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.

The Linux team has published an intermediate Linux 6.13-rc3 to fix a "hilarious" bug in the kernel-based Virtual Machine code — Phoronix reports, where CPUID updates are taking significantly longer time to be processed on Sapphire Rapids CPUs than they should be, resulting in much older Skylake CPUs being up to 4X faster in regards to CPUID cost.

Phoronix reveals that Google engineers originally discovered this technical issue in how Sapphire Rapids processes CPUID information in nested VMs. Specifically, Google's Sean Christopherson states, "On Intel's Emerald Rapids, CPUID is *wildly* expensive, to the point where recomputing XSAVE offsets and sizes results in a 4x increase in latency of nested VM-Enter and VM-Exit (nested transitions can trigger xstate_required_size() multiple times per transition), relative to using cached values. The issue is easily visible by running `perf top` while triggering nested transitions: kvm_update_cpuid_runtime() shows up at a whopping 50%..."

This performance bug/regression won't be purportedly fixed until Linux 6.14, but the issue is severe enough for the Linux devs to fix the outgoing Linux 6.13 kernel temporarily. Update 6.13-rc3 caches the CPUID data on Sapphire Rapid CPUs, speeding up this process. The Linux 6.14 implementation of the complete fix will result in all the CPUIDs nested in VM-Enter and VM-Exit being completely deleted.

For those unaware, CPUID is an instruction that lets software discover details of the processor it is running on. On Sapphire Rapids, the software reading CPU information through the CPUID instruction takes up to 4X longer than on much older Skylake CPUs. This only occurs inside nested virtualization transitions, which only happen when a VM runs inside a VM.

The Linux 6.13-rc3 comes with various other bug/regression fixes. However, the caching workaround for expensive CPUID handling appears to be one of the more critical updates to the kernel. Again, this caching technique is just a workaround, with Linux 6.14's complete fix for the issue.

Using its automated binary analysis system Eclypsium Automata, Eclypsium has uncovered the existence of high-impact security vulnerabilities in Phoenix SecureCore UEFI firmware used by a wide variety of motherboard providers and Intel CPUs spanning from 14th Gen to 6th Gen—all the "Lakes" in other words. This vulnerability also extends to several other UEFI BIOS vendors, including Lenovo, Intel, Insyde, and AMI. Phoenix is the latest to join the list.

The specific Phoenix SecureCore UEFI firmware vulnerability that prompted this posting is referred to as "UEFIcanhazbufferoverflow" by Eclypsium, which is just a funny way of pointing out that this is a buffer overflow exploit. The specific method in which the "UEFIcanhazbufferoverflow" exploit works is by using an unsafe call to the "GetVariable" UEFI service.

By making unsafe calls, a stack buffer overflow can be created, allowing for arbitrary code to be executed. In the BIOS or its modern counterpart, the UEFI, even a buffer overflow allows for full-system access and control to be gained very quickly, and the consequences of that happening can be challenging to remove from a PC permanently. Sometimes, it may even be impossible without replacing the machine entirely— and that's not counting passwords and such that may become compromised and still need changing between machines.

Any potentially impacted Intel user should update their BIOS to protect from this issue as soon as possible, though not before creating backups of important files and the original BIOS just in case something goes wrong. Since exploits impacting UEFI are as close to Layer 0 as they get with PC hardware, it's essential for all parties involved to act as quickly and safely as possible.

As noted by Eclypsium, this Phoenix vulnerability was discovered in a hands-off manner by its Automata security system, which is an automated binary analysis system using the research data of Eclypsium's own researchers. While there are certainly issues with things like AI-written code and AI "generated" art, it's always nice to see cutting-edge AI and machine learning tech put toward something useful for humanity.

Our CPU benchmark hierarchy provides a broad view of relative performance for the latest Intel and AMD processors. Over the last 30 years, Tom’s Hardware has been benchmarking CPUs, and we use the rankings here as the basis of our best CPUs for gaming and best budget CPU rankings. We run over 200 individual tests for each CPU we look at, and that comprehensive performance is condensed here for a high-level view of how CPUs compare across gaming, single-threaded, and multithreaded performance.

Each of our CPU benchmarks helps expose different aspects of performance, from heavily-threaded code compilation and data science workloads to lightly-threaded web apps and audio encoding. We’re currently in the process of the biggest refresh to our CPU benchmarks hierarchy ever, spanning over a decade of processor releases. The results here provide the first half of that testing, focusing on DDR5 platforms that span the AMD vs Intel product lineups. As we fill out our legacy benchmarks, you’ll see more CPUs added to our rankings. If you want to check the performance of older CPUs now, you can use the second page of this article to see our legacy benchmarks.

In games, AMD’s Ryzen 7 9850X3D is the fastest CPU on the market, though other Zen 5 X3D offerings like the Ryzen 9 9950X3D and Ryzen 7 9800X3D aren’t far behind. X3D chips dominate the charts for gaming at 1080p, with the other exception being the relatively unpopular (and expensive) Ryzen 9 7900X3D. Otherwise, Intel’s last-gen Core i9-14900K is the fastest offering from Team Blue, with the new Core Ultra 7 270K Plus coming in slightly behind with Intel’s new iBOT feature.

Intel pulls out strong positions in applications; however, with the Core Ultra 7 270K Plus topping the charts in single-threaded performance and coming in third in multi-threaded rankings. It’s only beaten by the Ryzen 9 9950X and its X3D variant, and only by a hair. Further, both of those CPUs cost about twice as much. AMD's recent Ryzen 9 9950X3D2 claims the top slot in overall performance, but at $900, it's too expensive for most buyers.

In each section below, we’ll show you the rankings for each CPU, as well as reveal what tests went into creating the rankings. We’ll also give you some pointers for benchmarking your own CPU to see how much performance an upgrade or overclock netted you, along with some common, easy-to-run benchmarks you can perform yourself.

CPU Benchmarks Rankings 2026

In the album above, you can see our master charts for gaming, single-threaded, and multi-threaded performance for CPUs. For games, all of our testing was done with an Nvidia RTX 5090 FE, and for applications, our testing was done with an Nvidia RTX 2080 Ti FE. For applications, no compute is actively running on the GPU; it’s a glorified display output that shares a driver with our gaming GPU. You can find a full breakdown of the test benches we used at the end of this article.

Gaming CPU Benchmarks Rankings 2026

You can see the relative score for AMD and Intel CPUs above, measured against the Ryzen 7 9850X3D, which is the fastest gaming CPU on the market, per our testing. So, the Ryzen 7 9800X3D offers 97.04% of the performance of the Ryzen 7 9850X3D, while the Ryzen 9 7900X offers 69.28% of the performance. You can set any CPU as a baseline for comparison with Bench, which is available in Tom’s Hardware Premium.

All of our gaming tests were run with the RTX 5090 FE at 1080p with a mixture of High and Ultra settings. We run each test multiple times — usually between three and five — and pick the median result. In other words, the results we use are real, recorded runs, not an average of several different runs. This is important as some games, such as Far Cry 6, show great CPU scaling but are otherwise inconsistent run-to-run.

In addition to consistent hardware (test benches at the end of this article), we use a consistent test image between platforms. That means the same GPU driver, the same Windows install, the game version, etc. We also tested with Virtualization-Based Security (VBS) turned off, Resizable BAR turned on, and automatic overclocking features disabled. That includes the Intel Extreme power profile and AMD’s PBO, both of which aren’t covered under standard warranty.

For this refresh, we tested 17 games and then calculated a geometric mean of the results. A simple average would provide skewed results with such a large test pool. A geomean provides a more realistic view of how each CPU compares to the others.

Here are the games that we used for testing:

• Counter-Strike 2
• The Last of Us Part One
• Cyberpunk 2077
• Starfield
• A Plague Tale: Requiem
• Hogwarts Legacy
• F1 24
• Marvel’s Spider-Man 2
• Baldur’s Gate 3
• Monster Hunter: Wilds
• Final Fantasy XIV
• Microsoft Flight Simulator 2024
• Doom: The Dark Ages
• Oblivion Remastered
• Far Cry 6
• Hitman 3
• Minecraft RTX

We’re constantly evaluating new games to include in our test suite — see our recent stories on Resident Evil Requiem CPU scaling and Crimson Desert CPU scaling — but we maintain a list of tried-and-true benchmarks for our hierarchy rankings. We want to avoid including brand-new titles, which may see many updates, to keep our rankings as true to reality as possible. If you want more about the rationale behind our game choices, see our behind the scenes look at our CPU hierarchy testing.

Single-Threaded CPU Benchmarks Rankings 2026

We run hundreds of tests for each CPU, but only a small subset of those tests factor into our single-threaded rankings. We use the mp3 encoder LAME with a single thread (both standard and extended), Cinebench 2026 and 2024’s single-threaded test, the ray-traced renderer POV-ray, and WebXRT4, which runs a series of browser-based applications written in various languages.

The fastest chip in the pool here is the Core Ultra 7 270K Plus, which scores 100%, with every other chip scored relative to it. The Core i9-14900K offers 95.4% of the single-threaded performance of the Core Ultra 7 270K Plus, the Ryzen 5 9600X offers 89% of the performance, and so on.

Most real-world workloads aren’t strictly single-threaded, which is why we include it on a subset of the total tests we run. The goal is to see what relative performance looks like in lightly-threaded applications, as well as look into the overall architecture of different CPUs. Single-threaded performance exposes a lot about the architecture in a way that heavily-threaded applications tend to mask.

We’re, of course, looking at performance on a single core, favoring high clock speeds and IPC (instructions per cycle). However, single-threaded performance also says a lot about what’s going on elsewhere inside the CPU, from the speed of the IMC (integrated memory controller) to the fabric/ring speed. That’s why we see things like the Core Ultra 7 270K Plus outperforming the Core Ultra 9 285K, despite the latter sporting higher clock speeds.

Multi-Threaded CPU Benchmarks Rankings 2026

Similar to single-threaded rankings, we use a subset of the total tests we run for CPU reviews in ranking multithreaded performance. Cinebench and POV-ray show up here again, this time using as many threads as possible, alongside VRay, four Blender tests, and Handbrake using various codecs. Although most applications will leverage multiple threads these days, we’re specifically looking at applications that will take as many threads as possible to maximize compute.

Compared to single-threaded workloads, heavily-threaded tasks are less concerned with clock speed and put a greater emphasis on interconnects and core-to-core latency. Core count is obviously important, as well, though it’s been somewhat undermined by Intel’s hybrid architectures over the last several generations.

Given that we’re spanning multiple nodes, core count alone isn’t indicative of higher multithreaded performance. Yes, higher core counts within the same generation will usually provide higher multithreaded performance, but a slew of other factors can increase performance, as well, from all-core and uncore frequencies to higher transistor density. Because of the wide swath of factors, you can see much more aggressive scaling with our multithreaded rankings compared to single-threaded rankings.

Integrated GPU Gaming CPU Benchmarks Rankings 2026

Here's our list of gaming performance with integrated graphics on several of the leading APUs available. We've split this into two different price ranges, so be sure to flip through all of the performance charts. For a bit of commentary and analysis of these results, head to our Ryzen 7 5700G, Ryzen 5 5600G, and Ryzen 3 5300G reviews. The most powerful chip gets a 100, and all others are scored relative to it.

How to Benchmark your CPU

It’s important to know how to benchmark your CPU. It gives you a way to compare performance after an overclock or a CPU upgrade, and it allows you to check if you’re getting the full performance out of your system. Maybe a poor CPU cooler mount is limiting your performance, or maybe your BIOS settings aren’t optimal. Using benchmarks to compare your results lets you see where your rig stacks up, not only for leaderboard purposes, but also basic troubleshooting.

The key to benchmarking your CPU is consistency. The only variable that should change is your CPU, be it a new CPU or an overclock/undervolt. Before starting, make sure to close any applications running in the background. That’s not only to net peak performance, but also to avoid any inconsistencies between runs. Background apps can gobble up threads inconsistently, making it difficult to compare your results from run to run.

If you want more consistency, you can optionally run the following command before benchmarking in an elevated command prompt:

Rundll32.exe advapi32.dll,ProcessIdleTasks

This will force Windows to perform the background tasks it normally does when your PC is idle. It’s not essential, but it’s a good sanity check to make sure there’s nothing interfering with your results.

For applications, you want to test the apps you actually use. If you use the Adobe suite, for example, you can download and use PugetBench for free and compare your results with Puget’s database. A lot of apps don’t have these easy-to-use benchmarking tools and databases, so you need to find a proxy. For instance, Procyon Office measures Microsoft Office performance, but a license costs nearly $1,600 per year. PCMark 10 Basic, which is free, measures open-source office applications. Below, we have some of our favorite free benchmarks for comparing CPU performance.

In games, you can take two approaches: manual or automated. Some modern games include built-in benchmarking tools, such as Cyberpunk 2077 and Doom: The Dark Ages, and although they aren’t perfect, they’re easy to run and highly repeatable. The best way to measure CPU gaming performance, however, is manual benchmarking.

That involves finding a scene where you can go over a specific path repeatedly. That could be starting from a specific checkpoint that you can reload or resorting to a manual save where you start from the exact same position. Regardless, it’s important to avoid randomness in your testing. Keep the path consistent — for example, a walking path through a town — and try not to swing the camera around.

For measuring performance in games, you’ll need a performance monitoring tool. There are simple apps like Nvidia’s FrameView, which logs a ton of information but is a little cumbersome to deal with; it exports data to spreadsheets. CapFrameX is a good alternative, which uses the same backend as FrameView (Intel’s PresentMon), but comes with a user-friendly GUI and extra features like the ability to generate charts right in the app.

After you run your benchmarks, you need a comparison point. Databases like Puget are your best resources on that front. If you’re comparing results to reviews, forum threads, or other systems, keep in mind the variables that can influence performance. It’s not a good idea to compare performance with uncontrolled variables unless you have a wide swath of comparison points.

Best CPU Benchmarks You Can Run

• Cinebench 2026 – Cinebench is the quintessential CPU benchmark, used almost universally in reviews, and it’s completely free to download and use.
• Geekbench 6 – Geekbench has a number of issues, but it offers a massive database for comparing your system against other similar systems. And it’s free to download and run.
• Blender – Blender has a benchmarking utility with a GUI that’s free to download, as well as a large database of results.
• PCMark 10 Basic – The main PCMark 10 benchmark is free to use with the Basic edition, allowing you to test productivity performance with open-source office apps, as well as compare your scores with UL’s database.
• Handbrake – Handbrake is a powerful, free, and open-source video transcoding tool, and it’s easy to run benchmarks with. Use any video file, make sure your settings are the same, and start a stopwatch to measure the time encoding takes. Lower is better.
• WebXPRT 5 – WebXPRT runs a variety of web applications directly in your browser, for free, and with a database to compare results to. It takes a while to run, however.
• JetStream 2 – JetStream is a faster browser-based benchmark, though it doesn’t have a database of results.
• CPU-Z – CPU-Z isn’t a reliable benchmark for real-world performance, but it includes single- and multithreaded tests, it’s easy to run, and you’ll find results online almost as commonly as Cinebench results.
• Y-cruncher – This test calculates Pi with digit extraction, and it’s accelerated with SIMD instructions like AVX. You can only run it from a command line, but it’s relatively straightforward.
• PugetBench – Puget includes benchmarks for the biggest apps in the Adobe suite, as well as DaVinci Resolve. The benchmark itself is free, and Puget maintains a large database. You’ll need a license for the applications it tests, however.

2026 CPU Benchmarks Test System and Configuration

• AMD Ryzen 7 7800X3D vs Intel Core i9-13900K vs Core i7-13700K: Big Gaming Punch, Smaller Price Tag
• AMD Ryzen 9 7950X3D vs Intel Core i9-13900K Faceoff: Battle of the Gaming Flagships
• Intel Core i5-13600K vs AMD Ryzen 7 7700X and Ryzen 5 7600X Face Off
• AMD Ryzen 9 7950X vs Intel Core i9-13900K Face Off
• Ryzen 7 5800X3D vs Core i7-12700K and Core i9-12900K Face-Off: The Rise of 3D V-Cache
• Intel Core i9-12900K vs Ryzen 9 5900X and 5950X: Alder Lake and Ryzen 5000 Face Off
• Intel Core i5-12600K vs AMD Ryzen 5 5600X and 5800X Face Off: Ryzen Has Fallen
• Intel Core i7-12700K vs AMD Ryzen 9 5900X and 5800X Face Off: Intel Rising
• Intel Core i5-12400 vs AMD Ryzen 5 5600X Face-Off: The Gaming Value Showdown

• MORE: Best CPUs for Gaming
• MORE: AMD vs Intel
• MORE: 13th-Gen Raptor Lake All We Know
• MORE: Zen 4 Ryzen 7000 All We Know
• MORE: How to Overclock a CPU
• MORE: How to check CPU Temperature
• MORE: All CPUs Content

2020 - 2022 CPU Benchmarks Hierarchy

You can find our rankings of the most current-gen systems on the previous page. The results below are from our legacy benchmarks, using a different GPU and test systems than our current CPU benchmark rankings. However, this provides great historical context and also includes other previous-gen CPUs not included in our new test suite. You'll also find our even older legacy rankings further below. These date back over the last decade.

Legacy: 2020 - 2022 CPU Benchmarks Rankings - Windows 10 and 11

Legacy: 2020 - 2022 Gaming CPU Benchmarks Ranking

These tests are from our 2022 test bench. We measured performance for the 1080p CPU gaming benchmarks with a geometric mean of Borderlands 3, Hitman 2, Far Cry 5, Project CARS 3, Red Dead Redemption 2, and Shadow of the Tomb Raider. We measured performance for the 1440p CPU gaming benchmarks with a geometric mean of Borderlands 3, Project CARS 3, Far Cry 5, Red Dead Redemption 2, and Shadow of the Tomb Raider. We conducted these tests in Windows 10.

Legacy: 2020 - 2022 Single-Threaded CPU Benchmarks Rankings

These results are from our 2022 test bench. We calculate the above single-threaded CPU benchmark rankings based on a geometric mean of the Cinebench, POV-Ray, and LAME CPU benchmarks. The most powerful chip gets a 100, and all others are scored relative to it. We conducted these tests in Windows 10.

Legacy: 2020 - 2022 Multi-Threaded CPU Benchmarks Rankings

These tests are from our 2022 test bench. The multi-threaded workload column is based on CPU benchmarks performance in Cinebench, POV-ray, vray, Blender (four tests - Koro, Barcellona, Classroom, bmw27), y-cruncher, and Handbrake x264 and x265 workloads. These CPU benchmarks represent performance in productivity-focused applications that tend to require more compute horsepower. The most powerful chip gets a 100, and all others are scored relative to it. We conducted these benchmarks in Windows 10.

Legacy: Pre-2018 Desktop CPU Benchmarks

Recognizing that a lot of older platforms are going to be paired with graphics subsystems multiple generations old, we wanted to define the top of our range to encourage balance between host processing and complementary GPUs. At this point, anyone with a Sandy Bridge-based Core i7 would realize a gain from stepping up to Coffee Lake or Kaby Lake, for example. And putting AMD's top FX CPUs next to a handful of Core i7s and those older Core i5s represents an upgrade to their status.

Currently, our hierarchy consists of 13 total tiers. The bottom half of the chart is largely outdated; you'll notice those CPUs dragging down performance in the latest games, whether you have one of the best graphics cards or not. If you own a CPU in that range, an upgrade could really take your experience to another level.

Really, it's the top five tiers or so that remain viable. And in that top half of the chart, an upgrade is typically worthwhile if it's a least a couple of tiers higher. Otherwise, there's just not enough improvement to warrant the expense of a fresh CPU, motherboard and RAM (not to mention the graphics card and storage solution you'd be considering as well).