In 2025, we introduced Tom's Hardware Premium, an even deeper dive into the world of tech and DIY computing, which you can access through our website for a small fee. While that includes exclusive news analysis, features and hardware roadmaps, the Premium offering also sports a large database of benchmark test results, named Bench.

Bench is a performance analysis tool that gives you deeper access into how a given product performs in real-world and synthetic benchmarks. Using it, you can compare two different products in a given category, offering you the ability to view benchmark data between products with a simple click.

Bench is a Tom's Hardware Premium exclusive feature, and you can get access to it — as well as everything THP has to offer — by subscribing now for as little as $3 a month.

With that out of the way, here's a sneak preview at what you can expect when subscribed. Let's take a look at a CPU comparison table to kick things off. Let's pit the AMD Ryzen 9 9950X3D against Intel's Core Ultra 9 285K.

First, all I have to do is hit the compare tab and select the category, which spans CPUs, GPUs, Laptops, Monitors, and SSDs. Once we're in the CPU category, we can select the two products. Then you'll see a comparison between all of the data points we recorded for those two CPUs in alphabetical order.

As well as the comparison tool, Bench offers tables of data per category. So, if you're looking to see which part performs the best in a certain workload, you can do that too.

For the next example, we're going to pick out a specific GPU workload and see which consumer graphics card performs best in that scenario. Let's take the Sony hit God of War: Ragnarok. If you are building a mid-range system, then 1440p or QHD is a good spot to select for this benchmark. Since we're focused on the mid-range, let's take a look at the middle of the table. Here, we can see that the Radeon 9060 XT is faster than Nvidia's RTX 5060. These two GPUs hit around the same price point, so if I were a God of War fanatic looking to save, that would be my pick.

How we run Bench

Bench offers tons of data to browse through; these are just two examples of how you can use the tool to browse through hardware benchmarks. While the database is primarily built from the work completed while reviewing a given product, the data that you see in reviews is often truncated into geomean values, or put into a certain workload "bucket."

Of course, while Bench has been live since last year, benchmarks don't stay up to date: Drivers change. New standards emerge. Therefore, we're undergoing a large-scale overhaul of our GPU and CPU tests, documenting exactly how we are benchmarking and what methodologies are being used.

Recording benchmark values and taking the time to run the tests in the first place is inherently time-consuming, and Bench is the perfect reason to maintain a source of truth for all enthusiasts.

As for our testing methodologies, here's a glimpse of what's next for our massive database.

Revamped data for GPU testing

On the GPU side of things, Senior GPU Analyst Jeffrey Kampman has the reins. He's worked out exacting testing methodologies and strategies and outlined them in a separate Tom's Hardware Premium article. For those who are not yet subscribed, we've explained why we're not using frame generation or upscaling technologies like DLSS and FSR, and how we conduct real-world benchmarks. A canned benchmark is handy, but it doesn't represent how it really feels to play a game on the hardware. And more important, it's a more accurate measure of what to expect when playing those titles for yourself.

As Jeff explains, he's spent years testing GPUs in both media and dedicated game testing labs, ensuring the integrity of the results on offer. The spread of titles being tested for 2026 is also representative of what players are actually playing, not merely popular benchmarking titles. For example, we've added the single-player epic Expedition 33, in addition to the multiplayer hit Arc Raiders. Adding newer titles to the list, across a range of engines and with Ray Tracing, allows us to build a cohesive outlook on how a given GPU performs in a wide range of scenarios.

Updates will be going in phases, first from current-generation hardware like Nvidia's RTX 5000 series and AMD's 9000 series, all the way back to RTX 3000 and the Radeon 6000 lineup. Of course, Intel's offerings are in the mix there too — no matter how small the market share of the dedicated GPU, it'll be being tested for 2026.

The three waves of updated tests will take around a month to fully complete, but will offer you a holistic look at the GPU market, and more importantly, offer you granular testing information that you simply can't get anywhere else.

A herculean CPU retesting effort

CPU Analyst Jake Roach is working alongside Editor-in-Chief and former CPU reviewer Paul Alcorn to deliver a gargantuan update to our CPU tests. Their goal is building a complete dataset of CPU releases that spans over a decade of hardware. For Premium subscribers, you can read Jake's entire methodology behind CPU testing here.

Between two and six machines are running concurrently, with two Golden OS images, frozen and duplicated across setups, using the same testing hardware, down to the TIM, to ensure consistent results. Of course, as we are currently testing different platforms, the motherboards used between CPUs differ, but are consistent between rigs.

While we are testing hundreds of CPUs, everything from the version of the software to the processes running in the background are identical. For the CPU effort, we'll be going from AMD Excavator and 7th-gen Intel to Zen 5 and Intel Arrow Lake. That's a lot of CPUs, and even more benchmarks.

Each modern CPU takes between 10 and 14 hours to test, with some older and slower models taking nearly an entire day, and with the number of chips on offer, it's going to take some time to get things up-to-date. As for the tests we run, things are split into several categories, such as applications, power/efficiency, and gaming workloads. While some tests are automated, each result is manually verified before being uploaded into our database and tools.

Application testing goes across a wide range of workloads, from rendering and encoding to code compilation and data science. We're testing power efficiency across benchmarks like Y-Cruncher, Handbrake, and Prime95, just to name a few. Gaming testing is hands-on where possible. The selection of titles is broad and represents real-world players and what people are playing.

As a result, no, you're not going to see something like Ashes of the Singularity, which no one plays, but is merely used as a benchmark. Instead, gaming benchmarks are selected across a range of popular titles like Monster Hunter: Wilds, DOOM: The Dark Ages, and Oblivion: Remastered, all of which use wholly different engines, and are being played by many today.

Results will begin to trickle out across the database over time, and given the scale of the tests on offer, we will continue to evolve the benchmarking suite, while acknowledging that we might not revisit some of the older architectures on this list again.

The benchmarks never stop

On the backdrop of a challenging year for consumer hardware, no thanks to the ongoing AI buildout and supply challenges. Getting the most out of what's available is imperative, and Bench gives you access to all of the data you need for whatever workload or game you're looking at within a handful of clicks.

We've put a lot of effort into building and maintaining this feature, and we will continue to do so, documenting the process and keeping things nice and fresh for Tom's Hardware Premium subscribers.

Maintaining Bench without Tom's Hardware Premium simply isn't possible. It offers us the bandwidth to test, retest, and refine our methodologies outside of a standard CPU cycle or hierarchy update. It also becomes an invaluable reference point for those of you who wish to seek out specific hardware workloads, going beyond our standard reviews and wading through the data yourself.

We're never going to stop testing, and Bench is the perfect place to compile our data. As we look forward to the next generation of PC hardware, we'll be sure to keep things up-to-date, giving you access to the cold, hard data behind our testing. In addition to our testing updates, we will also keep working behind the scenes to add new functionality to Bench in the coming months.

Tom's Hardware Premium is priced at $3 a month when taking out an annual plan, or $7 for a rolling monthly subscription. Bench is just a single piece of what's on offer. To take advantage of the ultimate resource for enthusiasts, be sure to check it out today.

At Future Publishing we rely on advertising to keep bringing you the content you love to read. The vast majority of the content on Tom's Hardware is created solely by our editorial team, but on occasion we also work with external partners to create content we hope our readers will find interesting and useful.

In some cases, advertisers support us in producing content. This content is labelled so you can see who has funded it and how it was created. We use the label on the page to clarify the advertiser’s involvement in the content.

“Sponsor Content Created With…”

Articles labelled ‘“Sponsor Content Created With…” are paid for and reviewed by a commercial partner. They may be produced by the client or by staff employed by Tom's Hardware. This is commercial content and so is subject to the Advertising Standards Authority regulations in the UK and Federal Trade Commission regulations in the US.

"Sponsored By…" and similar labels

Articles that are labelled "Sponsored By…" or similar labels are independent editorial articles, created by writers employed by Tom's Hardware that have been funded through the support of a commercial partner. When planning this content, the editorial team may find alignment with a funding partner on the topic and the headline of the article but the article is not subject to any client review in advance of its publish date. This content abides by the Editors’ Code of Practice from the Independent Press Standards Organisation in the UK and Federal Trade Commission regulations in the US.

If the commercial partner receives a sponsored section within a larger editorial article editorial article, that section will have a clear “Sponsored” label.

“Preferred Partner”

Articles that are labelled as "Preferred Partner" mean a commercial partner is offering a preferential affiliate rate to Future in exchange for greater prominence on the page, such as by highlighting a particular deal for a product that our journalists recommend.

Tom's Hardware will only write content featuring “Preferred Partners” when we feel the content or product is aligned with what our audience wants. It is not sent to the funding partner for approval. This content abides by the Editors’ Code of Practice from the Independent Press Standards Organisation in the UK and Federal Trade Commission regulations in the US.