The first benchmark results for the Intel Arc A580 RI (not sure what “RI” means) have been added to the Ashes of the Singularity database. This comes just hours after Intel released specifications for its upcoming Arc Alchemist graphics cards, including the mid-range Arc A580. This card should take on the mid-range GPUs in the GPU benchmark tier and could even replace one of the best. A graphics card, if priced appropriately.
Testing unreleased hardware is always difficult because the drivers may not be optimized and software tweaks may be required. But these Arc A580 performance numbers raise more questions than answers. The main question is whether the published results were actually obtained on the Arc A580 and whether they were running the final clocks and decent drivers.
@benchleaks Found 4 “Intel Arc A580 Graphics RI” benchmark results in AOTS database1, 2, 3, Four) obtained on a system based on an unreleased Intel 16-thread processor with a default clock rate of 2.50 GHz (and an unknown turbo frequency) and 16 GB of memory. Results were obtained with the Min_1080p preset. Comparing leaked results for Intel’s upcoming mid-range Arc A580 graphics board with Intel’s own test results for Arc A380, and results from Nvidia’s entry-level graphics card from his AOTS database. I made it
Note that some of the Arc A580 results were obtained using the DirectX 11 renderer and some using the DirectX 12 renderer. AOTS DX11 results are hardly worth considering given that Intel openly admitted that they did not optimize their discrete GPU drivers for the DirectX 11 application programming interface. Meanwhile, to show the difference, I’ve added the best Arc A580 results with the DX11 and DX12 renderers to the table.
Note: Nvidia graphics card performance numbers obtained on systems based on high performance CPUs are not included.
Currently, our own test results for the Arc A380 look pretty good when compared to Nvidia’s GeForce GTX 1650. This is logical as this is an entry-level gaming GPU based on an architecture from several years ago. On the other hand, the RTX 3050 results look bad as they use DX11 — RTX 3050 DX12 figures found did not.
So Jarred pulled out a few other GPUs and tested the same setup to put things in perspective, all on the same i9-12900K platform. You can see how things stack up and don’t lend much credence to the supposed A580 results, in these he AOTS benchmarks the Arc A580 seems to be slower than the Arc A380. There are several possible reasons for these strange results.
- It turned out that the CPU in the system used to benchmark the Arc A580 severely limited performance, slowing down the GPU, which should be 3x faster than the Arc A380.
- The driver used for Arc A580 is not optimized for ACM-G10 GPU.
- The Arc A580 RI is a power limited mobile part.
- This is not the Arc A580 and someone has fooled the benchmarks into thinking they are dealing with this card when in fact they are dealing with a different one.
Honestly, Ashes of the Singularity has not had good GPU benchmarks since the game launched in 2016 and should be fairly easy to crack on modern high-end graphics processors, especially with the Min_1080p preset. On the other hand, this benchmark favors AMD GPUs and makes great use of faster CPUs. CPU benchmarks are better than GPU benchmarks, especially at Min_1080p settings. Also, AOTS tends to perform better on his second run (but not always!). This suggests that run-to-run variability can grow if you are not careful.
In other words, all AOTS benchmark results should be viewed with a pinch of salt at the viewer’s discretion, perhaps with a slice of lemon and a shot of tequila. Bearing in mind that we are dealing with pre-release hardware with unknown drivers, it is very difficult to draw conclusions about Intel’s Arc A580 performance in AOTS and whether or not this is the actual Arc A580. I don’t even know. Based on what I’m seeing on the Intel graphics channel, I should have some real hardware on hand that I can run my own tests on in the not too distant future.
Jared Walton contributed to the story.