The retro community seems sharply divided over a decision by Commodore to lock down firmware access in its C64 Ultimate computer. In a post entitled Why We're Protecting Your Commodore 64 Ultimate FPGA, the iconic retro computing company telegraphs a change “preventing firmware not released by Commodore from being loaded onto the hardware.” This blog post seems to be designed to calm social media and forum commenters' concerns about the C64U breaking the spirit of the vibrant C64 scene, implementing an anti-tinkering change, and making a move that is really about IP control.

The latest firmware, version 1.1.0, was released a few days ago. You can see the announcement in the X post above, and can expand the embed to read some of the comments it has received.

Social media and forum posters started to grumble about the impending firmware locks soon after some users highlighted a section concerning firmware tinkering in the official 1.1.0 release notes. There are some very welcome changes in 1.1.0, including a new Left Arrow key implementation, USB mouse support for “most mice,” and enhancements for LED lighting interactivity – as well as a long list of fixes.

However, some enthusiasts found it. problematic that Commodore wrote that “A future update may introduce safeguards to help make sure incompatible firmware not released by Commodore does not damage your motherboard.”

Commodore and its supporters characterize the change as one safeguarding user hardware while official firmware updates still flow.

“The Commodore 64 Ultimate is not a static product," the blog post reads. There will be new hardware revisions, new components, and new capabilities! This is foundational to our roadmap and, frankly, core to the Commodore 64 Ultimate's value proposition." But then it warns about firmware built for different boards, causing issues on the C64U.

Probably most concerning for Commodore is that it says it has already seen non-functioning casualties of third-party firmware updates requiring support. This isn’t just hypothetical, insists Commodore. No company would find it sustainable to service “hardware returns and replacements due to actions entirely out of our control.”

It also uses the blog to make clear it isn’t intending to stomp on system patches, like the popular SPIFFY. The distinction is that SPIFFY is a community patch for the C64U - improving quality-of-life in many ways, and is not a replacement firmware. “This policy is not aimed at that kind of community-driven ingenuity,” Commodore makes clear.

A so-called ‘walled garden’ is also not on the menu, according to the new Commodore blog. In this section it warns that the C64U is different from the closely related Ultimate64 from the same designer/developer. Specifically, it says that “Our hardware roadmap for the Commodore 64 Ultimate includes board revisions and component changes that Gideon's Ultimate64 firmware has no reason to address, since it's built for his product, not ours.” In other words, as the systems diverge and develop, more and more problems could arise from swapping firmware files.

We don’t know if the Commodore Blog post is going to satisfy the sternest critics and worry worms on the great WWW. Surely it helps explain the thinking behind the upcoming restrictions, but folks will always hate restrictions and others deciding what is best for them.

On Friday, AMD and Raytheon said they would work together on a multi-chip package technology that could enable multi-chip solutions based on AMD devices and others. The technology is developed under a $20 million contract through the Spectrum Missions Advanced Resilient Trusted Systems (S2MARTS) consortium, and the multi-chip packages are projected to be used for processing data from 'ground, maritime and airborne sensors,' though this is an ambiguous description.

Based on the vague press release, the multi-chip package (or rather multi-chip packages, given three distinct natures of target applications, but we are speculating here) will convert RF data to a digital form and process it. These MCPs will rely on the latest industry-standard die-level interconnect technology, incorporate multiple chiplets to get new system capabilities, and cost-effectively enable high performance.

The multi-chip package is set to rely on a Raytheon-designed interposer and is manufactured in Lompoc, California. Meanwhile, given AMD's expertise in 2.5D and 3D direct bonding technologies, we imagine the Red Team will input this know-how into the project.

Yet, it should be noted that Raytheon has partnered with Xilinx (now part of AMD) for its field-programmable gate array (FPGA) solutions for decades, so it is not cast in stone that processing will be made on Zen-series general-purpose x86 cores and Ryzen CCDs and not on Xilinx FPGAs. FPGAs are highly suitable for RF signal processing tasks due to their inherent flexibility and parallel processing capability.

Being a military initiative, the nature of the technology set to be developed by Raytheon and AMD is not to be disclosed entirely for quite some time, which is why its description is indistinct to put it mildly.

"By teaming with commercial industry, we can incorporate cutting-edge technology into Department of Defense applications on a much faster timescale," said Colin Whelan, president of Advanced Technology at Raytheon. "Together, we will deliver the first multi-chip package that features the latest in interconnect ability – which will provide new system capabilities to our warfighters."

The press release by Raytheon, a part of RTX, a military contractor, is meant to show U.S. taxpayers how and where their hard-earned money is spent rather than give an idea to the enemy about the devices that AMD and Raytheon can produce. To that end, do not expect it to contain detailed information on how U.S. military devices are set to use technologies jointly developed by AMD and Raytheon.

AMD on Tuesday reported its financial results for the fourth quarter of 2023 as well as for the whole fiscal year. The company posted a strong end to the year as its data center GPU business rose on increased demand from the AI craze. However, the company posted a lower-than-expected guide, so its shares tumbled 6% in after-hours trading. An interesting wrinkle here is that the company expects to sell $3.5 billion worth of its MI300-series AI GPUs in 2024, though it says it will not be supply-constrained, which essentially means that demand for the new GPU is at least somewhat limited as it ramps — Nvidia, whose AI GPUs are in a constant state of shortage due to incredible demand, is said to have a 52-week wait time for a new GPU, so many would have expected AMD to have a similar problem. Regardless, AMD says that its MI300 sales are on pace to be the fastest revenue ramp of any product in the company's history as it increased its projected sales from a previous $2 billion estimate to $3.5 billion. 

AMD also posted a lower-than-expected guide, saying that its Q1 revenues from the client, embedded, and gaming segments are projected to decrease sequentially. In particular, revenue from semi-custom products (which mainly consist of console SoCs) is expected to experience a substantial double-digit percentage drop of "more than 30%." The company also forecasts that its data center segment revenue will remain flat in Q1 compared to the previous quarter, as a seasonal dip in server sales is likely to be offset by a robust increase in AI and HPC GPU sales, namely the Instinct MI300-series products.

Despite the looming challenges in the year ahead, AMD posted mixed results for the final quarter of its fiscal 2023. The company's fourth-quarter sales were $6.168 billion, up 10% year-over-year, whereas AMD's 2023 revenue totaled $22.68 billion, which is down 4% year-over-year. The company's results for the year were a mixed bag as, on the one hand, it increased its sales of data center processors, but on the other hand, shipments of its client platforms declined, just like sales of gaming hardware.

AMD received $6.168 billion in revenue for the fourth quarter of 2023 and earned $667 million in net income, up significantly year-over-year and quarter-over-quarter. The company's gross margin increased to 47%.  

When it comes to the full year, AMD earned $22.68 billion, down 4% year-over-year, whereas its net income totaled $854 million, down 35% year-over-year, which is the second consecutive yearly net income decline for the company. Meanwhile, AMD's gross margin increased by 1% to 46% year-over-year.  

"We finished 2023 strong, with sequential and year-over-year revenue and earnings growth driven by record quarterly AMD Instinct GPU and EPYC CPU sales and higher AMD Ryzen processor sales," said Lisa Su, chief executive of AMD. "Demand for our high-performance data center product portfolio continues to accelerate, positioning us well to deliver strong annual growth in what is an incredibly exciting time as AI re-shapes virtually every part of the computing market."

Datacenter Platforms Set Records

AMD's data center business unit earned $2.282 billion in revenue for Q4 2023 (up 38% YoY) and reported an operating income of $666 million (up 50% YoY) as the company ramped up shipments of its 4th Generation EPYC processors for datacenters and increased shipments of its Instinct GPUs for artificial intelligence and high-performance computing applications. This is the best quarterly result for AMD's data center business ever, though we should keep in mind that the results of this unit now include sales of Pensando hardware.

As for the whole year, net revenue of AMD's datacenter division amounted to $6.496 billion, up 7% year-over-year, which makes it the best year for the business unit. Meanwhile, operating income of the division dropped 31% to $1.267 billion.  

Sales of Client Platforms Decline, But Show Signs of Recovery

As far as sales of client platforms are concerned, AMD's client business unit managed to increase its Q4 2024 revenue to $1.461 billion, up 62% year-over-year as the PC market began to recover and AMD succeeded in boosting sales of its latest Zen 4-based Ryzen 7000-series processors. The business unit earned $55 million in operating income, up 136% YoY.

Yet, since the PC market was weak last year, AMD's client PC business declined by 25% year-over-year to $4.651 billion and lost $46 million for the whole year, which is a dramatic change as the firm's client PC unit's operating income was $1.19 billion in 2022.

Gaming Segment

AMD's gaming hardware business unit — which sells discrete Radeon graphics processors and system-on-chips for consoles — earned $1.368 billion for the fourth quarter of 2023, down 17% compared to the same quarter a year ago, while its operating income decreased to $224 million. AMD says that while sales of its Radeon GPUs increased, sales of game console SoCs declined.

For the whole year, the gaming hardware business unit earned $6.212 billion in revenue, down 9% YoY, and its operating income reached $971 million, up 2% YoY. Again, AMD attributes these revenue declines to decreasing sales of its processors for consoles, which have to get cheaper every year. Nonetheless, AMD's gaming unit was the company's second-largest business.

Embedded Segment

The revenue from AMD's embedded segment amounted to $1.057 billion, a decrease of 24% from the previous year, largely as a result of customers decreasing their stock levels, according to AMD. Yet, the unit remained profitable and earned $461 million in net income. 

As for the whole year, the revenue for the embedded segment reached $5.3 billion, marking a 17% increase from the previous year. This growth was chiefly attributed to the full-year contribution of revenues from the acquisition of Xilinx, which was finalized in February 2022. For the whole year the unit earned $2.628 billion, up 17% from the previous year.

Outlook

While, in general, AMD's results look good and predictable, the company provided a rather conservative outlook for the first quarter.

AMD anticipates its revenue for the first quarter of 2024 to be around $5.4 billion ± $300 million. The company forecasts that its data center segment revenue will remain flat compared to the previous quarter, as a seasonal dip in server sales is likely to be offset by a robust increase in AI and HPC GPU sales, namely the Instinct MI300-series products.

Meanwhile, revenues from the client, embedded, and gaming segments are projected to decrease sequentially. In particular, revenue from semi-custom products (which mainly consist of console SoCs) is expected to experience a substantial double-digit percentage drop. 

Intel showcased its Gaudi3 Processor for artificial intelligence (AI) workloads alongside the formal introduction of its 14th-Gen Core Ultra "Meteor Lake" processors and 5th-Gen Xeon Scalable CPUs for datacenters. The accelerator is set to arrive in 2024 and will offer a significant performance bump compared to its predecessor, the Gaudi2.

Intel CEO Gelsinger discussed the upcoming release of Intel's Gaudi3, which is scheduled for next year, and showed off the new AI accelerator aimed at deep learning and large-scale generative AI models for the first time. The new unit looks like a huge module (OAM, we presume) with a massive ASIC and multiple HBM3 (or HBM3E) memory stacks on it. The ASIC package looks significantly larger than the Gaudi2, so we presume that it is equipped with eight HBM3E stacks on it (rather then with six in case of Gaudi2). Based on a slide Intel presented at SC23, Gaudi3 is not a monolithic processor, but rather a dual-chiplet design that fuses together two processors.

In addition, Intel announced that the Gaudi3 will offer four times higher BF16 performance, two times faster networking performance, and 1.5X higher bandwidth compared to Gaudi2.

"Our Gaudi roadmap remains on track with Gaudi3 out of the fab, now in packaging and expected to launch next year," said Pat Gelsinger, CEO of Intel, at the company's latest conference call (via SeekingAlpha). In 2025, Falcon Shores brings our GPU and Gaudi capabilities into a single product."

Intel's Habana Gaudi2 already is a quite promising product, with 24 fully programmable Tensor Processor Cores (TPCs) and 96GB of HBM2E memory, capable of challenging Nvidia's H100 GPU for AI and HPC. Intel's Habana Gaudi3, which is expected to hit the market in 2024, will offer significantly improved performance ove its predecessor, Gelsinger said earlier this year.

Intel said the Gaudi line-up has experienced significant growth — attributed to its proven performance and competitive TCO, as well as its reasonable pricing. The company is confident that the rising demand for generative AI hardware will position Intel to secure a more substantial share of the accelerator market in 2024, primarily through its range of AI accelerators, which will be spearheaded by Gaudi.

"We are pleased with the customer momentum we are seeing from our accelerator portfolio and Gaudi in particular, and we have nearly doubled our pipeline over the last 90 days," said Gelsinger on the call.

Unlike its rival Nvidia, AMD has yet  to significantly capitalize on the rise of artificial intelligence (AI), but it looks like the company is going to win some major contracts with leading providers of cloud services. As it turns out, Oracle is set to use AMD's Instinct MI300X AI and HPC GPUs for its cloud services, reports MT Newswires (via Reddit) citing UBS. And IBM is expected to use AMD's Xilinx FPGA solutions for AI workloads, reports Medium (via Ming-Chi Kuo).

Oracle's cloud infrastructure is grappling with Nvidia GPU shortages, affecting its short-term growth, as noted by UBS. Despite this, Oracle remains optimistic, planning a considerable expansion of H100 GPU deployment by 2024 and deploying AMD's Instinct MI300X processors as an alternative solution.

Oracle has reportedly decided not to sidestep AI and HPC GPU shortages by starting its own silicon program, as it would take years to come to fruition. Instead, the company is setting its sights on AMD’s Instinct MI300X offerings, which promise to offer formidable performance level. The cloud giant is expected to be among the first to deploy these processors in early 2024.

Oracle is not the only company that is looking beyond Nvidia for its AI endeavors. IBM's new AI inference platform uses the NR1 AI chip from NeuReality, which is fabricated using TSMC's 7nm-class process technology. AMD plays an important role by providing the FPGA components essential for NeuReality’s AI solution, a contribution that is set to benefit owing to the robust infrastructure established by IBM's AI inference platform. Foxconn is expected to start production of AI servers employing this technology in the fourth quarter of 2023.

AMD chief executive Lisa Su along with two others notified Cisco that they would like to leave its board of directors. Su joined Cisco's board of directors in early 2020, but a lot of things have happened since then, which might be a reason why she decided to quit.

"On October 4, 2023, M. Michele Burns, Roderick C. McGeary, and Dr. Lisa T. Su each notified Cisco Systems, Inc. of their respective decision not to stand for re-election at Cisco's 2023 annual meeting of stockholders," Cisco's filing with SEC reads (via Don Clark). "Ms. Burns, Mr. McGeary, and Dr. Su will continue to serve as directors until the 2023 Annual Meeting." 

Neither disclosed reasons for leaving Cisco's board of directors, and we can only speculate about their intentions.

Of the three, Lisa Su is the only leader of high-technology company. When she joined Cisco's board-of-directors in 2020, AMD was beginning to recapture market share in PCs and servers from Intel and expansion prospects were rather foggy. But over the course of three years AMD got a lot bigger. 

The company purchased Pensando for $1.9 billion in 2022 to gain data processing units (DPUs), a type of processors that enables connectivity in datacenters and somewhat competes against Cisco's datacenter connectivity products. In addition, AMD recently launched its EPYC 8000-series 'Siena' processors that target communications and edge servers, thus also encroaching on Cisco's offerings. As a result, Su has at least two conflicts of interests when it comes to Cisco. 

Keep in mind that at this point that this is mere speculation. Before joining Cisco's board, Su left the board of Analog Devices.

AMD said Friday that over the next five years it will invest approximately $400 million in India. The company will greatly expand its presence in the country and will open its largest design center in Bengaloru, India, already in 2023, reports Reuters. The significant investment indicates that the company may be planning to expand its product line-up in the coming years. 

AMD's new design center campus is set to open in Bengaluru, the country's tech hub, later this year and is expected to create 3,000 engineering jobs within the next five years. The company disclosed that its India teams will be crucial in developing its high-performance (CPUs and GPUs) and adaptive (FPGAs) solutions for its global clientele. With this new investment, AMD will expand its office presence in India to a total of 10 locations, where it already employs over 6,500 people.

In total, AMD is committed to invest $400 million in India by 2028, Mark Papermaster, AMD's chief technology officer, revealed at a semiconductor conference in Gujarat.

India is well known for its talented software developers, but in the last couple of decades it became a major hub for chip designs. Companies like AMD develop plenty of chips in the country and there are also numerous contract chip designers in India.

Establishing its largest design center and investing $400 million in India over the next five years indicates that AMD not only intends to capitalize on the country's potential as a chip design hub, but it plans to significantly expand its chip design design prowess going forward. Meanwhile, for now it is impossible to make guesses how exactly it intends to spend the money.

AMD's move aligns with the Indian government's efforts to attract investment in the semiconductor sector and solidify its position in the global chip industry.

It is noteworthy that last month AMD also announced plans to invest $135 million in its Xilinx FPGA operations in Ireland over the next four years.

AMD transferred production of virtually all of its substantial products to TSMC in the recent years and it is in no hurry to add other manufacturing partners, according to Lisa Su, chief executive officer of AMD, who recently visited Taiwan. Despite rumors that AMD might shift some of its orders to Samsung Foundry, the head of AMD does not seem to be ready to make any concrete comments on the matter. 

"We typically do not publicly comment on the details of specific products and orders, but TSMC is an important partner," Su said in Taiwan when asked to comment on the rumors about working with Samsung Foundry, according to a report by BusinessKorea. "The Instinct MI300, a world-class accelerator for generative AI that AMD plans to launch later this year, has high complexity."

While on paper fabless chip designers can work with multiple foundries, this is usually associated with a lot of curbs. First, modern process technologies from different contract chipmakers are not compatible — a 5nm-class chip made by TSMC has to be redesigned to be produced at Samsung Foundry. When we are dealing with complex products such as AMD's Instinct MI300, such redesign costs may get too high. 

Second, fabless chip developers tend to reuse IP across different product lines to save costs and usage of multiple foundries limits this capability or at least reduces its appeal. Third, when you buy in huge volumes from one supplier, you are going to enjoy various benefits, including tailoring of process technologies for your needs and discounts. Of course, some companies — such as Qualcomm — can afford to work with multiple contract chipmakers and benefit from it, but this is because they buy huge amount of chips from everyone. 

"We will continue to work with our Taiwanese partners because we cannot launch this product without good partners like TSMC," Su added, underlining the importance of TSMC for the company. 

However, Su sounded less resolute when she visited Tokyo later that week. 

The head of AMD told Nikkei that the company would "consider other manufacturing capabilities" besides TSMC to produce its chips to "ensure that we have the most resilient supply chain."

Meanwhile, TSMC is building up fabs in the U.S. and Japan, which will inevitably ensure resiliency of its supply chain. Therefore, AMD may not be exactly inclined to work with TSMC's rivals like Samsung Foundry and Intel Foundry Services for the reasons mentioned above. At least for now.

Intel recently held a launch event for a new China-specific version of its Gaudi 2 processor for artificial intelligence (AI) workloads, and this new version is compliant with the latest US sanctions on China-bound AI and GPU processors. The new version cuts down the card's scale-up interconnect, which somewhat limits the performance of AI servers on its base. To some degree, this will help with the massive shortages of AI GPUs, such as Nvidia's A100 and H100, as well as the cut-down A800 versions that Nvidia has also created specifically for its China customers.

Intel has two Gaudi 2 versions that come in the 600W OCP model form factor. The regular version, called Gaudi 2 HL-225H, has 24 x 100Gbps RDMA (RoCE v2) interconnects, whereas the cut-down version of the product, called Gaudi 2 HL-225B, has 21 x 100Gbps RDMA (RoCE v2) scale-up interconnects. The reduced number of scale-up interfaces reduces the performance of AI servers based on this version of Gaudi 2, but keeping in mind the huge demand for AI capabilities in China, it will likely be popular regardless.

Intel describes the Gaudi 2 accelerator's version for the Chinese market as 'The HLB-225B Processor complies with US BIS regulations' for supercomputers and semiconductors.

While the availability of Intel's Habana Gaudi 2 in China will ease shortages of GPU servers, it should be noted that it isn't likely that applications that already use Nvidia's GPUs will be ported to Gaudi. Therefore, Nvidia's existing customers will have to keep buying from Nvidia.

"On July 11, Intel held an event for customers, partners and local media in the China market," an Intel spokesperson said. "As part of the event program, Intel provided attendees with updates on our AI strategy, our unique portfolio of AI products and announced the availability of Gaudi2 for customers in China. The availability of Gaudi2 in China continues Intel’s nearly 40-year history of delivering innovative yet legally-compliant products to this key growth market."

According to DigiTimes, TSMC is accelerating orders with backend equipment suppliers as it starts an expansion plan for its chip-on-wafer-on-substrate (CoWoS) packaging capacity. The shortage of compute GPUs for artificial intelligence and high-performance computing, which Nvidia largely dominates, is mainly attributed to TSMC's limited CoWoS packaging production capabilities.

Reports suggest that TSMC has plans to increase its current CoWoS capacity from 8,000 wafers per month to 11,000 wafers per month by the end of 2023 and then to around 14,500 – 16,600 wafers by the end of 2024. Previously Nvidia was rumored to increase its CoWoS capacity to 20,000 wafers per month by the end of 2024. Remember that the information comes from unofficial sources and may be inaccurate. 

Major tech giants like Nvidia, Amazon, Broadcom, Cisco, and Xilinx have all boosted their demand for TSMC's advanced CoWoS packaging and are consuming every wafer they can get. As a result, TSMC has been forced to renew orders for necessary equipment and materials, according to DigiTimes. The production of AI servers has significantly risen, fueling the already intense demand for these advanced packaging services. 

Nvidia has already booked 40% of TSMC's available CoWoS capacity for the coming year. However, due to the severe shortage, Nvidia has started exploring options with its secondary supplier, placing orders with Amkor Technology and United Microelectronics (UMC), although these orders are relatively small, the report claims. 

To cater to its increased CoWoS packaging needs, TSMC is partnering with multiple suppliers from around the globe, including U.S.-based Rudolph Technologies, Japan's Disco, and Germany's SUSS MicroTec, along with Taiwanese experts Grand Process Technology (GPTC) and Scientech. The suppliers are under pressure to provide almost 30 sets of relevant tools by mid-2024. 

TSMC has also begun implementing strategic changes, such as redistributing some of its InFO production capacity from its northern Taiwan site in Longtan to its Southern Taiwan Science Park (STSP). It's also fast-tracking the expansion of the Longtan site. Furthermore, TSMC is increasing its in-house CoWoS production while outsourcing part of its oS manufacturing to other assembly and test (OSAT) companies. For instance, Siliconware Precision Industries (SPIL) has been one of the beneficiaries of this outsourcing initiative.

TSMC opened its Advanced Backend Fab 6 facility last week. It is set to expand its advanced packaging capacity for its frontend 3D stacking SoIC (CoW, WoW) technologies and backend 3D packaging methods (InFO, CoWoS). For now, the fab is ready for SoIC. The Advanced Backend Fab 6 can process around one million 300-mm wafers per year and carry out over 10 million hours of testing annually, with cleanroom space that is larger than the combined cleanroom spaces of all other TSMC advanced packaging facilities.

Among the most impressive features of the Advanced Backend Fab 6 is the extensive five-in-one intelligent automated material handling system. The system controls the production flow and detects defects instantly, increasing yield. This is crucial for complex multi-chiplet assemblies like AMD's MI300, as packaging defects immediately render all chiplets unusable, leading to significant losses. With data processing capabilities 500 times faster than average, the facility can maintain comprehensive production records and track every die it processes.

Nvidia uses CoWoS for its highly successful A100, A30, A800, H100, and H800 compute GPUs. AMD's Instinct MI100, Instinct MI200/MI200/MI250X, and the upcoming Instinct MI300 also use CoWoS.

Just like other large chip designers, AMD has already started to use artificial intelligence for designing chips. In fact, Lisa Su, chief executive of AMD, believes that eventually AI-enabled tools will dominate chip design as the complexity of modern processors is increasing exponentially.

Su, believes that AI will dominate certain areas of chip design, she said at the 2023 World Artificial Intelligence Conference (WAIC) held in Shanghai. She also emphasized the need for interdisciplinary collaboration to enable better hardware design in the future, reports DigiTimes.

Previously both Jensen Huang, chief executive of Nvidia, and Mark Papermaster, chief technology officer of AMD, noted that chip development is an ideal application for AI. AMD is already utilizing AI in semiconductor design, testing, and verification. The company also has the intention to leverage generative AI more broadly in future chip design applications.

At AMD, AI is already in chip design, particularly in the 'place and route' stage, where sub-blocks of chip designs are positioned and optimized for better performance and lower energy consumption, Papermaster told Tom's Hardware in May. AI's ability to continuously iterate and learn from patterns greatly accelerates the process of achieving an optimized layout, thereby increasing performance and energy efficiency. Papermaster said that AI will even expand into more important aspects of chip design, like microarchitecture designs, particularly after certain hurdles are overcome to protect IP.

AI is also already employed in verification suites to reduce the time needed to detect bugs during the chip's development process, from conception to the verification and validation phases. Furthermore, AI assists in generating test patterns. With billions of transistors in a chip design, ensuring comprehensive test coverage is essential to guarantee that the product is flawless upon leaving the manufacturing floor. AI's ability to learn from each successive run, identify gaps in test coverage, and adjust the testing focus accordingly, significantly speeds up the process and enhances test coverage.

Artificial intelligence (AI) is increasingly playing a pivotal role in supporting and assisting chip designs. All three leading makers of electronic design automation (EDA) tools — Ansys, Cadence, and Synopsys — offer AI-enabled software to their clients, although Synopsys seems to be a bit ahead of its competitors when it comes to AI-enabled tools.

Earlier this year Synopsys launched Synopsys.ai, the first end-to-end AI-driven EDA solution. This enables developers to use AI throughout all stages of chip development, from architecture to design and manufacturing.

AMD on Wednesday said it would invest up to $135 million over the course of four years in its Xilinx operations in Ireland. The funding will be used to expand R&D and engineering operations at its sites in Dublin and Cork.

AMD indicated that the up to $135 million will be used to fund a number of strategic research and development initiatives, which will involve hiring 290 highly skilled employees in R&D and engineering along with various supporting workers. AMD's description of the strategic R&D initiatives in Ireland is rather vague as the company mentions next generation AI, datacenter, networking, and 6G communications infrastructure, which essentially means the majority of markets served by Xilinx’s adaptive computing solutions, such as FPGAs.

Meanwhile, AMD mentions that a number of Xilinx products, such as the AMD Zynq UltraScale+ RFSoC products, were developed in Europe.

"By further investing and expanding our presence in Dublin and Cork, we are committed to continuing to both drive innovation in Ireland and to support the European semiconductor ecosystem," said Ruth Cotter, senior vice president, Marketing, Communications and Human Resources at AMD. "Through this investment, our R&D teams in Ireland will design innovative high-performance and adaptive computing engines to accelerate data centre, networking, 6G communications and embedded solutions while taking a leadership position on artificial intelligence." 

The AMD Xilinx facility in Ireland was originally launched in 1994 and it was the company's first dedicated location outside the United States. The site is dedicated to production, engineering, operational aid, and management. After AMD purchased Xilinx in 2022, the campus in Ireland became one of the company's largest research and development centers in Europe.

"From the cloud and PCs to communications and intelligent end points, AMD’s high-performance and adaptive computing solutions play an increasingly larger role in shaping the future of computing today," added Cotter. "For nearly three decades, Ireland has been a flagship European R&D centre developing adaptive computing solutions, drawing from a strong and highly-skilled workforce." 

Intel today announced its Agilex 7 M-Series family of Field-Programmable Gate Array (FPGA) products, doubling down on its revitalized 10nm SuperFin fabrication node. Citing rising market demand for FPGA solutions as co-processors for Networking, Data Center, High Performance Computing (HPC)Cloud computing and other applications, Intel promises increased flexibility (mainly due to FPGA's inherently programmable nature) and higher scaling capabilities than ever before. Intel's Agilex 7 FPGAs introduce a new chiplet as part of their heterogeneous multi-die architecture, the R-Tile, which is responsible for delivering the latest connectivity technologies - namely PCIe 5.0 and CXL support - in hardware-accelerated, hard-coded IP blocks.

The new heterogeneous R-Tile chiplet is the star of the show for Intel, allowing it to claim the title of the only FPGA family that carries certification for the full PCI-SIG 5.0 x16 data rate. Xilinx, which is now folded into AMD, would be an example of another leading-edge FPGA developer, so there's the feeling of a race won for Intel here. 

Interestingly, Intel seems to be doubling down on a separation between FPGA and CPU products. Perhaps that's a result of its earlier inroads into integration, of which only silicon ghost stories remain. AMD seems to be much more confident it's cracked the case: the company is seemingly looking to incorporate FPGA capabilities into the company's EPYC CPUs as early as this year. Now, that doesn't cement a departure between both companies - AMD may be also looking at a chiplet-like integration, whether vertically via 3D stacking or by incorporating segregated FGPA-specific IP in a separate chip.

The general idea of FPGAs is their inherent flexibility, which allows developers to quickly iterate on circuit arrangement and processing blocks to adapt the FPGA for the specific workloads at hand. Being more specialized hardware, FPGAs can be used to accelerate workloads that aren't CPU-dependent, thus freeing up valuable CPU resources for their specific tasks (such as spinning out VMs in a cloud-based installation) instead of tapping into their lower power efficiency (the cost of their generalized processing capabilities).

Intel's R-Tile essentially brings the hardware-accelerated IP blocks responsible for handling the PCIe 5.0 and CXL 1.1/2.0 protocols into its Agilex 7 FPGA family. That  should bring about significant improvements in power efficiency and data throughput, which are important elements to lower Total Cost of Ownership (TCO) for high-performance installations. But with these choices, there are always trade-offs: Intel is adding yet another fixed-function hardware block onto a product whose desirability lies in it being programmable. Programmable die area is the FPGA buyer's mantra, after all.

It also bears mentioning that as a product, R-Tile is specifically saying "I'm here to reduce the load on your CPUs while delivering increased performance". But another answer to that need isn't about moving CPU functions towards an FPGA; it's to merely increase the number of available CPU resources. And that could be done with more CPUs (which might make sense to some installers) or additional CPU cores. Intel's Agilex 7 M-Series is specifically marketed to Intel's 4th Gen Scalable Xeons, after all, and those aren't the core-count kings in any shape.

Intel's bet is that the questions above have an answer, and that it knows the answer: that's why it introduced Agilex 7. Intel's answer is that consumers want to remove CPU overhead by moving it onto an FPGA packaging. They want the best performance/watt possible (one of the highest contributors to high TCO cost), so they'll move things to where they're fastest. Luckily, this move benefits Intel in other ways. For the company, it's a matter of its own efficiency - and as such, a matter of cost. 

That's where Intel's embedded multi-die interconnect bridge (EMIB) really shines. As the proverbial "glue" holding the disparate processing blocks together, EMIB allows Intel to further separate IP blocks at the manufacturing level, increasing die efficiency and reducing the overall cost-per-wafer (and resulting cost-per-chip) equation. 

For the consumer, it also theoretically lowers costs: Intel (and the industry's) dream is to be able to mix and match different hardware IP blocks (from the same vendor, or even multiple vendors and manufacturing processes), meaning that customers need only pay for the silicon they'll actually use, with the specs they'll actually use. In a sense, this would make every chip an FPGA.

With all that in mind, it's possible today's Agilex 7 with R-Tile is as much a new FPGA product as it is a new staple on Intel Foundry Services' (IFS) catalog. Either way, it moves Intel in the direction it wants - and needs - to go. And that's just good business.

A group of researchers with KAUST (King Abdullah University of Science and Technology) have announced a new, groundbreaking manufacturing technique for what is known as "memristors" - circuits that are one of the four fundamental electrical components, alongside resistors, capacitors, and inductors. The new technique has been shown to enable the creation of one of cryptography's essential components, a True Random Number Generator (TRNG).

True Random Number Generators are essential parts of cryptography, and perhaps unintuitively (after all, how hard is it to produce random numbers?), it's also one of the most prone to failure. That's because it's easy for a random distribution (that is, when all possible events have an equal chance of happening) to become a non-random distribution.

Usually, TRNGs are implemented at the silicon level, such as AMD's Ryzen and Epyc-bound Cryptographic Co-Processor (CCP) (now at iteration 5.0). One way to generate random numbers is to look at inherently random phenomena, such as the photoelectric effect that's the basis for our computers' operation. From these effects, random numbers are generated that then serve as the basis for an encryption operation - each random number translating to part of the encrypted message, in the process known as hashing. To better put the problem into perspective, consider that AMD's Xilinx division commercializes Field-Programmable Gate Arrays (FPGAs) whose aim is to serve as True Random Number Generators.

But electrical components have operational boundaries, and small voltage changes can introduce computational or photoelectric "errors" that form patterns. Of course, when patterns emerge in a pool of numbers that are supposed to be random, then it's not really random anymore. There's a pattern, a slightly different probability for one number to be chosen over the other. And if it isn't truly random, then the emerging patterns can be extracted, analyzed, and compared to the encrypted output... And the way is open toward the supposedly cryptographically-secure message.

Some patterns can emerge naturally, from certain imbalances in the system that push it away from its random "equilibrium" state (such as hardware degradation, which is partly responsible for CPUs and GPUs alike both seeing drops in the maximum sustained operational frequency as they age). We've seen those being exploited by researchers - exfiltrating data from patterns such as a system's fan speed, for instance. But others can be introduced by sophisticated-enough adversaries.

The work done by the KAUST researchers now unlocks memristor-based TRNG fabrication in a process not dissimilar to 3D printing. Except instead of the usual filament, atomically-thin layers of boron nitride and silver electrodes are deposited until the elements of a memristor all stack into place. Due to this specific fabrication process, the TRNG sips power compared to the usually CPU-integrated alternatives, built out of expensive circuits with millions of transistors (costly both in terms of power usage and the space they occupy on the accelerator's design).

“We fabricated a memristor using a novel two-dimensional layered material called hexagonal boron nitride, on which we printed silver electrodes using a scalable, low-cost inkjet printing technology,” said Pazos, a researcher within the KAUS team. “The unique properties of the 2D h-BN are maintained after the electrode has been printed, enabling superior power and random signal generation.”

The resulting TRNG generator was apparently in-line with the team's expectations: it showed the best performance of a TRNG in terms of stability of its random signal through time; it showed incredibly low energy consumption; and finally, easy and fast circuit readout, enabling the memristor-based TRNG to generate 7 million random bits per second.

“Furthermore, we demonstrated a built circuit that generates random numbers by interconnecting our memristor to a commercial microcontroller and making live experiments of random number generation on the fly,” Pazos added.

It also seems as if the technology is ready for prime time as is, as opposed to most other technological breakthroughs. The technology could be readily rolled-out to IoT (Internet of Things) applications and other edge devices, such as sensor node arrays.

“Our scalable low-cost fabrication method using inkjet printing not only enables excellent performance but is key to the successful integration of these devices into low-cost complex electronics,“ Pazos says. “This work demonstrates the potential of 2D materials like h-BN to underpin a revolution in solid-state micro- and nanoelectronic devices and circuits owing to their outstanding electronic, physical, chemical and thermal properties.”

The Occamy processor, which uses a chiplet architecture, packs 432 RISC-V and AI accelerators and comes with 32GB of HBM2E memory, has taped out. The chip is backed by the European Space Agency and developed by engineers from ETH Zürich and the University of Bologna, reports HPC Wire. 

The ESA-backed Occamy processor uses two chiplets with 216 32-bit RISC-V cores, an unknown number of 64-bit FPUs for matrix calculations, and carries two 16GB HBM2E memory packages from Micron. The cores are interconnected using a silicon interposer, and the dual-tile CPU can deliver 0.75 FP64 TFLOPS of performance and 6 FP8 TFLOPS of compute capability. 

Neither ESA nor its development partners have disclosed the Occamy CPUs' power consumption, but it is said that the chip can be passively cooled, meaning it might be a low-power processor.

Each Occamy chiplet has 216 RISC-V cores and matrix FPUs, totaling around a billion transistors spread over 73mm^2 of silicon. The tiles are made by GlobalFoundries using its 14LPP fabrication process. 

The 73mm^2 chiplet isn't a particularly large die. For example, Intel's Alder Lake (with six high-performance cores) has a die size of 163 mm^2. As far as performance is concerned, Nvidia's A30 GPU with 24GB of HBM2 memory delivers 5.2 FP64/10.3 FP64 Tensor TFLOPS as well as 330/660 (with sparsity) INT8 TOPS.

Meanwhile, one of the advantages of chiplet designs is that ESA and its partners from ETH Zürich and the University of Bologna can add other chiplets to the package to accelerate certain workloads if needed.

The Occamy CPU is developed as a part of the EuPilot program, and it is one of many chips that the ESA is considering for spaceflight computing. However, there are no guarantees that the process will indeed be used onboard spaceships. 

The Occamy design aims to support high-performance and AI workloads through a bare-metal runtime, but it is not yet clear whether the runtime will be at a container level or at the bare-metal level. The Occamy processor can be emulated on FPGAs. The implementation has been tested on two AMD Xilinx Virtex UltraScale+ HBM FPGAs and the Virtex UltraScale+ VCU1525 FPGA.

AMD on Tuesday posted its first quarterly loss in years due to weak sales of processors for client PCs. Overall, AMD's chip sales dropped 64%. AMD's data center and gaming hardware shipments remained strong and were flat year-over-year, which is quite an achievement given the slowing purchases of servers and weak demand for gaming hardware among consumers. While AMD's management expects the CPU market to start recovering in the second half of the year, the company's outlook for Q2 is not that optimistic.

Mixed Bag

In the first quarter of FY2023, AMD's revenue amounted to $5.353 billion, which is a 9% decrease compared to the same period in the previous year and a slight decrease compared to the previous quarter. Unfortunately, the company slipped into the red with a $139 million net loss as compared to a $786 million net income in Q1 FY2022. Additionally, AMD's gross margin decreased from 48% in Q1 FY2022 to 44% in Q1 FY2023.

"We executed very well in the first quarter as we delivered better than expected revenue and earnings in a mixed demand environment," said AMD Chair and CEO Dr. Lisa Su.

Indeed, AMD's results were a mixed bag as all of the company's business units except its Client Computing business remained more or less flat compared to the first quarter of FY2022, and even remained profitable. In fact, AMD's Data Center unit even managed to modestly increase its revenue, yet its profitability declined.

"Our strategically important Data Center and Embedded segments contributed more than 50% of revenue in the first quarter," said AMD EVP, CFO, and Treasurer Jean Hu. 

Consumer CPU Sales Nosedive, Data Center Hardware Remains Strong

AMD's Client Computing business unit earned $739 million in revenue in Q1 FY2023, a 65% drop from the same quarter in the previous year. The unit lost $172 million, which is in stark contrast to the $625 million profit it posted in the same quarter a year ago. For years, AMD's main source of revenue was from selling CPUs and chipsets for client PCs, but for the second quarter in a row, this unit suffered from a revenue decline and bled money. As it turns out, AMD's client PC business suffered more from the soft PC market than Intel's client computing group, which Q1 sales dropped 36% year-over-year.

"As we stated on our last earnings call, we believe the first quarter was the bottom for our client processor business," said Su. 

By contrast, AMD's Gaming Business unit's revenue in the first quarter reached $1.757 billion, down 6% year-over-year, which looks like a strong result given the ongoing turmoil. System-on-chips for Microsoft's and Sony's game consoles accounted for the lion's share of AMD's gaming business and increased YoY. Meanwhile, sales of discrete Radeon graphics processors dropped year-over-year. Still, they grew quarter-over-quarter as the company ramped up shipments of its Radeon RX 7900-series graphics offerings and offered its previous generation Radeon RX 6000-series products at competitive prices. Despite challenges, the division remained profitable and posted a net income of $314 million (down 12% YoY). 

"Semi-custom SoC revenue grew year-over-year as demand for premium consoles remain strong following the holiday cycle," said Su. "In gaming graphics channels sell through over Radeon RX 6000 and Radeon RX 7000 series GPUs increased sequentially. We saw strong sales of our high-end Radeon RX 7900 XTX GPUs in the first quarter, and we are on track to expand our RDNA 3 GPU portfolio with the launch of new mainstream Radeon RX 7000- series GPUs this quarter."

In Q1 2023, AMD's Data Center Business generated $1.295 billion, which is almost flat compared to the same quarter in the previous year ($1.293). Meanwhile, the profitability of the unit decreased 65% year-over-year and totaled $148 million, which is not surprising as large cloud services providers are reconsidering their server purchases these days, and therefore AMD and other players have to offer them better pricing to keep volumes high.

It is noteworthy that when compared to Intel's data center and AI group, AMD's server business performed somewhat better as it remained flat. By contrast, the blue company posted a 39% revenue drop for its data center parts. 

AMD's Embedded Business, which mostly sells products developed by Xilinx and some CPUs designed by AMD, reported earnings of $1.562 billion and profits of $798 million, a positive highlight in AMD's financial report. 

Cautious Outlook

AMD expects its revenue in the second quarter to stay more or less flat with the first quarter and total around $5.3 billion ± $300 million. 

Such a cautious outlook is a bit surprising as AMD is ramping up production and sales of its Ryzen 7000-series processors for laptops, with its Phoenix chips already on the way to OEMs, and these units tend to be sold at a premium, so the company expects its client PC revenue to be up quarter-over-quarter. In addition, the company is on track to release its new Radeon 7000-series graphics processors in Q2, which are also expected to boost sales of the company's gaming hardware, but this might be offset by weaker sales of console SoCs.

Furthermore, AMD plans to formally introduce its EPYC 'Bergamo' CPUs for cloud data centers and EPYC 'Genoa-X' for high-performance technical computing applications. While ramps of server CPUs tend to be slow, these parts will still allow AMD to increase its sales of data center hardware slightly. 

"For the second quarter, we expect sequential growth in our Data Center and Client segments offset by modest declines in our Gaming and Embedded segments," said AMD's chief financial officer. "We remain confident in our growth in the second half of the year as the PC and server markets strengthen and our new products ramp."

 Meanwhile, AMD remains optimistic about strong demand for its products in the second half and beyond. For example, the firm expects its data center sales to beat 2022 this year. 

"Looking longer term, we have significant growth opportunities ahead based on successfully delivering our roadmaps and executing our strategic datacenter and embedded property have priorities led by accelerating adoption of our AI products," Su said.

AMD announced last month that it had missed its original ship date for its flagship Ryzen 7040HS laptop processors, also known as the "Phoenix" series, but that it would ship those processors in April. However, it is now May, and we haven't seen the Phoenix chips in the market at retailers. Today we had a chance to ask AMD for comment on the matter. A company representative told us that the company's original statement referred to the timeline for shipping the chips to OEMs, who will then deliver the finished systems to retail. AMD says this is already well underway.

"We did start to ramp production of those Ryzen 7040 Series [chips]. Our OEM partners have started shipping those systems. Largely, the 7040HS systems. So those should hit etail and retail over the next several weeks. We started shipping last quarter, and those systems should be available in the next several weeks." — AMD representative to Tom's Hardware.

As always, the definition of a shipping processor can be a bit fuzzy, and here we see that AMD says the goal was to have the processors shipping to OEMs by April. AMD originally announced the Ryzen 7040HS laptop processors at CES 2023, and they were slated for release in March, but now it is apparent they are finally working their way to market. 

Fuzzy timelines aside, the Ryzen 7040HS processors are plenty impressive. The Phoenix chips are the first to integrate AMD's new RDNA 3 graphics engine for mobile, along with being the first AMD consumer chips to incorporate a Xilinx XDNA-powered AI Engine. The chips will leverage AMD's Zen 4 architecture fabbed on the 4nm process and come with robust connectivity options, like DDR5 memory and the PCIe 4.0 interface. 

The Phoenix chips represent AMD's most powerful chips for ultra-thin laptops and operate in the 35W to 45W TDP range. AMD has yet to share more details on just which systems are tipped for release in the coming weeks, but several models have been spotted for preorder at JD.com, a Chinese retailer. Given the timelines, we expect to hear more from AMD soon about the specific Phoenix-powered laptops that will be on offer.  

Sony became AMD's largest customer last year, accounting for 16% of the company's revenue as its PlayStation 5 increased its lead over competitors. Sravan Kundojjala, a semiconductor industry analyst, noted that if Xilinx results are excluded, Sony accounts for 20% of AMD's revenue, probably making it the company's largest customer in recent history. 

Indeed, the gaming business unit was AMD's largest revenue generator, which indicates that sales of system-on-chips for Microsoft's Xbox Series X|S consoles were also strong. AMD sold Sony some $3.776 billion worth of chips for PlayStation 5 game consoles in 2022, which accounted for 16% of the company's revenue for the year, according to the company's filing with the SEC. 

Sony confirmed at CES that it had sold over 30 million PlayStation 5 game consoles. VGChartz estimates that by now, life-to-date sales of PS5 exceeded 31.77 million units. Microsoft’s Xbox Series X|S consoles are by far not as successful as Sony’s systems, with LTD shipments of around 20.68 million units, according to VGChartz. 

In fact, AMD's gaming business unit — which sells discrete graphics processors for desktop graphics cards and notebooks as well as SoCs for game consoles — earned $6.805 billion in earnings and $953 million in profits last year and was the company's main source of revenue. Keeping in mind that in recent quarters the unit sales of AMD's standalone GPUs have declined (based on data from Jon Peddie Research), consoles SoCs accounted for the lion's share of AMD's gaming business unit revenue. 

"Gaming revenue declined 7% year-over-year to $1.6 billion [in Q4 2022] as lower gaming graphics sales more than offset higher semi-custom revenue," said Lisa Su, chief executive of AMD, at the company's most recent earnings call (via SeekingAlpha). "Semi-custom SoC revenue grew year-over-year as demand for game consoles remained strong during the holidays. Gaming graphics revenue declined year-over-year as we further reduced desktop GPU downstream channel inventory." 

But AMD's earnings from PlayStation 5 and Xbox Series S|X SoCs will likely decline in 2023 and onwards since sales of consoles usually peak in their third year and platform holders tend to renegotiate components pricing after that. Therefore, AMD's gaming revenue is set to drop in 2023 compared to 2022. 

"Given where we are in the cycle, we would expect gaming to be down on a year-over-year basis," said Su.  

Analysts and investors tend to praise AMD's data center EPYC processors as they have been the company's key profit driver for about six years now. In 2022, AMD's data center unit earned $6.043 billion in revenue and was behind the company's gaming unit ($6.805 billion) as well as client computing group ($6.201 billion). Yet, based on softening demand for the PC and game console cycle, it looks like AMD's data center business will become the company's main source of revenue in 2023.

AMD published its financial results for the fourth quarter and full year of 2022. Due to strong demand for AMD's data center EPYC processors, the company posted an all-time record revenue for the full year. Meanwhile, sales of consumer CPUs and GPUs declined significantly in the fourth quarter, which clearly limited the company's growth in Q4.

Record Year, Not-So-Record Quarter

AMD's Q4 2022 revenue totaled $5.6 billion, up 16% year-over-year and flat with Q4 2022. Meanwhile, the company's net income collapsed to $21 million, 98% lower than in the same period a year ago. As for gross margin, it dropped to 43% from 50% in Q4 2021.

As for the full year 2022, AMD earned $23.601 billion, up a whopping 44% from $16.434 billion in 2021. This was the highest yearly result that AMD has ever posted, yet it could have been higher if demand for consumer processors and graphics cards had been stronger in the third and fourth quarters.  

"2022 was a strong year for AMD as we delivered best-in-class growth and record revenue despite the weak PC environment in the second half of the year," said AMD Chair and CEO Dr. Lisa Su. "We accelerated our data center momentum and closed our strategic acquisition of Xilinx, significantly diversifying our business and strengthening our financial model. Although the demand environment is mixed, we are confident in our ability to gain market share in 2023 and deliver long-term growth based on our differentiated product portfolio." 

Because of softening PC market, the year 2022 was somewhat of a mixed bag for AMD since demand for its products met expectations in the first half, but demand for consumer PC hardware dropped sharply in the second half, affecting the full-year results.

Data Center Hardware on the Rise, Consumer Hardware Sales Craters

Consumer CPUs and chipsets used to be AMD's bread and butter some five or six years ago, but that is no longer true. In Q4, the revenue of AMD's Client Computing Business dropped to $0.903 billion, or 51% year-over-year. The business unit lost $152 million, whereas in the same quarter a year before, it posted a $530 million profit.

Perhaps the most alarming detail about AMD's consumer CPU business is that in a quarter when the company introduced its brand-new Ryzen 7000-series processors, the average selling prices of its CPUs were flat year-over-year, suggesting tepid sales of the latest parts. 

When it comes to AMD's Gaming Business, there was a mixed bag again. On the one hand, the unit earned $1.6 billion in revenue (down 7% year-over-year) and $266 million in profits (down from $407 million in Q4 2021). But on the other hand, this strong result was achieved primarily because AMD managed to sell loads of system-on-chips for consoles, whereas the sales of its graphics processors for discrete desktop PC GPUs were down year-over-year.

Having earned $1.7 billion in revenue, an increase of 42% year-over-year, AMD's Datacenter Business was the main source of income for the company in Q4 2022, outstripping the Gaming and Embedded businesses. Despite lower margins (27% in Q4 2022 vs 32% in Q2 2022), the group's operating income totaled $444 million, up from $369 million in the same period a year before. 

As for AMD's Embedded Business — which mostly sells products designed by Xilinx and some chips designed by AMD — posted earnings of $1.4 billion and earned $699 million in profits, marking another bright spot in AMD's financial report.

Cautious Outlook for Q1, Lack of Outlook for Whole 2023

AMD expects its first-quarter revenue to be approximately $5.3 billion (±$300 million), a decline of about 10% YoY mainly because of lower consumer CPU and gaming GPU and SoCs sales. The company expects shipments of its data center solutions, such as EPYC CPUs and Pensando DPUs, FPGAs, and embedded products to continue increasing. 

Meanwhile, just like some other high-tech companies, AMD didn't provide an outlook for the whole year due to a lack of proper visibility for the second half of 2023. 

AMD revealed a cadre of laptop chips today in Vegas at CES 2023, including its first chiplet-based processors for laptops, the 5nm Dragon Range chips that come with the Zen 4 architecture. AMD also introduced its new 4nm Phoenix lineup that comes with the Zen 4 architecture and RDNA 3 graphics engine paired with AMD’s first Ryzen AI engine built on the new XDNA architecture. This new advance uses the tech from AMD’s Xilinx acquisition one short year after the purchase to infuse a dedicated AI engine inside AMD’s latest laptop chips.

AMD also introduced several other new designs, with the 6nm Mendocino, 7nm Barcelo-R, and 6nm Rembrandt-R series all coming to market early this year. The overhaul marks the broadest laptop portfolio AMD has ever had as it continues to chew away market share from Intel, so there’s a lot to cover. Let’s dive in.

AMD Ryzen 7045 Series: Dragon Range

The 5nm Dragon Range HX-Series marks the debut of AMD’s chiplet architecture on a laptop platform, crafting a new line of chips for ‘extreme gaming’ and creators. This new lineup of chips leverages AMD’s existing chiplets from its Ryzen 7000 CPUs but infuses them into a smaller BGA package specifically for laptops. This creates a 55W+ processor that wields the Zen 4 CPU core architecture and RDNA 2 integrated graphics, with the latter being used for light display tasks only -- these chips will come in the latest high-performance laptops for gaming, so they’ll always come in systems that have a discrete GPU.

The halo Ryzen 9 7945HX model tops out at 16 cores and 32 threads, just like the desktop PC models, giving AMD the highest ‘performance core’ count on the market. (Intel leads with 24 total cores with its recently launched laptop chips, but eight of those are cut-down efficiency cores.) This chip boosts up to 5.4 GHz, comes armed with an incredible 80MB of total cache, and has an official TDP rating of 55-75W+.

AMD tells us that these processors can be fed up to 100W in select laptop designs -- it’s up to the OEM to dictate the peak power delivery -- but still not have egregiously bad sub-4-hour battery life. Regardless, we expect you won’t want to stray too far from a wall plug with these laptops. AMD also has 12-, 8- and 6-core Dragon Range models that largely mirror what we see in the Ryzen desktop PC lineup, much like Intel does with its own HX series that is also derived directly from its desktop Core processors.

AMD provided benchmarks, but as with all vendor-provided test data, approach it with caution. The company claims the 7945HX is from 29% to 62% faster in CPU bound games than the Ryzen 9 6900HX. AMD also claims the 7945HX is 18% faster in single-threaded work and 78% faster in multi-threaded than its previous-gen halo chip, the Ryzen 9 6900HX.

AMD says these chips will not require an external discrete chipset due to the capabilities built right into the I/O die inside of the processor, though OEMs will need to use an external chip for USB.

Dragon Range is specifically for the highest-end extreme machines, so AMD says there won’t be a proliferation of designs on the market. The leading-edge designs arrive in February 2023. Meanwhile, the Phoenix chips will be the real workhorse of the new lineup.

AMD Ryzen 7040 Series: Phoenix

The Ryzen 7040 HS-Series ‘Phoenix’ chips slot in as the mainstay for ultrathin notebooks. These 35-45W chips come with the Zen 4 CPU architecture on the 4nm process paired with AMD’s RDNA 3 graphics engine, both debuting in the company’s mainstream laptop lineup for the first time. The flagship model wields eight cores and 16 threads that operate at up to 5.2 GHz. The chips also have 40MB of total cache. AMD also has a model with 6 cores and 12 threads.

The Zen 4 cores and RDNA 3 graphics in the Phoenix chips are plenty impressive, but the new XDNA architecture steals the spotlight. AMD has rapidly integrated the FPGA technology it acquired with Xilinx last year to integrate a new FPGA-based AI engine directly into the die of the new 7040 series processors. This engine can handle up to 4 concurrent AI streams, though it can be rapidly reconfigured to handle varying amounts of streams. AMD claims the fully programmable “Ryzen AI Engine,” which is based on the XDNA architecture, is faster than even Apple’s neural engine in its M2 processors.

AMD is delivering this hardware early - there aren’t a lot of workloads that benefit yet – but is working on an API to further software development work. Additionally, Microsoft is in deep collaboration with AMD to fully exploit the new engine in Windows, with several new features slated to be available soon, like camera tracking and eye focus features.  AMD tells us that it is fully committed to an AI roadmap with its XDNA architecture – there will be an XDNA 2 and XDNA 3, for instance. AMD expects other use-cases to emerge quickly, including in gaming, security, predictive UIs, and collaboration work.

The Phoenix processors arrive in March 2023.

AMD Ryzen 7035 and 7030 Series: Rembrandt-R and Barcelo-R

The 6nm Rembrandt-R and 7nm Barcelo-R chips are a refresh generation that come with the Zen 3+ and Zen 3 architecture, respectively. Rembrandt wields the RDNA 2 graphics engine, while Barcelo comes with Vega.

AMD also has the Mendocino chips that come with the Zen 2 architecture paired with the RDNA 2 graphics engine, all fabbed on the modern 6nm process.

AMD CEO Lisa Su will take to the CES 2023 stage tonight in Las Vegas at 6:30 pm PT to fill us in on the company's newest products. AMD is largely expected to reveal a slew of new Ryzen 7000 processors, including the hotly-anticipated X3D models that use the radical new L3 cache stacking tech to deliver explosive gains in gaming performance. These chips are expected to help AMD retake the gaming crown from Intel's Raptor Lake processors. 

AMD is also expected to give us an update on several of its other technologies, like the rumored reveal of a new line of Dragon Range laptop processors, but anything goes in Vegas. Who knows, we might even hear about a new line of Ryzen Threadripper processors. 

Pull up a seat to see our live coverage as Lisa Su's keynote progresses. You can also follow along and watch the keynote video live here on YouTube. 

UPDATE: AMD has made several big announcements so far, which we've covered here: 

AMD Instinct MI300 Data Center APU Pictured Up Close: 13 Chiplets, 146 Billion Transistors

AMD Unveils Three Ryzen 7000X3D V-Cache Chips, Three New 65W Non-X CPUs, Too

AMD Brings Chiplets, Zen 4, RDNA 3 and XDNA AI to Laptops: 5nm Dragon Range and 4nm Phoenix Arrive

AMD Launches RX 7600M GPUs: Mobile RDNA 3 With RTX 3060 Performance

The show is still live, so check for updates below as well. 

AMD always draws a crowd -- there are currently over 5,000 people watching the livestream with 20 minutes to go before showtime. 

The show has begun, just a few minutes late, with a pre-recorded intro video. 

The head of the Consumer Technology Association, the group behind CES, Gary Shapiro, is making a short introductory speech to welcome AMD CEO Lisa Su to the stage. 

Lisa Su has taken to the stage after a short intro video that focused on the 'AMD, Together We Advance' branding. 

Su said she is ready to dive right into the new products that cover gaming, AI, sustainable computing, and other areas. 

Su is beginning with AI, which she says is the most important megatrend in tech, allowing us to make predictions about future outcomes. 

Lisa Su described the company's new XDNA architecture which encompasses the AI technology that the company purchased with Xilinx. AMD will use XDNA in its Ryzen Mobile 7040 series chips first. The new engine is called "Ryzen AI" and delivers up to 12 TFLOPS of performance. 

The 7040 processors, codenamed Phoenix, come with the Zen 4 CPU architecture and RDNA 3 graphics engine. It has 25 billion transistors. 

AMD presented benchmarks that show the 7040 series has a faster AI engine than the Apple M2 processors, and mentioned that Intel doesn't have an inbuilt AI engine to compare to. AMD also showed the Ryzen system being 35% faster than Apple and 45% faster than Intel's chips in Blender. 

Su says the 7040 series offers up to 40 hours of battery life. 

Su said that the company will need deep partnerships to develop the AI ecosystem. She invited Microsoft EVP Panos Panay to the stage to talk about the collaboration between the two companies. He complimented her jacket profusely, drawing laughter from the crowd and Lisa Su.  

Panoy discussed the partnership and other initiatives that AMD and Microsoft have worked on together, like bringing the Pluton security chip to AMD's processors. 

Panoy demoed some of the software that Microsoft has developed to use AMD's new Ryzen AI engine. That includes features in Microsoft Studio, like portrait blur, eye contact, and auto framing, all of which requires trillions of operations per second. These functions can run on the AI engine at milliwatts of power. This doesn't drain the battery while keeping the CPU and GPU free for more important work.

Lisa Su prompted Panos to tell us what the future of AI might look like. He says that AI will redefine what we do with Windows, but that it requires new compute capabilities and an operating system that blurs the line between the edge and the cloud. And with that, Panos took his leave from the stage. 

Su switched gears to talk about hybrid and remote work, with more than 80% of employees wanting to work from home, while more than half say that connectivity issues limit their work. AMD is working on making hybrid work 'frictionless.' Lisa Su called Enrique Lores, the CEO of HP, to the stage to talk about their efforts in this area. 

Yes, we're also ready for the talk to turn back to more new silicon. 

Lores says that 50% of the workforce will be freelancers in five years. Interesting.

HP announced its Dragonfly Pro laptop today, and it's powered by AMD and targeted for remote freelance workers. This laptop delivers 40% higher Passmark scores than Apple's M1, according to HP. 

Now Lisa Su is switching gears to gaming. Su says there are 3.2 billion gamers in the world, and most play on PCs and consoles. AMDs CPUs and GPUs power all sorts of gaming products, including handhelds, consoles, PCs, and even cars. 

Su introduced the new Ryzen 7450X CPUs, the first chiplet-based processors for laptops. They have up to 16 cores and 32 threads with the Zen 4 architecture and RDNA 2 graphics. These are basically Ryzen 7000 desktop CPUs crammed into a laptop form factor. 

The 79450HX offers up impressive gains over the previous-gen. 

AMD introduced the Radeon RX 7600M XT mobile laptop graphics. Up to 26% faster framerates than the competition, according to AMD. These come to market in February. 

Matt Zielinski from Lenovo came to the stage to talk about their partnership. 

Lenovo is unveiling its first Legion Pro Series laptops tomorrow with the 7045HX processors. He says it is the most powerful gaming laptop Lenovo has ever created. 

Lisa is moving on to desktop CPUs now! She's talking about the Ryzen 7000X3D processors. AMD is bringing 3D V-Cache to Ryzen 7000 processors. This tech stacks more L3 on top of the chip to boost gaming performance. 

The Ryzen 7 7700X3D delivers on average up to 15% more gaming performance than the 5800X3D. 

AMD is also bringing V-Cache to Ryzen 7 and Ryzen 9 processors, with the 16-core Ryzen 9 7950X3D coming with 16 cores and 32 threads. The Ryzen 9 7900X will have 12 cores and 24 threads. 

We now have an article on the site covering the 3D V-Cache chips. AMD is also launching the locked 65W Ryzen 7000 CPUs, three models, and low-cost budget motherboards. You can read about all of that here. 

Lisa Su is now talking about how computing can help to revolutionize health care and other fields through adaptive computing. 

Here are the details on the new Ryzen 7000 65W CPUs. These three new 65W Ryzen 7000 “non-X” models span the Ryzen 5 to Ryzen 9 families, thus providing a new lower level of entry to the AM5 platform to address some of the pricing issues associated with the company’s new AM5 platform. These chips will arrive January 10.

Lisa Su pivoted to the metaverse now, bringing Magic Leap CEO Peggy Johnson to the stage. AMD's chips power the Magic Leap devices. Though Magic Leap never lived up to the hype, it is now used in several scenarios, like surgery. 

Magic Leap 2 uses a custom processor from AMD paired with some of the company's own IP, thus creating the most advanced augmented reality system in the industry, at least according to Magic Leap. 

Former NASA astronaut Dr. Cady Coleman has come to the stage to talk about AMD's efforts in space, largely through its Xilinx arm. In the meantime, here's something about chips:

Here is some more info on the AMD mobile chips. 

Here's a die shot of the Ryzen 9 7950X3D. We took this pic during a briefing earlier today.  

Here are even more chips we pictured earlier today. I wrote labels for the chips you might not recognize. 

AMD is back on track, talking about its datacenter CPUs.

AMD launched its fourth-gen EPYC Genoa CPUs last November, which we reviewed here. Su touted its 300 world records and amazing power efficiency. 

Lisa Su says that 5 AMD servers can replace 15 Intel servers, thus saving incredible amounts of power. There are over 15 million servers worldwide. If they were all EPYC, they would save 52 billion k/Wh of power, and 26 million tons of C02. 

Su is announcing the company's new Alveo V70 inference accelerator. This is a Xilinx product. 

Lisa Su talked about how AMD now powers the fastest supercomputers in the world. AMD has now developed the first chip with both CPU and GPU and memory all into a single chip, the MI300. This has the CDNA3 architecture paired with 24 Zen 4 cores, all paired with 128GB of HBM3 memory. AMD is using 3D stacking to connect multiple GPU and CPU die together. 

Su showed us the chip, it's a beast! I held this chip earlier today and have other pics. This has 146 billion transistors. 

MI300 can reduce the time it takes to train the largest models, like ChatGPT, from months to weeks. This chip will launch in the second half of 2023. 

Lisa Su thanked the audience and closed the show. 

That's all, folks! Be sure to check out our writeups above. 

AMD on Tuesday revealed its financial results for the third quarter of 2022. The company's revenue totaled $5.6 billion primarily because of weak sales of AMD's CPUs and GPUs for the client segment. Meanwhile, sales of AMD's data center products set another record as demand for EPYC CPUs, Pensando DPUs, as well as Xilinx field programmable gate array products remained strong.

"Third quarter results came in below our expectations due to the softening PC market and substantial inventory reduction actions across the PC supply chain," said Dr. Lisa Su, the head of AMD. "Despite the challenging macro environment, we grew revenue 29% year-over-year driven by increased sales of our data center, embedded and game console products. We are confident that our leadership product portfolio, strong balance sheet, and ongoing growth opportunities in our data center and embedded businesses position us well to navigate the current market dynamics."

Strong Data Center and Gaming Revenue

AMD's Q3 2022 revenue reached $5.565 billion, up 29% year-over-year but down $985 million compared to the previous quarter. The company's gross margin dropped to 42%, down from 48% in the same quarter a year ago. As for net income, it cratered to $66 million from $923 million in Q3 2022, whereas the company's earnings per share declined to $0.04, down from $0.75 a year ago. AMD stressed that its profitability decreased due to the amortization of Xilinx acquisition-related intangible assets and increased R&D investments, partially offset by a $135 million tax benefit in the quarter.

Demand for client processors and chipsets got significantly lower in the third quarter, which is why AMD's Client Computing revenue fell to $1.022 billion in Q3 2022, or a whopping 40% year-over-year. In addition, the company's client business unit lost $26 million during the quarter as the company's clients were trying to clear out inventory. Therefore AMD needed to offer specific incentives to sell products sitting in the company's stock.

AMD's Gaming business — which includes client GPUs and console system-on-chips — continued to demonstrate mixed results during the quarter. The business unit earned $1,631 billion during the third quarter, up 14% year-over-year. But AMD's Gaming BU thrived not because demand for its Radeon RX 6000-series graphics processors was particularly strong but because Microsoft and Sony prepared for the holiday season and bought a boatload of system-on-chips for their Xbox and PlayStation game consoles in Q3 2022.

AMD's Datacenter business unit has been the company's main success story in recent years. The third quarter was not an exception; the BU's sales totaled $1,609 billion, up 45% year-over-year, whereas its operating income topped $505 billion, a 64% YoY increase. It is particularly important that while AMD began shipments of its next-generation EPYC 'Genoa' processors to select customers in Q3, those volumes were barely significant. The company said that its Datacenter business unit was supply constrained during the quarter and that in Q4 2022, those constraints will be less severe, allowing the company to grab some additional market share from its arch-rival Intel.

AMD's Embedded business, which sells products developed by AMD and various solutions designed by Xilinx, earned $1.3 billion, up 1,549% year-over-year since in Q3 2021, AMD's Embedded BU only sold the company's CPUs and GPUs, which are not particularly popular.

Cautious Outlook

Although AMD is optimistic about the performance of its data center and embedded businesses in the fourth quarter, it does not expect much from Q4 2022. First, the fundamental weakness of the client PC market will hardly let the company sell loads of its client CPUs. Meanwhile, sales of console SoCs will be down sequentially. Higher sales will unlikely offset them if AMD's discrete graphics processors as the company typically ramp up its GPUs very slowly.

For Q4 of 2022, AMD expects revenue to be approximately $5.5 billion ± $300 million, an increase of around 14% YoY and flat sequentially. For the full year 2022, AMD projects revenue to be approximately $23.5 billion ± $300 million, an increase of about 43% over the previous year, primarily driven by the success of AMD's data center and embedded products.

An FPGA - that’s a field-programmable gate array, a sort of reconfigurable microchip - has been shown to run a 3D, ray-traced game written in C 50 times more efficiently than an x86 CPU while using a fraction of the energy and perhaps pointing the way to future programming efficiency gains. The claims are made in a white paper [PDF] by Victor Suarez Rovere, a developer from Argentina, and Julian Kemmerer, a systems engineer from Pennsylvania, and brought to our attention by CNX Software.

The FPGA in question is the Arty A7, a Xilinx Artix-7 100T FPGA development board that sells for around $280 and features 101,440 logic cells (an FPGA’s logic cells contain a look-up table that can implement any logic function, giving the chip its programmability) on a 28 nanometer process, and which pulls less than a watt of power. The CPU it was pitted against (without, it must be said, troubling the chip’s iGPU) was a Ryzen 7 4800H, an eight-core 16-thread laptop processor that was built on a 7 nm process and has a default TDP of 45W. That's a laptop chip that's not available on its own, but the R7 4700G is currently available for about $240.

The game that was compiled to run on the two very different platforms is "Sphery Vs Shapes," and doesn’t appear to contain much in the way of plot, characters or actual gameplay, but does have lots of ray-tracing, as a shiny metallic ball bounces its way across a chessboard-like environment, which is reflected in its shiny spherical surface.

Both platforms rendered the game at 1080p and 60 frames per second without a problem, but the FPGA did it using 660 mW, while the R7 needed 35W, a difference of 53x. It is speculated that, were the FPGA to use the same 7nm process as the CPU, this figure could be six times higher.

The keys to the whole thing are Pipeline C, an invention of Kemmerer’s, and CflexHDL from Suarez. You can find them both on GitHub. "The game’s pixel rendering and animation logic is based on floating point and vector math operations. All of the game code is expressed using a clean syntax that translates directly to a digital circuit. The current target of this design is a FPGA board with Full HD digital video output, and the workflow also allows running the game in realtime on a regular PC using the unmodified source," they write in their paper. "This allows for much faster development-test iterations than with traditional hardware design tools. For the same workload, the computing efficiency resulted in more than 50X better than using a modern CPU, in a chip an order of magnitude smaller."

"Sphery Vs Shapes" stands up pretty well as a graphics demo, but what it means for the future of programming is more interesting - especially as FPGAs are going to start appearing in AMD chips. There are plans to port the whole thing to RISC-V, and to design an open-source ASIC (application-specific integrated circuit) that supports the pipeline, and there are possibilities for the world of microcontrollers too." The code can be translated to a logic circuit, run on a[n] off-the-shelf CPU, or on a microcontroller to develop hardware/software peripherals without changes to the code,” Suarez and Kemmerer write in their conclusion. "The results we obtained are readily reproducible, as materials are easy to obtain and not expensive."

Computational storage has its first set of official specifications following the publication of version 1.0 of The Storage Networking Industry Association’s Computational Storage Architecture and Programming Model. As reported by The Register, the model is aimed at developing this promising new tech by allowing different manufacturers’ designs to operate together.

‘Computational storage’ is a new buzzword that doesn’t exactly mean that we’re going to turn SSDs into processors, but that processing and storage will be more tightly bound together, with processors embedded in the storage to share the load of storage-heavy tasks. At the moment, this means an FPGA running encryption, compression and erase tasks under the CPU’s supervision, and could mean an Arm CPU running Linux left to its own devices in the near future. 

Products already shipping include Samsung’s SmartSSD CSD (computational storage drive, pictured above), which uses Xilinx Versal Adaptive SoCs from AMD and, according to Samsung’s own figures, can slash CPU utilization by up to 97 percent. Scaleflux also has a product, the third-gen CSD3000, which uses an eight-core Arm chip.

What computational storage has lacked, until now, is standardization, and the SNIA’s (Storage Networking Industry Association) Computational Storage Technical Working Group has been working since 2018 on the new specs. Now that they’re here, hardware vendors can begin the process of building their own interoperable computational storage systems, and software types can get stuck into drivers with common terminology and a discovery process so computers faced with a new CSD know how to begin interfacing with their resources.

The specifications themselves come to a hefty 71-page PDF, but for those looking for lighter reading the SNIA has published a blog post explaining the important parts of the new computing model.

“Having an industry developed reference architecture that hardware and application developers refer to is an important attribute of the 1.0 specification,” wrote SNIA Computational Storage Architecture and Programming Model editor Bill Martin in the post, “especially as we get into cloud to edge deployment where standardization has not been as early.  Putting compute where data is at the edge – where data is being driven – gives the opportunity to provide normalization and standardization that application developers can refer to contributing computational storage solutions to the edge ecosystem.”

AMD on Tuesday reported its revenue of $6.6 billion for the second quarter of 2022, its highest quarterly revenue ever. In addition, all AMD products, including CPUs and GPUs, increased in Q2 2022. Nonetheless, shipments of products for client PCs started to demonstrate softness because of growing inflation and a challenging macroeconomic environment.

AMD's revenue for the second quarter of 2022 increased 70% year-over-year due to the company's products success and adding products from Pensando and Xilinx to AMD's family. The company posted a gross of 46%, net income of $447 million, and earnings per share of $0.27. While AMD's net income dropped 37% year-over-year, it was primarily due to the amortization of intangible assets associated with the Xilinx acquisition, the chip designer explained.

AMD's Client Computing segment — which includes desktop and notebook CPUs as well as chipsets — remained the company's main cash cow in Q2 with $2,152 billion revenue (up 25% year-over-year) and $676 million operating income (a YoY growth). In addition, the chipmaker noted solid sales of its mobile Ryzen CPUs and their increased average selling prices (compared to the same quarter a year ago), which helped boost client computing segment revenue.

Sales of Datacenter products — CPUs, GPUs, Xilinx FPGAs, and Pensando DPUs — surged to $1,486 billion compared to Q2 2021, an increase of 83%. To a large degree, AMD's data center business unit posted a massive gain as it added FPGAs and GPUs to its lineup. Still, AMD's EPYC processor shipments increased as the company continued to win designs with server makers.

AMD's Gaming business — which includes client GPUs and console system-on-chips — demonstrated mixed results during the quarter. On the one hand, AMD's gaming hardware sales boosted 32% year-over-year to $1,655 billion. But on the other hand, the operating income of the business unit increased by only 7%, to $187 million. AMD said that mixed results resulted from slowing demand for consumer graphics cards by gamers and miners following strong 2021, slowing demand for client PCs, and dropping graphics card prices.

Shipments of AMD's Embedded products (which includes embedded solutions both from AMD and Xilinx) totaled $1,257 billion, an increase of a whopping 2,228% year over year primarily because of the low base effect as AMD's shipments of embedded products were not exceptionally high in the recent years.

AMD expects its revenue to increase to $6.7 billion ± $200 million in Q3 2022, which is 55% higher than Q3 2021. The company expects its revenue growth to be driven primarily by expensive data center products and embedded products from Xilinx. While the company confirmed plans to release its Ryzen 7000-series 'Raphael' processors this fall and Radeon RX 7000-series GPUs based on its RDNA 3 architecture later this year, it did not quantify how these launches improve the results of its Client Computing and Gaming businesses. Meanwhile, AMD must maintain its $26.3 billion ± $300 million revenue guidance for 2022, an increase of 60% over 2021.

Samsung's revised take at a Computational Storage Device (CSD) may be just another key element in High Performance Computing (HPC) environments. Powered by an AMD Xilinx Versal FPGA, Samsung promises its new product will lead to 50% reduced computational times, a 70% lower power consumption, and a reduction in CPU utilization in the order of 97% compared to the cost of running conventional SSD devices for frequently-accessed data. The SmartSSD has just become smarter.

Enabled by the recent AMD-Xilinx marriage (which closed on February 14th just like any mainstream love story), Samsung's second-gen SmartSSD can process data itself. While this may seem like a case of finding a solution for a non-existent problem, it actually solves one of the most pressing bottlenecks in HPC: the flow of data between different elements of the compute path.

Because data is processed directly within the SSD, there's no need for it to be shuffled back and forth between storage banks. Not only does this free up bandwidth for the flow of other data, it also aids on cutting latency, as there's no additional trips between the central processing unit and the data storage device itself. It also frees up the system's CPUs towards more performance-critical tasks rather than processing the read and writing of data towards the CSD's memory banks.

It's a cost that may be negligible for the end-user in 99% of scenarios, but as the amount of processed data increases, so does its reading and writing apply a bigger load on the system. With the extreme compute density of the latest systems being what it is, diverting even a tiny bit of processing towards a component that's comparatively passive when compared to the CPU also aids in better dispersing the generated heat.

Samsung's second-gen SmartSSD features in-built Arm cores - especially well-geared towards the low-power, low-temperature environment of an SSD deployment within a supercomputer.

According to Samsung, the CSD is especially geared towards the AI/ML inferencing market, Business Intelligence analysis, big data analytics, video transcoding, financial services, genomics, search queries and transparent compression.

"Commercialization of the first-generation SmartSSD, in collaboration with AMD, established that the computational storage market has great potential,” said Jin-Hyeok Choi, Executive Vice President and Head of Memory Solution Product & Development at Samsung Electronics in a press release. “With the upgraded processing functionality of the second-generation SmartSSD, Samsung will be able to easily address increasing customer needs in the database and video transcoding sectors, as we expand the boundaries of the next-generation storage market.”

At the very least, it aids in liberating the CPU cores for those extra percentage points of work. With power budgets for supercomputers being what their are, the benefits of that 97% reduction in CPU load compared to a typical SSD will allow more work to be done within the same CPU or GPU.

Mind you, this isn't the first computational SSD we've covered - that honor belongs to the NGD Systems' Chia-AutoPlot CSD, which aimed to give Chia users a way to achieve the same macro-level results as Samsung's SmartSSD, but aimed at mining the Proof of Time and Space Chia cryptocurrency.

According to a report on Nikkei Asia, Intel has informed its clients about its plans to raise prices on its processors and peripheral chips this fall. "According to three industry executives with direct knowledge", the alleged root cause of rising chip production costs are the surging costs of electricity, raw materials, transportation, and labor. Inflation is also seen as another factor that requires Intel to hike its prices. Take the news with a pinch of salt for now.

According to the report, Intel plans to raise prices on virtually all of its products, including processors for client PCs and servers, Wi-Fi chips, and various controllers. The possible hikes could vary on a product by product basis. From a minimal single-digit increase to 10% – 20% on some SKUs. 

Among the products that could be more expensive this fall are 'flagship' CPUs for desktop and notebook PCs (which are among the best CPUs for gaming) and servers, which will have a significant impact on PC makers, server suppliers, and even operators of hyperscale datacenters that buy tens of thousands of processors.

According to Nikkei Asia, Intel confirmed that it had started to inform its customers about impending price hikes, but declined to comment how significantly the company will increase them.

"On its Q1 earnings call, Intel indicated it would increase pricing in certain segments of its business due to inflationary pressures," a statement by Intel published by Nikkei reads. "The company has begun to inform customers of these changes." 

Intel produces the majority of CPUs for client PCs, embedded systems, and servers. In addition, the company is a leading supplier of Wi-Fi, Ethernet (both for client systems and datacenters), FPGAs, and many other products. A major price hike by Intel will inevitably affect the whole industry given the wide range of products its supplies to thousands of its customers worldwide. 

What remains to be seen is whether other industry leaders will also raise their prices following Intel and citing similar reasons. Companies like AMD, Nvidia, Xilinx (now part of AMD), Marvell, Realtek, Phison, Silicon Motion, and many others serve the same or adjacent market segments and experience the same problems as Intel. So, if the giant company increases its quotes, its rivals may choose to maintain their prices, sell more and grab some market share from Intel; or increase their pricing without any fear of losing market share to the blue chipmaker. 

Intel reports its Q2 2022 on July 28, so this is when we might here some additional information about incoming price increases.

According to a new report from Northland, the sales of AMD's processors and graphics chips for client PCs may decline significantly next year. The estimate echoes today's story claiming that AMD will reduce orders to TSMC in the coming quarters due to demand concerns. However, the Northland analyst believes sales of game consoles and AMD's SoCs for PlayStation and Xbox systems will grow in 2023.

Sales of AMD's PC CPUs (which include chips for desktops and laptops) will decrease by 6% year-over-year in 2023, whereas GPU revenue will fab by 7% YoY next year, analyst Gus Richard wrote in a note to investors on Friday, reports SeekingAlpha. In total, AMD will lose $675 million of sales in 2023 due to softening demand for client systems. 

Meanwhile, sales of system chips for Microsoft's Xbox Series X|S and Sony's PlayStation 5 will increase by $400 million (up from the previous forecast of $740 million) to around $1.140 billion next year. This number reflects the fact that sales of game consoles usually triple in their third year of lifecycle and 2023 will be their third year.

On the enterprise side of matters, Northland expects sales of AMD's EPYC server CPUs to continue climbing while gaining share. The investment firm estimates AMD's server revenue to increase by 55% yet-over-year (or by $3.1 billion) in 2023. Any impact from other markets will be offset by sales of server parts, the analyst notes. Shipments of Xilinx field-programmable gate arrays (FPGAs) are also expected to drop by 6%, which is natural for post-merge companies. 

"We scrubbed our AMD numbers and lower our [2023] estimates to comprehend a global economic slowdown that will likely impact PC CPUs, GPUs, game consoles, and XLNX," Richard wrote in a note to clients," wrote Richard.

In general, companies seem to be quite cautious about sales of their client PC hardware due to global inflation, uncertainties caused by the ongoing bloody Russian-Ukrainian war and global geopolitical tensions. Therefore, it is not surprising to see an investment firm report like this one. However, nobody knows what the future brings, so one should take predictions with a grain of salt or sugar, depending on whether you are an optimist or pessimist. 

Law Street Media reported that Japanese professor Masahiro Iida had sued Intel for infringing U.S. Patent No. 6,812,737. The complaint accuses Intel of manufacturing, using, and selling Field Programmable Gate Array (FPGA) and System-on-Chip (SoC) chips that employ Adaptive Logic Modules (ALM), a patent that Iida has held from 2004 to 2014.

When Iida was a doctoral student back in 2001, he had discovered a method to configure large look up tables (LUTs) so that a single M-input N-output LUT can operate as a single “whole” LUT or as a group of “fractured” LUTs. His discovery reportedly helped substantially reduce the implementation area and power consumption for chips that leveraged the innovation.

Iida filed for a patent application on June 29, 2001, and consequently did so in the United States on June 28, 2002. As a result, the United States Patent and Trademark Office (USPTO) issued the '737 patent entitled “programmable logic circuit device having to look up table enabling to reduce implementation area” to Iida on November 2, 2004, which he has been the sole owner since then until October 1, 2014.

Altera and Xilinx are two the big players in the FPGA market. Intel bought Altera for $16.7 billion in 2015, and AMD consequently acquired Xilinx for $54 billion this year. Iida’s complaint pertains to Altera’s Stratix II line of FPGA chips launched in 2004 using ALMs. The company continued to utilize ALMs in its subsequent Stratix chips, including the Stratix III, Stratix IV, Stratix V, and Stratix 10, and in some of its other Arria and Cyclone product lines. After the Altera acquisition, Intel continued to manufacture and commercialize the Stratix, Arria, and Cyclone lineups. In addition, the chipmaker’s Agilex chips also utilize ALMs.

According to the filing, Intel may have benefited significantly from the products, with up to 80% of its annual revenues coming from selling FPGA and SoC that used ALMs. The filing estimated at least $11.5 billion from sales of the accused products over six years (Q3 2016 to 2022).

Professor Iida’s counsel sent a certified letter to Intel’s General Counsel concerning the ’737 patent infringement. Intel has infringed at least claim 1 of the ’737 patent. Despite the plaintiff’s notice, Intel continued to make, offer to sell and sell the accused FPGA chips without a license from Iida.

Iida is looking for monetary compensation equal to or greater than a reasonable royalty that he should have received if Intel had licensed his patent. He also requested that the Court grant him up to three times that amount found by the jury for Intel’s infringement pursuant and his attorney’s fees.

AMD is extending its custom chip design services even as it aims to offer a portfolio that's close and personal to its client's requirements. During its latest analyst day meeting, AMD Chief Technical Officer (CTO) Mark Papermaster spoke of recent advancements in semiconductor fabrication and chip interconnect technologies that would enable a modular future for the company's chip design. It seems that AMD might be looking at carving another strategy for the decades to come: something like "The Future is Custom."

Building upon work that started with custom silicon designs for the Xbox 360 and PlayStation 4, AMD's Semi-Custom Solutions Division has quickly become one of the company's main strengths. AMD adapted its mainstream architectures according to Microsoft and Sony's power, performance, and cost requirements. Even Steam's runaway success, the handheld Steam Deck, has embraced AMD chip designs. But the company wants to take it further.

"We are focused on making it easier to implement chips with more flexibility," said Papermaster, who then expanded on the new philosophy by adding that customers would be invited to bring their own choice of chips towards a tile-like, manufacturable chip structure that can incorporate multiple chiplets wrapped up in a custom chip package. This means AMD customers would be able to pick and choose amongst the company's impressive IP portfolio - ranging from x86 CPUs, GPUs, Arm-based FPGAs through its merge with Xilinx, and even Arm-based networking chips from its Pensando acquisition - while also being able to deploy their own intellectual property.

The new strategy builds upon the recently-introduced Universal Chiplet Interconnect Express (UCIe) standard, which has garnered support from AMD, Intel, Arm, Google, Meta, and others. These companies see the future as an amalgamation of disparate, complementary IPs rather than what each player can singularly offer in-house. UCIe enables a standardized connection between chiplets, like cores, memory, and I/O, in both an in-die and off-die manner that enables performance and latency specifications that are high enough for HPC scaling while relying on well-known protocols like PCIe and CXL.

This makes sense considering the increasing prevalence of custom chip design - bolstered by royalty-free architectures such as RISC-V. Interestingly, RISC-V still hasn't integrated the UCIe standard. Nvidia hasn't either, but the company likely wants to push its proprietary NVLink interconnect as much as possible.

"We're going to make it much easier to add third-party IP as well as customer IP to that chiplet platform," said AMD CEO Lisa Su during the conference. "We've gotten a lot of positive customer engagement so far when you think about hyperscalers, when you think about 5G in automotive," she added. "These are big opportunities where people want to customize and we want to be their partner of choice."

AMD's chiplet-based approach to custom manufacturing is based on its strong relationship with TSMC and leverages the Taiwanese foundries' CoWoS (Chip-on-Wafer-on-Substrate) technology. The same is true for Apple's CoWoS-derivative UltraFusion interconnect. Nvidia's own Hopper GPU is also rumored to be based around that particular bit of TSMC's tech. AMD, naturally, will also be looking to leverage its Infinity Fabric 4.0 architecture, which has been developed for on-die connectivity and hyperscaling in HPC environments.

In what might be one of the most interesting tidbits regarding the future of computing, saying that the company is happy to be ISA (Instruction-Set Architecture) agnostic. Of course, it's easier to be agnostic when a company has the resources, licenses and engineering capabilities to have its feet firmly planted in both dominating and emerging architectures all at once. AMD is capable of this feat thanks to its nearly immaculate execution since Su became CEO.

It does mark a difference in stance compared to the walled garden approach historically preferred by chipmakers. To be fair, there's still a wall - there's a reason only AMD and Intel manufacture high-performance X86-x64 chips. But the gates are now ever so slightly ajar. In the meantime, we await the arrival of AMD's Ryzen 7000 processors and other new products on its consumer CPU roadmaps. 

AMD recently shared a roadmap of its desktop and mobile client processors at Financial Analyst Day 2022. However, it didn't take long for hardware leaker Tum_Apisak to dig up one of the chipmaker's upcoming processors, presumably Phoenix Point.

Phoenix Point is the successor to AMD's Ryzen 6000 (Rembrandt) mobile APUs. Although AMD didn't reveal the exact time frame for Phoenix Point's release, the next-generation chips should hit the market next year. Phoenix Point will see AMD's jump from TSMC's 6nm manufacturing process to the cutting-edge 4nm manufacturing process. The latter offers a 6% smaller die area and will empower chipmakers to optimize their chips' performance and power consumption.

Besides the transition to a smaller process node, Phoenix Point will also wield AMD's mighty Zen 4 cores, which reportedly bring a 10% IPC uplift and more than 35% higher overall performance. However, the more compelling trait will be the APU's integrated graphics. AMD will incorporate its RDNA 3 graphics into Phoenix Point, potentially allowing the APUs to compete with entry-to mid-range discrete mobile graphics. That's big news, considering Phoenix Point competes in the 35W to 45W category, powering thin and light gaming laptops.

As expected, Phoenix Point arrives with support for PCIe 5.0 and DDR5 memory, specifically LPDDR5. Nonetheless, AMD may extend memory support to include other formats in the future. In addition, with Phoenix Point, we'll see the first inclusion of AMD's Artificial Intelligence Engine (AIE), an IP the chipmaker obtained after absorbing Xilinx for $54 billion.

The AMD engineering sample (ES) sports the 100-000000709-23_N codename and belongs to AMD's Family 25, which houses both Zen 3 and Zen 4 chips. Hardware detective Benchleaks found the same chip in the MilkyWay@Home project on the BOINC platform. It's weird, but AMD's unreleased Ryzen processors have started to see a cozy home at the MilkyWay@Home project. Back in January, two Ryzen 7000 (Raphael) showed up in the same project, indeed a weird place to be since it isn't even a benchmark.

According to the entry, which carries today's date, the Phoenix Point chip in question wields 16 threads, meaning it's likely an eight-core part with simultaneous multithreading (SMT). Other than that, the report doesn't tell us anything useful.

We'll likely hear more about Phoenix Point in the upcoming months since the Zen 4 and RDNA 3 APUs will, in all likelihood, debut in 2023 to challenge other chips on our list of best CPUs for gaming.

In the first quarter of 2022, the Top 10 fabless chip designers increased their cumulative sales to $39.43 billion, or by a whopping 44% year-over-year, according to TrendForce. Qualcomm and Nvidia continued to remain on top of the list, but AMD — now in the fourth place — is slowly but surely increasing its sales and has all chances to become the third largest fabless chip designer in the coming quarters.

Being the world's leading supplier of smartphone SoCs and RF modules, Qualcomm enjoyed the natural growth of these businesses in Q1 2022. In addition, the company experienced growth in its automotive and IoT sales, so Qualcomm's earnings for the first quarter totaled $9.548 billion (excluding its licensing business), up 52% from the same quarter a year ago.

Nvidia continued to benefit from the growing demand for discrete graphics processing units for client PCs and datacenter GPUs for AI and high-performance computing applications. Hence, its revenue for Q1 reached $7.904 billion, up 53% annually. By contrast, Broadcom was not that lucky, so its sales 'only' increased by 26% YoY to $6.11 billion.

AMD replaced MediaTek from the No.4 spot in Q1 mainly because it added Xilinx sales to its revenues. Still, AMD's CPU and GPU sales significantly increased in the first quarter due to strong demand for high-performance processors and graphics cards. As a result, AMD's earnings exceeded $5.887 billion in Q1 2022, up 71% from the same quarter last year. Once the company adds Pensando to its results and ramps up production of CPUs, GPUs, and console SoCs due to seasonality later this year, it will probably outpace Broadcom in terms of earnings. It will become one of the world's Top 3 fabless designers of chips, a landmark event for a company that almost went bankrupt about half of a decade ago. Of course, it remains to be seen how smooth integration of Pensando and Xilinx will proceed, but for now, the sky is blue for AMD. 

While MediaTek switched its place with AMD, its Q1 revenue grew a healthy 32% year-over-year to $5.007 billion. The company continues to address all kinds of smartphones and tablets with its mobile SoCs. As its portfolio expands and the performance of application processors increases, the company can charge more significant premiums for its products. 

Marvell was another company in Q1 2022 to demonstrate an over 70% revenue growth year-over-year (to $1.412 billion) mainly due to its acquisition of Innovium, a cloud and edge data center networking solutions designer, in October 2021. Meanwhile, Marvell was not the only company to improve its business results by adopting an M&A strategy. For example, Cirrus Logic (No.10) took over Lion Semiconductor to boost its mixed-signal business in mid-2021, so its Q1 2022 sales reached $490 million, up 67% year-over-year. 

Demand for different chips continues to remain strong, particularly in 5G, automotive, AI, high-performance computing, and edge computing spaces. But the geopolitical and macroeconomic situation may pose challenges even for the seemingly resilient semiconductor market. As a result, it will be fascinating to watch how the leading chip companies will address these challenges in the coming quarters.

AMD announced during its earnings call that it will infuse its CPU portfolio with Xilinx's FPGA-powered AI inference engine, with the first products slated to arrive in 2023. The news indicates that AMD is moving swiftly to incorporate the fruits of its $54 billion Xilinx acquisition into its lineup, but it isn't entirely surprising — the company's recent patents indicate it is already well underway in enabling multiple methods of connecting AI accelerators to its processors, including using sophisticated 3D chip stacking tech.

AMD's decision to pair its CPUs with in-built FPGAs in the same package isn't entirely new — Intel tried the same approach with the FPGA portfolio it gained through its $16.7 billion Altera purchase in late 2015. However, after Intel announced the combined CPU+FPGA chip back in 2014 and even demoed a test chip, the silicon didn't arrive until 2018, and then only in a limited experimental fashion that apparently came to a dead-end. We haven't heard more about Intel's project, or any other derivatives of it, for years. 

AMD hasn't revealed any specifics of its FPGA-infused products yet, but the company's approach to connecting the Xilinx FPGA silicon to its chip will likely be quite a bit more sophisticated. While Intel leveraged standard PCIe lanes and its QPI interconnect to connect its FPGA chip to the CPU, AMD's recent patents indicate that it is working on an accelerator port that would accommodate multiple packaging options.

Those options include 3D stacking chip tech, similar to what it currently uses in its Milan-X processors to connect SRAM chiplets, to fuse an FPGA chiplet on top of the processors' I/O die (IOD). This chip stacking technique would provide performance, power, and memory throughput advantages, but as we see with AMD's existing chips that use 3D stacking, it can also present thermal challenges that hinder performance if the chiplet is placed close to the compute dies. AMD's option to place an accelerator atop the I/O die makes plenty of sense because it would help address thermal challenges, thus allowing AMD to extract more performance from the neighboring CPU chiplets (CCDs).

AMD also has other options. By defining an accelerator port, the company can accommodate stacked chiplets on top of other dies or simply arrange them in standard 2.5D implementations that use a discrete accelerator chiplet instead of a CPU chiplet (see above diagrams). Additionally, AMD has the flexibility to bring other types of accelerators, like GPUs, ASICs, or DSPs, into play. This affords a plethora of options for AMD's own proprietary future products and could also allow customers to mix and match these various chiplets into custom processors that AMD designs for its semi-custom business.

This type of foundational tech will surely come in handy as the wave of customization continues in the data center, as evidenced by AMD's own recently-announced 128-core EPYC Bergamo CPUs that come with a new type of 'Zen 4c' core that's optimized for cloud native applications.

AMD already uses its data center GPUs and CPUs to address AI workloads, with the former typically handling the compute-intensive task of training an AI model. AMD will use the Xilinx FPGA AI engines primarily for inference, which uses the pre-trained AI model to execute a certain function.

Victor Peng, AMD's president of its Adaptive and Embedded Computing group, said during the company's earnings call that Xilinx already uses the AI engine in image recognition and "all kinds" of inference applications in embedded applications and edge devices, like cars. Peng noted that the architecture is scalable, making it a good fit for the company's CPUs.

Inference workloads don't require as much computational horsepower and are far more prevalent than training in data center deployments. As such, inference workloads are deployed en masse across vast server farms, with Nvidia creating lower-power inference GPUs, like the T4, and Intel relying upon hardware-assisted AI acceleration in its Xeon chips to address these workloads.

AMD's decision to target these workloads with differentiated silicon could give the company a leg up against both Nvidia and Intel in certain data center deployments. Still, as always, the software will be the key. Both AMD CEO Lisa Su and Peng reiterated that the company would leverage Xilinx's software expertise to optimize the software stack, with Peng commenting, "We are absolutely working on the unified overall software enabled the broad portfolio, but also especially in AI. So you will hear more about that at the Financial Analyst Day, but we are definitely going to be leaning in AI both inference and training."

AMD's Financial Analyst Day is June 9, 2022, and we're sure to learn more about the new AI-infused CPUs then. 

AMD has patented a processor featuring a machine learning (ML) accelerator that is stacked on top of its I/O die (IOD). The patent indicates that AMD may be planning to build special-purpose or datacenter system-on-chips (SoCs) with integrated FPGA or GPU-based machine learning accelerators.

Just like AMD can now add cache to its CPUs, it might add an FPGA or GPU on top of its processor I/O die. But, more importantly, the technology allows the company to add other types of accelerators to future CPU SoCs. As with any patented work, the patent doesn't guarantee that we'll see designs with the tech come to market. However, it gives us a view into what direction the company is moving with its R&D, and there is a chance we could see products based on this tech, or a close derivative, come to market. 

Stacking AI/ML Accelerator on Top of an I/O Die

AMD's patent titled 'Direct-connected machine learning accelerator' rather openly describes how AMD might add an ML-accelerator to its CPUs with an IOD using its stacking technologies. Apparently, AMD's technology allows it to add a field-programmable processing array (FPGA) or a compute GPU for machine learning workloads on top of an I/O die with a special accelerator port. 

AMD describes several means of adding an accelerator: one involves an accelerator with its own local memory, another implies that such an accelerator uses memory that's connected to an IOD, while in the third scenario, an accelerator could possibly use system memory, and in this case, it does not even have to be stacked on top an IOD. 

Machine learning techniques will be used extensively by future data centers. However, to be more competitive, AMD needs to accelerate ML workloads using its chips. Stacking a machine-learning accelerator on top of a CPU I/O die allows it to significantly speed up ML workloads without integrating expensive custom ML-optimized silicon into CPU chiplets. It also affords density, power, and data throughput advantages. 

The patent was filed on September 25, 2020, a little more than a month before AMD and Xilinx announced that their management teams had reached a definitive agreement under which AMD would acquire Xilinx. The patent was published on March 31, 2022, with AMD fellow Maxim V. Kazakov listed as the inventor. AMD's first products with Xilinx IP are expected in 2023.

We do not know whether AMD will use its patent for real products, but the elegance of adding ML capabilities to almost any CPU makes the idea seem plausible. Assuming that AMD's codenamed EPYC 'Genoa' and 'Bergamo' processors use an I/O die with an accelerator port, there could well be Genoa-AI and Bergamo-AI CPUs with an ML accelerator. 

It is also noteworthy that AMD is rumored to be looking at a 600W configurable thermal design power (cTDP) for its 5th Generation EPYC 'Turin' processors, which is more than two times higher than the cTDP of the current-generation EPYC 7003-series 'Milan' processors. Furthermore, AMD's AMD's SP5 platform for 4th Generation and 5th Generation EPYC CPUs provides up to 700W of power for very short periods to processors. 

We do not know how much power AMD's future 96 – 128 (Genoa and Bergamo) CPUs will need, but adding an ML accelerator into the processor package will certainly increase consumption. To that end, it makes great sense to ensure that next-generation server platforms will indeed be capable of supporting CPUs with stacked accelerators.

Building Ultimate Datacenter SoCs

AMD has talked about data center accelerated processing units (APUs) since it acquired ATI Technologies in 2006. Over the last 15 years, we have heard of multiple data center APU projects integrating general-purpose x86 cores for typical workloads and Radeon GPUs for highly parallel workloads.  

Neither of these projects has ever materialized, and there are many reasons why. To some degree, because AMD's Bulldozer cores were not competitive, it didn't make a lot of sense to build a large and expensive chip that could be in very limited demand. Another reason is that conventional Radeon GPUs did not support all data formats and instructions required for data center/AI/ML/HPC workloads, and AMD's first compute-centered CDNA-based GPU only emerged in 2020.  

But now that AMD has a competitive x86 microarchitecture, a compute-oriented GPU architecture, a portfolio of FPGAs from Xilinx, and a family of programmable processors from Pensando, it may not make much sense to put these diverse IP blocks into a single large chip. Quite the contrary, with today's packaging technologies offered by TSMC and AMD's own Infinity Fabric interconnection technology, it makes much more sense to build multi-tile (or multi-chiplet) modules featuring general-purpose x86 processor chiplets, an I/O die as well as GPU or FPGA-based accelerators.  

In fact, it makes more sense to build a multi-chiplet data center processor rather than a large monolithic CPU with built-in diverse IP. For example, a multi-tile datacenter APU could benefit from a CPU tile made using TSMC's N4X performance-optimized node as well as a GPU or FPGA accelerator tile produced using a density-optimized N3E process technology.

Universal Accelerator Port

Another important part of the patent is not a particular implementation designed to accelerate machine learning workloads using an FPGA or a compute GPU, but the principle of adding a special-purpose accelerator to any CPU. The accelerator port will be a universal interface presented on AMD's I/O dies, so eventually, AMD could add other types of accelerators to its processors aimed at client or data center applications.  

"It should be understood that many variations are possible based on the disclosure herein," a description of the patent reads. "Suitable processors include, by way of example, a general-purpose processor, a special purpose processor, a conventional processor, a graphics processor, a machine learning processor, [a DSP, an ASIC, an FPGA], and other types of integrated circuit (IC). […] Such processors can be manufactured by configuring a manufacturing process using the results of processed hardware description language (HDL) instructions and other intermediary data including netlists (such instructions capable of being stored on a computer readable media)." 

While FPGAs, GPUs, and DSPs could be used for a variety of applications even today, things like data processing units (DPUs) for data centers will only gain importance in the coming years. DPUs are essentially an emerging application that AMD now happens to have. But as the data center transforms to process even more types of data and faster (and so do client PCs, like how Apple integrates application-specific acceleration, like for ProRes RAW, into its client SoCs), accelerators are becoming more common. That means there must be a way to add them to any, or almost any, server processor. Indeed, AMD's accelerator port is a relatively simple way of doing so.

In recent years one of AMD's top priorities was its datacenter business powered by its highly competitive EPYC processors. To further boost the business , AMD acquired Xilinx, a leading developer of field programmable arrays (FPGAs). In another move to boost the company's datacenter business AMD announced [PDF] it had reached an agreement to buy Pensando, a designer of fully programmable processors in a deal valued at $1.9 billion. 

So what is a fully programmable processor and how different it is from an FPGA?
The answer is in proprietary software and hardware, it is both complex and easy. With an FPGA, you have a lot of flexibility for managing security networking or data processing, but for specific special-purpose tasks you need additional hardware. This is where Arm cores and fixed-function hardware comes in Xilinx FPGAs (and AMD now has Xilinx). With Pensando, AMD gets another type of IP: programmable processors and applicable software that can operate it.  

In fact, software is perhaps the key reason for the acquisition. AMD says that the solutions platform that it gets includes a "programmable packet engine and complete software stack that accelerates networking, security, storage and other services for cloud, enterprise and edge applications." 

AMD says that with the addition of Pensando, the company will have the capability to innovate at the "chip, software and platform level and deliver optimized solutions with unmatched performance and value for our cloud and enterprise customers." 

"The combination of our CPUs, GPUs, FPGAs, and adaptive computing engines with Pensando's packet processor and software technologies will enable AMD to offer a broad portfolio of compute engines that have been optimized for different workloads," a statement by AMD says. "We believe this is going to be a critical requirement to deliver the performance, power efficiency and capabilities required to power the next generation of accelerated data centers at scale."

While we can only speculate about the chip level innovation that AMD suggests, there is a clear strategic move on AMD's side: 

"To build a leading-edge data center with the best performance, security, flexibility and lowest total cost of ownership requires a wide range of compute engines," said Dr Lisa Su, chief executive of AMD. "All major cloud and OEM customers have adopted EPYC processors to power their data center offerings. Today, with our acquisition of Pensando, we add a leading distributed services platform to our high-performance CPU, GPU, FPGA, and adaptive SoC portfolio. The Pensando team brings world-class expertise and a proven track record of innovation at the chip, software and platform level which expands our ability to offer leadership solutions for our cloud, enterprise and edge customers." 

There are other reasons why AMD is particularly interested in Pensado, as it is interested in any other enterprise-related companies: margins, quotes, and allocations at foundries. 

Selling high-end CPUs for servers and desktops enables AMD to post a ~50% gross margin, something that is good enough to compare to other fabless companies, but not good enough to compare to firms like Nvidia. Datecenter products bring in higher margins and higher prices, which is why AMD wants Pensando.

Intel announced its new Agilex M-Series of FPGA (Field-Programmable Gate Array) products, and at the same time gave some very interesting positioning toward cryptocurrency mining. Intel is claiming several "world-firsts" with the Agilex M-series, including usage of the Intel 7 (previously Intel 10 nm) manufacturing process, support for up to 32 GB HBM2e DRAM (via twin memory stacks), and a 60% performance increase compared to previous-generation Stratix FPGAs. No claims on cryptocurrency mining performance were made by Intel, however.

Designed by Intel's Programmable Solutions Group (PSG), which incorporated the talent and design IP from Intel's $16.7 billion Altera acquisition back in 2015, the new Agilex M-Series promise to offer the highest memory bandwidth in the industry. This is thanks to a pair of HBM2e stacks configurable up to 32 GB, which provide up to 820 GBps in memory bandwidth (410 GBps per stack). That could prove particularly useful for Dagger Hashimoto / Directed Acyclic Graph (DAG) workloads frequently associated with cryptocurrency mining, e.g. Ethereum.

Additionally, intel says its new Agilex M-Series are the first FPGAs supporting cache and memory coherency with Intel's Xeon CPUs via the Compute Express Link interconnect standard. Agilex-M FPGAs are also the world's only solutions capable of operating with three types of memory: HBM2e, DDR5, and Intel Optane — though to be fair, Intel has locked other vendors out of Optane implementations in the past.

The new Agilex M-Series FPGAs offer the industry's highest memory bandwidth for an FPGA, thanks to the inclusion of HBM2e memory and an external DDR5/LPDDR5 memory bank, all controlled via Agilex M's hardened memory controllers. Intel also claims the industry's highest Digital Signal Processing (DSP) compute capabilities in an HBM-enabled FPGA, though note that Intel doesn't claim the highest compute density in non-HBM-enabled FPGAs.

The same is true regarding Intel's efficiency claims. The company says its Agilex M products offer over twice the fabric performance per watt vs competitive 7nm FPGAs (read, Xilinx's Versal). The devil, as always, is in the details: Intel isn't claiming twice the performance per watt of its competitor's offerings, only fabric efficiency, likely boosted by the company's usage of its EMIB technology.

This is the first time an Intel FPGA product is being marketed toward cryptocurrency-related workloads, and it's the second Intel product in a single month looking to enter the blockchain space, after Intel's Bonanza Mine ASIC (Application-Specific Integrated Circuit) was made public. Intel seems to be adopting a broader strategy when it comes to blockchain workloads, and the company will be offering at least three distinct product segments that serve this space. While the Bonanza Mine ASIC and its subsequent iterations are a given, the new Agilex M-Series FPGAs are also joined by Intel's upcoming Arc Alchemist, which, as the company pointed out, will offer full performance in cryptocurrency mining workloads.

That Intel is marketing FPGAs for mining workloads doesn't detract from the more conventional use-cases for these chips, which include compute, cloud-to-edge infrastructure, pervasive connectivity, and AI. FPGAs are particularly desired due to their high flexibility, which allows for businesses to quickly iterate on semiconductor designs on the same piece of silicon, instead of spending the resources in multiple tape-outs of different approaches, solutions, and stages of a given chip's design.

It is interesting that Intel is broadening the appeal of its FGPAs toward markets they previously weren't in. This is happening against the backdrop of the apparently diminishing competitiveness of its Altera-based products compared to those made by Xilinx, which was recently acquired by AMD, turning it into a bigger company (by market capitalization) than even Intel.

By itself, Xilinx showcased an extremely impressive 2021, raking in $3.15 billion in revenue for a cool 20% increase YoY. Intel's PSG group, in contrast, brought in $1.9 billion in revenue, with only a 4% revenue increase YoY, signaling that the market found Xilinx's offerings to be preferable over Intel's solutions. However, Intel claimed that its PSG group could have brought in an additional $500 million in revenue were it not for the supply constraints felt throughout the semiconductor market at large — supply constraints that interestingly weren't mentioned in Xilinx's own earnings report.

It remains to be seen if Intel's "new" FPGA market, blockchain, will account for anything more than a product release marketing move, or if the company's solutions can indeed bring added value to a market dominated by ASICs.

AMD is now a bigger company, by market capitalization, than Intel. It may be close, but AMD has passed Intel for the first time in the company's history with a ~$197.75 billion market cap at the close of the market on February 15 compared to Intel's $197.24 billion. AMD's sudden market cap surge comes on the back of its blockbuster $49 billion acquisition of Xilinx, the largest semiconductor acquisition in history.

AMD's acquisition of Xilinx triggered the conversion of 248.38 million Xilinx shares into 428 million new AMD shares (a process that is ongoing). Added to AMD's existing 1.2 billion shares, that brings the company's overall share count to 1.628 million, giving AMD a market cap of $197.75 billion that squeaks past Intel by a mere $51 million. (There might be some variance in calculations from the various third parties, but all should place AMD in the lead over Intel at current stock valuations.) 

That marks a stark reversal for a company that was teetering on the brink of bankruptcy a mere six years ago when it first unveiled its revolutionary Zen CPU microarchitecture. Six years of relentless execution later, AMD now has an all-time high share of the CPU market, giving it the cash to ink the biggest semiconductor deal in history. 

However, we do have to remember that Intel is the larger company by far and commands ~75% of the overall x86 market share. Intel also generates far more revenue and profit per year than AMD and is coming off its sixth year of record revenue. That's not to mention that the firm makes its own chips and owns a global network of fabs, whereas AMD designs chips and outsources production.

Taken in context, AMD's valuation signals that the market generally has more optimism about the firm's growth prospects than Intel's. There are plenty of reasons to be bullish: AMD's recent acquisition of FPGA-maker Xilinx brings a broad portfolio of differentiated silicon products under the company's umbrella, opening up new, profitable climes like automated driving, aerospace, 5G/communications, and IoT markets, to name a few. 

Xilinx is a healthy company, too, as its recent 3Q earnings included a record quarterly revenue of $1 billion, a 26% increase year-over-year. Meanwhile, Intel's comparable unit, its Programmable Solutions Group (PSG) that sprung from its $16.7 billion Altera acquisition, has had more than its fair share of struggles over the last few years as Xilinx has eaten away at its market share. Intel also revealed during its last earnings call that its PSG unit has been abnormally impacted by supply constraints over the course of 2021, causing a predicted loss of $500 million in additional revenue.

Meanwhile, AMD is hitting the ground running with Xilinx — the company expects its first 'blended' products, meaning devices with both AMD and Xilinx logic, to come to market in 2023.

Times are good for AMD, it also surpassed Qualcomm on the market cap list, but Nvidia is also having a stellar run as well — the company's $662.38 billion market cap exceeds both AMD and Intel, combined.

AMD's new market cap is a far cry from July 2020 when the company's stock price surpassed Intel's for the first time in 15 years but it only had a market cap of $74 billion compared to Intel's $260 billion. Back then, Intel's stock was worth $61.57 compared to AMD's $61.79, but the picture is quite a bit different now with Intel's stock weighing in at $48.44 compared to AMD's $121.47.

Intel certainly isn't content to see its market share being whittled away and has shown signs of a comeback in the client market. Intel CEO Pat Gelsinger has also bet the future of the company on a new vision that includes producing chips for other firms via its IDM 2.0 strategy and has boosted the company's R&D spending to recent highs. We'll learn more about those efforts in a few days during Intel's investor event on February 17.

AMD on Monday completed the acquisition of Xilinx, creating a company that can offer various types of compute devices, including CPUs, GPUs, and FPGAs. The move will help AMD to continue expanding its presence in the datacenter sector and offer unique solutions that will combine IP ingredients designed by the two companies. Interestingly, the first fruits of the deal are expected to materialize next year. 

AMD CEO Dr. Lisa Su told analyst Patrick Moorhead that the first processor combining Xilinx technologies will arrive in 2023, which is in contrast to the company's previous integration efforts. After AMD bought ATI Technologies in 2006, it took the company five years to build its first accelerated processing units (which included AMD's x86 cores and ATI's GPU). This time AMD inked a long-term development pact with Xilinx and was able to work collaboratively even before the regulators approved the transaction. It remains to be seen what exactly AMD plans to offer, but it is reasonable to expect the new processor to feature AMD's x86 cores and Xilinx's programmable engines.  

The merger between AMD and Xilinx is one of the cases when the combined company represents more than the sum of its parts. AMD sells CPUs for client PCs and servers, graphics processors for client PCs and HPC/datacenters, semi-custom system-on-chips for game consoles, and semi-custom SoCs for embedded applications in the automotive and entertainment industries. AMD's products rarely overlap with those from Xilinx, which are used in aerospace, analytics, AI inference, automotive, communications, datacenter, HPC, industrial, and many other verticals.  

There are certain industries when AMD's high-performance CPUs or GPUs compete against Xilinx's FPGAs or special-purpose ASICs, but in most cases, AMD's CPUs and Xilinx chips/software can complement each other in one system. Now that AMD can offer CPUs, GPUs, FPGAs, and adaptive SoCs, its commercial platforms can become significantly more competitive. 

 "The acquisition of Xilinx brings together a highly complementary set of products, customers and markets combined with differentiated IP and world-class talent to create the industry's high-performance and adaptive computing leader," said Lisa Su, chief executive of AMD. "Xilinx offers industry-leading FPGAs, adaptive SoCs, AI inference engines and software expertise that enable AMD to offer the strongest portfolio of high-performance and adaptive computing solutions in the industry and capture a larger share of the approximately $135 billion market opportunity we see across cloud, edge, and intelligent devices." 

Of course, it will take time to integrate Xilinx into AMD and build platform offerings comprising both companies' hardware. But even the journey of a thousand miles begins with the first step, which is what the two companies formally start today. Surprisingly, the two companies expect quite rapid progress in developing products encompassing IP from AMD and Xilinx, which is a result of collaborative work that spanned for well over a year. Meanwhile, AMD's and Xilinx's roadmaps will not be affected by the transaction, so all the products previously announced in one way or another will be developed and will come to market.

The Xilinx business will become AMD's Adaptive and Embedded Computing Group (AECG), led by former Xilinx CEO Victor Peng. As a result, Xilinx will maintain its leadership for at least a while. Furthermore, AMD's embedded business will cease to be a part of the company's enterprise and semi-custom unit and will merge into AECG, which might be good news as executives from the enterprise division will now spend more time on EPYC CPUs. 

In a statement today, AMD confirmed it "has received approval from all necessary authorities to proceed with the acquisition of Xilinx." All that remains now is the minor housekeeping referred to as "customary closing conditions," before the deal is done. The final price of the all-stock deal weighs in at $54 billion at current valuations. The transaction is expected to close on Valentine's Day. That's Monday, Feb 14, to those not au fait with the calendar of romance.

We have been watching AMD's pursuit of Xilinx with interest since the original announcement in October 2020. There was a brief delay to proceedings at the end of last year, but the deal has been relatively plain sailing compared to Nvidia's failed effort to purchase Arm. The final hurdle was successfully taken in stride a month ago.

$35 Billion All Share Deal Now Worth $54 Billion

Thanks to AMD's significant share price appreciation since the all-share deal was announced, it is worth another look at the sums that will be involved in the close. Back in October 2020, we reported on a deal valuation in the region of $35 billion. Xilinx shareholders will be receiving 1.7234 shares of AMD common stock for each share of Xilinx common stock they hold at the closing of the transaction.

This originally equated to $143 per share of Xilinx common stock, but AMD stock at the time was only worth around $85 per share, now it is worth over $130, which is a 53% increase. If we apply this increase to the $35 billion, the deal will, as it closes, cost AMD about $54 billion. This sum makes Xilinx a significantly more expensive acquisition than we could have known or would have expected back in October 2020.

Our readers are very familiar with AMD, it's one of the holy trinity of the PC industry, but many will be less familiar with Xilinx. The soon-to-be-acquired tech company is primarily known for its FPGA chips, which will add an important string to AMD's bow, and is a great complementary technology to fold in alongside its CPUs, GPUs, APUs, and SoCs.

When the deal was first announced, AMD CEO Dr. Lisa Su set out her vision concerning the intended purchase. She said that the acquisition could "establish AMD as the industry's high-performance computing leader." Moreover, she described the future merged entity as "truly a compelling combination that will create significant value for all stakeholders."

Xilinx has worked closely with AMD on multiple previous products, and this looks like it will be a blissful marriage, so tying the knot on Valentine's Day seems rather apt.

AMD's plans to acquire Xilinx appear to have crossed a significant hurdle as China's market regulator has given its conditional approval. The news, which comes via Reuters, means that with this decision, the $35 billion all-stock deal looks like it will now be able to speed towards completion. Last month, we reported the timeline had slipped to Q1 2022, noting that the Chinese regulators were looking very closely at anti-monopolistic conditions they might apply before approval.

The regulator, China's State Administration for Market Regulation (SAMR), has said that the deal will be approved on the condition that AMD and Xilinx do not force or discriminate against customers that buy certain products.

The table below collates the conditions which AMD will have to agree with to ensure its Xilinx acquisition is rubber-stamped. There are five key obligations imposed by SAMR:

The above table is based upon the machine-translated SAMR news post, so there's a possibility for a slight misinterpretation in the obligation column. However, we can summarize that China wants the merged entity to continue business as it did before the tie-up. That means sales of the respective parts will continue to be available in China without any restrictive bundling. Moreover, an important point it wishes to press is that AMD doesn't restrict or obstruct Xilinx's continued work with other CPU and GPU architectures in any way.

AMD must accept the legally binding conditions laid out by SAMR, and the merged entity will be watched closely through the lens of China's anti-monopoly laws by SAMR.

Xilinx's purchase by AMD should be broadly welcomed by investors, as it may help AMD compete better in the data center market, to name one important business segment. However, the China market regulator's approval looks like it was already priced in by AMD investors (who haven't had the best couple of months), as the share price hasn't really reacted in pre-market trading.

AMD has also weighed in on the developments, saying: 

We continue making good progress on the required regulatory approvals to close our transaction. While we had previously expected that we would secure all approvals by the end of 2021, we have not yet completed the process and we now expect the transaction to close in the first quarter of 2022. Our conversations with regulators continue to progress productively, and we expect to secure all required approvals.

There are no additional changes to the previously announced terms or plans regarding the transaction and the companies continue to look forward to the proposed combination creating the industry’s high-performance and adaptive computing leader.

The company also filed an 8K, saying, "AMD and Xilinx currently anticipate that the closing of the Merger will occur in the first quarter of 2022, subject to the expiration of the waiting period under the HSR Act and the satisfaction (or, to the extent permitted by applicable law, waiver) of the conditions set forth in the Merger Agreement that by their nature are to be satisfied at the closing of the Merger."

AMD's $35 billion acquisition of Xilinx is now expected to close in the first quarter of 2022, which is a few months after AMD and Xilinix's originally proposed closing of the deal before the end of 2021.

AMD has cleared regulatory hurdles in all relevant agencies except China's SAMR. AMD entered the second phase of the approval process in China in July. Earlier this month, news filtered out that AMD had filed certain undefined 'behavioral remedies' to appease Chinese regulators. By mid-month, news emerged that China's antitrust agency, the State Administration for Market Regulation (SAMR), was still market-testing AMD's proposed remedies.

AMD CEO Lisa Su has expressed confidence that the deal would close in 2021, telling us during an interview earlier this year that completing the Xilinx deal was one of her key goals for 2021. She has reiterated that optimism over the last few months, and regulators in the UK and the US have approved the deal. For now, the slight delay appears to be more a matter of timing rather than any structural problems with the proposed acquisition that would scuttle the deal.

AMD's acquisition of Xilinx comes during a time of heightened tension between the US and Chinese governments as they remain embroiled in a trade war. Still, China's recent approval of Intel's sale of its NAND and SSD businesses to SK hynix indicates that such mergers are still possible, albeit with some restrictions. Meanwhile, Nvidia's proposed blockbuster $40 billion acquisition of Arm is all but dead due to regulators, but that deal faces challenges in nearly every geography due to very broad concerns of potential anti-competitive behavior. AMD's proposed deal hasn't faced any opposition approaching the challenges surrounding the Nvidia/ARM tie-up.  

Here's AMD's statement on its adjusted expectations for the acquisition:

"SILICON VALLEY, Calif. — Dec. 30, 2021 — AMD (NASDAQ: AMD) and Xilinx, Inc. (NASDAQ: XLNX) today released the following statement related to the status of global regulatory approvals for AMD's proposed acquisition of Xilinx.

"We continue making good progress on the required regulatory approvals to close our transaction. While we had previously expected that we would secure all approvals by the end of 2021, we have not yet completed the process and we now expect the transaction to close in the first quarter of 2022. Our conversations with regulators continue to progress productively, and we expect to secure all required approvals."

There are no additional changes to the previously announced terms or plans regarding the transaction and the companies continue to look forward to the proposed combination creating the industry's high-performance and adaptive computing leader."

A new world record for the densest arrangement of RISC-V cores (measured by the CoreScore benchmark) has been achieved by pairing 6,000 RISC-V SERV cores and one of Xilinx's most powerful designs with an onboard FPGA, the VCU128 evaluation board that comes with a VU37P FPGA. The benchmark simulates how many SERV cores can be deployed on a single piece of silicon, and the Xilinx's Virtex UltraScale+ VU37P FPGA can fit as many as 6,000 SERV cores via its internal reconfiguration. The previous record-holder had a total of 5,087 cores hosted on Xilinx's VCU118.

FPGAs (Field-Programmable Gate Array) are exotic pieces of hardware because they have very few fixed-function elements. Instead, they are built to be programmable on the fly (or in the field). This essentially allows FPGAs to be adaptive processing electronics, changing from moment to moment according to the workload at hand (this is a simplified explanation).

"What do you do when you have the award-winning SERV, the world's smallest RISC-V CPU?" asks Olof Kindgren, designer of both the SERV core and the CoreScore benchmark. "Well, among other things, we, of course, want to see how many SERV cores you can fit into various devices. This is what CoreScore is for. And on top of that list of currently 30 boards, we can now find Sylvain Lefebvre and his Xilinx VCU128 board that fits 6000 SERV cores."

These cores aren't what you'd typically find on your best CPUs for gaming from Intel or AMD; they are stripped-down, barebones bit-serial work units that include as few extraneous functions as possible. That approach minimizes the total die space occupied by each core. The design achieves performance via workload parallelization, not from the obvious processing grunt from each core.

"We are nearing the max," Lefebvre says of his 6,000-core record, "with 98.5% LUTs [Lookup Tables] (and 100% BRAM [Block RAM]) of the VCU128 FPGA utilized. It's been great fun working with Olof Kindgren on this, and it was a perfect intro to our Xilinx VCU128 monster." 

DigiTimes reports that AMD and Mediatek may be in talks to establish a Joint Venture (JV). According to the publication, the JV would be focused on SoC-level data transmission technologies, such as 5G, Wi-Fi and even wired data transmission controllers. Should the joint venture materialize, we would be looking at Mediatek IP being integrated into AMD SoCs - with the focus apparently being placed on mobility hardware solutions such as laptops.

Sources close to DigiTimes reportedly point towards a 2024 market availability for any SoC designs that result from this joint venture. Lending credence to this report is the fact that both AMD and Mediatek have collaborated in the past: AMD integrated Mediatek's Wi-Fi 6 technology into their AMD RZ608 chip, which debuted in the AYANEO handheld games console. While the module is AMD-branded, it is in fact a rebranded version of Mediatek's MT7921K wireless communications module.

According to DigiTimes, AMD is looking to make its notebook solutions more attractive by integrating data transmission technologies; the Taiwanese Mediatek, which has become one of the foremost mobile SoC developers and which holds a rich 5G portfolio, seems like a natural alternative to Intel-developed solutions - competition between AMD and Intel is fierce enough that it's only natural that the Red Team wouldn't want to improve Intel's bottom-line by adopting the company's solutions.

Another interesting element to point out is that AMD's acquisition of Xilinx also opens up the door for IP collaborations in the wired and wireless high-speed data transmission technology world; perhaps the joint venture would see AMD and Mediatek leverage IP from both parties. DigiTimes' sources do seem to think that the joint venture between both companies would also be looking to sell their developed SoC solutions to third parties, instead of limiting its results to in-house products. However, Mediatek is one of the foremost providers of mobile SoCs (it surpassed Qualcomm in shipments for the global chipset market in 2020), and has a much deeper understanding of Wi-Fi 6 and related chip integration designs, as well as data on the manufacturing and supply infrastructure for these technologies. AMD would be able to cut out Intel-made data communications solutions, and MediaTek would increase its foothold in the notebook segment - an apparent win-win for both companies.

AMD has in the meantime responded to the DigiTimes report, saying it does not comment on market speculation; MediaTek, on the other hand, overtly denied the report - both standard responses when it comes to industry insider leaks.

If Samsung has its way, the memory chips in future desktop PCs, laptops, or GPUs could think for themselves. At Hot Chips 33, Samsung announced that it would extend its processing-in-memory technology to DDDR4 modules, GDDR6, and LPDDR5X in addition to its HBM2 chips. Earlier this year, Samsung announced its HBM2 memory with an integrated processor that can compute up to 1.2 TFLOPS for AI workloads, allowing the memory itself to perform operations usually reserved for CPUs, GPUs ASICs, or FPGAs. Today marks more forward progress with that chip, but Samsung also has more powerful variants on the roadmap with its next-gen HBM3. Given the rise of AI-based rendering techniques, like upscaling, we could even see this tech work its way into gaming GPUs.

Today's announcement reveals the official branding of the Aquabolt-XL HBM2 memory, along with a reveal of the AXDIMM DDR4 sticks and LPDDR5 memory that also come with embedded compute power. We covered the nitty-gritty details of the first HBM-PIM (Processing-In-Memory) chips here. Put simply, the chips have an AI engine injected inside each DRAM bank. That allows the memory itself to process data, meaning that the system doesn't have to move data between the memory and the processor, thus saving both time and power. Of course, there is a capacity tradeoff for the tech with current memory types, but Samsung says that HBM3 and future memories will have the same capacities as normal memory chips.

Samsung's Aquabolt-XL HBM-PIM slots right into the company's product stack and works with standard JEDEC-compliant HBM2 memory controllers, so it's a drop-in replacement for standard HBM2 memory. Samsung recently demoed this by swapping its HBM2 memory into a standard Xilinx Alveo FPGA with no modifications to the card, netting a 2.5X system performance gain with a 62% reduction in energy consumption.

While Samsung's PIM tech is already compatible with any standard memory controller, enhanced support from CPU vendors will result in more performance in some scenarios (like not requiring as many threads to fully utilize the processing elements). Samsung tells us that it is testing the HBM2-PIM with an unnamed CPU vendor for use in its future products. Of course, that could be any number of potential manufacturers, be they on the x86 or Arm side of the fence — Intel's Sapphire Rapids, AMD's Genoa, and Arm's Neoverse platforms all support HBM memory (among others).

Naturally, Samsung's PIM tech is a good fit for data centers, largely because it is a great fit for memory-bound AI workloads that don't have heavy computation, like speech recognition. Still, the company also envisions the tech moving to more standard climes too. To that end, the company also demoed its AXDIMM, a new acceleration DIMM prototype that performs processing in the buffer chip. Like the HBM2 chip, it can perform FP16 processing using standard TensorFlow and Python code, though Samsung is working feverishly to extend support to other types of software. Samsung says this DIMM type can drop into any DDR4-equipped server with either LRDIMMs or UDIMMS, and we imagine that DDR5 support will follow in due course.

The company says its tests (conducted on a Facebook AI workload) found a 1.8X increase in performance, a 42.6% reduction in energy consumption, and a 70% reduction in tail latency with a 2-rank kit, all of which is very impressive—especially considering that Samsung plugged the DIMMs into a standard server without modifications. Samsung is already testing this in customer servers, so we can expect this tech to come to market in the near future.

Samsung's PIM tech is transferable to any of its memory processes or products, so it has even begun experimenting with PIM memory in LPDDR5 chips, meaning that the tech could come to laptops, tablets, and even mobile phones in the future. Samsung is still in the simulation phase with this tech. Still, its tests of a simulated LPDDR5X-6400 chip claim a 2.3X performance improvement in speech recognition workloads, a 1.8X improvement in a transformer-based translation, and a 2.4X increase in GPT-2 text generation. These performance improvements come paired with a 3.85X, 2.17X, and 4.35X reduction in power, respectively.

This tech is moving rapidly and works with standard memory controllers and existing infrastructure, but it hasn't been certified by the JEDEC standards committee yet, a key hurdle that Samsung needs to jump before seeing widespread adoption. However, the company hopes that the initial PIM spec is accepted into the HBM3 standard later this year.

Speaking of HBM3, Samsung says that it will move forward from the FP16 SIMD processing in HBM2 to FP64 in HBM3, meaning the chips will have expanded capabilities. FP16 and FP32 will be reserved for data center usages, while INT8 and INT16 will serve the LPDDR5, DDR5, and GDDR6 segments.

Additionally, you lose half the capacity of an 8GB chip if you want the computational power of HBM2 PIM, but there will be no such capacity tradeoffs in the future: The chips will have the full standard capacity regardless of the computational capabilities.   

Samsung will also bring this capability to other types of memory, like GDDR6, and widen the possible applications. CXL support could also be on the horizon. Samsung says its Aquabolt-XL HBM2 chips are available for purchase and integration today, with its other products are already working their way through the developmental pipeline.

Who knows, with the rise of AI-based upscaling and rendering techniques, this tech could be more of a game-changer for enthusiasts than we see on the surface. In the future, it's plausible that GPU memory could handle some of the computational workloads to boost GPU performance and reduce energy consumption.

AMD's proposed $35 billion acquisition of Xilinx is reportedly moving through the Chinese regulatory process. Market reporting organization MLex reported that the deal is progressing to Phase Two of the State Administration for Market Regulation (SAMR) review process.

MLex's report arrived shortly after the UK Competition and Markets Authority (CMA) approved the AMD-Xilinx deal. The acquisition won't be able to move forward until it's approved by other regulators, however, most notably in the U.S. and China.

SAMR's process involves up to three review periods: Phase One lasts 30 days, Phase Two lasts 60 days, and Phase Three lasts 90 days. The regulator decides at the end of each phase if it's ready to make a decision or if it will move on to the next phase.

Moving on to Phase Two doesn't necessarily spell trouble for AMD; it simply means SAMR wants more information about the deal's potential impact before it reaches a decision. AMD reportedly still believes the deal will close by the end of the year.

There shouldn't be many remaining obstacles to the acquisition. AMD and Xilinx said last year that both of their boards of directors unanimously approved the deal, and  their shareholders also voted in favor of it when it was put to a vote in April.

AMD said the combined company would employ more than 13,000 engineers and invest more than $2.7 billion annually in R&D. Xilinx is expected to contribute its expertise with deep learning to AMD's efforts to expand in that sector.

The UK Competition and Markets Authority (CMA) today announced its approval of AMD's plan to acquire Xilinx in an all-stock transaction worth roughly $35 billion.

The CMA didn't offer many details about its decision at time of writing. "The CMA has cleared the anticipated acquisition by Advanced Micro Devices, Inc. of Xilinx, Inc.," it said in today's update. "The full text of the decision will be available shortly."

AMD announced its plans to acquire Xilinx in October 2020. The deal would give AMD shareholders a 74% stake in the combined company; Xilinx shareholders would own the other 26%. Both companies' boards of directors unanimously approved the deal.

The companies then announced in April that their shareholders "overwhelmingly" approved of the deal, and AMD said it expected regulators to clear the acquisition by the end of the year. The CMA's decision brings that expectation closer to reality.

AMD has said the combined company would invest over $2.7 billion in R&D annually to continue development of CPUs, GPUs, FPGAs, SoCs, and other technologies with more obscure initialisms. Xilinx's experience with deep learning—on which the companies have collaborated in the past—is likely vital to AMD's plans for the future.

This isn't the only major acquisition the CMA has considered this year. The regulator is also supposed to prepare a report on Nvidia's proposed acquisition of Arm by July 31. Let's see if recent support for that acquisition will lead to a similar decision.

The UK Secretary of State for Digital, Culture, Media and Sport on Monday began to probe AMD's proposal to acquire Xilinx for $35 billion. The CMA launched an inquiry inviting interested parties to notify it if the potential transaction can negatively impact UK's consumers or businesses. The deadline for the first phase of the investigation is July 6, 2021. 

"The Competition and Markets Authority (CMA) is considering whether it is or may be the case that this transaction, if carried into effect, will result in the creation of a relevant merger situation under the merger provisions of the Enterprise Act 2002 and, if so, whether the creation of that situation may be expected to result, in a substantial lessening of competition within any market or markets in the United Kingdom for goods or services," the statement by the CMA reads.  

Last October AMD and Xilinx announced that they reached a preliminary agreement for AMD to acquire Xilinx for $35 billion in an all-stock transaction. Shareholders of both companies approved the deal this April. 

AMD needs Xilinx, a leading developer of FPGAs, to better compete with Intel, which owns Altera. Both AMD and Xilinx could survive and prosper independently, but the management of these companies believes that the combined company will be stronger than the two separate businesses. 

With the takeover of Xilinx, AMD will be able to offer three types of compute devices used in datacenters: CPUs, GPUs, and FPGAs. Furthermore, it will be able to develop hybrid products combining general-purpose processor cores, compute GPU capabilities and field-programmable transistors.

AMD and Xilinx announced Wednesday that their respective shareholders "overwhelmingly" approved of AMD's plan to acquire Xilinx by the end of the year.

Those shareholders essentially followed the companies' board of directors, both of which unanimously approved the deal when it was announced in October 2020. Now the deal's main obstacle is regulatory approval, which the companies expect to get.

The acquisition would be an all-stock transaction through which AMD shareholders would own 74% of the combined company and Xilinx shareholders would own 26%. AMD's share price was about $143 when the deal was announced; it's currently $82.

AMD said yesterday that the deal "will bring together two industry leaders with complementary product portfolios and customers, combining CPUs, GPUs, FPGAs, Adaptive SoCs and deep software expertise" in a single market-leading company.

The companies have worked together on deep learning projects before, but this acquisition would reportedly lead to an increased annual R&D spend of $2.7 billion, and the combined company would employ an estimated 13,000 engineers.

The deal is also expected to give AMD more leverage with TSMC by making it an even bigger part of the foundry's business. Apple would still be TSMC's largest customer, but AMD could cement its place as second-largest, and the Xilinx deal would help.

"The closing of the transaction remains subject to the satisfaction of other customary closing conditions," AMD said Wednesday, "including the receipt of required regulatory approvals. The companies continue to expect that the transaction will be completed by the end of the 2021."

We had a chance to talk with AMD CEO Lisa Su to discuss its future plans and how it plans to address several new challenges in 2021. The discussion took place in a roundtable format. While we're not allowed to post the full transcript, we do have plenty of new information to share about the company's challenges in 2021, like the ongoing chip shortages that Su says will persist until the second half of the year, rising prices due to tariffs, the company's progress on the GPU side of its business, its Xilinx acquisition, and the quest for more cores. Su’s comments have been edited for clarity. (We also have an interview with Intel CEO Bob Swan.)

AMD's progress throughout 2020 has been nothing short of phenomenal: The company launched its new line of Zen 3-powered Ryzen 5000 desktop CPUs that have finally taken the performance crown from Intel for the first time since the Athlon 64 days, and the company's Radeon 6000 GPUs have given the company it's best competitive footing against Nvidia's gaming GPUs in the last six years. 

AMD has also made momentous progress in the laptop market, with its Ryzen 4000 Mobile processors powering their way to its highest notebook market share in its history. We expect that momentum to accelerate with the Ryzen 5000 Mobile processors the company launched today. Things are humming along on the datacenter side of its business, too, as it continues to chew away market share from Intel, and the recent launches of the new Microsoft Xbox Series X|S and Sony PS5 signals plenty of demand in 2021.

But even though AMD is firing on all cylinders as it enters 2021, plenty of challenges, many of them unprecedented, loom ahead. Here are Lisa Su’s explanations of how the company is addressing those problems, and where it sees new opportunities in 2021.

Chip Shortages to Improve 

The PC market grew faster in 2020 than it has in the last decade, with demand spurred by a shift to working from home during the most unprecedented pandemic to grip the globe in modern times. But while the PC industry is in the midst of record demand, factors related to both the US-China trade war and the pandemic have roiled supply chains across the globe, strangling supply. Like many other companies, AMD has seen rolling shortages of its products as a result, and exasperated customers have been unable to buy both CPUs and GPUs for the last several months.

Lisa Su discussed the timeline for a recovery, saying, "Paul, we are shipping a lot of parts. Volumes are continuing to increase, and that's across gaming graphics as well as CPUs. We expect that to continue to happen through 2021. I think there will be tightness, certainly through the first half of the year, but we continue to ship more into our OEM partners, as well as our general partners, to increase the overall supply. We completely [understand] why consumers want more, and so it's very high on the priority list to have supply catch up to demand." Su also clarified that the company isn’t prioritizing OEM customers over the PC builders. Instead, the company is prioritizing in real time to balance supply.

"I do want to be very specific, and I want to say to our fans and enthusiasts: I get it, I completely understand that there's a huge desire for more Ryzen 5000 and Radeon 6000 graphics cards," she said. "What I can tell you is we've shipped a lot into the channel, but it takes some time for it to work itself through, and that was some of the logistics I was talking to you about. There will be more; you will continue to see refreshes as we go into the first quarter and into the first half. I will say that it will still be tight. But there is a lot of product coming to the market, and we appreciate that there's so much interest and desire for these products. And we look forward to getting more into the hands of our users."

AMD and Nvidia seem to have been disproportionately impacted by the shortages, partially due to their respective rollouts of incredibly powerful chips that have spurred demand, but that’s led to questions about whether the companies’ status as fabless chipmakers contributes to the supply issues. Su commented, "I would say we're very happy with our manufacturing strategy and our manufacturing partners. I think it's been a competitive advantage for us. There is tightness in the supply chain, particularly around some of the consumer PC products. I think that's really a result of the overall demand environment, and not necessarily any issues from a manufacturing standpoint, as it relates to semi's."

Reports of ABF substrate shortages, a key commodity in chip manufacturing that impacts nearly every chip made for PCs, have been prevalent over the last few months. The shortage seems to have impacted a broad cross-section of the industry, even impacting chip supply for auto manufacturers that idled several manufacturing plants last week.

"I think it's fair to say you've seen some reports of substrate shortages," Su said. "And we also see tightness in the substrate market. I think, again, this is more a function of the demand that has outstripped overall worldwide capacity. You do see that more capacity is being invested in coming online, including AMD investments, but it takes some time to get that online. I think the industry is overall reacting quite broadly, because it is across the industry, by ensuring that we put more capacity online. I expect that that will continue to happen in 2021."  

Those same shortages may have an impact on some of the most important products AMD has helped to bring to market this year: The Microsoft Xbox Series X|S (see where to buy the Xbox Series X) and Sony PS5 (see where to buy Sony PS5). However, those consoles have generated an unprecedented amount of demand, which contributes to the ongoing shortages.

"We're really thrilled with how the console launches went," Su said. "You've heard separately from Sony, and Microsoft, their discussions about sort of the size of the launch and the reception of the products. From our standpoint, if you think about it, with just the amount of new hardware that had to come into place, and the millions and millions of units of both consoles, or all three consoles, frankly, that needed to ship. I think it actually came together very nicely."

"In terms of what have we learned? We have learned that there's higher demand than we thought, and we're trying to put more capacity in place for that. But we're very, very happy with the launches, and very happy with the partnership with both Sony and Microsoft […] We believe that this is a big cycle. And that says a lot about just how much technology we've been able to integrate into the console form factor."

US-China Trade War and Tariffs 

Even as the industry was already in the midst of supply shortages, several key tariffs, which come as a result of the US-China trade war, expired at the beginning of the year. That led several third-party graphics card makers (AIBs) to announce price hikes, but at least for now, those price increases seem to have come to Nvidia’s graphics cards while AMD remains unaffected.

Su acknowledged the tariff issues, saying, "As we came into the new year, there were some changes in tariffs policies. We have spent quite a bit of time to ensure that we have a very flexible supply chain, so I don't think that's a significant issue as it relates to AMD."

In terms of the supply and demand disparities that have resulted in price hikes, regardless of the impact of the tariffs, Su said:

"I will say, more broadly, that we're very committed to trying to keep the GPU pricing as close to suggested pricing as possible from an overall fairness standpoint. One of the things we’ve done, for example, is that typically for when we start our GPU launches, we will have our own Radeon Graphics MBAs [Made By AMD cards], and then we phase those out and go to AIBs [third-party GPUs]. We actually aren’t phasing out our MBAs with the purpose of trying to ensure that, as stock becomes available, we will offer them on amd.com at suggested pricing. And we’ll encourage our partners to do that as well.

"I will say that there are some COVID related logistics and other commodity components that have increased in pricing. And I think some of that is what's going through. And, again, these are things that we're living within the current situation that hopefully, as we get to a more normal environment in the second half of the year, we'll see some improvements."

GPU Progress 

AMD has made tremendous progress on the GPU front. However, the newly-released ‘Cezanne’ Ryzen 5000 Mobile processors still come with the company’s previous-gen Vega graphics instead of its newer solutions. That’s led to some speculation that AMD is design-resource limited, but Su says the decision was more about assuring the correct product timing.

"[…]When we think about product cadences and roadmap cadences, for example, one of the things that was very important to us with Cezanne was actually shipping and production early in 2021,” she said."And the reason for that is, if you think about the entire OEM cycle, we have a whole bunch of platforms that will now launch throughout the first half of 2021. And that's a nice way to build cadence."

"As it relates to a choice on, do we put the latest generation graphics or not? I really think it's really a choice. And it's a matter of the timing of where we want to be on that particular cadence. So nothing fundamental, and we’re not design resource-limited. It’s more on the notion of what we think is needed at any given point in time. Some people might have expected that we would have left Renoir in the marketplace a little bit longer because it's, frankly, a fantastic product. But we thought that there would be high demand for Zen 3 in the normal form factor, and so we prioritized the Ryzen 5000 series."

AMD continues to forge ahead, even as it has lost key executives like Raja Koduri, to expand its competitive stance.

"I have to say; I'm very proud of our engineering teams on the CPU side. Mark Papermaster, Mike Clark, and the team have just done a phenomenal job on a very ambitious up roadmap, and the team has executed very well,” she said. “We also know, though, that we're as good as our current architecture. And so, as excited as we are about Zen 3, and getting it out into the marketplace, all of the focus is on Zen 4 and Zen 5 and ensuring that those are also extremely competitive.

"On the GPU side, I’m very pleased with the work that David Wang and the team have done. Our focus from an engineering side is to set up long-term roadmaps, and we're looking at a five-year roadmap and how to pick the right mix. You have to take some risks to get the innovation where it needs to be. But you also want to be very predictable about when the products are going to come out. It’s really that give-and-take of the bets you make. And how do you make sure that you track progress? 

"But I think we have a very strong team on both the CPU and the GPU side. And our aspirations are continuing. I know many of you asked me about our progress on GPU architecture. But I'm extremely happy with the progress we've made with RDNA2 in terms of performance per watt and overall performance, and there’s a lot of focus on our RDNA3, and beyond, to ensure that we continue to drive those leading architectural capabilities."

2021 Goals and Xilinx 

Su explained that AMD’s three key goals for 2021 revolve around overall customer adoption, particularly in the enterprise and commercial markets (with a heavy focus on the server market), establishing a steady supply of processors, and closing its pending $35 billion Xilinx acquisition.

AMD’s Xilinx acquisition is a needed step to broaden the company’s portfolio and addressable markets. Still, many have opined that an acquisition of this size could serve as a distraction for AMD that might hamper its steady string of on-time execution. Su explained that the company has a solid plan to remain on track:

"Well, we are continuing to stay laser-focused on execution. So I want you all to know that is ‘job one’ overall. I think we do have a very talented management team, and I think we have the desire to have a much bigger footprint in this industry,” she said. “And so Xilinx is the right next step for us. I'm quite confident that we can both execute on the base AMD business, as well as bring over Xilinx. The fact that Victor Peng, the current CEO of Xilinx, will join us is part of that strategy to ensure a seamless transition. These are the things that leaders have to do. We have to expand and scale, and we have the capability to do that. I very much think that we can do both."

More Cores 

AMD’s disruptive impact on the CPU market has been propelled by a pretty simple philosophy: More of everything. That approach has manifested itself primarily in the company’s push to higher core counts, forcing Intel to follow along in order to remain competitive. That trend has decelerated with the company’s latest Zen 3 processors, though, which come with the same core counts found with their predecessors. 

This raises questions about whether it’s feasible to infuse more cores within the electrical and thermal confines of today’s systems, or if software even requires a jump to higher core counts. Su says that AMD will continue to expand core counts in the future:

“If you look at what we've done between Zen 2 and Zen three, as well as between the second- and third-generation EPYC and Ryzen 4000  and 5000, we really focused on increasing single-threaded performance, as well as Improving some of the latencies and overall systems, such that we've gotten tremendous gen-on-gen performance within the same process technology,” she said. 

“So, all of that is in 7nm products, and we’ve probably been able to increase performance by 20%+, depending on which metric you're looking at. There will be more core counts in the future. I would not say that somehow 64 cores is the limit, but I think they will come as we scale other parts of the system, as well.”

This week Dr. Lisa Su, chief executive of AMD, received the 2020 Robert N. Noyce Award in recognition of her achievements in the semiconductor industry. The award is considered to be one of the most prestigious in the semiconductor world. 

The Robert Noyce award was created in 1990 by the SIA Board of Directors to honor the memory of Robert Noyce (who passed away earlier that year), who co-founded Fairchild Semiconductor in 1957, Intel in 1968, and the Semiconductor Industry Association in 1977.  

Lisa Su is the 30th semiconductor industry executive to receive the Robert Noyce award and she is the second AMD CEO to get the award after Jerry Sanders in 1998. Over the years, four Intel executives, three IBM executives, three Texas Instruments executives, two Applied Materials executives, and two Micron executives have received the award. Officials from other companies received the Robert Noyce award only once.

Dr. Su joined AMD in 2012 and has served as president and CEO of AMD since October 2014. During her tenure as the head of at AMD, Lisa Su has significantly changed the way how the company executes and has managed to launch a number of successful CPU families and microarchitectures that enabled the company to turn around from a struggling designer of processors into a rapidly growing company. The most recent achievement of Dr. Su is the agreement to acquire Xilinx and transform AMD into an industry giant with a portfolio of CPUs, GPUs, and FPGAs. 

Prior to AMD, Dr. Su served in a variety of high-level roles at Freescale Semiconductor, IBM, and Texas Instruments. 

“It is a tremendous honor to receive the Noyce Award and join the distinguished list of previous recipients, many of whom I’m fortunate to count as mentors, role models, colleagues, and friends,” said Su. “I look forward to continuing to push the envelope of what we can accomplish in the semiconductor industry, which has become so instrumental to every aspect of our lives.”