressure isn't as much of a consideration for the highest-end systems, but you should factor it in if you're struggling to cobble together a system in this price range. The AM5 platform is new, and the Ryzen 7000 chips aren't backward compatible with older AM4 motherboards. However, AMD will support the AM5 platform until 2025+.

The 7800X3D supports the auto-overclocking Precision Boost Overdrive (PBO), Curve Optimizer, and fabric and memory overclocking. Unfortunately, direct frequency overclocking isn't possible due to a voltage limitation for the 3D tech.

Intel's $589 sixteen-core Core i9-13900K comes with eight hyper-threaded P-cores and 16 single-threaded E-cores, for a total of 32 threads. That's an increase of eight additional E-cores over the previous-gen Core i9-12900K, equating to more threaded horsepower. You can also save some cash with the $564 Core i9-13900KF, which has identical specs but lacks the integrated GPU.

The P-cores come with a 3.0 GHz base, but, more importantly, a 5.8 GHz boost that's a whopping 600 MHz increase over the prior gen. Meanwhile, the E-cores now have a 3.0 GHz base (+600 MHz) and stretch up to 4.3 GHz (+400 MHz). The Core i9-13900K also has 32MB of L3 and 36MB of L2 cache.

The $417 Core i7-13700K comes with eight p-cores and eight e-cores, for a total of 16 cores and 24 threads. The reduced number of e-cores means this chip doesn't provide as much threaded horsepower as the 13900K, but its 5.4 / 2.4 GHz boost/base clocks keep performance snappy in latency-sensitive work, like gaming, while the e-cores are also effective in threaded work with a 4.2 / 2.5 GHz boost/base. 24MB of L2 and 30MB of L3 round out the package. You can also save some cash with the graphics-less $384 Core i7-13700KF.

The Raptor Lake chips use a newer revision of the 'Intel 7' node to push clock speeds higher while improving power efficiency. Intel's chips now have a Processor Base Power (PBP) value instead of a TDP metric, and a secondary Maximum Turbo Power (MTP) value that denotes the highest power level during boost activity. Be aware that modern Intel chips can always operate at the MTP when under load, though the actual power use will vary with application intensity. The 13900K and 13700K both come with 125W PBP (base) and 253W MTP (peak) power ratings, but in practice, the 13700K sucks less power.

The Raptor Lake chips support DDR4 or DDR5 memory, which can save you money on both the motherboard and memory. Both Raptor Lake chips drop into existing 600-series motherboards or the new 700-series models, but you'll need to ensure a 600-series board has the right BIOS or BIOS flashback.  Intel allows full overclockability, but only when you pair the chips with a Z-series motherboard.

✅ Winner: Tie

The Core i9-13900K vs Ryzen 9 780X3D vs Core i7-13700K battle yields a tie in this category. All three chips support DDR5 and the PCIe 5.0 interface, meaning none have an inherent connectivity advantage.

Intel has the advantage of backward compatibility for DDR4 and the 600-series chipset, enabling lower-cost memory and motherboards for value seekers (caveats apply). Intel also has its Raptor Lake Refresh chips on the horizon, meaning there will be one more generation of upgrades for the LGA 1700 platform.

AMD's new AM5 platform generally carries a premium over the Intel motherboard ecosystem, but it does afford extended forward compatibility — AMD plans on supporting the AM5 platform until 2025+.

Given its new hybrid architecture with two types of cores, Intel's core counts aren't directly comparable to AMDs due to their different capabilities. Our performance results will dictate the performance value of the overall designs.

Gaming Benchmarks and Performance: AMD Ryzen 7 7800X3D vs Intel Core i9-13900K vs Core i7-13700K

Remember, this article is an overview of our much more in-depth testing in our AMD Ryzen 7 7800X3D Review. Below is the geometric mean of our gaming tests with the Core i7-13700K vs AMD Ryzen 7 7800X3D vs Intel Core i9-13900K at 1080p and 1440p, with each resolution split into its own chart. As usual, we're testing with an Nvidia GeForce RTX 4090 to reduce GPU-imposed bottlenecks as much as possible, but differences will shrink with lesser cards or higher resolutions. 

The 3D V-Cache tech puts AMD’s X3D chips in a league of their own in gaming, so the $449 7800X3D contends with the $699 16-core Ryzen 9 7950X3D for the lead — a victory it wins by a mere 1% in its stock config. The 7800X3D increases its lead over the 7950X3D to 2.5% at 1440p, but this is still a very close contest that falls into the imperceptible range. The price tag and number of cores is the biggest difference between these two chips; the 7950X3D costs $250 more and has twice the number of cores, which will help in productivity workloads.

The Ryzen 7 7800X3D is 12% faster in 1080p gaming than the $580 Core i9-13900K, and overclocking only narrows that to 8%. The Core i9-13900K requires exceptionally aggressive accommodations for overclocking. In contrast, the Ryzen 7 7800X3D never exceeded 90W in our testing. Overall, the 7800X3D will run faster and cooler than the 13900K in every gaming scenario, but the 13900K is far more potent in productivity-oriented tasks.

The $417 Core i7-13700K trails the 7800X3D by 14%, but it costs ~$30 less, and its more balanced performance in productivity apps makes it attractive if you're looking for a better all-rounder. The 13700K really takes the shine off the 13900K for gaming-focused rigs — the 13900K is only 1.5% faster in gaming but costs 39% more cash (an extra $163).

Moving over to 1440p pushes the bottleneck more towards the GPU, so the difference between the chips shrinks. For example, the Ryzen 7 7800X3D is 9% faster than the Core i9-13900K at 1440p with both stock and overclocked settings. It's logical to assume that most high-end gaming systems will use higher-resolution monitors, so the differences could be slight, but the 7800X3D will give you more headroom for future GPU upgrades.

As a reminder, the 3D V-Cache doesn’t accelerate all titles equally. Therefore, you should also look at the individual game benchmarks in the above album to assess the overall trend. As always, the AMD vs Intel competition can vary based on the title and the GPU you use, so it's best to make an informed decision based on the types of titles you play frequently.

Be aware that large deltas in individual game titles, as we see with the Ryzen 7 7800X3D, impact cumulative measurements. You can see examples of that in the above album in titles like Microsoft Flight Simulator 2021, Far Cry 6, and F1 2021. Our test suite heavily favors the improvements from 3D V-Cache, so we also included AMD’s own benchmarks in a broad range of games at the end of the above album to give you an idea of performance gains in other titles. As with all vendor-provided benchmarks, take AMD's slides with a grain of salt.

✅ Winner: AMD Ryzen 7 7800X3D | All-Rounder Alternative — Core i7-13700K
It may be by the slimmest of margins over its own sibling, but on average, the Ryzen 7 7800X3D is the fastest gaming chip money can buy, dethroning the 7950X3D at a much friendlier price point. It is also much faster than both of the Intel contenders.

In our test suite, the Ryzen 7 7800X3D is 12% faster than the Core i9-13900K at 1080p gaming and 9% faster at 1440p. After overclocking, those deltas shrink to ~9% at 1080p and 1440p.

Ultimately, despite its lower-than-suggested pricing of $580, we can't give the 13900K the nod in this competition — instead, the Core i7-13700K gets an honorable mention. The 13700K delivers 98.5% of the 13900K's gaming performance despite the fact that the 13900K costs $163 more (39%).

The Ryzen 7 7800X is 14% faster than the 13700K, but you will pay higher up-front pricing to build a 7800X3D system. The 13700K is competitive with the 7800X3D in our system-level and fps-per-dollar pricing calculations below, leaving it as a competent alternative.

Productivity Performance: AMD Ryzen 7 7800X3D vs Intel Core i9-13900K vs Core i7-13700K

We can boil down performance in productivity apps into two broad categories: single- and multi-threaded. The first slide above shows the geometric mean of performance in several of our most important tests in the single-threaded category, but be sure to look at the expanded results in the above album.

The Core i9-13900K beats the 7800X3D easily: it's 32% faster in single-threaded work, while the Core i7-13700K is 22% faster. Naturally, those performance deltas won't carry over to every type of workload. However, flipping through the album of test results reveals that the Intel processors excel in single-threaded work, thus providing more responsive performance than the 7800X3D in light applications.

The first slide in the above album contains a geometric mean of performance in several of our multi-threaded benchmarks. In multi-threaded work, the $580 Core i9-13900K is a whopping 89% faster than the $449 Ryzen 7 7800X3D at stock settings, while the $417 Core i7-13700K is 51% faster. That's an easy win for the Intel chips.

✅ Winner: Intel Core i9-13900K and Core i7-13700K

The Core i9-13900K and 13700K have a more balanced profile in both single- and multi-threaded productivity applications than the 7800X3D, delivering solid performance in both disciplines. In contrast, the Ryzen 7 7800X3D suffers mightily in these workloads compared to the Intel silicon, highlighting that this is truly a highly specialized chip for gaming that won't offer as much performance in standard workloads as the regular Ryzen 7000 processors. 

The implication here is clear — if performance in productivity workloads is important to you, select the Intel processor that best matches your needs in terms of both performance and pricing. The Ryzen 7 7800X3D is specifically designed for gaming-centric rigs, so you should steer clear if productivity apps are a big part of your daily usage.

Overclocking: AMD Ryzen 7 7800X3D vs Intel Core i7-13700K vs Core i9-13900K

We have long since reached the land of diminishing returns for overclocking the highest-end chips from AMD and Intel, like the Ryzen 9 and Core i9 families. Because AMD has pulled Intel into a dogfight for performance superiority on the high end, much of the overclocking frequency headroom is rolled into standard stock performance. However, there are still at least some gains to be had, and both Intel and AMD expose tunable parameters and sophisticated software overclocking utilities, like Intel's XTU and AMD's Ryzen Master.

Like the rest of AMD's new X3D chips, the Ryzen 7 7800X3D doesn't support overclocking via the CPU multiplier, so you can't change the core clocks or voltages directly. However, AMD now allows changing the power limits (PPT, TDC, EDC) through the auto-overclocking Precision Boost Overdrive (PBO) feature, and you can also now undervolt with the Curve Optimizer. These features weren’t supported on the previous-gen models. You’ll find plenty of examples of PBO+UV performance in our Ryzen 7 7800X3D review. As before, you can overclock the memory and fabric, too, though we've seen limited gains with the 7800X3D from these approaches.

Meanwhile, the Intel Core i9-13900K and Core i7-13700K expose every knob possible for tweaking and tuning. However, the 13900K's aggressive out-of-the-box power and Adaptive Boost Technology already consume most of the overclocking headroom you could achieve with conventional cooling. The 13700K leaves slightly more room than the 13900K for manual tuning due to its less-aggressive tech, but you shouldn't expect massive improvements. Sub-zero overclockers with exotic cooling solutions can reach much higher with the Intel chips than the 7800X3D, but that isn't as valuable to the overwhelming majority of users. 

As usual, performance gains from overclocking can boil down to your luck in the silicon lottery. As you can see above, overclocking with conventional cooling for both the Intel and AMD processors results in fairly minimal gains in applications. In fact, the 13900K actually suffers from lower performance in single-threaded work because its 5.6 GHz overclock is lower than its stock 5.8 GHz boost. The Core i9-13900K does gain 6% in gaming, a higher percentage gain than the 13700K and 7800X3D, but it still lags behind the 7800X3D's peak gaming performance.

✅ Winner: Tie

The payoff for overclocking is extremely limited with today's high end chips. Still, both platforms have plenty of knobs for enthusiasts, their respective overclocking advantages, and a suite of auto-overclocking and software utilities. Of course, dedicated tuners could find larger performance boosts by overclocking the core, fabric, and memory more aggressively than we have, but bear in mind that you're always at the whims of the silicon lottery when it comes to overclocking.

AMD's 3D V-Cache technology prevents direct voltage and frequency manipulations, but AMD has unlocked memory, fabric, Precision Boost Overdrive, and undervolting via the Curve Optimizer, all of which combine to provide plenty of avenues for boosting performance via various tuning techniques. In fact, the 7800X3D often delivers a similar amount of performance uplift as the Core i9-13900K, showing the value of its somewhat unconventional approaches. Additionally, some motherboard vendors, like MSI, have baked in new tuned overclocking profiles directly into the motherboard firmware, promising up to a 10% improvement in performance. But, again, your mileage will vary.

Meanwhile, Intel offers more overclocking functionality and the highest peak frequencies from direct tuning. However, much of that capability is already exposed right out of the box via the company's aggressive boosting technologies and sky-high power limits, thus reducing its value to casual overclockers. 

Power Consumption, Efficiency, and Cooling: Intel Core i7-13700K vs AMD Ryzen 7 7800X3D vs Core i9-13900K

The Ryzen 7 7800X3D is incredibly power efficient, drawing far less power than the competing Core i9-13900K or 13700K, thus delivering superb power efficiency that ranks among the best we’ve ever tested. A quick glance at the HandBrake renders-per-day-per-watt metric shows that the efficiency competition isn't even close — the 7800X3D is the hand's down winner.

In fact, the Ryzen 7 7800X3D appears to fall into the standard 65W profile that specs a maximum of 88W of power consumption (PPT). AMD has spec'd a higher 120W TDP rating for the 7800X3D, but it appears to be running closer to the 65W TDP envelope during normal use — given our measurements, the Ryzen 7 7800X3D's 120W TDP rating seems far overprovisioned. This means the chip is basically operating in Eco Mode, which confers massive power consumption and efficiency benefits.

In contrast, Intel's chips consume far more power in every type of workload, resulting in higher thermal output.

✅ Winner: Ryzen 7 7800X3D

Intel has reduced its power consumption metrics, but the Ryzen 7 7800X3D is still far more power efficient. The 7800X3D consumes much less peak power and also delivers more work per unit of power consumed. That results in a win in power consumption, efficiency, and thermal output, ultimately giving you a cooler and quieter system.

None of these chips comes with a bundled cooler, but Intel's higher power consumption means you'll need more aggressive, and thus more expensive, cooling to extract the peak performance. In contrast, we don't think most users will have problems cooling the Ryzen 7 7800X3D with the recommended 280mm or greater cooler. 

Pricing: Intel Core i9-13900K vs AMD Ryzen 7 7800X3D vs Core i7-13700K

Tallying up the pricing numbers to compare the Core i7-13700K vs Ryzen 7 7800X3D vs Core i9-13900K can get incredibly complex once we start factoring in different classes of motherboards, so we included as much information as possible (links in the text below).

The Ryzen 7 7800X3D stands out as the fastest gaming chip money can buy and as the best gaming value in this price range. At $2.00 per frame, the 7800X3D is a far better value than the 13900K, which works out to $2.90 per frame. Of course, the Core i7-13700K is slower than the 7800X3D, but it is within striking distance in our value calculation at $2.11 per frame.

As you can see when we calculate the total platform cost above, plenty of other cost factors are in play, like memory and motherboards. In the end, provided you ensure it has adequate power delivery (which is easy for the 7800X3D), you can drop the 7800X3D into a B650 board and pay $2.97 per frame. That's the best value of the potential configurations. The 13700K with DDR4 memory on a B760 motherboard is close at $3.21 per frame, but it still lags slightly at $3.21.

Surprisingly, after equipping both chips with DDR5, we basically have a tie with the 13700K on a Z790 board and the 7800X3D on an X670. The Core i9-13900K simply isn't worth the cash if you're aiming solely for gaming, but it has its own attractions on the productivity side.

We used both the B650 and X670 chipsets for modeling the Ryzen 7 7800X3D's total system price, and tallied different price points based on DDR4, DDR5, Z790, and B760 motherboards for the Intel processors. Note that we didn't include a B760/13900K combo because that doesn't make sense with that class of chip, and you'll need to be careful to ensure adequate VRMs when pairing the 13700K with a B760 board. You also can't overclock the cores with a B-series board. Finally, we reduced the fps value for the Intel chips by 4% for the DDR4 calculations, as our testing shows that to be the average performance loss with the higher-end chips from stepping back from DDR5. You will retain the full performance in productivity applications, though.

The graphics-less $551 Core i9-13900KF saves you $30, and the $392 Core i7-13700KF saves you $25. We did not include those reductions in our calculations. The Core i9-13900K and 13700K are also compatible with previous-gen motherboards if they have the right BIOS, and performance will be the same between 600- and 700-series if the power delivery is identical.

AMD's X670/E motherboards currently start at roughly $80 more than Intel's Z790, but AMD B650 boards start at $126. However, you'll only find a single B650 board for around $130, while seven Intel B760 boards are either at or below $130. This is due to the lower pricing for Intel's DDR4-equipped motherboards; we've priced those differences into our full-system measurements.

Memory pricing is also tricky. DDR4 memory isn't an option with the AM5 platform that houses the 7800X3D, and while the deltas between DDR5 and DDR4 are becoming smaller over time, they still exist: You'll pay ~$60 bucks at a minimum for 32GB of DDR4, but pricing starts at $90 for 32GB of DDR5. That gap increases with higher capacities — it's $100 for the least expensive 64GB DDR4 kit but $195 for a 64GB DDR5 kit. That's nearly twice the price. How much weight you should give to the difference in memory pricing will be impacted by the speed and capacity of the kit you select. Additionally, perhaps the best memory kit is the one you don’t have to buy — many upgraders likely already have a DDR4 kit, which is something to consider if you have a kit on hand. 

✅ Winner: Ryzen 7 7800X3D and Core i7-13700K

The Ryzen 7 7800X3D has a lower price tag than the $580 Core i9-13900K, but much higher performance in games. Overall, the 7800X3D has a commanding lead in both raw and system-level fps-per-dollar metrics, thus giving it the nod in the pricing department.

The Core i9-13900K is a superior offering if you're interested in performance in productivity applications. Still, the Core i7-13700K offers the lion's share of the 13900K's performance at a lower price point. In gaming, the Core i7-13700K offers similar raw and system-level pricing to the 7800X3D when you factor in the fps-per-dollar, thus also earning it a win in the pricing category.

The higher cost of AMD's motherboards generally adds more cost to your build than Intel's platforms, and the price premium for DDR5 also continues to add cost: At higher capacities, you'll pay nearly twice as much for DDR5 memory as DDR4 memory, and DDR5 is a strict requirement for AMD's 7800X3D.

However, the Ryzen 7 7800X3D has the performance lead in gaming by a large enough margin that it holds the lead in terms of overall value for builds focused strictly on gaming, an advantage that is most apparent against the 13900K. Additionally, the 7800X3D is the fastest gaming chip on the market, a privilege that most are typically willing to pay a premium for, yet it still beats the more value-centric 13700K, albeit by smaller margins, in our pricing efficiency metrics. 

Bottom Line: AMD Ryzen 7 7800X3D vs Intel Core i9-13900K vs Core i7-13700K

We have two separate battles in this contest: The Ryzen 7 7800X3D vs Core i7-13700K ends in a tie, with both chips getting the nod in five categories. Meanwhile, the Ryzen 7 7800X3D vs Core i9-13900K matchup tallies up to a 5-to-3 advantage in the 7800X3D's favor. 

For gaming-centric rigs, the Core i9-13900K doesn't offer enough gaming performance to justify its price premium over the 7800X3D, and its other deficiencies in pricing and power consumption hurt its broader overall score. However, the 13900K vs 7800X3D contest isn't as much of a blowout win as the score implies. In the end, both of these processors have their own distinct advantages and disadvantages depending on your target use case.

If you're concerned about building a gaming-focused rig and everything else is a secondary concern, the Ryzen 7 7800X3D is by far the superior chip over the 13900K in terms of both outright performance and value. However, the Core i9-13900K is a better option than the 7800X3D if you use productivity applications frequently, and ensuring the highest level of performance is paramount. You should also consider that AMD has both the Ryzen 9 7950X3D and Ryzen 97900X3D on offer, both of which offer more cores for the productivity minded while maintaining 13900K-beating levels of gaming performance. Just be prepared to pay for the privilege of faster gaming performance.

When it comes to competing in gaming, the 13900K's biggest foe comes in the form of the Core i7-13700K — the 13900K is only 1.5% faster in gaming but costs 39% more cash (an extra $163). That makes the 13700K the obvious Intel pick for gaming-focused rigs. Additionally, although the 7800X3D beats the 13700K by 14% in gaming, the 13700K is competitive in terms of both raw and system-level fps-per-dollar pricing metrics, though you might have to make a few reasonable tradeoffs.

The Ryzen 7 7800X3D beats the 13700K by 14% in gaming but suffers from much lower performance than the standard Ryzen 7000 models in productivity applications, so you'll need to be aware of the tradeoffs. If you're looking for strong gaming performance paired with the ability to cut through heavier workloads, the Intel Core i7-13700K is a solid all-rounder at $417. For instance, the 13700K is 22% faster in single-threaded and 51% faster in multi-threaded work than the 7800X3D, and you can opt for the graphics-less $390 13700KF to save some extra cash.

AMD's 3D V-Cache tech doesn't accelerate all games evenly, so the lead could vary depending on your gaming selection. In either case, the Ryzen 7 7800X3D is the current gaming champ, and it often isn't a close contest. That's especially true in some game titles, like Microsoft Flight Simulator 2021 and F1 2021. The chip also drops into AM5 motherboards that support the latest connectivity tech, like PCIe 5.0 and DDR5, and AMD plans to support the platform until 2025+. In contrast, Intel’s current round of motherboards will only support one more generation of chips, so upgradeability remains an advantage for AMD.

The 7800X3D also holds the advantage in all power consumption metrics, including peak power and efficiency, ranking among the best we’ve ever tested. That results in more forgiving cooling requirements and a cooler and quieter system. On the other end of the spectrum, overclocking is becoming increasingly less interesting on the high end for most enthusiasts — you'll get a better payoff in the mid-range and low-end of the market. The Intel and AMD processors all provide at least some uplift after overclocking, making that contest a wash.

Given AMD's history of lower-than-MSRP pricing several months after launch, we think the Ryzen 7 7800X3D will eventually become another great chip for value seekers, particularly as platform-level costs continue to recede. If you're after an all-rounder, the Core i7-13700K is the better choice. If you're after the fastest gaming chip on the market, the Ryzen 7 7800X3D offers a lower-cost pathway to the very best that AMD's 3D V-Cache has to offer than you'll pay with the Ryzen 9 X3D chips. If you're willing to accept the tradeoffs, the Ryzen 7 7800X3D is the fastest gaming chip money can buy, earning a leading spot on our list of the best CPUs for gaming.

• MORE: AMD vs Intel
• MORE: Zen 4 Ryzen 7000 All We Know
• MORE: 13th-Gen Intel Raptor Lake All We Know 

System-on-chips based on Arm instruction set architecture now command a 14% market share, but they are poised to at least double their share in several years, reports Counterpoint Research. Analysts note two main advantages that Arm-powered SoCs have over their x86 counterparts for now: lower overall power consumption and customization enabling richer features and higher performance. 

Arm-based SoCs will command a 15% notebook processor revenue share of this year, but their share could increase to 25% in 2027. Meanwhile, Intel's share will decrease from 68% this year to 60% in 2027, whereas AMD's share will drop from 16.7% to 14.4%. 

Nowadays, the vast majority (90%) of Arm-based notebooks come from Apple, limiting their growth as Microsoft's Windows is the world's most popular operating system. However, Counterpoint believes that once Qualcomm develops competitive Arm SoCs compatible with Windows, Arm's market share in the PC world will increase. 

Arm-based SoCs have several benefits over x86 CPUs, such as reduced power consumption and better thermal management. Designed with energy efficiency in mind, these SoCs can be tailored to specific power needs, making them perfect for mobile devices and laptops where battery life is vital. Also, Arm-based SoCs can integrate a larger number of high-performance CPU cores and tightly integrated memory, something that not all x86 CPUs can offer. In addition, custom cores in these SoCs provide advanced features that off-the-shelf x86 processor cores do not offer these days, leading to enhanced hardware and operating system integration, the analysts note. 

Also, Counterpoint mentions the built-in AI capabilities of Apple's SoCs and believes that those capabilities can be used for such workloads as image and video encoding and compression, noise cancellation, image recognition and vocal enhancement. 

Counterpoint Research believes large PC vendors can introduce custom Arm SoCs tailored to their needs and featuring differentiation. For example, lower-power Arm processors can be used to build tablet-notebook hybrids that are lighter and longer lasting compared to today's 2-in-1 and convertibles. 

As more makers with knowledge of Arm SoCs enter the market (e.g., smartphone suppliers), they contribute their know-how in Arm-based hardware and software, further enhancing the appeal of Arm-based PCs. The growing availability of native Arm-based applications will also improve user comfort and familiarity with the platform. The shift towards Arm-based PCs is anticipated to persist, and their market share is projected to rise substantially in the upcoming years. 

What is noteworthy is that for now, Counterpoint Research makes no predictions about Arm adoption by desktops. While Arm SoCs have an indisputable advantage for laptops, desktops have different requirements. For example, they take advantage of high performance, and power consumption is not always crucial for desktops. Therefore, whether Arm will gain a significant share of the desktop space remains to be seen.

Just two weeks after retiring from Intel, Raja Koduri joined the board of directors of Tenstorrent, an AI and high-performance RISC-V CPU company that intends to challenge the blue giant in the coming years, as noticed by @SquashBionic. A renowned developer of GPUs for graphics and compute, Raja Koduri will be joining his former colleagues from AMD, ATI, and Intel.

Tenstorrent is developing datacenter solutions comprising of RISC-V-based AI/ML accelerators as well as high-performance RISC-V general purpose processors. The company was established in 2016 by Ljubisa Bajic, and it is presently led by Jim Keller, a renowned CPU architect who has spearheaded the creation of revolutionary processor architectures at companies such as Apple, AMD, and DEC. (We recently outlined Tenstorrent's short and mid-term plans.)

Truth to be told, a position in the board of directors does not imply that Raja will be able to influence development of CPUs or AI accelerators. Yet, he will have his word when setting Tenstorrent's strategic goals and strategic roadmap, something that is particularly important. Being developer of AI accelerators and high-performance RISC-V CPUs, Tenstorrent is essentially travelling in uncharted watersm since AI-related technologies are changing rapidly, whereas the open-source RISC-V microarchitecture is poised to change fast in general. 

To a large degree, development of AI and RISC-V hardware these days resembles development of GPUs back in the 1990's and 2000's when companies tried a variety of approaches and usually changed microarchitectures radically once in a year or two. That's something that we barely see these days in the fields of x86 and Arm CPUs as well as GPUs from companies like Arm, AMD, Nvidia, and Intel.

Koduri has already worked both with Jim Keller and Ljubisa Bajic while at AMD (with both), Intel (Keller), and Apple (Keller). Most recently Koduri and Keller worked together at Intel developing high-performance compute GPUs and CPUs, respectively. While the two have very different backgrounds and fields of expertise, both have plenty of experience in building high-performance hardware, something that Tenstorrent can certainly take advantage of. 

Having assembled a team of top-notch AI and CPU engineers, start-up Tenstorrent, helmed by industry icon Jim Keller, has huge plans that involve both general-purpose processors and artificial intelligence accelerators.

At present, the company is working on the industry's first 8-wide decoding RISC-V core capable of addressing both client and HPC workloads that will be first used for a 128-core high-performance CPU aimed at data centers. The company also has a roadmap of several more generations of processors, which we'll cover below. 

Why RISC-V?

We recently spoke with Wei-Han Lien, the chief CPU architect at Tenstorrent, about the company's vision and roadmap. Lien has an impressive background, with stints at NexGen, AMD, PA-Semi, Apple, and is perhaps best known for for his work on Apple's A6, A7 (world's first 64-bit Arm SoC), and M1 CPU microarchitectures and implementation. 

With many world-class engineers with vast experience in x86 and Arm designs, one may ask why Tenstorrent decided to develop RISC-V CPUs considering that the data center software stack for this instruction set architecture (ISA) is not as comprehensive as that for x86 and Arm. The answer Tenstorrent gave us is simple: x86 is controlled by AMD and Intel, and Arm is governed by Arm Holding, which limits the pace of innovation. 

"Only two companies in the world can do x86 CPUs," said Wei-Han Lien. "Due to the x86 license restriction, innovation is basically controlled by one or two companies. When companies get really big, they become bureaucratic hierarchically, and the pace of innovation [slows]. […] Arm is kind of the same thing. They claim they are like a RISC-V company, but if you look at their specification, [it] becomes so complicated. It is also actually kind of dominated by one architect. […] Arm kind of dictating all the possible scenario even to architecture [license] partners." 

By contrast, RISC-V is developing quickly. Since it is an open-source ISA, it is easier and faster to innovate with it, particularly when it comes to emerging and rapidly developing AI solutions, according to Tenstorrent. 

"I was looking for a companion processor solution for [Tenstorrent's] AI solution, and then we wanted BF16 data type, and then we went to Arm and said, 'Hey, can you support us?' They said 'no,' it requires like maybe two years internal discussion and discussion with partners and whatever," explained Lien. "But we talked to SiFive; they just put it in there. So, there is no restriction, they built it for us, and it is freedom." 

On the one hand, Arm Holding's approach ensures high quality of the standard as well as a comprehensive software stack, but it also means that the pace of ISA innovation gets slower, which might be a problem for emerging applications like AI processors that are meant to be developed quickly. 

One Microarchitecture, Five CPU IPs in One Year

Since Tenstorrent is looking forward and addressing AI applications at large, it needs not only different system-on-chips or system-in-packages but also various CPU microarchitecture implementations and system-level architectures to hit diverse power and performance goals. This is exactly the department of Wei-Han Lien. 

A humble consumer electronics SoC and a mighty server processor have little in common but can share the same ISA and microarchitecture (albeit implemented differently). This is where Lien's team comes in. Tenstorrent says that the CPU crew has developed an out-of-order RISC-V microarchitecture and implemented it in five different ways to address a variety of applications. 

Tenstorrent now has five different RISC-V CPU core IPs — with two-wide, three-wide, four-wide, six-wide, and eight-wide decoding — to use in its own processors or license to interested parties. For those potential customers who need a very basic CPU, the company can offer small cores with two-wide execution, but for those who need higher performance for edge, client PCs, and high-performance computing, it has six-wide Alastor and eight-wide Ascalon cores.  

Each out-of-order Ascalon (RV64ACDHFMV) core with eight-wide decode has six ALUs, two FPUs, and two 256-bit vector units, making it quite beefy. Considering that modern x86 designs use four-wide (Zen 4) or six-wide (Golden Cove) decoders, we are looking at a very capable core. 

Wei-Han Lien was one of the designers responsible for Apple's 'wide' CPU microarchitecture, which can execute up to eight instructions per clock. For example, Apple's A14 and M1 SoCs feature eight-wide high-performance Firestorm CPU cores, and two years after these were introduced, they are still among the most power-efficient designs in the industry. Lien is probably one of the industry's best specialists in 'wide' CPU microarchitecture, and, as far as we understand, the only processor designer who leads a team of engineers developing an eight-wide RISC-V high-performance CPU core. 

In addition to a variety of RISC-V general-purpose cores, Tenstorrent has its proprietary Tensix cores tailored for neural network inference and training. Each Tensix core comprises of five RISC cores, an array math unit for tensor operations, a SIMD unit for vector operations, 1MB or 2MB of SRAM, and fixed function hardware for accelerating network packet operations and compression/decompression. Tensix cores support a variety of data formats, including BF4, BF8, INT8, FP16, BF16, and even FP64.

Impressive Roadmap

Right now, Tenstorrent has two products: a machine learning processor called Grayskull that offers performance of around 315 INT8 TOPS that plugs into a PCIe Gen4 slot, as well as a networked Wormhole ML processor with approximately 350 INT8 TOPS of performance and uses a GDDR6 memory subsystem, a PCIe Gen4 x16 interface and has a 400GbE connection to other machines. 

Both devices require a host CPU and are available as add-in-boards as well as inside pre-built Tenstorrent servers. One 4U Nebula server containing 32 Wormhole ML cards offers around 12 INT8 POPS of performance at 6kW.

Later this year, the company plans to tape out its first standalone CPU+ML solution — Black Hole — that combines 24 SiFive X280 RISC-V cores and a multitude of 3rd Generation Tensix cores interconnected using two 2D torus networks running in opposite directions for machine learning workloads. The device will offer 1 INT8 POPS of compute throughput (approximately three times performance uplift compared to its predecessor), eight channels of GDDR6 memory, 1200 Gb/s Ethernet connectivity, and PCIe Gen5 lanes.  

In addition, the company is looking forward to adding a 2TB/s die-to-die interface for dual-chip solutions as well as for future use. This chip will be implemented on a 6nm-class fabrication process (we would expect it to be TSMC N6, but Tenstorrent has not confirmed this), yet at 600mm^2, it will be smaller than its predecessors produced on TSMC's 12nm-class node. One thing to remember is that Tenstorrent has not taped out its Blackhole yet, and its final feature set may differ from what the company discloses today. 

Next year the company will release its ultimate product: a multi-chiplet solution called Grendel that features its own Ascalon general-purpose cores featuring its own RISC-V microarchitecture with eight-wider decoding as well as a Tensix-based chiplet for ML workloads.  

Grendel is Tenstorrent's ultimate product set to be released next year: the multi-chiplet solution comprises an Aegis chiplet featuring high-performance Ascalon general-purpose cores and a chiplet or chiplets with Tensix cores for ML workloads. Depending on business requirements (and the financial capabilities of the company), Tenstorrent may implement an AI chiplet using a 3nm-class process technology and therefore take advantage of higher transistor density and Tensix core count, or it can keep using Black Hole chiplet for AI workloads (and even assign some work to 24 SiFive X280 cores, the company says). The chiplets will communicate with each other using the aforementioned 2TB/s interconnect.   

The Aegis chiplet with 128 general-purpose RISC-V eight-wide Ascalon cores organized in four 32-core clusters with inter-cluster coherency will be made using a 3nm-class process technology. In fact, the Aegis CPU chiplet will be among the first to use a 3nm-class fabrication process, something that will probably put the company on the map when it comes to high-performance CPU designs. 

Meanwhile, Grendel will use an LPDDR5 memory subsystem, PCIe, and Ethernet connectivity, so it will offer tangibly higher inference and training performance than existing solutions from the company. Speaking of Tensix cores, it is necessary to note that while all of Tenstorrent's AI cores are called Tensix, these cores actually evolve.

"The [Tensix] changes are evolutionary, but they are definitely there," explained Ljubisa Bajic, the company's founder. "[They add] new data formats, change ratios of FLOPS/SRAM capacity, SRAM bandwidth, network-on-chip bandwidth, new sparsity features, and features in general."

It is interesting to note that different Tenstorrent slides mention different memory subsystems for Black Hole and Grendel products. This is because the company is always looking at the most efficient memory technology and because it licenses DRAM controllers and physical interfaces (PHY). Therefore it has some flexibility when choosing the exact type of memory. In fact, Lien says that Tenstorrent is also developing its own memory controllers for future products, but for 2023 ~ 2024 solutions, it intends to use third-party MCs and PHYs. Meanwhile, for now, Tenstorrent does not plan to use any exotic memory, such as HBM, due to cost concerns. 

Business Model: Selling Solutions and Licensing IP

While Tenstorrent has five different CPU IPs (albeit based on the same microarchitecture), it only has AI/ML products in the pipeline (if fully configured servers are not taken into account) that use either SiFive's X280 or Tenstorrent's eight-wide Ascalon CPU cores. Thus, it is reasonable to ask why it needs so many CPU core implementations.  

The short answer to this question is that Tenstorrent has a unique business model that includes IP licensing (in RTL, hard macro, or even GDS forms), selling chiplets, selling add-in ML accelerator cards or ML solutions featuring CPU and ML chiplets, and selling fully configured servers containing these cards.  

Companies building their own SoCs can license RISC-V cores developed by Tenstorrent, and a broad portfolio of CPU IPs allows the company to compete for solutions requiring different levels of performance and power.  

Server vendors can build their machines withTenstorrent's Grayskull and Wormhole accelerator cards or Blackhole and Grendel ML processors. Meanwhile, those entities that do not want to build hardware can just buy pre-built Tenstorrent servers and deploy them. 

Such a business model looks somewhat controversial since, in many cases, Tenstorrent competes and will compete against its own customers. Yet, at the end of the day, Nvidia offers both add-in cards and pre-built servers based on these boards, and it doesn't look like companies like Dell or HPE are too worried about this because they offer solutions for specific customers, not just building blocks.

Summary

Tenstorrent jumped onto the radar about two years ago with the hire of Jim Keller. In two years, the company recruited a host of top engineers who are developing high-performance RISC-V cores for data center-grade AI/ML solutions as well as systems. Among the development team's achievements is the world's first eight-wide RISC-V general-purpose CPU core, as well as an appropriate system hardware architecture that can be used for AI as well as HPC applications. 

The company has a comprehensive roadmap that includes both high-performance RISC-V-based CPU chiplets as well as advanced AI accelerator chiplets, which promise to enable capable solutions for machine learning. Keeping in mind that AI and HPC are major megatrends poised for explosive growth, offering AI accelerators and high-performance CPU cores seems like a very flexible business model. 

Both AI and HPC markets are highly competitive, so getting some of the world's best engineers onboard is a must when you want to compete against the likes of established rivals (AMD, Intel, Nvidia) and emerging players (Cerebras, Graphcore). Like large chip developers, Tenstorrent has its own general-purpose CPU and AI/ML accelerator hardware, which is a unique advantage. Meanwhile, since the company uses RISC-V ISA, there are markets and workloads that it cannot address for now, at least as far as CPUs are concerned. 

A 'Qualcomm Snapdragon 8cx Next Gen' processor has appeared in Geekbench’s online result browser. The listing was unearthed by Windows on Arm (WoA) developer Gustave Monce. Subsequent raw data digging performed by Kuba Wojciechowski revealed that this 'Next Gen' part is the Snapdragon 8cx Gen 4 (if a genuine result). This Qualcomm processor, codenamed Hamoa, is notable as it is expected to be the product which debuts the Nuvia team's CPU cores for a significant boost in performance.

We've previously discussed the impact Nuvia’s CPU core designs are expected to have on the world of PCs powered by the Arm architecture. To recap: the Nuvia team is headed up by three senior Apple execs who departed the Cupertino company after the first Apple Silicon (M1) projects had been successfully delivered. 

Before they were bought up by Qualcomm, Nuvia published a series of slides touting the astounding performance of their CPU cores. You can look back at previous articles for the chart(s) but to quickly summarize: they showed the expected performance per watt of the Nuvia CPU cores cast a very long shadow on the contemporary x86 competition (AMD, Intel), as well as Apple’s best Arm architecture SoCs at the time. 

In a nutshell, Qualcomm’s release of Nuvia core CPUs for Windows PCs should have a similar disruptive impact as the launch of the original Apple Silicon (M1) — but on the more expansive world of Windows PCs.

Circling back to today’s Geekbench leak, let's take a look at the CPU information section, to start. According to this data, the Snapdragon 8cx Gen 4 is a 12-core SoC with a hybrid 8+4 core configuration. Previous leaks indicate, more specifically, that there are eight performance cores and four efficiency cores. 

The data shows the SoC’s base frequency was 2.38 GHz. Previously, we heard the P-cores could boost to around 3.4 GHz, with the E-cores somewhat slower at around 2.5 GHz. An 'Adreno' GPU is mentioned in the system information, but it does not otherwise improve on earlier reports, which pointed to the presence of an integrated Adreno 740 GPU.

These Geekbench results — assuming they're genuine — are, unfortunately, nothing to get excited about. Monce reckons this is partly due to the chip under test being an "early engineering sample." Other reasons for the clock/performance shortfalls could be the influences of early firmware and driver software. The Qualcomm Snapdragon 8cx Gen 3 (3.0 GHz octa-core) easily beats this Gen 4 chip in both the 1T and nT Geekbench tests.

While it's great to see indicators that suggest Snapdragon 8cx Gen 4 chips are 'in the wild,' it's hard to believe the hype. We hope things will change (and that leaks will be more revealing and impressive) in upcoming months. Remember, there's still plenty of time for optimization, as Qualcomm and partners aren't expected to launch products until the end of 2023 or in early 2024.

Intel and the Gordon and Betty Moore Foundation announced that Intel co-founder Gordon Moore died peacefully this evening, surrounded by his family in Hawaii, at the age of 94. Moore co-founded Intel with his longtime friend Robert Noyce in 1968, but is perhaps best known as the creator of the eponymous Moore's Law. Intel has posted his obituary here.

Moore's Law, originally penned in 1965, stated that the number of transistors on an integrated circuit would double every year. This law served as a goal that helped drive the semiconductor industry forward. In 1975, Moore revised this law to predict a doubling of transistors every two years, which, despite plenty of debate, has largely held true to this day.

After being part of the founding group of Fairchild Semiconductor, Moore and Noyce founded Intel in July 1968. Their first hire was Andy Grove, and the trio steered Intel to success.

Intel originally began with $2.5 million from investors and produced Static RAM (SRAM) memory. The company later shifted to processors, shipping the Intel 4004, the first commercially available microprocessor, in 1971. Intel later pioneered the x86 instruction set in 1978 when it released the now-legendary Intel 8086, the first building block of its empire based on x86 architectures. The company still focuses on CPUs today and is now valued at ~$167 billion. 

Moore initially served as the executive vice president until 1975, then as president of the company until he was named the CEO and chairman of the board in 1979. He relinquished the role of CEO in 1987. Moore continued to serve as the chairman until 1997, when he was appointed the chairman emeritus. He held that position until he stepped down in 2006.

In his later years, Gordon Moore was an active philanthropist, especially in the fields of environmental conservation, science, and patient care improvements. Moore and his wife of 75 years established the Gordon and Betty Moore Foundation, which has donated more than $5.1 billion to charity since its inception in 2000. 

Pat Gelsinger, Intel CEO, said, “Gordon Moore defined the technology industry through his insight and vision. He was instrumental in revealing the power of transistors, and inspired technologists and entrepreneurs across the decades. We at Intel remain inspired by Moore’s Law and intend to pursue it until the periodic table is exhausted. Gordon’s vision lives on as our true north as we use the power of technology to improve the lives of every person on Earth. My career and much of my life took shape within the possibilities fueled by Gordon’s leadership at the helm of Intel, and I am humbled by the honor and responsibility to carry his legacy forward.” 

Moore was a recipient of the National Medal of Technology in 1990 and the Presidential Medal of Freedom, the nation’s highest civilian honor, in 2002. He is survived by his wife of 75 years, Betty Moore, sons Kenneth and Steven, and four grandchildren.  

Nvidia teased its forthcoming Arm-based Grace CPU at GTC 2023, but the company's announcement that systems will now ship in the second half of this year represents a delay from its original launch timeline that targeted the first half of 2023. We asked Nvidia CEO Jensen Huang about the delay during a press question and answer session today, which we'll cover below. Nvidia also showed its Grace silicon for the first time and made plenty of new performance claims during its GTC keynote, including that its Arm-based Grace chips are up to 1.3X faster than x86 competitors at 60% of the power, which we'll also cover.

I asked Jensen Huang about the delay in delivering the Grace CPU and Grace Hopper Superchip systems to the end market. After he playfully pushed back about the expected release date (it was undoubtedly 1H23, now 2H23), he responded:

"Well, first, I can tell you that Grace and Grace Hopper are both in production, and silicon is flying through the fab now. Systems are being made, and we made a lot of announcements. The world's OEMs and computer makers are building them." Huang also remarked that Nvidia has only been working on the chips for two years, which is a relatively short time given the typical multi-year design cycle for a modern chip.

Today's definition of shipping systems can be fuzzy — the first systems from AMD and Intel often ship to hyperscalers for deployment long before the chips see general off-the-shelf availability. However, while Nvidia says it is sampling chips to customers, it hasn't said Grace is being deployed into production yet. As such, the chips are late according to the company's projections, but to be fair, perennially late chip launches from companies like Intel aren't uncommon. That highlights the difficulty of launching a new chip, even when building around the dominant x86 chips with established hardware and software platforms built upon for decades.

In contrast, Nvidia's Grace and Grace+Hopper chips are a ground-up rethinking of many of the fundamental aspects of chip design with an innovative new chip-to-chip interconnect. Nvidia's use of the Arm instruction set also means there's a heavier lift for software optimizations and porting, and the company has an entirely new platform to build.

Jensen alluded to some of that in his extended response, saying, "We started with Superchips instead of chiplets because the things we want to build are so big, And both of these are in production today. So customers are being sampled, the software is being ported to it, and we're doing a lot of testing. During the keynote, I showed a few numbers, and I didn't want to burden the keynote with a lot of numbers, but a whole bunch of numbers will be available for people to enjoy. But the performance was really quite terrific."

And Nvidia's claims are impressive. For example, in the above album, you can see the Grace Hopper chip that Nvidia showed in the flesh for the first time at GTC (more technical details here).

During the presentation, Huang claimed the chips are 1.2X faster than the 'average' next-gen x86 server chip in an HiBench Apache Spark memory-intensive benchmark and 1.3X faster in a Google microservices communication benchmark, all while drawing only 60% of the power.

Nvidia claims this allows data centers to deploy 1.7X more Grace servers into power-limited installments, with each providing 25% higher throughput. The company also claims Grace is 1.9X faster in computational fluid dynamics (CFD) workloads. 

However, while the Grace chips are ultra-performant and efficient in some workloads, Nvidia isn't aiming them at the general-purpose server market. Instead, the company has tailored the chips for specific use cases, like AI and cloud workloads that favor superior single-threaded and memory processing performance in tandem with excellent power efficiency.

"[..]almost every single data center is now powered limited, and we designed Grace to be extraordinarily performant in a power-limited environment," Huang told us in response to our questions. "And in that case, you have to be both really high in performance, and you have to be really low in power, and just incredibly efficient. And so, the Grace system is about two times more power/performance efficient compared to the best of the latest generation CPUs."

"And it's designed for different design points, so that's very understandable," Huang continued. "For example, what I just described doesn't matter to most enterprises. It matters a lot to cloud service providers, and it matters a lot to data centers that are powered unlimited."

Energy efficiency is becoming more of a concern than ever, with chips like the AMD EPYC Genoa we recently reviewed and Intel's Sapphire Rapids now pulling up to 400 and 350 watts, respectively. That requires exotic new air cooling solutions to contain the prodigious power draw at standard settings and liquid cooling for the highest-performance options.

In contrast, Grace's lower power draw will make the chips more forgiving to cool. As revealed at GTC for the first time, Nvidia's 144-core Grace package is 5" x 8" and can fit into passively-cooled modules that are surprisingly compact. These modules still rely upon air cooling, but two can be air-cooled in a single slim 1U chassis. 

Nvidia also showed its Grace Hopper Superchip silicon for the first time at GTC. The Superchip combines the Grace CPU with a Hopper GPU on the same package. As you can see in the album above, two of these modules can also fit into a single server chassis. You can read the deep-dive details about this design here.

The big takeaway with this design is that the enhanced CPU+GPU memory coherency, fed by a fat low-latency chip-to-chip connection that's seven times the speed of the PCIe interface, allows the CPU and GPU to share information held in memory at a speed and efficiency that's impossible with previous designs.

Huang explained that this approach is ideal for AI, databases, recommender systems, and large language models (LLM), all of which are in incredible demand. By allowing the GPU to access the CPU's memory directly, data transfers are streamlined to boost performance.

Nvidia's Grace chips may be running a bit behind schedule, but the company has a bevy of partners, with Asus, Atos, Gigabyte, HPE, Supermicro, QCT, Wiston, and Zt all preparing OEM systems for the market. Those systems are now expected in the second half of the year, but Nvidia hasn't said whether or not they will come towards the beginning or end of the second half.