Why you can trust Tom's Hardware
Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test.
(Image credit: Tom's Hardware)
I've been using a Core i9-13900K testbed since 2023, and that means it was time for an upgrade. But if you're planning on running an Arc B580 or similar level of GPU, your CPU likely won't present too much of a bottleneck. I've tested the B580 on both my old 13900K test PC as well as the new test system, which will be used for the next year or more.
Along with new hardware, I've completely overhauled my test suite. There are a few games that are sticking around, but most of the benchmarks have been updated, and of course newer drivers and software updates mean I had to retest everything regardless. So I updated the OS to 24H2, installed the latest motherboard firmware, grabbed AMD's 24.12.1 drivers, Nvidia's 566.36 drivers, and Intel's preview 6251 drivers (which appear to be a mix of old and new fixes relative to the public 6319 drivers), and set about testing.
Note that this means all the results from our GPU benchmarks hierarchy, while still valid for when they were run, need to be refreshed. I'll be working on a revised GPU hierarchy in the coming weeks, but it will be a bit before that's fully ready — we want at least all the current generation cards to be included.
Our PC is hooked up to a Samsung Odyssey Neo G8 32, one of the best gaming monitors around, allowing us to potentially experience some of the higher frame rates that might be available on the fastest GPUs. Most games can't get anywhere close to the 240 Hz limit of the monitor, especially not with budget to midrange hardware like the Arc B580 and its direct competition.
The new GPU test suite consists of 23 games. Yup, we're going whole-hog for now! I may drop some of these over time, but for the Arc Battlemage launch in particular I wanted a wider selection of games. We've also toned down on ray tracing tests, mostly because outside of a few select games, it often seems to kill performance for debatable image quality upgrades. So, while more of the games have RT support, it's only enabled in six of the games — and even then, the visual upgrades are only really noticeable in three of the games.
The remaining 17 games are run in pure rasterization mode. Also, all 23 games were tested without any upscaling or frame generation. We'll see about doing additional XeSS testing on the B580 soon, but trying to compare DLSS, FSR 2/3, and XeSS performance without accounting for differences in image quality strikes us as a bad baseline way of measuring performance. Plus, we'd rather the default in games be native rendering, leaving upscaling and framegen as true performance boosting options — so you can break 120 fps or 144 fps, rather than just trying to get to 60 fps.
All the games are tested using 1080p 'medium' settings (the specifics vary by game and are noted in the chart headers), along with 1080p, 1440p, and 4K 'ultra' settings. Some may wonder about the reasoning behind the selected settings, so let's quickly elaborate.
What we want to show with graphics cards is how performance scales. We include 1080p medium as a baseline "everything released in the past few years ought to handle this" setting. Then moving to 1080p ultra provides enough of a gap to be interesting — sometimes it's still only 10% slower, but other games it might be half as fast as medium settings. If we tested 1080p high instead, that's potentially one less useful piece of information.
Going beyond 1080p ultra, I don't want to change both the resolution and the settings, as there's going to be a lot of overlap between 1440p medium and 1080p ultra as an example. So I just test 1440p and 4K ultra, at least where it makes sense. And keep in mind that today's ultra is tomorrow's high, the next day's medium, and next week's low — except it's more like a year or so between each level.
The end result is that our tests will show both how GPUs run at comparable settings, where some designs may have shortcomings (e.g. insufficient VRAM or bandwidth), and provide ways for people to extrapolate how things would run at other settings. While we don't test 1440p or 4K at medium settings, if you check the 1080p medium to ultra scaling on a slower GPU from the same vendor, that should also apply (roughly) to a higher tier GPU at higher resolutions.
Due to the need to retest everything, the number of GPUs we'll have in these charts will be more limited than in the past. I've got the most direct modern competion with Nvidia's RTX 4060 and AMD's RX 7600 and 7600 XT. I also tested the Arc A770 and A750, plus the Arc B580 on both the 9800X3D and 13900K. But I do plan on testing some additional cards in the coming days and I'll update the charts when I'm done. The plan is to include the RTX 3060 12GB, maybe RTX 2060 to show where an even older GPU lands, and then the RTX 4060 Ti 8GB/16GB and RX 7700 XT to show a step up from the B580 level.
But the primary competition is for GPUs that cost closer to the $250 asking price of the Arc B580. If you want to get an idea of where other GPUs might land, check out our full GPU benchmarks hierarchy — and then use the percentage increase in the hierarchy and apply that to the test data from this review.
Our test PCs are now running Windows 11 24H2, with all the updates applied. We're also using Nvidia's PCAT v2 (Power Capture and Analysis Tool) hardware, which means we can grab real power use, GPU clocks, and more during our gaming benchmarks. We'll cover those results on our page on power use.
Finally, because GPUs aren't purely for gaming these days, we've run some professional and AI application tests. We've previously tested Stable Diffusion, using various custom scripts, but to level the playing field we're turning to standardized benchmarks. We use the Procyon AI Vision test and Stable Diffusion (1.5 and XL) tests; MLPerf Client 0.5 preview for AI text generation; SPECworkstation 4.0 for Handbrake transcoding, AI inference, and viewperf professional applications; Geekbench AI for a quick and dirty AI test suite; 3DMark DXR Feature Test to check raw hardware RT performance; and finally Blender Benchmark 4.3.0 for professional 3D rendering.
Jarred Walton is a senior editor at Tom's Hardware focusing on everything GPU. He has been working as a tech journalist since 2004, writing for AnandTech, Maximum PC, and PC Gamer. From the first S3 Virge '3D decelerators' to today's GPUs, Jarred keeps up with all the latest graphics trends and is the one to ask about game performance.
looks like a good value. i'm in the market for my next card but seems like waiting a little bit to see what AMD does next is not too bad of an idea. i hate waiting to see what the next best thing is but this close it seems like prudent advice.
side note: it does look like you forgot to replace the place holders on the power consumption paragraph.
"On average, the B580 used xxxW at 1080p medium, xxxW at 1080p ultra, xxxW at 1440p, and xxxW at 4K. As you'd expect, power use typically increases at higher settings and resolutions."
I am happy to have competition in the market. I imagine next gen AMD and Nvidia cards will be stronger competitors but they might be priced poorly again. Price to performance is key here. We'll see in a month or so.
Great review, against relevant parts for this price class too :D
Given that Steam shows the 3 most popular GPUs are the 3060 discrete, 4060 laptop, and 4060 discrete, Intel now has a GPU that competes in the largest part of the segment - and leads it in both value and performance.
Granted AMD and Nvidia are about to release new GPUs, but let's also note that the 5060 / 8600 aren't likely to show up until late 2025 or early 2026 if they follow their normal pattern.
Great review, I'm in the market for a SFF two slot low power card for a living room system. The APU can only do so much and I'm starting to hit walls with it lately so a lower power dGPU might be the only real answer.
I guess it will come down to an availability issue. I doubt we will see superior cards by AMD or nVidia in this price bracket by end of Q1 2025. We will certainly see lots of benchmarks blowing these early battlemage offerings in January, but nothing ready for purchase. Later battlemage offerings are in my best guess going to be in the $400 range, likely beating 7800 and 4070, but again probably not until mid-late Q1. If AMD and nVidia drop anything crushing a 4090 you can bet it will be in the $700+ range. Is it fair to say that something beating the B580 readily available in April for $250 is fair competition now? I don't know. Maybe not if a normal consumer can actually get B580 silicon before Christmas.
Great review, I'm in the market for a SFF two slot low power card for a living room system. The APU can only do so much and I'm starting to hit walls with it lately so a lower power dGPU might be the only real answer.
I would probably still lean towards an RTX 3050 6GB for that. B580 is still a little power hungry for the job.
I use an A380, and that isn't ideal either, since it still needs an 8-pin (at least that model). Though supposedly still only a 75W GPU.
Great review, Jarred! The elaboration on your thinking and updating of your test bench's hardware and software is appreciated.
It'll be some time before AMD and NVIDIA (green wants it written this way, by the way: http://http.download.nvidia.com/image_kit/LG_NVCorpBadge.pdf ... was curious as I noticed they have it written that way on their website) have new budget GPU's in this price class, so I myself wouldn't really recommend that a prospective owner waits. Of course, a lot of it also depends on if building new or upgrading (and upgrading from what). There's a huge user base at this price point, so I do imagine that Intel will get some market penetration for end users, not just prebuilds. This, particularly since day 1 drivers are already fairly stable overall and Intel now has the value leader at this price point; Intel didn't mess up this launch, whereas botched launches can tarnish audience sentiment of the product for months and years, if not permanently.
Transistor density pretty close to AMD.
Transistor density of 7600XT is 65.2 M/mm2 and B580 is 72.1 M/mm2 per Techpowerup. Sure B580 is on 5nm while 7600XT is on 6nm but that isn't that big a difference. Definitely moving in the right direction. And more importantly performance per mm2 is much closer to AMD. And performance per watt.
