Intel's latest Core Ultra 200 Plus (Arrow Lake refresh) and Core Ultra 300 (Panther Lake) chips support a new feature called Binary Optimization Tool (BOT) that improves performance in some apps. We've tested it extensively and found it to be largely consistent, but Geekbench developer Primate Labs says it "paints an unrealistic picture" in terms of daily usage. The latest update, Geekbench 6.7 came out today with BOT detection built-in to ensure benchmark results with it enabled are flagged as invalid.

The saga started last month when Primate Labs first put out a warning against BOT, claiming it can increase Geekbench workload scores by up to 40%. Then, it actually investigated the tool and saw a 5.5% improvement in both multi- and single-core scores. However, some specific tests, like HDR processing, achieved 30% better results with BOT enabled, showing how it can selectively favor certain workloads over others.

The issue is how biased these performance uplifts can be since BOT isn't supported in every app. Geekbench argues its benchmark is designed to evaluate against a broad range of workloads that accurately represent real-world performance; any test with BOT enabled goes against that by replacing the existing varied code with specifically tuned binaries that only two desktop CPUs (currently) benefit from.

BOT runs a checksum on every executable to identify and compile optimized binaries for it — that's how and why the previous version was affected. It seems BOT doesn't yet have the tuned binaries for Geekbench 6.7, so you'll see no difference in performance. Even when they're ready to apply, runs with BOT enabled will be marked invalid and incomparable to other CPUs in the database.

The update also brings some other welcome changes, such as improved SoC identification on Android that now reports the actual model of the chip along with its architecture. Similarly, Geekbench 6.7 will now display the name of RISC-V processors instead of just their ISA string. And lastly, Arm-based Linux systems should be more stable running Geekbench's multi-threaded workloads.

A few days after reviews of Intel's Core Ultra 7 270K Plus and Core Ultra 5 250K Plus rolled out, Geekbench said it would invalidate all results recorded with the two CPUs. That's because it's the only non-gaming application that currently supports Intel Binary Optimization Tool, or iBOT, which modifies a binary to optimize it for a specific Intel architecture. A week later, Geekbench has published its findings after investigating what iBOT is doing behind the scenes, and attributed an uplift of up to 30% in certain workloads to newly-vectorized instructions.

Overall, Geekbench found a 5.5% increase in both single and multithreaded performance with version 6.3 run on the MSI Prestige 16 AI+ with an Intel Core Ultra 9 386H. Those results aren't dissimilar to what we saw when testing iBOT with the 270K Plus and 250K Plus. Several of Geekbench's subtests saw no performance benefit, but some saw outsized performance increases, namely object removal at a 24.6% jump and HDR processing at a 28.5% jump. Geekbench chose to dig deeper into the HDR subtest to see what was going on.

With iBOT enabled, Geekbench saw a 14% reduction in overall instructions and a 62% drop in scalar instructions. However, it saw a 1,366% increase in vector instructions. To see which instructions were executing, Geekbench used Intel's Software Development Emulator, or SDE.

With iBOT disabled and after 100 runs of the HDR subtest, Geekbench saw a total of 220 billion scalar instructions and 1.25 billion vector instructions. With iBOT on, that went to 84.6 billion scalar and 18.3 billion vector. By vectorizing a large number of the instructions in this subtest, iBOT is able to significantly improve performance, relying on SIMD (single instruction, multiple data) rather than a linear pipeline (single instruction, single data, or SISD) of scalar instructions.

The change in instruction mix is what's interesting here. Geekbench's conclusion is what you probably expect; it doesn't appreciate an optimization that only applies in a small list of applications. "[iBOT] undermines this by replacing that varied code with processor-tuned, fully optimized binaries, measuring peak rather than typical performance."

Geekbench's view is rather negative, and understandably so, but the peek behind the curtain here has a lot of implications for the future of iBOT. Vectorized instructions on modern CPU architectures can vastly improve performance with a relatively small hit to power consumption — just look at Zen 5's performance in an AVX-512 workload like Y-Cruncher. This investigation shows that Intel is able to do that on the backend with a shipping binary.

There are downsides here, however. Geekbench noted a 40-second startup delay in its initial testing with iBOT, which shrunk to a consistent two-second delay on subsequent passes. There was no delay with iBOT disabled. Additionally, it found no performance improvement with Geekbench 6.7. iBOT computes a checksum against the executable, which means it's trying to find out if a specific binary is optimized.

Intel didn't talk about Panther Lake performance in detail when it took the wraps off its first 18A processors a bit ago, but a possible flagship SKU from the lineup — the Core Ultra X9 388H — has just appeared on Geekbench, and the scores are quite impressive. They not only one-up Intel's outgoing chips, but match AMD's top-end Strix Halo offerings as well.

The Core Ultra X9 388H got 3,057 points in the single-core test and 17,687 points in the multi-core test, both figures that put it way ahead of the Arrow Lake-H Core Ultra 9 285H, but also on par with the beefier Core Ultra 9 275HX.

That all is impressive enough, but compared to Ryzen AI Max+ 395, AMD's best mobile chip right now, the X9 388H's multi-core score is within the margin of error, but the single-core numbers are 8.7% ahead. The Geekbench listing also shows the CPU boosting to 5.1 GHz. Check out the table at the end of the article for a detailed breakdown.

These are exceptional values for a CPU that fits into an even tighter thermal envelope than Strix Halo. The X9 388H has a 45W default TDP versus the Ryzen AI Max+ 395's 55W default.

For some more context, the X9 388H is the highest-spec'd model from Panther Lake we've seen so far, featuring 16 cores in a 4P + 8E + 4 LP-E config, with no Hyper-Threading. The SKU also has the "X" designation because it features 12 Xe3 iGPU cores, but that's beyond the scope of this leak, though we've covered Panther Lake graphics before.

Panther Lake isn't supposed to come out until early next year, but excitement for the release is already dimmed. As it stands right now, surging memory prices may result in insane markups on laptops and other mobile devices featuring Panther Lake parts, no matter how impressive the SoC itself may or may not be.

Also, keep in mind that each Geekbench listing is different, so we have to wait for more benchmarks to come out before an average can be formed, but the first impressions with this leak sure do look great.

Intel's Arc Pro B50 has appeared on a benchmark listing for the first time, providing a realistic idea of its performance. Spotted by Benchleak, the B50 was seen on Geekbench, where it scored 78,661 points in the Vulkan GPU test and 69,890 points in the OpenCL test. Those are relatively modest numbers, but somewhat expected given the card's specifications. The B50 is the lowest-end SKU Intel is currently producing, featuring only 16 Xe Cores, two fewer than the gaming-oriented B570.

Therefore, these results place the B50 behind the Arc B570 by approximately 15% in OpenCL and 20% in Vulkan, but ahead of the previous Arc Pro GPU, the Pro A50, by more than 40% when compared with the best A50 score we could find on Geekbench. Keep in mind that Geekbench results vary significantly, so direct comparisons are unreasonable when one GPU has multiple runs and the other only has one. Therefore, don't take these synthetic tests at face value.

Regardless, the benchmark was conducted on an instead "normal" setup, suggesting it may have been internally done at a manufacturer rather than in a professional environment where the GPU is actually in use. The configuration consisted of a Colorful CVN X870 Ark Frozen motherboard, running a Ryzen 7 9800X3D processor with 32GB of 6400MT/s DDR5 memory.

Apple just launched a new generation of the Mac Studio, which is now powered by either an M4 Max or M3 Ultra chip. Both versions are now up for pre-order, but they won’t start shipping until March 12.
Despite that, it seems that one benchmark result has appeared on Geekbench, as shared by X user @jimmyjames_tech, and the numbers seem quite impressive. According to the Geekbench 6.4.0 for macOS AArch64 test, the Mac running an Apple M3 Ultra chip equipped with 256 GB of memory hit a single-core score of 3,221 and a multi-core score of 27,749. Treat the scores with some skepticism for now.

The single-core score is not as impressive, especially as the newer Apple M4 chips could hit higher scores. The top-end M4 Max hit 3,925 points in single-core, with even the entry-level M4 attached to an iMac beat the M3 Ultra with a score of 3,699. That is to be expected, of course, especially given that the Apple M4 has a higher frequency.

However, the M3 Ultra blows away the competition in multi-core operation with its massive 32-core CPU with up to 24 performance cores. It beat the M4 Max’s 25,647 score by about 2,102 points in the multi-core test on Geekbench, giving it a performance uplift of around 8% over the newer-generation chip.

The M3 Ultra is bound to defeat the M4 Max in multi-core performance, especially as it’s the equivalent of two M3 Max chips connected together via an interposer. Apple did not announce an M4 Ultra SoC, but developing these two-in-one chips takes time. The company usually launches the top-end version some time after the arrival of the base variant. Furthermore, Apple said that not every Apple Silicon generation will get an Ultra variant, although it hasn’t skipped one yet. So, only time will tell if we’ll get an M4 Ultra chip.

The new Mac Studio is set to be made publicly available from March 12, and in the days after we will see a slew of benchmarks appear. Right now these numbers seem to track with Apple’s announcements. But, if you want to be sure that the new Mac Studios are worth upgrading to, you should wait a few days for before making an informed purchase.

As we approach the official launch next month, the first wave of leaked RTX 5070 benchmarks has started to arrive on Geekbench. According to these figures, across the OpenCL and Vulkan APIs, the RTX 5070 fails to definitively beat the RTX 4070 Super. Sadly, that has been the norm with Blackwell and budget counterparts seemingly are no exception. However, Geekbench results are extremely inconsistent and don't always reflect how the GPU will hold up in real-world scenarios. Thus, it's advised you season this leak with an unhealthy sprinkling of salt.

The RTX 5070 is powered by a slightly cut-down GB205 die under the hood with 48 SMs or 6,144 CUDA cores and 12GB of GDDR7 memory. Jensen is now infamously quoted stating the RTX 5070 is equivalent to the RTX 4090 (with MFG); quite misleading for the average consumer. In raw-raster, we speculate the RTX 5070 may struggle to surpass even the RTX 4070 Super, given the reduced shading units. The GPU carries a modest 250W TBP (Total Board Power) and an MSRP of $549, though launch prices and availability might be extremely volatile.

An identical setup used for both benchmarks suggests they were likely conducted by the same reviewer who made the results public by mistake. The test bench is outfitted with the AMD Ryzen 7 9800X3D atop the Asus ROG Crosshair X870E Hero motherboard, side-by-side with 32GB of DDR5 memory. We've gathered publicly available OpenCL and Vulkan results at Geekbench, but bear in mind these benchmarks might be subject to inconsistencies and variations.

The alleged RTX 5070 managed to score 187,414 and 188,712 points in OpenCL and Vulkan respectively, marking an up-to 20% lead over the base RTX 4070. Versus the RTX 4070 Super, Vulkan sees the RTX 5070's lead drop to around 5%, while it ends up slower in OpenCL. Provided the massive difference in specs and pricing, the RTX 5070 Ti is in another league, as its non-Ti counterpart trails by 18%.

AMD's RDNA 4 is expected to shake up the $500-$600 market within the next couple of days. Leaked benchmark results intended for the press suggest the RX 9070 non-XT to be 21% faster than the RX 7900 GRE in 4K; in the same performance range as the RTX 4070 Super. The RX 9070 XT on the flipside allegedly scores a 42% uplift, putting it near an RX 7900 XTX or the unnecessarily verbose RTX 4070 Ti Super. All that's left is the pricing and we'll soon learn more from AMD on February 28.

Nvidia's RTX 50 series will debut with the RTX 5090 and RTX 5080 starting January 30. While official reviews, including ours, are under embargo, a potential leak has surfaced that suggests a modest 30% improvement over the last generation. The RTX 5090 has reportedly been benchmarked in Geekbench 5 (via Benchleaks) using the CUDA API, likely by a reviewer who inadvertently made the test results public. As always, spread some salt over this leak, even though the performance claims are similar to what Nvidia has depicted in its slides.

The test bench features AMD's Zen 4-based Ryzen 9 7900X with 12 cores / 24 threads and an Asus ProArt X870E Creator WiFi motherboard. The system has 32GB of DDR5-6000 memory and uses Windows 11 Pro as the Operating System for this test. The benchmark was carried out using the CUDA API, the results of which are hard to come by as Geekbench does not publicly maintain a database for CUDA benchmarks.

We've gathered a few RTX 40 numbers to give you a general performance overview; however, your results might differ slightly. Remember that Geekbench is just a synthetic benchmark and may not accurately reflect how this GPU performs in real-world scenarios.

The RTX 5090 amasses 542,157 points in the CUDA API, landing a solid 27% lead over its predecessor, the RTX 4090. This isn't a massive leap generation-on-generation as we're used to seeing, plus the RTX 5090 has 32% more CUDA cores than the RTX 4090. Then you have the apparent elephant in the room: the pricing. Nvidia's $1,999 price tag for the RTX 5090 makes it roughly 25% more expensive than the RTX 4090, assuming you got one at MSRP. Again, this is leaked information, so almost everything you see here is subject to change.

The GB202, which powers the RTX 5090, is 744mm2 and has roughly 92 billion transistors. This equates to around 123 million transistors per mm2 on the updated 4NP process, similar to the 4N used on Ada Lovelace. The L1 and L2 cache sizes per SM also show slight improvement, compensated by the faster GDDR7 memory. Nvidia claims a 2x increase in performance over the RTX 40 series, suggesting that an RTX 5070 equals an RTX 4090, but this requires enabling Multi Frame Generation.

Conversely, the RX 9070 XT from AMD has been rumored to match the RTX 4080 Super in raster performance. All that power has been crammed into a near-390mm² Navi 48 chip, expected to be priced around $500. AMD has been tight-lipped about RDNA 4, though we might hear more from them in February when the mid-ranged RTX 5070 hits shelves and AMD finalizes its pricing strategy.

Intel's upcoming entry-level Arc B570, which competes against the best graphics cards, has been benchmarked in Geekbench's new AI benchmark. Discovered by Tomasz Gawronski on X, the B570 GPU is reportedly 10% slower than the B580 in this specific benchmark.

Using the OpenVINO framework, the B570 graphics card produced a single precision result of 20,213 points, 35,819 half-precision points, and 38,717 quantized points. Gawronski shared two B580 OpenVINO results for comparison. One showed a single precision score of 22,337 points, a half-precision result of 38,752 points, and a quantized score of 42,201 points. The latter was within 150 points of the other's results in all three metrics.

Overall, both B580 scores were 8-10% quicker than the B570's AI score. However, as with all Geekbench scores, take this information with a great deal of salt. Geekbench scores alone won't tell the whole story of a CPU or GPU's real-world performance.

However, the results align with educated guesses in the industry that the B570 will likely be "just" 10-15% slower than the B580. The B570 is the lower-end counterpart to the B580 and is very similar spec-wise, with the B580 only having 11% more cores, though it has 20% more memory bandwidth and 33% more cache. Featuring 18 Xe-Cores, 2,304 shader cores, 144 AI cores, 18 Ray Tracing cores, 80 ROUs, 144 TMUs, and a memory sub-system comprised of a 160-bit memory bus and 10GB of VRAM featuring 380 GB/s of memory bandwidth.

We will have to wait for third-party reviews of the B570, including ours, to see where it truly stands. Intel decided to withhold B570 performance numbers from its B-series GPU announcement, only issuing specs of the upcoming GPU to the public.

Based on Intel's B580 benchmark figures, if the B570 turns out to be 10% slower than the B580, it will likely feature RTX 4060 performance but at a lower price of $219.

A rack comprised of two Hygon server-based processors was spotted on the Geekbench browser featuring a run of Geekbench's new AI-based benchmark. Benchleaks on X reports that the chip's CPUID name is 900F22.

Sadly for the company, the dual-CPU configuration scored abysmal results in Geekbench's AI benchmark. The single precision score was 1,412 points, half precision score 531, and the quantized score was 1,523. The two CPUs operated at a base frequency of 3GHz flat and were paired with 64GB of memory.

By comparison, Intel's nearly 10-year-old Skylake-based Core i7-6700HQ quad-core mobile CPU was barely any slower than the dual 32-core chips, featuring a single precision score of 1,113 points, half-precision score of 589 points and a quantized score of 1,394 points.

Compared to a much more modern CPU, the Hygon-based server rack is (expectedly) vastly outperformed by mid-range desktop CPU hardware. For instance, one AMD Ryzen 5 7600X user report we pulled up scored 3,542 points in single precision, 1,686 points in half-precision, and 6,281 points in quantized, representing a 2.5x to 4x performance improvement over the Hygon server CPUs, despite having only a fraction of the number of cores.

We can't be sure what exact Hygon CPU model was tested, but regardless, Hygon only utilizes AMD's original Zen architecture from 2017 in all of its server chips — in-fact, it has figured out a way to port AMD's Zen architecture into AMD's latest SP5 socket, which powers its latest Zen 5-based chips. This is why Hygon's CPUs suffer a serious performance deficit compared to all modern CPUs, despite having many cores.

Hygon is a fabless Chinese-based chip maker, and due to U.S. sanctions, it can't use anything beyond AMD's original Zen architecture (at least for now), which is the company's Achilles heel for competitive generational performance improvements. The only way it has been able to boost performance is by adding more cores to its CPUs and adding more CPUs to its platform, which does help boost multi-core performance but critically doesn't help in most other areas such as single-core IPC and latency-related bottlenecks, not to mention missing out on the latest CPU instruction sets such as AVX-512.

A new 16-core processor from Hygon—a fabless chip maker in China that uses AMD's first-gen Zen IP—has popped up on Geekbench. It offers 60% better performance than a similar eight-core Hygon processor we covered in the past. The performance bump isn't substantial when you consider the chip has twice the number of cores, but since the manufacturer is stuck on a now seven-year-old architecture—Zen 1—there isn't much that can be achieved apart from increasing core counts.

Looking at the Geekbench listing, the chip was benchmarked using the Hygon XHVTBST board under the openKylin 2.0 Operating System (based on Linux). The CPU is named Hygon C86-4G - alternatively dubbed C86 3490 - a step higher than the eight-core C86 3350 - belonging to the same Hygon C86-3000 family. The CPU offered a rather low base frequency of 2.8 GHz and was equipped with around 32GB of RAM. Since an older release of Geekbench was used for this benchmark, we've listed a few other x86 CPUs running the same version for comparison.

Packed with 16 cores and 32 threads, the Hygon C86-4G is a direct contender to the Ryzen Threadripper 1950X, and the performance delta isn't far off from what you'd expect. The abysmally low single-core score of 1073 points puts it on par with older Haswell and Skylake processors. In the multi-threaded department, it bodes fairly well at 8811 points as it inches ever so close to the Zen 1-based Threadripper 1950X. Still, modern-day CPUs such as the Ryzen 5 5600 (Zen 3) put it to shame with less than half the cores.

US sanctions have barred chip makers from exporting high-performance devices to China and this is likely the reason we haven't seen a Hygon chip with Zen+ or a newer design. Likewise, scaling has not improved at all since 2021 when two C86 3185 chips (16 cores in total) were needed to beat one Ryzen 5 5600X. Reports suggest that the chip maker has found a way to allegedly port the Zen 1 architecture to AMD's latest SP5 socket for more I/O and better compatibility.

Besides that, Zen 2 was the defining architecture that introduced a paradigm shift in AMD's design philosophy as the company shifted to MCM - decoupling I/O and CPU cores into separate chiplets. So there is still a lot of headroom left for these chips. However, since Hygon relies completely on AMD's design, the pathway for future releases of these chips is narrowing.

With less than one month left to CES 2025, leaked benchmarks have started to surface starring AMD's flagship Strix Halo APUs. The Ryzen AI MAX+ Pro 395 - now that's a mouthful - has emerged on Geekbench packing 16 CPU cores, 64GB of RAM, and a massive 40 CU (Compute Units) Radeon RX 8600S iGPU (Integrated GPU). Despite the apparent low score - set to improve with further optimizations - the real juicy details lie in the specifications.

For the uninitiated, Strix Halo or the Ryzen AI MAX 300 APUs are said to be AMD's top-of-the-line offering for workstations and laptops next year. The test bench utilizes the "AMD MAPLE-STXH" reference board, designed for the FP11 socket. Likewise, Ryzen AI MAX+ Pro 395—which we'll call Ryzen AI 395 for simplicity's sake—is paired with 64GB of RAM, going as high as 128GB per shipping manifests. The performance isn't anything to write home about at this stage, with the Vulkan score on par with an RTX 2060 resulting from early silicon.

In line with previous rumors, the Ryzen AI 395 features 16 cores based on the Zen 5 microarchitecture and 32 threads. The suspected dual-CCD design lands it 32MBx2 (64MB) of L3 cache alongside 16MB of L2 cache (1MB per core). Geekbench lists the maximum clock speeds at 4.4 GHz, which will likely improve with time.

Strix Halo's graphics solution will fall under AMD's Radeon 8000S umbrella of iGPUs: the Radeon 8060S and Radeon 8050S are based on RDNA 3.5. On an architectural level, the iGPU likely resides in the large SoC die with upwards of 32MB of MALL or Infinity Cache to avoid memory bottlenecks. Geekbench doesn't explicitly list the CU count, but previous leaks hint towards a 40 CU configuration for all Radeon 8060S models—25% more than the RX 7600.

Strix Halo is designed to compete against Apple's M-series silicon and Nvidia's dedicated GPUs in the laptop segment. It may not rival the RTX 4090 laptop GPU, but it could give mid-ranged offerings like the RTX 4070 laptop a run for its money.

Pricing is a valid concern since Strix Point - the Ryzen AI 300 family - is limited in availability and costs a pretty penny if you want the best. Nonetheless, we expect AMD to unveil its Strix Halo lineup of APUs at CES 2025 in addition to Krackan Point and the Radeon RX 8000 series.

An upcoming Intel laptop processor has surfaced on Geekbench, revealing intriguing details about its performance and specifications. The Core 7 240H will potentially be part of Intel’s rumored Core 200H lineup, boasting ten cores (6-Performance + 4-Efficient cores) and 16 threads. Notably, the Geekbench listing (via Benchleaks) suggests the codename as ‘Raptor Lake,’ which could also mean that it will be more or less a Raptor Lake refresh.

The Geekbench listing was part of an Acer Nitro ANV15-52 gaming laptop with a discrete RTX 4050 laptop GPU. The Core 7 240H operates with a base frequency of 2.5 GHz, up to 5 GHz boost clock, and 24MB of L3 cache. The Core 7 240H configuration suggests continuing Intel’s hybrid core architecture, which debuted with the Alder Lake series.

The approach combines high-performance and energy-efficient cores for optimal multitasking and power efficiency. This chip includes DDR5 memory support and advanced connectivity options that meet modern laptop requirements.

In terms of performance, the Core 7 240H was benchmarked more than once, achieving a maximum score of 2,689 points in single-core and 13,330 points in multi-core benchmarks on Geekbench 6. These scores are quite competitive, considering their segment.

For comparison, the Acer Nitro V 15 with an Intel Core i7-13620H delivered 2,571 points in single-core and 13,373 points in multi-core benchmarks on Geekbench 6 in our testing. Notably, the Core i7-13620H offers a similar 10-core architecture as the Core 7 240H, with slightly different boost and base clock speeds.

Coming to the iGPU of the leaked Core 7 240H is based on Intel's Xe architecture, equipped with 64 Compute Units, equivalent to 8 Xe Cores, and operates at a maximum clock speed of 1.5GHz. The OpenCL benchmark delivered scores of 13,362 and 13,478 across two tests, demonstrating decent performance for integrated graphics.

Of course, it will not be able to compete with GPUs like the Nvidia RTX 4050 mobile, which has an average score of over 80,000 points. This gap highlights the iGPU’s focus on efficiency and general-purpose tasks rather than high-end gaming, for which dedicated GPUs are more suited.

While Intel has not officially announced the Core 7 240H, these leaks provide a glimpse into what consumers can expect from the company's upcoming mobile CPU lineup. The official Core 200H announcement is rumored to be scheduled for next year, probably around CES.

According to a benchmark at Geekbench, we have more information regarding the alleged specifications of Intel's upcoming Arc B580 GPU — based on Team Blue's Xe2 "Battlemage" architecture. The Arc B580 was spotted in a few preliminary listings last week at Amazon from ASRock, which were promptly taken down. This benchmark seemingly reaffirms that the B580 will carry 12GB of VRAM alongside 20 Xe cores, though initial performance in OpenCL leaves more to be desired. Keep in mind that this is not an official benchmark, and that Geekbench OpenCL can be a terrible way of measuring performance, so reserve judgment until review units are available.

The test bench features a Z890 AORUS MASTER motherboard with the flagship Intel Core Ultra 9 285K and 32GB of DDR5-6400 memory. Do note that the benchmark doesn't explicitly mention that this is an Arc B580. This is not an exact confirmation of the specifications, though there is a strong correlation with previous leaks.

The Arc B580 in this benchmark is listed with 160 Compute Units which really are just Xe Vector Engines. Based on the Xe2 architecture's core division, this would yield 20 Xe cores (1 Xe Core = 8 XVEs) or 2,560 ALUs (1 XVE = 16 ALUs). For context, the last-gen Arc A580 offered 24 Xe cores with 3,072 ALUs. On the memory side of things, the B580 packs 12GB of GDDR6 VRAM on a 192-bit memory bus. At least in this benchmark, the GPU was running at a maximum clock speed of 2.85 GHz — 42.5% faster than the A580.

If you calculate the teraflops on tap, the higher clocks should make up for the shader ALU deficit. The A580 has 10.4 TFLOPS of FP32 based on its 1700 MHz boost clock (which is still conservative). With a 2.85 GHz clock speed, the B580 has 14.6 TFLOPS of FP32. So, it should be faster, but again this is Geekbench OpenCL.

The performance end of things is slightly disappointing since the B580's subpar 78,743 points land it slower than the A580. But don't let that disappoint you because Battlemage has always been slower in synthetic workloads than Alchemist.

Case in point, the Core Ultra 7 155H (Meteor Lake) is 20% faster than the Core Ultra 7 258V (Lunar Lake) in the same test with the same Xe core counts. Depending on real-world performance, Intel may need to price these GPUs aggressively if it plans to up its market share this generation. While we expect a price tag in the ballpark of $200–$250, high production costs associated with 4nm technology could hurt Intel's bottom line.

In any case, all indicators show that Intel is prepping to reveal Battlemage in December, ahead of Nvidia's Blackwell RTX 50-series and AMD's RDNA 4 RX 8000-series GPUs. Initial supply could be limited to the B580 as the high-end BMG-31-based Arc B770 appears to be coming later.

A new unreleased Core Ultra 200S series product has hit the Geekbench browser. Benchleaks on X shared a Geekbench result of the upcoming mid-range Core Ultra 5 225F 10-core chip, with performance numbers similar to Intel's Core i5-13600 Raptor Lake CPU.

The Core Ultra 5 225F reached a single-core result of 2,653 points and a multi-core result of 13,028 points. The CPU is armed with six P-cores, four E-cores, 20MB of L3 cache, and a purported max frequency of 4.887GHz during the Geekbench run.

Compared to the closest Core Ultra 200S series CPU available for purchase now, the Core Ultra 5 245K, the Ultra 5 225F, the 245K is 16% faster in single-core performance on average and a whopping 44% quicker in multi-core. (Average results for the 245K were taken from the first page of user-published results in the Geekbench browser, resulting in an average score of 3,087 points for single-core and 18,882 points for multi-core).

The Core i5-13600 is the closest CPU in Geekbench scores to the Core Ultra 5 225F, with 2516 points in the single-core test and 12,375 points in the multi-core test. The Ultra 5 225F is barely faster than the i5-13600, beating the Raptor Lake chip by 5% in single and multi-core performance. However, the 225F achieves this feat with four fewer cores and just half the thread count.

Compared to the 225F's direct predecessor, the Core i5-14400F, the 225F is 13% faster in single and multi-core. In the AMD camp, the Ryzen 7 9700X is 29% faster in single-core performance and 38% faster in multi-core performance (single average score of 3,435 and average multi-core result of 18,042). The Ryzen 5 9600X is 32% faster in single-core and 23% faster in multi-core despite having a severe core count disadvantage (with an average single-core result of 3,521 points and multi-core result of 16,072).

As per usual, take this Geekbench score with a pinch of salt. Results can vary depending on how the CPU is configured, and we could see much faster results in the future, especially since Intel revealed serious performance issues with Arrow Lake that will be rectified in a future firmware (or series of firmware) update(s).

A new hexa-core processor under AMD's Krackan Point (KRK) lineup of APUs has emerged at Geekbench - thanks to Olrak at X. The CPU wields a rather unconventional 3+3 hybrid configuration - taking a different approach than AMD did with its Phoenix 2 APUs.

Krackan Point is AMD's cheaper alternative to Strix Point - packing upwards of eight cores and an eight Compute Unit (CU) RDNA 3.5-based iGPU (Integrated GPU). Krackan Point utilizes a hybrid-core design to save valuable die space, featuring up to four Zen 5 and four Zen 5c cores. The six-core offering has an OPN code of "100-000001600-40_Y," corresponding to a Ryzen 5 APU per previously leaked shipping manifests. The benchmark, being an AI test, isn't that exciting once you consider that Krackan employs the same 50 TOPS XDNA 2 NPU as Strix Point.

Going over the listed specs, the budget Ryzen 5 300 APU packs three Zen 5 and three Zen 5c cores alongside 12 threads - quite different from the Phoenix 2-based Ryzen 5 7545U - which had two Zen 4 and four Zen 4c cores. Moreover, the APU was being tested in an Asus validation platform and had a relatively low base frequency of 2 GHz - but that's likely a result of early silicon.

AMD reintroduced a dual-CCX design with Strix Point - splitting up the Zen 5 and Zen 5c cores across different rings - with separate L3 caches for both core types. Interestingly, for the Krackan Point APU in question, Geekbench divides the core count into two clusters. However, Olrak suggests that AMD might package all six cores together - connected through a unified ring bus - sharing a coherent L3 cache. To chip in, the Ryzen 5 7545U (Phoenix 2) also employed a single CCX, but the final decision rests on AMD.

Rest assured, laptops powered by AMD's Krackan Point offerings are set to arrive in early 2025 - starting at $799 if we go by leaks. Overall, CES 2025 will be jampacked with next-gen launches from all major players. Expect to see AMD unveil Krackan Point alongside Strix Halo APUs, Fire Range CPUs, and Radeon RX 8000 "RDNA 4" GPUs.

A new benchmark has surfaced on Geekbench showcasing Intel's upcoming mobile Core Ultra 9 285H in action. Initial scores put it a smidgeon behind Lunar Lake in terms of single-core performance. Against Meteor Lake, the CPU manages to pull ahead by 19% at roughly the same power envelope, giving us cause for some optimism.

Intel's new Core naming schematic has some weird nuances. To keep things short, any CPU without the "Ultra" moniker is equipped with Raptor Lake / Alder Lake silicon. Hence, the Core Ultra 9 285H is part of the Arrow Lake-H family and should feature a TDP of 45W.

Spotted in the upcoming Dell Pro Max 16 (MC16250) laptop, the Core Ultra 9 285H offers 16 cores (six P + eight E + two LPE) and 16 threads. The laptop hosts 64GB of DDR5-6400 memory - one of the highest speeds we've seen for SODIMMs, assuming the laptop doesn't use LPDDR5 memory. The CPU achieved a maximum clock speed of 5.4 GHz, just 100 MHz shy of the desktop Core Ultra 7 265K.

The Core Ultra 9 285H managed to score 2,665 and 15,330 points in the single-core and multi-core categories, respectively. Don't get us wrong - the performance is acceptable at roughly 19% faster than the Core Ultra 9 185H from the last generation but fails to match Lunar Lake's Core Ultra 9 288V (30W). The worst part is that the Core Ultra 9 288V scores higher at just 5.1 GHz, meanwhile, the 285H pushes upwards of 5.4 GHz so something is clearly wrong with this benchmark.

Since we are unaware of the operating environment and temperature, it is not possible to draw any judgments yet. Case in point - the leaked 65W Core Ultra 9 285 non-K scored 3,247 points (single core) landing it 21% faster than the 285H. Hence, this is likely an engineering sample or a result of poor testing conditions.

Nonetheless, we expect this CPU to come paired with an RTX 40, maybe even an RTX 50 laptop GPU in gaming notebooks next year. Despite the initial disappointment, Arrow Lake scales pretty well at lower power limits so there is a lot of room for improvement. Intel is rumored to announce its Core Ultra 200H/HX/S non-K CPUs at CES 2025, or in just two months.

Benchmarks of Apple's newly launched M4 Max have started to surface at Geekbench. Unsurprisingly, Apple has retained its performance crown and has taken over at the helm of Geekbench as the fastest chip. Even the multi-core performance puts Intel's and AMD's latest offerings to shame, that too at a fraction of the power.

The test bench features the new 16-inch MacBook Pro, which hosts the M4 Max in all its glory. Apple's latest and updated design allows the M4 Max to blaze through the benchmark—scoring 4,060 points in the single-core test and 26,675 points in the multi-core test. The M3 Max results were taken from the Geekbench 6 database, while the remaining results are from our reviews of the processors.

The M4 Max has secured the single-core crown in Geekbench 6. The chip was roughly 30% faster than last year's M3 Max in single-core performance and 27% in multi-core performance.

On the x86 end, AMD and Intel pale in contrast. The M4 Max handily keeps up even in multi-core at a fraction of the power—beating Intel's Core Ultra 9 285K by around 19% in the single-core category and 16% in the multi-core category. Compared to the Ryzen 9 9950X, the M4 Max showed 18% higher single-core performance and 25% higher multi-core performance.

The M4 Max is Apple's flagship SoC targeted at data scientists, 3D artists, and professionals. The top-end configuration packs 16 CPU cores (twelve performance and four efficient) and 40 GPU cores alongside up to 128GB of unified memory - accessible by both the CPU and GPU. Apple has also kitted its new MacBook Pro lineup with support for Thunderbolt 5, delivering speeds of up to 120 Gb/s.

Apple's new M4 lineup is a solid response to the latest AI PC wave from Intel, AMD, and Qualcomm. While the performance outlook is excellent, pricing remains a concern since Apple has always charged a premium for its products. The full-fat M4 Max configuration in question will set you back $3,999. Content creators and productivity-centric users might eye laptops with dedicated graphics solutions at that price point.

Geekbench isn't the best benchmark for comparing chips. Therefore, it'll be interesting to see how the M4 Max performs in something like Cinebench or HandBrake to see whether Apple's latest chip still beats the competition. The M4 Max-powered MacBook Pro 2024 deliveries are set to begin on November 8, so we'll soon see what the M4 Max can do.

The Intel Core Ultra 9 285 is neck and neck against the unlocked Core i9-14900K, based on leaked benchmarks over at Geekbench. As highlighted by Benchleaks on X, the Core Ultra 9 285 is an upcoming Arrow Lake processor with a TDP of 65W, falling in the non-K category of CPUs. Despite being power-limited, the Core Ultra 9 285 easily catches up to the last generation's flagship Core i9-14900K.

The Core Ultra 9 285 packs 24 cores, divided into eight P-cores based on Lion Cove and 16 E-cores based on Skymont. Much like its unlocked counterpart, the 285 hosts a total of 76MB of total cache (40MB L2 + 36MB L3). In line with a previous benchmark, the Core Ultra 9 285 has a boost clock of 5.6 GHz, which is quite the feat as it is just 100 MHz lower than its K-equivalent.

The test bench sports GIGABYTE's Z890 UD motherboard and 32GB of DDR5-5600 memory, which should increase to DDR5-6400 if you opt for CUDIMM sticks, so there's still room for improvement.

The Core Ultra 9 285 amasses 3,247 points in the single-core category, beating the Core Ultra 7 265K. Compared to its predecessor, the Core i9-14900, the Core Ultra 9 285 is roughly 12% faster and 5% faster than the Core i9-14900K. In multi-core, the Core Ultra 9 285 breaches the 20k barrier by scoring 20,204 points, a 13% lead over its last-generation equivalent.

This is an awe-inspiring result, no matter how you look at it. The Core Ultra 9 285 has a PL1 of 65W, which can temporarily reach 182W (performance preset PL2). Even so, Intel's pivotal selling point with Arrow Lake is efficiency, and initial results hold onto that promise. These non-K CPUs should be great for power-conscious gamers and OEM machines. Likewise, seeing how well Arrow Lake scales with the 35W "T" family will be exciting whenever it launches or gets leaked.

It also indicates that Arrow Lake might have a lot of room for tuning and undervolting, giving enthusiasts another fun toy with which to tinker. The Core Ultra 200S K-series lineup will hit shelves on October 24. Following tradition, Intel might be eyeing CES 2025 to launch the mobile and non-K Arrow Lake lineups.

A Geekbench 5 OpenCL (GPU) score was uploaded and shared on Twitter for the yet-unannounced Intel Core 3 N350, which seems to be an 8-core processor using last-gen Intel graphics.

The total OpenCL scoring of 6,191 points places the Core 3 N350's iGPU within range of Intel UHD Graphics (6,316, previously highest-end Intel iGPUs before integrated Arc/Iris graphics), Intel Iris Graphic Plus Graphics 650 (6,295) and the Nvidia GeForce 940M (6,244) discrete mobile GPU. Now, it's worth noting that even if these benchmarks reflect final performance, and there's a fair chance they're not, synthetic benchmarks are fundamentally flawed at communicating the actual end-user experience one should expect from a given product, particularly in pre-release.

However, if these results hold in the long term, they speak to Intel's non-Ultra Core 3 CPUs being left with somewhat anemic integrated graphics solutions. Now, no available onboard NPU (the main difference between Core and Core Ultra per the original name change announcement) probably won't matter much to most of you. However, entry-level users like decent iGPUs, and this move from Intel points toward entry-level iGPU users still gravitate toward AMD systems for their vastly superior integrated graphics.

For users who don't bother with integrated graphics, the Intel Core 3 N350 will likely still find a comfy enough spot within the market, even just as a cheap entry point to the ecosystem. The Intel Core 3 N350 designation also suggests that this design will likely utilize E-cores solely like the Intel Core N100 and N305. We can't ascertain the value of the CPU part here until we have benchmarks of those cores and not just the iGPU, though.

In any case, the appearance of the Intel Core 3 N350 on Geekbench running at up to 3,886 MHz from a 1 GHz base clock rate does seem attractive. The iGPU appears to reach a maximum frequency of 1,350 MHz, and the most curious part of the benchmarks identifies that the iGPU only has an allocation of 1.34GB— with the device totaling 3.78GB RAM, implying some harsh limits on RAM capacity for this testing. Since RAM capacity and speed tie directly to iGPU performance, there's still a chance that a better setup would likely improve these scores if they are legitimate to start.

The vanilla version of Intel's flagship Arrow Lake Core Ultra 9 part has been spotted in the Geekbench browser. Discovered by Benchleaks on X, Intel's 24-core Core Ultra 9 285 was benchmarked in Geekbench 6.3.0, showing strong single-core performance but strange multi-core results.

The Ultra 9 285 achieved a score of 3,081 points in Geekbench's single-core benchmark and just 14,150 points in the benchmark's multi-core counterpart. In the single-core department, the chip holds equal performance to many of its Raptor Lake-based predecessors, including the Core i7-13700KF, Core i7-14700K/KF, Core i9-13900K, and Core i9-14900K/KF, which have single-core scores in the 2,900 - 3,000 point range. However, the Arrow Lake chip is noticeably behind the Core i9-13900KS, which has a score of 3,134 points (Geekbench doesn't have any official results of the newer i9-14900KS except for user-generated results).

However, multi-core-wise, the Core Ultra 9 285 is disappointing. Its score of 14,150 points is very weak, even compared to Intel's previous-generation outings. The Arrow Lake chip doesn't even outperform the 13th Gen Core i7-13700, which has a multi-core score of 14,862 points, despite the fact that it sports both an older architecture and fewer cores than the 285.

For example, against its direct predecessor the Core i9-14900, the Raptor Lake part is 24% faster. This is very different from other Arrow Lake models that we have compared to in the past, where the multi-core performance was far more in line with the Raptor Lake parts, if not faster than its predecessors. For instance, in late July we covered the Core Ultra 7 265KF in Geekbench 5 which pumped out a multi-core result that was 6% faster than the Core i9-14900K. We can't compare the results directly to the 285 since the 265KF was benchmarked in Geekbench 5. However, the fact that the 265KF, which is a Ultra 7 part, was able to outperform the Core i9-14900K suggests where the Core Ultra 9 285 should land.

As with most of these Geekbench benchmark results before launch, we have to take this information with a grain of salt. The Core Ultra 9 285 benchmarked here may likely be a pre-production sample with different behavior than the production model, and Geekbench is just one benchmark that won't show the full capabilities of the CPU.

The strong single-core performance accompanied by the unexpectedly weak multi-core results suggests Intel could be shifting the power limits for its Arrow Lake processors down, at least for the non-K variants. This would fit previous leaks that have suggested as much. But we will have to wait and see if that is the case once Arrow Lake launches and we get our hands on these chips. 

AMD's flagship Ryzen AI 9 HX 375 CPU was benchmarked in Geekbench 6 with potent results. Initially discovered by HXL on X, the Zen 5 CPU's single-core and multi-core results were high enough to outperform most of Intel's fastest mobile CPUs, featuring the same benchmark results as a desktop Core i5-13600K.

The HX 375 produced a single-core result of 2,864 points and a multi-core result of 15,012 points on an HP OmniBook Ultra Laptop 14 paired with 32GB of DDR5 memory. The Zen 5 chip's results were good enough to outperform Intel's flagship CPUs from its Meteor Lake lineup and almost good enough to surpass Intel's Raptor Lake Refresh mobile flagship.

The Ryzen AI 9 HX 375 was 5% faster than the Core i9-14900HX in Geekbench's single-core benchmark. However, the Core i9-14900HX was 7% faster in the multi-core benchmark. Compared to the Core Ultra 9 185H, the HX 375 lead was much more substantial, 27% faster in the single-core benchmark and 25% faster in the multi-core test.

We don't have official Lunar Lake Geekbench 6 benchmarks yet, but based on some leaked results, the flagship 288V appears to do single-core results in the 2,900 range, which would outperform the HX 375 by a couple of percent. However, multi-core results are generally much worse, in the 10,000 - 11,000 range.

Regardless, the Ryzen AI HX 375 can, at worst, match the performance of Intel's fastest mobile CPUs and, at best, vastly outperform its primary Intel competitors, Lunar Lake and Meteor Lake, specifically in the multi-core results. For a desktop comparison, the HX 375 matches the Core i5-13600K in terms of multi-core performance and the Core i9-14900 in terms of single-core performance.

The Ryzen AI 9 HX 375 is the flagship chip from AMD's Ryzen AI 300 lineup. The chip shares the exact specifications as the HX 370, except for the XDNA2 NPU, which has been improved by 10% and features 55 TOPS of performance rather than 50. Expect to see the Ryzen AI 9 HX 375 inside gaming laptops and mobile workstation devices.

One of AMD's Ryzen AI Max 300 (Strix Halo) APU has gone through Geekbench AI, the AI-focused counterpart to Geekbench 6. Discovered by Everest on X, the Ryzen AI Max 390 (AMD Eng Sample: 100-000001421-50_Y) was tested, but only on the Zen 5 CPU cores, not the GPU or NPU.

As always, take the Geekbench result with a pinch of salt since we don't know the state of the silicon. According to the submission, the Ryzen AI Max 390 is inside HP's ZBook Ultra 14-inch G1a, which appears to be a mobile workstation device, so the cooling should be adequate.

Using all 12 CPU cores, the Ryzen AI Max 390 outputted a single precision score of 4,733 points, a half-precision score of 4,944 points, and a quantized score of 13,944 points. Testing was accomplished on Geekbench AI 1.1.0 using the OpenVINO framework.

Since the benchmark was relegated entirely to the CPU cores, AI performance was not as good as what the chip would be able to do on the GPU or NPU. For example, the chip's quantized score was inferior to that of Intel's entry-level Arc A310 graphics card, which outputted a score of 15,453 points. It was also worse than some Core Ultra 7 Meteor Lake integrated graphics chips, with an even higher score. A similar story also goes for the single-precision and half-precision scores.

The Strix Halo chip failed to impress when looking specifically at CPU scores in an apples-to-apples comparison. Out of the relatively few OpenVINO CPU scores I was able to find, the Ryzen AI Max chip outperformed AMD's previous generation eight-core Ryzen 7 7840HS Zen 4 mobile CPU, which outputted a single precision score of 5,099 points, a half-precision score of 5,118 points, and a quantized score of 14,680 points.

I also took the liberty of testing my personal Ryzen 7 5800X3D to provide a desktop reference. It also outperformed the Strix Halo APU (for the most part). The 5800X3D featured a single precision score of 5,428 points, a half-precision score of 5,522 points, and a quantized score of 8,115 points. The quantized results were the only area where the Strix Halo chip was at an advantage.

From what I could find, the only chips that the Ryzen AI Max 390 can consistently outperform are Intel's Meteor Lake notebook chips, such as the Core Ultra 7 165H. Two listings revealed single-precision and half-precision scores below 3000 points and quantized scores below 7200 for this CPU model.

Looking at CPU-only results for AI performance is not as useless as it might seem. When AI workloads can tax the entire chip (CPU, NPU, and GPU), having capable AI performance on the CPU can be beneficial. Regardless, we are disappointed the benchmarker did not test the NPU and monster-integrated GPU inside the Strix Halo chip, which is far more critical for AI-focused workloads.

The Ryzen AI Max 300 series, aka Strix Halo, is an upcoming APU lineup focused on high-performance computing for mobile devices. Strix Halo is slated to arrive in early 2025.

Apple's latest A18 (non-Pro) application processor for smartphones not only challenges AMD's Ryzen 9 9950X and Apple's M4 in single-thread workloads in Geekbench 6, but it seemingly beats all previous-generation flagship CPUs for desktops and laptops. But when it comes to multi-thread, the limited amount of cores certainly makes an impact on performance. 

Apple's vanilla A18 processor packs two high-performance cores operating at up to 4.0 GHz (the first time the company's smartphone processor hits such a high frequency) and four low-power cores. The new A18 system-on-chip (SoC) scores 3409 points in a Geekbench 6 single-thread benchmark and 8492 points in Geekbench 6 multi-thread benchmarks, both are very good results. In fact, they are 15.5% higher for single-thread performance and 16.6% higher for multi-thread performance when compared to its direct predecessor, the A16 Pro. The new chip is also tangibly faster than Qualcomm's Snapdragon 8 Gen 3, at least on this one test.

When compared to PC-grade processors, A18 looks pretty good too. It beats Apple's M3, AMD's Ryzen 9 7950X, and Intel's Core i9-14900KS processors in the single-thread Geekbench 6 benchmark. It also challenges AMD's Ryzen 9 9950X, but falls behind Apple's own M4 (which has 400 MHz higher clocks) both CPUs have considerably higher power budgets. As you might expect, then it comes to multi-thread performance, A18 with its six cores falls significantly behind PC-grade processors. Hence, the intrigue is whether the eight-core A18 Pro can challenge Apple's M3.

Keeping in mind that the A18's frequency increase, when compared to its predecessor is 6%, the majority of single-thread and multi-thread performance improvements should be attributed to microarchitectural enhancements.  

While the performance increases look modest, this isn't particularly surprising as Apple's A18 is made on TSMC's N3E (3nm-class) process technology that was architected to reduce costs compared to N3B (3nm-class), at the cost of transistor density rather than to tangibly increase performance or lower power consumption (even though N3E enables a bit higher clocks and reduces power compared to N3B). In other words, Apple had a limited transistor budget to add here. 

One thing to keep in mind about Geekbench 6 results is that we are dealing with a synthetic benchmark, and the real-world performance of Apple's A18 and A18 Pro will be different. Nonetheless, Geekbench 6 results seemingly demonstrate that Apple's own performance estimates are accurate (who said they are not based on Geekbench 6 though?) and that the company's new processors can offer about 15% performance improvements over the A17 Pro in best-case scenarios. 

AMD's bleeding-edge Instinct MI300A data center APU has been tested in Geekbench 6.3.0. Discovered by BenchLeaks on X, the pair of MI33A APUs performed seemingly worse than a Core i5-14600K across three runs.

The APU showed up as "AMD Proto Sample: SH5-MI300A-A0," somewhat implying that it could be a prototype sample. According to the submissions, the MI300A was tested on a platform called "AMD Corporation Eviden2p," consisting of a dual-socket design. Therefore, the system had 48 cores and 96 threads, as each MI300A wields 24 Zen 4 execution cores. The MI300A operated at 3.7 GHz, the rated CPU peak clock speed for the data center APU.

The MI300A scored 1,798 points in the single-core benchmark and 13,888 points in the multi-core test for the first run. The second run saw scores of 1,938 and 14,694 points. The third run saw scores of 1,992 and 15,085 points, respectively.

Those familiar with Geekbench 6 scores will immediately know how horrible all three benchmark runs are for a chip on the scope and scale of the MI300A. By comparison, Intel's mid-range Core i5-14600K desktop CPU outperforms the MI300A in the same test. The 14600K features a single-core score of 2,806 points and a multi-core score of 15,977 points in Geekbench 6, surpassing the MI300A by a whopping 47% in single-core and 9% in multi-core.

Also armed with 24 Zen 4 cores, the Ryzen Threadripper 7960X was seemingly faster than the MI300A. In Geekbench 6, the Ryzen Threadripper 7960X is well ahead of the MI300A, featuring an average single-threaded score of 3,050 points and an average multi-core score in the 25,000-point range based on test results in the Geekbench browser. The result was to be expected in a way since the Ryzen Threadripper 7960X has a faster base (4.2 GHz) and boost clock (5.3 GHz) than the MI300A.

Given the abysmal results, it's clear that the MI300A wasn't performing to its full potential. We wouldn't be surprised if the chip is not supported correctly on Geekbench 6 or if the cooling could have limited the MI300A. Remember that the MI300A has a peak TDP of 760W.

The MI300A is AMD's flagship APU for the data center; the chip's primary competition is Nvidia's Grace Hopper Superchip, which is currently one of the fastest data center chips in the world. Geekbench 6 targets desktops, laptops, and smartphones. Benchmarking a data center chip with Geekbench 6 is pointless since specialized workloads exist. Nonetheless, sometimes, it's amusing to see how exotic CPUs perform on ordinary benchmarks.

Early this morning, a series of Geekbench OpenCL benchmarks for an AMD Radeon RX 8000-series GPU, which will rival the best graphics cards, was uploaded and subsequently caught the attention of leakers and outlets across the web. Hardware detective BenchLeaks discovered multiple entries of this mysterious "gfx1201" GPU that expose some of its specifications.

The gfx1201 device ID had previously appeared in AMD's ROCm Validation Suite in April. The same patches confirmed Navi 48 as "gfx1201" and Navi 44 as "gfx1200", but these benchmarks all only seem to be for "gfx1201".

Geekbench listed the gfx1201 GPU with 28 RDNA 4 Compute Units, but they're probably the number of Work Group Processors (WGPs). Assuming that RDNA 4's layout hasn't changed (two CU per WGP), the configuration equals 56 CUs or 3,584 Stream Processors (SPs), depending on how you want to look at it. That core count would place the mysterious gfx1201 GPU between a Radeon RX 7700 XT (54 CUs) and a Radeon RX 7800 XT.

The Navi 48-based graphics card, with a very conservative clock speed, completed the benchmark at 2.1 GHz. The graphics card is probably an engineering sample, which would explain the low clock speed. Some existing Radeon RX 7000-series graphics cards feature a boost clock speed of up to 2.5 GHz. RDNA 4, potentially on a newer node, should have no problems reaching the same heights, if not surpassing it. The gfx1201 graphics card seems to have 16GB of memory. Unfortunately, Geekbench doesn't specify the memory type or speed, but it is likely GDDR6. Now, 16GB of GDDR6 implies that the gfx1201 should have a 256-bit memory interface.

Sadly, the benchmark results don't tell us anything conclusive. Where things get a little odd, though, is in the spread of the results—four of the five results are somewhat expected variance, but the earliest, lowest result (12,962) exhibits less than half the performance of the same hardware in the latest four results (ranging from 31,041 to 33,241). 

Geekbench's OpenCL benchmark leaderboard places this ~30K range of OpenCL benchmark scores in line with AMD's mobile Radeon 780M iGPU or old models like the Radeon RX 570 or Nvidia GTX 1060. These aren't the comparisons one would necessarily expect from the AMD RDNA 4-based Radeon RX 8000-series graphics card and could boil down to a mobile or engineering sample.

The fact that we're seeing benchmarks of RDNA 4 graphics cards shows that the arrival of AMD's next-generation Radeon RX 8000 series may not be far off. AMD may use CES 2025 to make an official announcement, but we'll have to wait and see.

Intel's upcoming Core Ultra 7 265KF and Core Ultra 5 245KF, which vie for a spot on the list of best CPUs, have been benchmarked in Geekbench 6 and Geekbench 5, respectively. Benchleaks on X discovered the Core Ultra 7 265KF was up to 4% faster than Intel's outgoing Core i9-14900KF flagship in Geekbench 6 and only 7% slower in multi-core performance. However, the chip showed unimpressive performance against its direct predecessor, the Core i7-14700KF.

The Core Ultra 7 265KF was benchmarked on a Gigabyte Z890 Aero G motherboard utilizing Geekbench 6.3.0. In the Geekbench 6 single-core benchmark, the Arrow Lake processor scored 3,219 points. In the multi-core benchmark, the chip scored 19,433 points. The same CPU and motherboard combo was also tested in Geekbench 6.3.0's GPU test, featuring an OpenCL score of 235,327 with an RTX 4080.

The Core Ultra 5 245K was benchmarked on a Colorful iGame Z890 Ultra motherboard but in a much older version of Geekbench—version 5.4.5. The mid-range Arrow Lake chip scored 2,248 points in the single-core benchmark and 18,354 points in the multi-core benchmark.

The Core Ultra 7 chip's score largely resembles Intel's Core i9-14900K flagship, boasting an official score of 3,085 and 20,763 points in the Geekbench 6 single and multi-core benchmarks. The Core Ultra 7 265KF managed to outpace the 14900K by 4% in single-core performance, but the 14900K can still outpace the 265KF in multi-core performance by 7%. Still, it is impressive to see a Core Ultra 7 product performing similarly to a Core i9.

However, against its direct predecessor, the Core i7-14700KF, the Arrow Lake's scores are less impressive. The i7-14700KF scored 3,004 points in the Geekbench 6 single-core test and 19,583 points in the multi-core test. This results in the Arrow Lake Ultra 7 part beating the 14700KF 7% in single-core performance but unable to win it in muti-core performance, where the 14700KF outperforms the Ultra 7 265KF by a fraction of a percentage.

Unfortunately, we weren't able to directly compare the Core Ultra 5 part to its predecessor or any chip. Geekbench has apparently removed its Geekbench 5 results, making it impossible to compare numbers to this older version of Geekbench.

Regardless, the Core Ultra 7's performance in Geekbench 6 is not promising. The Arrow Lake chip was unable to effectively outperform its predecessor in multi-core performance.

However, these results need to be taken with a grain of salt since these CPUs are possibly engineering or qualification samples that might not have the same clock speeds as their official production-ready counterparts.

Geekbench is moving its artificial intelligence benchmark to prime time. Developer Primate Labs is pushing the newly renamed Geekbench AI (previously called 'Geekbench ML' while in preview) to version 1.0. The benchmark is available for Windows, Linux, and macOS, as well as for Android in the Google Play Store and for iPhones and iPads in Apple's App Store.

Like regular Geekbench, which provides two scores for the CPU, Geekbench AI will use multiple scores. In this case, the software will use three scores that range in precision: single-precision data, half-precision data, and quantized data. Primate Labs claims that this is "to better describe the multidimensionality of AI performance and the impact of different hardware designs."

But there's another component: correctness. Another measurement for each test sees how close the workload delivers what's expected. Primate Labs explained using an example of a model that detects the presence of a hotdog:

"Your hotdog object detection model might be able to run very, very quickly, but if it can only accurately detect a hotdog 0.2% of the time one is actually present, it’s not very good," it explained. "This accuracy measurement can also help developers see the benefits and drawbacks of smaller data types — which can increase performance and efficiency at the expense of (potentially!) lower accuracy. Comparing accuracy as part of performance through the use of our database can also help developers estimate relative efficiency."

A whole bunch of new frameworks are available with the AI 1.0 launch. It now supports OpenVINO, ONNX, and Qualcomm QNN on Windows, OpenVINO on Linux,  and "vendor-specific TensorFlow Lite Delegates" including Samsung ENN, ArmNN, and Qualcomm QNN on Android. This release will also be adding more datasets, the company says.

Scoring and comparing

Primate Labs wants to prevent any gaming of the system, or, more politely stated: "vendor and manufacturer-specific performance tuning on scores[.]" All workloads run for at least one second, which should ensure devices have the time to reach peak performance levels during testing.

Primate Labs suggests this will also help show the difference between phones and desktops, or AI GPUs. If a test is completed too fast the device will underreport its performance, the company says.

Like regular Geekbench, Geekbench AI will report to the Geekbench browser (and potentially provide fodder for upcoming hardware leaks — who knows). The highest-performing devices will be here, while the most recent results will be here. 

Currently, there are few ways to measure AI performance that aren't from the big hardware manufacturers, so this is a welcome addition. Primate Labs said in a blog post that Samsung and Nvidia, among others, are using the software, and the company suggests the software will evolve quickly, with lots of future releases.

But it's difficult right now to evaluate real-world AI performance, as there are limited use cases. So there's a lot of weight riding on just a few benchmarks to tell us what's what.

It looks like those who received their AMD Ryzen 9000 samples early are continuing to test them, as an interesting new benchmark result has been shared online. Twitter’s HXL, an Asus system packing an AMD Ryzen 9 9950X engineering sample (ES), was checked in Geekbench 6. Moreover, this Zen 5 16C/32T consumer flagship was overclocked to 5.95 GHz. That’s a modest 300 MHz faster than the stock maximum clock.

The overclocked Zen 5 desktop chip achieved a 3,706-point single-core score and 26,047 in the multi-core tests built into Geekbench 6. The system included 32GB of DDR5 memory, and as we see, the test run was completed yesterday.

Over recent weeks, we have seen many leaked AMD Ryzen 9 9950X tests. One Anandtech forum user has been busy benchmarking an ES model using Cinebench R23 and a wide range of power targets. It has shed light on this chip’s potential and the diminishing benefits of applying higher power targets.

Back to today’s result from an AMD Ryzen 9 9950X ES overclocked to 6 GHz, we are going to assume that this wasn’t an LN2-cooled feat but someone more casually pushing their chip with an off-the-shelf air or AiO cooler. Importantly, this may be a better indication of the overclock ability and performance that the average buyer will experience. However, please add salt for a leaked benchmark and understand that this is just a sample that may not represent the average retail released 9950X.

AMD’s Ryzen 9 9950X, overclocked to 6 GHz, is at the top of the table, and it certainly looks like it will be an impressive performer. The second result in the table came from a July 10 entry to the Geekbench database, and its 5.7 GHz max clock precipitates a 10% slower single-core score and a gaping 26% slower multi-core score. Something else must be going on, not mentioned in the Geekbench 6 data we can see, that creates such a gulf between the systems.

Another enlightening view of the AMD Ryzen 9 9950X came from an official source, the firm’s in-house Extreme Overclocking team. They cooled the upcoming Zen 5 flagship with LN2 to demonstrate impressive performance in popular benchmarks like Cinebench R23 and R15. The team broke records along the way, thanks to the Zen 5 architecture IPC benefits and lashings of LN2 enabling clock speeds up to 6.75 GHz.

Sadly, we will have to wait a little longer than expected to get chips like the AMD Ryzen 9 9950X in Tom’s Hardware labs. Yesterday the news broke that these Zen 5 desktop chips would start to roll out in August due to unspecified quality issue(s).

Geekbench shared the percentage of processors that ran its Geekbench 6 benchmark in the past 30 days. Out of the 81,081 tests done on a Windows PC between June 16 and July 16, 2024, 5,304 were conducted on a device running the Snapdragon X SoC. It might be small compared to Intel’s 48,970 and AMD’s 26,807, but we must remember that these laptops launched on June 18.

The Snapdragon X isn’t the first chip designed for Windows on Arm, but it’s the first to gain massive support from Microsoft and its partners like Acer, Asus, Dell, HP, Lenovo, and Samsung. Furthermore, these chips are touted to have better efficiency than Intel and AMD offerings, with testing showing them to last more than 15 hours on a single charge.

Apple’s 14-inch and 16-inch MacBook Pros have lasted longer, at 17 hours, but they have larger batteries, 70 and 100Whr (compared to the 66 and 54Whr on the 15-inch and 13.8 Surface Laptops). Aside from the longer battery life, Qualcomm Snapdragon X laptops are the first to get the Copilot+ PC mark from Microsoft.

Although AMD’s upcoming Ryzen AI 300 processors exceed the 40 TOPS minimum requirement that Microsoft set, it’s still not getting the Copilot+ branding at launch. Instead, Microsoft’s AI features will arrive on Ryzen AI 300-powered laptops later this year via a free Windows update.

However, the Snapdragon X launch didn’t go without a hitch. The most interesting Copilot+ feature, Windows Recall, was removed before launch because of its lack of privacy and security. Snapdragon X Elite laptops suffer compatibility issues, with Intel’s integrated Arc graphics performing way better. Many software developers still do not support on-device AI processing, making the Snapdragon X’s (and most new processors’) NPU cores moot and academic.

Nevertheless, the entrance of a new player in the Windows processor market is crucial for consumers. Much like how Apple silicon transformed the laptop market, pushing AMD and Intel to produce more efficient chips, the Snapdragon X will move the two largest processor makers to make more innovations for users who want a Windows PC; otherwise, they risk being overtaken by Qualcomm.

Qualcomm is still far from delivering the performance many enthusiasts and power users expect from their computers. But they’re making strides in that direction. Unless the Snapdragon X encounters some nasty issues that Qualcomm can’t address quickly, we hope the market share between these companies will eventually be balanced.

More Geekbench benchmarks for AMD's all-new Radeon 890M RDNA 3.5 integrated graphics flagship have surfaced. As discovered by Benchleaks over on X (Twitter), new Geekbench Vulkan and OpenCL scores have placed the latest AMD iGPU on the same performance scale as Nvidia's GTX 1070 from eight years ago.

Performance results come from two separate benchmarks, one using OpenGL and one utilizing Vulkan. Both were using the same machine: an Asus ProArt P16 laptop featuring an AMD Ryzen AI 9 HX 370 with Radoen 890M integrated graphics.

The system scored 42,932 points in the OpenCL score and 46,298 points in the Vulkan score. The Radeon 890M's OpenCL score is right ahead of AMD's Radeon RX 580, which scored 41,991 points in the Geekbench browser, and right in line with the Radeon RX 5500, which scored 42,923 points. Performance was also close to the GTX 1650 TI (mobile) and GTX 1070, which scored 44,708 and 45,011 points, respectively.

Thanks to a more modern design, the RDNA 3.5 iGPU gets a performance boost in the Vulkan benchmark. In this benchmark, the Radeon 890M is right on top of the GTX 1070 in performance, with the GTX 1070 boasting a score of 46,299 points—one point ahead of the 890M. Performance is also better than that of the GTX 1650 Ti.

It's worth mentioning that the GTX 1070 results probably come from its laptop counterpart. Regardless, it is impressive to see integrated graphics engines now competing with discrete GPUs (mobile or desktop) from several years ago.

Another highlight is that the Radeon 890M is only 15% slower than the GTX 1650 Super in Geekbench's Vulkan database and faster than the fastest GTX 1650 result we could find in the same database. That is arguably just as impressive, if not more impressive, than the fact it is competing with the GTX 1070. The GTX 1650 ranks as the second most popular GPU in the Steam database, which means the Radeon 890M has performance comparable to not just any discrete GPU but performance similar to one of the most popular GPUs on the market today (at least according to Steam's database).

We must always take Geekbench results with a pinch of salt. Performance implications from these results are not always indicative of real-world performance. We will have to wait for third-party reviews of the Radeon 890M in actual gaming benchmarks to see where it stands.

But if these results indicate the Radeon 890M's performance, it'll be a fantastic little iGPU for light gaming on the go.

 Intel is continuing the trend of replacing its entry-level processors with the generic "Intel Processor" nomenclature. A new Geekbench listing discovered by BenchLeaks on X reveals a new entry-level chip, dubbed the Intel Processor 310. It will feature two performance cores and four threads, similar to Intel's Celeron and Pentium processors of old.

The 310 is unique in that it is only one of two Intel Processor models (so far) sporting no efficiency cores and classified as a Raptor Lake CPU (but still uses an Alder Lake die). In Geekbench, the chip's performance reflects this, boasting very good single-core performance compared to other Intel Processor variants, such as the N200 series. Multi-core performance is also quite decent, thanks to its use of HyperThreading technology. It enables the chip to approach the multi-threading power of a Core i3-N305, which comes with eight Gracemount efficiency cores.

 The Geekbench listing reveals that the Intel Processor 310, paired with a single 8GB stick of DDR4 memory, achieved a single-threaded score of 2,152 points and a multi-core result of 4,254 points. The next closest chip in the Intel Processor lineup is the Core i3-N305, which comes with eight Gracemount efficiency cores sporting a single-core result of 1,431 points and 5,538 multi-core result.

The closest Raptor Lake processor to the Intel Processor 310 is the Core i3-13100 series, which features four cores, and much higher frequencies. The vanilla variant of the 13100 has a 2,253 single-core result and a 7,368 multi-core result. Single-core performance is comparable to the 310, but the Core i3 destroys the 310 in multi-core performance due to the extra cores.

Unfortunately, we can't read into these results with incredible detail since the 310 was operating with just a single stick of memory. This inevitably hampers the 310's score significantly since the CPU's second memory channel wasn't used. The good news is that we can expect this chip to perform much better in future benchmarks where two DRAM sticks are enabled, but don’t expect the chip to get close to the Core i3-13100’s multi-threaded performance regardless due to the core count deficit.

The Intel Processor 310 is part of Intel's initiative to remove the Celeron and Pentium sub-brands altogether. The 310 and the vanilla 300 will be taking their place, representing the cheapest modern Intel processors you can buy compatible with Intel's LGA 1700 desktop sockets. The main difference between these low-end chips and the Core i3 is core count, with the 310 only having two cores, and the Core i3 boasting four cores. Clock speed is also major difference that favors the core i3 parts, with the Core i3’s having turbo-boosting technology. According to the 310 Geekbench listing, the chip's maximum recorded clock was just shy of 4.1GHz, while the Core i3-13100 boasts turbo frequencies of up to 4.5GHz.

The 310's power rating is unknown, however based on the Intel Processor 300's base power of 46W, we suspect the 310 will be the same. If so, power consumption will also be a big differentiation between the two chips, with the i3-13100 able to suck down 60W at a minimum and up to 110W through Intel's modern PL1/PL2 system.

Pricing and availability are unknown, but now that its existence has been revealed, we can expect this chip to debut soon.

Intel’s Lunar Lake processors aren’t slated to launch until this September, but we’re now seeing trickles of alleged benchmarks appear online. Benchleaks just shared Vulkan test results marked as Intel Core Ultra 7 258V, with the first one getting 31,560 points, followed by a second test after 15 minutes showing 34,181 points.

The tests list the following CPU information: Intel Core Ultra 7 258V 2.2 GHz with eight cores and eight threads. The system used for the tests shows 32GB of RAM with Windows 11 Enterprise OS, and the involved processor could hit a maximum frequency of 4.8 GHz on the Balanced power plan.

The results put the Intel Core Ultra Series 2 in the same neighborhood as AMD’s Radeon 780M found in the Ryzen 7 8700G processor, the entry-level Intel Arc A380, and Nvidia’s GTX 1060. These are the numbers from Geekbench 6 Vulkan tests.

We can see that Intel’s latest laptop processor is nearly on par with the last-gen Radeon 780M integrated GPU from AMD and is already within spitting distance of the entry-level discrete GPUs from Team Blue, Team Green, and Team Red. This could conceivably allow laptops that run these chips without a discrete GPU to do some light gaming, or it could even drive manufacturers to use Intel’s chips to build handheld consoles.

However, you should note that these are just alleged test results. We cannot verify if these come from valid systems, especially as Intel’s partners are likely contractually bound to stay silent until after these chips officially launch. And even if they come from laptops with the Intel Core Ultra 7 258V, we won’t know if the specs noted in the tests will be the same as on the final retail version. After all, these tests were likely done under lab conditions to test and validate the performance of their engineering samples.

One more thing you should note about synthetic benchmarks is that they do not indicate the real-world performance you will get from a particular system. While they may show ballpark figures versus how these chips compare against other graphics processors, the only way to know how these will perform is after we run them through actual gaming benchmarks and see the FPS they deliver. So, before purchasing (or preceding) these new Intel processors, please wait for the review results and see how they work out.

AMD's upcoming Ryzen 9000 chips aren't expected to arrive until July 31, but we're seeing some early benchmarks appear here and there. The @Benchleaks bot recently shared Ryzen 7 7900X and Ryzen 5 9600X results on X, revealing performance higher than AMD's and Intel's top previous-generation chips.

These leaked performance numbers usually come from PC manufacturers conducting final testing of their retail units before they go on sale. However, you should still take these results with a grain of salt, as these companies can change the final hardware configurations before they hit store shelves. Also, the results come from just a single test of each processor; it will take several more tests before we can see the general performance trends for AMD's latest chips.

Nevertheless, the two tests show the processors paired with 32GB RAM DDR5 RAM, with the 9700X mounted to an Asus ROG Crosshair X670E Gene and the 9600X on an X670E Hero motherboard.

Geekbench reports that the Ryzen 7 9700X has eight cores and 16 threads, a base frequency of 3.8 GHz, a max frequency of 5.54 GHz, and 32MB of L3 cache. On the other hand, the Ryzen 5 9600X has six cores and 12 threads, a 3.9 GHz base frequency, and a 5.47 GHz max frequency. It also has the same 32MB L3 Cache.

These are the Geekbench 6 results of other processors so we can see how these new processors compare:

Both Ryzen 9000-series processors outperform the previous generation AMD and Intel chips in single-core score performance despite having a lower base or boost clock, probably owing to AMD’s new Zen 5 architecture. Nevertheless, Intel’s 14th-gen chips still outperform them in the multi-core department, especially with their P-core / E-core architecture.

While the benchmarks may show the general theoretical performance of a processor, how fast your computer will be will still vary with the components you get, plus your luck with the silicon lottery. And before you make a long-term purchase of such a significant PC part, you should always wait for benchmarks that put these chips through their paces in both theoretical and real-world applications. 

A purported AMD Ryzen 9 9900X Geekbench 6 CPU test run has been surfaced by Benchleaks. The single-threaded score of 3,401 points is remarkable and, if genuine, would go straight to the top of the Geekbench 6 processor benchmark charts. A reported multi-threaded benchmark score of 19,756 points is less impressive for a 24T part, but isn’t shamed by the best consumer 24T CPUs from Intel.

The Geekbench result you see above appears to come from a user testing a new Asus system earlier today. In addition to the new AMD ‘Granite Ridge’ desktop CPU, the system uses the ROG Crosshair X670E Gene motherboard as its foundation. An ample 32GB of DDR5 RAM is installed in the system.

Looking closer at the CPU specifications surfaced by Geekbench 6, the chip under scrutiny is a Ryzen 9000 part using the Zen 5 microarchitecture. It wields 12 cores and 24 threads, and Geekbench 6 reports that the processor’s base frequency is 4.40 GHz, while it can boost to 5.66 GHz. The chip is also running with a TDP of 120W.

For some perspective, let us compare the new AMD ‘Granite Ridge’ desktop CPU against some well-known recent Intel rivals, as well as one of the best Zen 4 beasts…

It is worth reiterating the significant jump in single threaded benchmark test results we are seeing with the new Zen 5 sample. In this part of the Geekbench 6 test it is 16.5% faster than the AMD Ryzen 9 7950X3D, and against the prior single-core champ, the Intel Core i9-13900KS, it is 9.5% faster.

We of course have to add a pinch of salt to benchmark results that are unearthed in online databases, as there are known cases of people uploading deceptive results, but the AMD Ryzen 9 9900X results are not outlandish. This result also helps build excitement for the upcoming Ryzen 9000 desktop processors, as the tested sample seems to have performed so well with a very respectable 120W TDP.

AMD will ship these new Zen 5 architecture desktop CPUs later this month and we can’t wait to put them through our extensive testing.

AMD's two new Ryzen AI 9 HX 300 processors are the only Zen 5-based mobile chips on the market so far. But that could change soon; Benchleaks on X (Twitter) has discovered the first non-Ryzen 9 AI-series CPU from AMD in the Geekbench browser, featuring the Ryzen AI 7 Pro 160 with just eight Zen 5/Zen5c cores.

The new CPU is undoubtedly one of AMD's upcoming lower-end SKUs that will join the CPU manufacturer's lineup of outgoing Ryzen AI 300 series processors. Adding the 'Pro' nomenclature confirms that AMD will make professional/business variants of the AI 300 series, just as it has done with previous CPU generations. The 100 series nomenclature is a naming scheme AMD officially changed at the last minute before the Ryzen AI 300 series debut. As a result, we can expect this new chip to be changed to 300 series nomenclature when it hits the market.

According to the Geekbench spec sheet, the Ryzen AI 7 Pro 160 sports eight cores in two clusters, one holding three cores and the other five cores. The first cluster houses Zen 5 cores, while the other houses more power-efficient Zen 5c cores. The chip also comes with Radeon 870M integrated graphics, a new mid-range RDNA 3.5-based iGPU that will compliment the flagship Radeon 890M launched with the Ryzen 9-based AI 300 series variants.

One strange addition is the L3 cache capacity, which comes in at just 8MB. This could be an error on Geekbench's part, but if this information is correct, it represents a significant reduction in L3 cache compared to previous generation parts which had 16MB (desktop variants have up to 32MB).

Geekbench performance is within striking distance of the previous generation Ryzen 9 8945HS, which also comes with eight cores. The 'AI Pro 160' chip yielded a single-core result of 2,514 points and a multi-core score of 11,772 points, respectively. The Ryzen 9 8945HS yielded a slightly slower score of 2,387 points and 11,654 points, respectively, in the single and multi-core Geekbench benchmarks. This represents a 5% advantage and 1% advantage in the single -and multi-core results in favor of the Ryzen 7 AI Pro 160.

Assuming these benchmark results showcase the real-world performance of the new AI pro 160 chip, these results provide evidence that AMD's Zen 5-based Ryzen 7 parts will perform very similarly to its previous generation Ryzen 9 chips. Users will be able to opt for a Ryzen AI 300 Ryzen 7 part (when they come out) and know they will get previous-generation Ryzen 9 performance.

Compared to Intel, the new Ryzen 7 part outperforms Intel's Core Ultra 9 185H but doesn't have enough juice to outperform Intel's more performance-oriented mobile processors (based on Raptor Lake Refresh), such as the Core i5-14500HX and Core i7-14700H. But these Raptor Lake parts are much more inefficient than AMD's Zen 5 processors, so it isn't an all-out win for Intel. AMD's Ryzen AI 300 series CPUs (including the new Ryzen 7 part) all target a balance of power efficiency and performance to work best inside thin and light laptops. By contrast, Intel's Raptor Lake Refresh parts are mainly aimed at high-performance laptops, such as big bulky gaming laptops. The best comparison is from the Core Ultra 9 185H, Intel's closest competitor (until Lunar Lake debuts) to the new Ryzen AI 300 series.

As previously stated, we expect this Ryzen AI 7 Pro 160 processor to arrive in the future but rebranded to AMD's official 300 naming scheme, which the company switched to right before Computex 2024. With eight core Ryzen 7 Pro processors on the horizon, we can expect mainstream Ryzen 7 parts, as well as Ryzen 5 and Ryzen 3 parts, to come in the not-too-distant future.

Two new Intel Lunar Lake results have cropped up on the Geekbench browser (as spotted by @BenchLeaks on X), featuring the upcoming Core Ultra 7 268V with four non-Hyper-Threading P-cores and four E-cores. If we take these results at face value, the new CPU's performance in Geekbench is a mixed bag. It boasts superior single-core performance but not enough multi-threaded juice to defeat Intel's Meteor Lake Core Ultra 7 parts. That's a solid result, though, as Lunar Lake is designed for lower-TDP thin-and-lights and compact notebooks than the higher-tier Meteor Lake processors.

Two listings featuring the Core Ultra 7 268V have appeared on Geekbench; the first reveals a score of 2,713 points and 10,036 points for the Lunar Lake chip in the single and multi-core benchmarks. The other shows a slightly better single core but a slightly inferior multi-threaded score of 2,739 and 9,907 points, respectively. 

For simplicity's sake, we will use the average of both scores to better compare the chip to other chips further down the article. Our simple calculation results in scores of 2,726 and 9,972 for the Lunar Lake part, in single and multi-threaded workloads.

Compared to the Ultra 7 268V's outgoing Meteor Lake-based predecessors, its performance in Geekbench is mixed. The chip is 15% faster in single-core performance compared to the Core Ultra 7 155H, but it is 19% slower in Geekbench's multi-core benchmark. Our Ultra 7 155H result was taken from an HP Spectre laptop sporting a single-core score of 2,356 and an 11,926 multi-core score. There are certainly inferior scores in the Geekbench browser, but we found this one in particular to be a good balance between the fastest and slowest Core Ultra 7 155H results in the online results.

It will be interesting to see if this behavior exists in subsequent benchmarks and with production-ready devices packing Intel's Lunar Lake chips. One of the biggest changes Intel has made is removing its Hyper-Threading technology from its Lunar Lake chips in an effort to boost power efficiency. In addition, the Ultra 7 268V only has four E-cores, while the Core Ultra 7 155H has eight. 

Intel will rely entirely on its substantially faster E-cores in Lunar Lake to combat the architecture's E-core deficit compared to Meteor Lake and even Raptor Lake / Alder Lake. Intel's upcoming Lunar Lake chips come with upgraded E-cores based on the Skymont architecture, sporting up to a 68% IPC gain compared to Meteor Lake's Crestmont LPE cores.

Again, it'll be very interesting to see if Intel's new E-core architecture can rectify Lunar Lake's disadvantage in core count compared to Meteor Lake. However, we must note that a major focus of Intel's Lunar Lake is reduced power consumption for thin-and-lights, so it still impresses in the multi-threaded tests if you take into account the comparative thread counts and target TDPs (reported default of 17W for most models, with 30W peak). In contrast, Meteor Lake processors are designed to scale to higher TDPs. Intel has yet to introduce its successor for the higher-powered 45W/115W Meteor Lake chips.

In conclusion, Geekbench is only one benchmark (and this is a leak), so we'll have to see how Lunar Lake performs in other benchmarks/apps before we get a clearer picture.

Igalia, the free software consultancy perhaps best known for its work on the Raspberry Pi's GPU, has revealed that it is investigating NUMA (Non-Uniform Memory Access) emulation for ARM64 devices. The investigations have so far yielded a potential and significant performance uplift for the Raspberry Pi 5, discussed on a Linux kernel list via a message from Tvrtko Ursulin.

The patch details were posted to the mailing list, and it appears to be around 100 lines in length. However, those 100 lines potentially have a big impact on the Raspberry Pi 5 and many other ARM64 devices. 

According to the post. "This series adds a very simple NUMA emulation implementation and enables selecting it on arm64 platforms."
This improves single-core performance by 6% and multi-core performance by approximately 18%. These figures were determined using Geekbench 6 test runs.

Ursulin explains in a little more depth: "[...] splitting the physical RAM into chunks and utilizing an allocation policy such as interleaving can enable the BCM2721 memory controller to better utilize parallelism in physical memory chip organization."

What could this mean for the Raspberry Pi 5? Overall better performance from an already performant 2.4 GHz Arm CPU, which can be easily overclocked to 3 GHz or more.

The code is out for review, and with a little luck and hard work from the Linux Kernel developers, this patch could add even more performance to the Raspberry Pi 5 and many other ARM64 devices.

NUMA emulation, mainly used in systems with multiple processors, is a computer memory design where memory access times depend on the memory location that is relative to a processor. In simple terms, NUMA allows each CPU to have its own bank of locally attached memory while still having access to the memory directly connected to other processors in the system. This results in fast latency for 'near' memory (locally attached) but slightly slower latency for 'far' memory (memory directly attached to other processors in the system).  

The Linux Kernel documentation page goes into NUMA with a little more depth when it comes to the Linux software stack. "Linux divides the system’s hardware resources into multiple software abstractions called “nodes.” Linux maps the nodes onto the physical cells of the hardware platform, abstracting away some of the details for some architectures. As with physical cells, software nodes may contain 0 or more CPUs, memory and/or IO buses. And, again, memory accesses to memory on “closer” nodes–nodes that map to closer cells–will generally experience faster access times and higher effective bandwidth than accesses to more remote cells."

The patch claims, "Code is quite simple and new functionality can be enabled using the new NUMA_EMULATION Kconfig option and then at runtime using the existing (shared with other platforms) numa=fake= kernel boot argument."

We'll investigate this and see if we can reproduce Igalia's results.

AMD's fire-breathing MI300X GPU has made its official debut on Geekbench 6 OpenCL, outpacing previous chart-toppers such as the RTX 4090. However, despite being one of the fastest GPUs on the Geekbench 6 charts, the AMD GPU's score does not reflect its real performance and shows why it's a terrible idea to benchmark data center AI GPUs using consumer grade OpenCL applications (which is what Geekbench 6 is).

Let's get the benchmark numbers out of the way, though. The MI300X boasts a score of 379,660 points in Geekbench 6.3.0's GPU-focused OpenCL benchmark, making it the fastest GPU on the Geekbench browser to date. (Note that it's not listed on the official OpenCL results page yet.) That gives it the pole position, ahead of the second highest score that goes to, ironically, another enterprise GPU, the Nvidia L40S. The L40S managed 352,507, which in turn beats the RTX 4090's 319,583 result by 10%.

So, the MI300X beats all contenders right now, outpacing the RTX 4090 (the fastest consumer GPU on the list) by 60,077 points or 18.8%. Clearly, other factors are holding back some of these GPUs, as Nvidia's H100 PCIe also shows up on the list with a meager score of only 281,868. Don't use Geekbench 6 OpenCL as a measuring stick for enterprise-grade hardware, in other words. It's like driving a Formula One car in a school zone to check the acceleration and handling of the vehicle.

We should also discuss price. RTX 4090 is easy enough to pin down, with a $1,599 MSRP and a current lowest price of $1,739.99 online. AMD MI300X is a different story, as you generally buy those with servers and support contracts. However, a quick search gives a suggested price of anywhere from $10,000 to $20,000 per GPU — we can't say for certain how accurate that data is, as the companies actually buying and selling the hardware generally don't reveal such information, but obviously the MI300X plays in a completely different league than consumer hardware. You also can't just run out and buy an MI300X to slot into your standard desktop PC; you'll need a server with OCP accelerator module support.

You only need to look at the rankings to quickly see that all may not be right with this benchmark. The RTX 4090 outpaces the RTX 4080 Super by 28% — not an entirely out of the question result, but that's using the same architecture. The RTX 4080 Super meanwhile beats AMD's top consumer GPU, the RX 7900 XTX, by 21%. If this were a ray tracing or AI performance test, that wouldn't be out of the question, but in general FP32 compute performance the 7900 XTX tends to be far closer than these results would suggest. And again, that's not even looking at the often terrible results from data center GPUs like the MI300X and H100.

Spec-wise, the AMD MI300X GPU should be in a league of its own. It has 192GB of HBM3 memory with 5.3 TB/s of bandwidth, paired with 304 CDNA3 Compute Units (CUs) and 163.4 TFLOPS of FP32 performance. And that's not even its strong suit. As an AI GPU, it also boasts 2.6 petaflops of FP16 throughput, as well as 2,600 TOPS of inference performance — sort of puts the 40 TOPS requirement of Copilot+ to shame. The MI300X also comes with severe power requirements to match, with a peak 750W power rating.

The MI300X is AMD's latest enterprise GPU, designed to compete with the likes of Nvidia's H100 and H200 AI GPUs. The GPU takes advantage of AMD's CDNA 3 graphics architecture and heavily utilizes 3D-stacking technologies. In fact, the GPU itself is so large that it does not come in a traditional PCIe graphics card form factor. In proper AI-based benchmarks, the MI300X is purportedly up to 60% faster than Nvidia's H100, let alone the RTX 4090.

By contrast, the RTX 4090 is barely half as powerful as the AMD chip in FP32 perforamnce. It features 24GB of GDDR6X and 1TB/s of memory bandwidth, 128 SMs with 82.6 TFLOPS of FP32 compute, and 1,321 TOPS of AI performance. Power consumption is also substantially lower at 450W.

The MI300X's Geekbench 6 debut reveals just how poor such a test is for measuring higher performance GPUs. Sure, sometimes the results aren't terribly out of whack, but OpenCL driver optimizations alone likely account for a large amount of the potential performance. The test can run on a wide range of hardware — Qualcomm's Snapdragon X Elite as an example posts a score of 23,493 — but it's clearly not tuned for all potential workloads. Like most synthetic benchmarks, it only looks at a very narrow slice of the potential performance on tap.

And that's fine, just as long as people looking at the benchmarks know what they mean. We're pretty certain the MI300X result is more just someone with access to AMD's MI300X having some fun seeing what would happen on Geekbench 6, rather than a serious effort to evaluate the GPU. We can't wait to see how 1.2 million GPUs in a supercomputer cluster rate in the same test.

The Ryzen AI 300 series still has a few weeks until its release, but new Geekbench scores bode well for the chips. Two new engineering samples of the Ryzen AI 9 HX 370 APU were found on Geekbench today, reaching scores beating the highest-end Zen 4 mobile processors, and running at their full speed.

The last time Strix Point (AMD's codename for the upcoming Ryzen AI 300 series) was seen in Geekbench leaks, it turned in impressive test scores but was running in Silent Mode. This power-saving meant the chip only reached a core clock of 3.67 GHz, well shy of its recorded boost clock of 4.2 GHz. Today's test scores let the processor fully loose, with two chips hitting 4.5 and 4.8 GHz. The 4.5 GHz chip still holds the outdated "Ryzen AI 9 HX 170" name for the 370, while the other is an unnamed engineering sample. 

As in our previous coverage, below is a table of the Ryzen AI 9 HX 370's best result in this latest batch of results (the 4.5 GHz test) compared with the June 7th results, as well as its predecessor, the Ryzen 9 8945HS, and AMD's fastest available mobile processor, the Ryzen 9 7945HX3D. 

In just the two weeks between the Silent Mode benchmark and today's results, Strix Point has seen some improvement. Turning off Silent Mode and likely some level of driver optimization and development contributed to an 11% jump in single-core scores and a 4% gain in multi-core. With an expected due date of July 15th, AMD and its OEM partners are likely to continue optimizing performance until better results are unlocked alongside the retail launch of the 300 series. 

As expected from a generational leap, the HX 370 soundly beats the 8945HS, giving a 19% increase in single-core performance and a 25% boost in multi-core. This is more in line with the 16% improvement in IPC between Zen 4 and Zen 5 CPU architecture that AMD has been advertising. 

More impressively, the Ryzen AI 9 HX 370's single-core results squeezed past the Ryzen 9 7945HX3D, beating its 2,820 score with a 2,833. This is a bit of a shock, considering the 7945HX3D contains AMD's 3D V-Cache technology, which powers AMD's best CPUs for desktops ( for more on 3D V-Cache, see our explainer) and Dragon Range's higher TDP over Strix Point. What's unsurprising is the 7945HX3D's 11% lead in multi-core testing, thanks to its four additional cores. 

Geekbench GPU results also popped up today for the integrated graphics within the Ryzen AI 9 HX 370 APU, though the Radeon 880M is named rather than the 890M which is confirmed to be in the HX 370. Results in Geekbench 6's OpenCL and Vulkan tests are in line with what we reported last week on the Radeon chip's early TimeSpy results. With a Vulkan test score of 33,849, the "880M" falls in line with the performance of discrete GPUs like the GTX 1650 Max-Q (34,169) and beats Intel's desktop Arc A380 (31,766). 

We would assume that the integrated graphics were listed incorrectly in the Geekbench tests, as the HX 370 has the Radeon 890M inside, but engineering sample weirdness or some other problem could be to blame. The RDNA 3.5-based Radeon 890M GPU has 16 compute units, which allows it to soundly beat its predecessor, the Radeon 780M. 

AMD's Strix Point gets better every time we see it. Even with a confusing last-minute name change from the 100-series to 300-series to have a bigger number than Intel, the benchmark results it turns in are good enough to forgive AMD's clout chasing. The Ryzen AI 9 HX 370 will reportedly be released in laptops starting July 15th. And if launch performance is anything like what we're seeing today, it should be a big win for AMD over Intel in the laptop market in both graphics and processing power.

More performance numbers surrounding the new Ryzen AI processors leaked following AMD's official Computex 2024 announcement. Discovered by Benchleaks, a Ryzen AI 9 HX 370 12-core CPU was benchmarked on Geekbench 6 and sports some very impressive numbers.

The CPU was tested on an Asus ProArt P16 prosumer laptop sporting 64GB of memory. Crucially, the chip was benchmarked using the laptop's silent power plan, which reduces system performance and fan speed to maintain quiet operation even under load. It's worth mentioning that Geekbench 6 classifies the CPU as the "AMD Ryzen AI 9 HX 170" with a 100 series nomenclature. However, this is inaccurate and resulted from AMD switching the branding name to the 300 series at the last minute.

Running in a neutered power state, the Ryzen AI 9 processor scored 2,544 points in the single-core benchmark and 14,158 in the multi-core test. Compared to its predecessor, the Ryzen 9 8940HS, the AI-branded chip is 6.5% faster in the single-core test and 17% faster in the multi-core test. The Ryzen 9 8940HS pumped out scores of 2,380 and 11,775, respectively, for the single and multi-core tests.

The Ryzen AI 9 HX 370's performance is so good that it impressively approaches the performance of AMD's Dragon Range 16-core mobile processors. Looking at the Ryzen 9 7945HX3D benchmark numbers with 3D-VCache technology, the Ryzen AI 9 CPU is only 11% slower in Geekbench's single-core test and 16.2% slower in multi-core performance. The 7945HX3D's Geekbench scores were 2,820 and 16,460 respectively.

These numbers are pretty impressive, all things considered; the fact that we are seeing anything close to Dragon Range levels of performance in an efficiency-optimized power state makes us wonder how much more potent the Ryzen AI 9 HX 370 will be when it is kicked up to its maximum thermal and power envelope. The new chip is part of AMD's latest Ryzen AI 300 series processors sporting its bleeding edge Zen 5 CPU architecture, RDNA 3.5 integrated graphics, and XDNA2 neural engine.

The Ryzen AI 9 HX 370 is a hybrid processor sporting 12 cores in total, consisting of four high-performance Zen 5 cores and eight compact Zen 5c cores. The TDP is rated at 28W, but the configurable TDP ranges between 15W and 54W. As a result, performance will vary wildly depending on what configurable TDP configuration each laptop maker chooses.

We have to take these results with a pinch of salt since Geekbench 6 is just one synthetic test out of many. But, if Geekbench 6's results are indicative of the CPU's real-world performance, AMD's new Zen 5 chips will be really fast, even at lower power envelopes.

China's homemade CPUs have not been very competitive compared to Western competitors, but that appears to be changing. According to IT Home, a desktop sporting a Hygon C86 3350 CPU was benchmarked by a Bilibili content creator on Geekbench, showing Zen 2 levels of performance.

The chip was benchmarked inside a Tsinghua H880-T3M desktop PC aimed at government and enterprise customers. The unit sports 8GB of RAM, a 256GB SSD, and a mysterious 2GB graphics card. Strangely, the Geekbench results say the system has 64GB of memory, so take the RAM rating with a grain of salt. The processor itself is an 8-core SKU featuring a base frequency of 3GHz.

In Geekbench 6.3.0, the Hygon C86 3350 managed a single-core score of 984 and a multi-core score of 5,379 points. The chip's performance is very close to the Ryzen 5 3500 in the same benchmark, particularly its multi-threaded performance. The Ryzen chip features a score of 1,500 points for single-core performance and 5,380 for multi-core performance.

Single-core performance is obviously a weak spot for the Chinese CPU, but its multi-core performance is at least comparable to the Ryzen 5 3500. That might sound generous given that we are comparing an eight core CPU to a six core part, but you'll see in a moment why this is still a performance win for the Chinese chip.

When we last looked at a Hygon CPU, the company barely achieved first-generation Zen performance levels. In fact, this previous chip, with its eight cores, was so bad that it took two of these CPUs (the C86 3185) to outperform the Ryzen 5 5600X in multi-core testing. Even then, the performance gap was not much, given that the two Hygon CPUs had 32 cores combined, compared to the Ryzen 5's six.

The Hygon C86 3350 is a much different story. Its multi-threaded performance is comparable to Zen 2's with almost the same core count. On top of this, it attains this performance level with just a single chip rather than two. This is a big step forward and shows that Hygon is progressing in its CPU development.

Sadly, due to Chinese intervention, getting more details about Hygon's chips is nearly impossible beyond the most basic specifications. However, it is interesting to see how the CPU manufacturer is developing despite its limited scope.

Apple yesterday announced the latest iteration of its own silicon with the M4-powered iPad Pros. The company claims that the CPU is 50% percent faster than the M2 on the 12.9-inch iPad Pro (6th-generation), sporting two more cores (4 performance + 6 efficiency) over the last generation’s eight cores (4 performance + 4 efficiency). Although we cannot yet confirm these claims, the M4’s Neural Engine was tested today on Geekbench, with machine learning (ML) benchmarks posted under iPad 16,3. As ever with leaks and unverified benchmarks, take the news with a pinch of salt.

The results do not explicitly say iPad Pro (13-inch), but the specifications noted on the benchmark match that of the higher-end M4-powered iPad: a 10-core ARM processor with 16GB RAM. We also compared it with the 12.9-inch iPad Pro (6th generation)’s machine learning test results to see how it would stack up against, and the results show an incremental improvement over the last gen’s ML performance.

The M4 iPad Pro hit a Geekbench ML 0.6.0 score of 9234, around 22.9% better than the M2-powered iPad Pro’s 7511. We also compared the M4 iPad Pro’s ML performance with an M3 14-inch MacBook Pro, where its gains were a little more modest at 10.4% (8365 vs. 9234). However, the M3 MacBook Pro might have a slight advantage over the M4 iPad Pro owing to its larger chassis and battery.

The original M1 Apple silicon was a true generational leap over Intel’s x86 processors. However, the subsequent M2, M3, and M4 chips are just building on the M1’s architecture instead of offering massive gains. Nevertheless, the M4 focuses heavily on AI processing, with its 16-core neural engine supporting up to 38 trillion operations per second (TOPS). In comparison, the previous-gen M3 also has a 16-core neural engine. Apple's M3 was rated for 18 TOPS at FP16 precision, but the M4 is rated for 38 TOPS with INT8. That means, if equalized to INT8 precision, we're looking at a 5% improvement in TOPS for the M4 over the M3.  

However, the May 7 presentation mainly focused on hardware — the new iPad Air, iPad Pro, and the Apple Pencil Pro. We’ve only received a few snapshots of what AI features the M4 silicon will bring us, mostly on its use with Final Cut Pro for iPad 2 and Logic Pro for iPad 2. Aside from that, there wasn’t much mention of AI; we expect that to happen during WWDC 2024 instead.

Apple still has the upper hand in mobile computing, especially as Windows laptops still lag behind in terms of efficiency. But Qualcomm isn’t far behind, with the Snapdragon X Elite and X Plus chips beating AMD and Intel’s top laptop processors. Assuming Windows 11 (or Windows 12) can take proper advantage of these new Arm processors, we’re entering a new golden age of powerful yet highly efficient laptops, and Apple has to step up its game to remain in the lead.

An engineering sample of a Zen 5 Strix Point CPU has made an appearance on Geekbench 6. Discovered by @haurkaze5719 on X (formerly Twitter), the CPU features 12 cores, 16MB of L3 cache, and a 28W TDP. Strix Point is the codename for AMD's upcoming Zen 5-based Ryzen mobile processors.

Unfortunately, the benchmark results on Geekbench 6 don't show us the chip's full potential. In fact, its performance is drastically lower than today's Ryzen 8000 mobile processors, likely due to the fact that it's an engineering sample. The CPU only achieves a multi-core score of 8,016 points and a single-core score of 1,217 points. By comparison, a six-core Ryzen 5 8640U utilizing AMD's Zen 4 architecture archives well over 8,500 points in the multi-core test and over 2,000 points in the single-core test, in most cases.

The Strix Point sample's slow performance is caused by the chip's extremely slow clock speeds —  the engineering sample was running at just 1.4GHz throughout the whole test.

That said, this new engineering sample reveals that AMD is currently working on Zen 5-based mobile solutions. In fact, the sample has the same FP8 nomenclature that we saw on a shipment of CPUs one month ago. So it's been around for some time.

The core CPU specs align with speculation that Strix Point will feature upgraded core counts compared to AMD's current Ryzen 7040 and 8040 series mobile processors. There's a chance Strix Point could feature 12 P-cores, but it's more likely that it will use a mixture of P-cores and Zen 5c E-cores to hit the 12-core target. Unfortunately, it appears the L3 cache and L2 cache are still unchanged, coming in at 16MB and 1MB respectively. We have yet to see how this will impact performance, but it appears Zen 5 will be getting most of its performance prowess from its Zen 5 cores alone.

What we do know, aside from the Zen 5 aspect of the chip, is that Strix Point will come with a more potent RDNA 3+ integrated graphics unit, as well as a significantly more powerful XDNA 2 NPU. AMD confirmed that XDNA 2 will offer twice the performance of its predecessor, translating to over 70 TOPS of performance — well above Microsoft's 40 TOPS requirement for AI PCs.

Zen 5 is expected to feature a 15% IPC improvement just from the architecture itself. With the addition of 12 Zen 5 cores on top of the GPU and NPU improvements, Strix Point is shaping up to be a significant upgrade over AMD's existing low-power Ryzen 8000 series CPUs.

@techinmul on X (Twitter) shared performance figures for Qualcomm's new Snapdragon X Elite CPU in Geekbench 6. The ARM CPU's results are incredibly potent, performing on par with Intel's Core i7-13700HX, Core i9-13900H, and AMD's latest Ryzen 9 8945HS mobile flagship. If these numbers match real-world performance, Qualcomm's new ARM-based laptop chip will be able to perform on par with some of the best hardware Intel and AMD have to offer today.

The Snapdragon X Elite scored 2,427 points in Geekbench 6's single-core test and 14,254 points in the benchmark's multi-core test. Compared to Intel's plethora of high-end Core i7 and Core i9 Raptor Lake mobile CPUs, and AMD's top Ryzen 8000-series mobile CPUs, Qualcomm's new chip outperforms them.

In the single-core test the Snapdragon chip outperformed the i9-13900H by 3.7%, the i7-13700HX by 5.9%, and Ryzen 9 8945HS by 2.2%. In the multi-core test, the Snapdragon X Elite beat the 13900H by 25.7%, the 13700HX by 18.4%, and the 8945HS by 18.3%.

Assuming these results showcase the Snapdragon's real-world performance; Qualcomm has an incredibly potent chip on its hands. We already knew from previous official coverage that the Snapdragon X Elite would be very fast with its 12 Oryon cores, 42MB of cache, and 4.3GHz peak turbo frequency. These Geekbench 6 results serve as additional confirmation that Qualcomm has reached its performance goals with the Snapdragon X Elite.

Qualcomm is targeting "best-in-class" multi-threaded performance with the Snapdragon X Elite. Unlike its previous laptop chips, Qualcomm is aiming to compete with the best laptop CPUs in the industry. Qualcomm has already revealed that its new chip is 60% more performant than the i7-13800H, and is twice as fast as Intel's i7-1355U and i7-1360P while sipping just a third of the power.

Qualcomm is also going after Apple, which is its main competitor when it comes to ARM-based laptop CPUs. Qualcomm states that the Snapdragon X Elite provides 50% better peak multi-threaded performance than the Apple M2.

We do need to temper these results somewhat, as Geekbench 6 isn't always the best indicator of real-world performance. Still, these results suggest Qualcomm can certainly compete with the best x86 chip makers in the PC landscape, just like Apple. Apple showed that ARM CPUs can be fast and efficient, and Qualcomm is now providing those capabilities in a mainstream Windows 11 solution.

A leaked Geekbench 6 result for the Snapdragon X Elite shows the Qualcomm CPU performing better than the Ryzen 9 7940HS - AMD's flagship laptop APU. The Snapdragon X Elite was faster in both the single- and multi-core scores. If genuine, these scores indicate that the upcoming Arm-based CPU will be more than a match for x86 silicon in laptops and maybe even desktop PCs.

The leaked result apparently came from a test run on a "Qualcomm ZH-WXX" PC, which is presumably the name of a reference PC for the Snapdragon X Elite. It's hard to say if this reference PC shows performance consistent with what the final product is expected to have, however.

The Snapdragon X Elite is highly anticipated as it should be Qualcomm's fastest Arm chip ever, featuring 12 Oryon cores developed by Nuvia, which Qualcomm acquired in January 2021. The CPU is fabbed on TSMC's 4nm node and can boost to 4.3 GHz on two cores, with the other cores being limited to 3.8 GHz. On paper, the CPU doesn't seem like it would lead the competition, since both AMD and Intel's top mobile offerings can break 5 GHz and come with more threads, and more cores in Intel's case too.

Despite its on-paper specifications, the Snapdragon X Elite is able to exceed the performance of AMD's Ryzen 9 7940HS APU, which is AMD's top-end mobility-focused APU. Of course, AMD also has its 7045-series APUs, which are basically Ryzen 7000 desktop chips ported to a laptop form factor. The flagship Ryzen 9 7945HX beats the Snapdragon here, but not by a huge margin.

The Snapdragon could even be a potent desktop CPU if Qualcomm wishes, as it was able to match the Ryzen 5 7600 in multi-core performance and come within striking distance of its single-core performance. To be clear, the Snapdragon X Elite probably won't be competing with the likes of the Ryzen 9 7950X or Intel's Core i9-14900K, but it would nevertheless be one of two other CPUs in the world able to compete with midrange x86 chips (the other being Apple's M3).

Of course, there's one important question: is this benchmark legit? A couple of areas in the Geekbench 6 results are a little weird, with the reported base CPU clock speed being 4.01 GHz rather than 3.8 GHz and there being two CPU clusters of eight and four cores each, which is strange though it does at up to 12.

Despite these oddities, there's good reason to believe this benchmark is the real deal. Searching "Qualcomm Oryon" on Geekbench 6 will show you tons of other results for what is ostensibly the Snapdragon X Elite, including six Windows results tested on the Qualcomm ZH-WXX PC, such as the one we just looked at. The scores were all pretty consistent too, which is one point that indicates the result's authenticity. The earliest result for Windows was done on October 31, which incidentally was just after Qualcomm hosted a press event in late October to preview the performance of the Snapdragon X Elite.

Additionally, at that very same press event, Qualcomm was showcasing the Snapdragon X Elite in Geekbench 6 for Windows. That alone strongly indicates the leaks could be real, but Qualcomm also made hard performance claims for Geekbench 6's Windows version: a single-core score of roughly 2,900 and a multi-core score of about 15,000 per Anandtech. The publicly available scores fall short of those claims, but since this is all pre-launch silicon, there could be several reasons why this is the case, such as a lack of optimization or misconfigured settings. 

What should be one of the most powerful upcoming Intel Meteor Lake processors — an Intel Core Ultra 9 185H — was spotted (by Benchleaks) in Geekbench's online database. The CPU's single-thread and multi-thread Geekbench 5 scores of 1803 and 9806, respectively, aren’t much to get excited about. However, this is the first Core Ultra 9 we've seen in a benchmark leak, and it's particularly interesting to see the core config and clocks revealed.

We must add salt to these kinds of leaks, but let's take the results at face value for now. If the results are genuine/correct, the Intel Core Ultra 9 185H has/was tested with the following specs: 

• 16-core CPU with 22 threads
• Cores include six P-cores, eight E-cores, and two SoC-cores
• Reported CPU base / boost frequencies are 2.5 / 5.1 GHz
• The system used to test this chip was of an unknown brand
• 64 GB of RAM installed
• Windows 11 installed

We've only seen lower-tier Meteor Lake leaks, including the Intel Core Ultra 7 165H and the Intel Core Ultra 7 1002H, until now. Both the Core Ultra 7 165H and the Core Ultra 7 1002H were reported by Geekbench as having 16 cores and 22 threads. However, the Core Ultra 9 185H that appeared today has the highest reported boost clock seen so far.

Moving onto the benchmark results, we looked at other Geekbench 5 results for a close 1T performance comparison. It turns out the Intel Core i5-13400 and AMD Ryzen 7 7840U have very similar 1T performances compared to the leaked chip. As for nT performance, the Core Ultra 9 185H performs a lot like the Intel Core i7-1360P and — again — the AMD Ryzen 7 7840U. These aren't stellar results for Intel's next great mobile CPU, but remember these are early results and we may be looking at an ES chip in a test system, with tuning yet to be completed. It's also a sample of one (and this is assuming it's all real, of course). 

For more information on Meteor Lake, check out our deep dive into the architecture. While we know a lot of official details about the technologies that Meteor Lake will deliver, no price or SKU details have been confirmed yet. Intel is preparing to launch its Core Ultra processors on Dec. 14, which seems just right for a flood of new Intel laptops at CES 2024.

What appears to be a test run of an “Intel Corporation Meteor Lake Client Platform” has appeared in the Geekbench online results database. Spotted by Twitter/X user Benchleaks, said system was tested today. The purported new Intel Core Ultra 7 1002H powered computer, perhaps a whitebook test platform, scored 2439 and 12668 in Geekbench 6 single- and multi-threaded tests, respectively.

According to the Geekbench database, the Intel Core Ultra 7 1002H features 16 cores and 22 threads. Two core clusters were recorded by the benchmarking software: one with six cores, the other with 10. If the first set of cores are performance cores with Hyperthreading, that would tally with the 16C / 22T configuration observed. Other specs revealed during the software’s probing of the system were that the Ultra 7 1002H ran at a base / boost of 3.4 / 5.0 GHz.

Using the Geekbench online UI, we made a comparison between the Core Ultra 7 1002H and what we think may be its direct predecessor, the Core i7-13700H (in a Gigabyte Aero laptop). Both are 7-series and H-series laptop chips, so this should be a fair comparison generation over generation. 

The system RAM reading of the Meteor Lake test laptop was unusual. Geekbench 6 said the system was packing 11.69 GB of RAM, but had no information about the memory type, transfer rates, or channels. Meanwhile, the rival Gigabyte Aero we selected for comparison had 16GB of DDR5-5200 in multi-channel configuration. We wonder if the Meteor Lake platform RAM detection issue was because of an esoteric new feature, like on-package LPDDR5X memory.

The single thread (1T) score of the purported new Intel Core Ultra 7 1002H is better than the Core i7-13700H laptop we put up for comparison. However, there are lots of other Core i7-13700H with better (and worse) Geekbench 1T scores. Similarly, there are Core i7-13700H laptops in the Geekbench database with better and worse multithreaded (nT) scores than the new MTL-H challenger. It is probably most disappointing that the Meteor Lake chip performed worse in the nT tests, as the design appears to have more cores at its disposal.

Overall, with just the single Core Ultra 7 1002H sample, there aren’t many conclusions that can be drawn. This is even more clearly the case as the results are so similar to a multitude of RPL-H predecessors in the Geekbench database.

Remember, Meteor Lake is Intel’s first foray into a mass produced consumer CPU which is chiplet-based. The strategy of splitting the 14th Gen Core processors between Raptor Lake Refresh for desktops and Meteor Lake for laptops hints that Intel had some difficulties affecting its original plans. However the iconic chip maker is making very confident noises about Meteor Lake.

Intel and partners are widely expected to be preparing to launch Meteor Lake chips and laptops at the impending Innovation event. Intel’s event runs Sept. 19 - 20, 2023, at the San Jose Convention Center in California.

The frequency of Intel 14th Gen Core ‘Raptor Lake Refresh’ processor leaks is increasing as more insiders play with test samples before launch. Today, Benchleaks on Twitter/X shared three results from a purported Core i9-14900K with scores from single- and multi-threaded testing in Geekbench v6. Sadly for enthusiasts, the test results are not inspiring.

The results unearthed by Benchleaks appear to be from a user repeatedly running the same system through the suite of performance tests. This is typical behavior as someone makes system tweaks and then wishes to see if performance has changed. However, the spread of results is relatively narrow, which indicates any adjustments had little or no effect. In other words, the three sets of results are within a range that could be expected from repeatedly testing an unchanged system. For example, there is less than a 1% difference in the multi-threaded (nT) test scores.

Another comment we have on the supposed 14900K chip scores is that they don’t look impressive compared to the current-gen Core i9-13900K. The Geekbench result browser says that, in Geekbench 6, the 13900k scores 2953 and 19911 in 1T and nT tests, respectively. That 13th Gen multi-threaded score is better than all three Core i9-14900K runs on this system, built upon a Biostar Z790A-Silver motherboard with 16 GB of DDR5-4800 RAM. The current-gen Raptor Lake chip achieves about a 4% faster nT score than the best leaked Raptor Lake Refresh score today.

News for those with an eye on a Raptor Lake Refresh upgrade is more encouraging if we look at the single-threaded test results. Add more salt here, but the Core i9-14900K seems consistently better in the 1T tests. The best 14900K 1T result is over 6% faster than the Geekbench normalized 13900K score. It also easily beats the 13900KS.

Wider Perspective

The above leaks are interesting and might warn potential new PC builders and upgraders against waiting for Raptor Lake Refresh for substantially better performance. Put bluntly, the uplift in performance in the leaked Geekbench 6 tests is poor to non-existent.

A recent leak of CPU-Z benchmark scores painted a nicer picture for Intel regarding multi-threaded gains. Also, about a month ago, we saw some interesting Crossmark tests, which also poured cold water on hopes of a big performance lift coming via the 14th Gen family.

We hope the rumored Raptor Lake Refresh price hike of about 15% vs. the current gen doesn't materialize, as it looks increasingly difficult to justify.

For the ground truth about Intel’s 14th Gen Core ‘Raptor Lake Refresh’ processors, you are advised to wait until the hotly tipped launch at Intel Innovation from September 19 to 20.

A fresh benchmark run featuring a purported Intel Arc A580 graphics card has been spotted. According to a screenshot shared by ITHome (Chinese language) the A580 scores nearly 83K in Geekbench 5 OpenCL tests which is over 10K clear of scores achieved by systems relying on Nvidia's GeForce RTX 3050 graphics card.

The Geekbench database sighting of this 'lost' Intel Arc graphics card appears to be from a benchmark run earlier today. This particular Arc A580 is part of a well-specified Windows 11 desktop PC system which also features an Intel Core i7-12700 CPU, an Asus ROG Strix Z690-E motherboard, and 32 GB of DDR5 RAM. Take the Geekbench results with a pinch of salt for now.

We have already mentioned a Geekbench 5 Open CL score comparison in the intro (vs RTX 3050), but let's build a fuller table, including the established Intel Arc cards that we are now quite familiar with.

*= leaked single result

In September 2022 we covered, in depth, the technical details of the Intel Arc A580 alongside stable mates and now well known desktop GPUs like the Arc A770, Arc A750, and Arc A380. In summary, the GPU configuration of the A580 is derived from a cut-down ACM-G10 GPU, and is much closer in spec / performance to the A750 than the little A380. This fact is quite immediately evident not just from benchmarks, but from the Arc A-Series Desktop GPUs infographic published by Intel, reproduced below.

It is curious that Intel has not yet launched its Arc A580 graphics card, as it was announced alongside the above products nearly a year ago. 

After a long gap in any leaks or spills regarding the Arc A580 we saw it pop up in GFXBench 5.0 results in mid-July, so perhaps Intel and partners are at last readying this model for release. In our recently updated GPU pricing Index feature, we noted that the best price of an Arc A750 is around $200, and the A380 around $100. If we had to guess the Arc A580's price it would probably not be right down the middle, but nearer to the A750 price level.

Intel's Alder Lake-N family of processors aimed at entry-level PCs has its own hierarchy and includes multiple options, including the flagship eight-core Core i3-N305, mid-range quad-core N97, and low-end dual-core N50. The latter belongs to a segment that represents little interest to performance-minded users, which is why such chips never get reviewed. But someone shared test results from the Intel Processor N50 in Geekbench (via @BenchLeaks).

Intel's Processor N50 features two general-purpose cores based on the Gracemont microarchitecture. Operating at 3.40 GHz, the chip comes with UHD-badged Xe-based graphics core with 16 EUs at 750 MHz and supports up to 16GB DDR4, DDR5, and LPDDR5 memory. With a thermal design power of 6W, the CPU can serve various applications, including laptops, thin clients, compact desktops, and many other low-power things. What the SoC is not meant to offer is, of course, high performance.

Indeed, even single-core results of the Intel Processor N50 in Geekbench 6 CPU tests are significantly lower than those of the Processor N97 and Core i3-N305, which is unsurprising given its lower frequency. The multi-core score is predictably over two times lower than the N97 since this is a dual-core CPU. 

As far as OpenCL compute performance is concerned, the Processor N50 comes with a severely cut-down integrated GPU that only has 16 execution units and runs at only 750 MHz. It cannot even compare to iGPUs of the N97 and N305 since they have more EUs that run considerably faster.

Of course, Intel's N50 can run productivity applications just fine, especially if it does not need to compute large XLS files or perform other resource-demanding tasks. Still, this CPU is generally meant to sit in a very simplistic machine that uses cloud resources or processes light workloads.

Geekbench 6 is a synthetic benchmark that does not necessarily represent performance in real-world applications, but when a CPU is times behind even entry-level offerings, it is evident that it is slow.