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.
Intel Arc A580 AI Performance
GPUs are also used with professional applications, AI training and inferencing, and more. Along with our usual professional tests, we've added Stable Diffusion benchmarks on the various GPUs. AI is a fast-moving sector, and it seems like 95% or more of the publicly available projects are designed for Nvidia GPUs. Those Tensor cores aren't just for DLSS, in other words. Intel has XMX (Xe Matrix eXtensions) for a similar performance boost, while AMD has "AI Accelerators" in RDNA 3 to help boost FP16 performance. Let's start with our AI testing and then hit the professional apps.
We're using Automatic1111's Stable Diffusion version for the Nvidia cards. Intel also has an OpenVINO fork of Automatic1111's webui, which shows about a 50% improvement over our previous testing. Finally, AMD has a DirectML fork of Automatic1111, though there are some limitations: You have to use SD 1.5, not SD 2.1, which means 768x768 image isn't using 768x768 training data; you can only generate one image at a time (we normally do a batch of 24); and using a batch size other than 1 also fails.
The DirectML fork seems to provide a good boost in performance to RDNA 3 GPUs, but RDNA 2 GPUs are slower, so we're using a recent Nod.ai Shark variant for RDNA 2 and the DirectML fork for RDNA 3.
With the latest OpenVINO fork of Stable Diffusion, Intel's GPUs look quite impressive. The A580 nearly ties the RTX 3060 at 512x512 generation, though 768x768 results aren't quite as high. AMD's GPUs meanwhile still look relatively tame. The Olive-ONNX DirectML fork does boost the RX 7600 up to RTX 3050 levels, though, which is much better than what we saw previously.
Nvidia pulls slightly ahead with 768x768 image generation, at least with the RTX 4060. AMD's GPUs don't look nearly as competitive at higher resolutions either, and there are plenty of issues that still need to be worked out (see above). If OpenVINO and OneAPI gain support, though, Intel's Arc GPUs could provide a real alternative to Nvidia for AI workloads.
Intel Arc A580 Professional Workloads
SPECviewperf is a different story, and overall the Arc A580 ends up as the slowest GPU in our selection of cards. It's just a hair behind the RTX 3050, while the 3060 and all of the AMD GPUs easily pull ahead. Even Intel's top Arc A770 16GB can't match the RTX 3060 or RX 6600.
As usual, the overall score doesn't really get into all of the details, so check the individual charts for any professional applications you might actually use. Arc does decently in Catia, Maya, and Medical. SNX meanwhile is a horrible showing. SolidWorks, Energy, Creo, and 3DSMax all put Arc at or near the bottom, but the margins are quite as severe as in SNX.
Still, unless Intel really changes things around with future drivers, it's difficult to imagine any professionals taking the Arc A-series seriously. Intel does have Arc Pro offerings, but we're not sure if they're truly tuned for professional applications or if they just have "Pro" tacked on to their product name — we've looked around and can't find any SPECviewperf tests of the Arc Pro A60, sadly.
For GPU-accelerated 3D rendering, the only app that currently supports all three GPU vendors is Blender. It now leverages the ray tracing hardware to boost performance, and that generally puts Nvidia in the pole position. Still, Intel's Arc offerings aren't that far behind.
We're now using Blender Benchmark, version 3.6.0. The Nvidia results haven't changed much with recent updates, but 3.6.0 gave a significant boost to the Intel Arc results. Oddly, the A770 and A750 are effectively tied now, while the A580 falls a bit behind. The A580 still beats AMD's RX 6700 10GB by 50%, and nearly doubles the performance of the RX 6600. At the same time, if you're serious about 3D rendering with Blender, you'd probably want something faster from the RTX 40-series lineup.
Intel Arc A580 Content Creation Summary
Intel's Arc GPUs can be very hit or miss for content creation, depending on what you want to do. Stable Diffusion worked well, Blender was okay, but professional applications are a weak spot.
Something else to consider is the video encoding/decoding performance and quality. We didn't retest anything for this review, but we looked at AMD, Intel, and Nvidia video encoding back in March. Nvidia came out on top, but Intel wasn't far behind. The A380 and A770 had effectively identical results, and we expect the A750 and A580 to perform the same.
Other aspects of video encoding with Arc again show driver immaturity. For example, the "Highlights" feature (Alt+H by default) captures video, but only at up to 1080p and 60 Hz. Furthermore, there was a clear drop in performance when using the function while playing a game. There are other missing aspects as well, like you only get to specify one file name — AMD and Nvidia smartly create process and time/date stamped videos with their equivalent tools.
- MORE: Best Graphics Cards
- MORE: GPU Benchmarks and Hierarchy
- MORE: All Graphics Content
