Please note that this post is tagged as a rumor.
New desktop Arc GPU, but it is not A580
According to the CompuBench website, Intel might be working on a new graphics card with 16 Xe-Cores.
Thus far only the mobile Arc A550M features 16 Xe-Cores, and it is based on ACM-G10 GPU. A new card that has just been spotted on benchmark website is apparently clocked at much higher speed, so clearly not a mobile variant.
As a reminder, Intel is yet to launch its Arc A580 desktop graphics card. This model was announced months ago and there have been no updates on this graphics card ever since. This time, however, the company was careful not to mention any release date, therefore no one except Intel knows when it launches.
Meanwhile, it looks like Intel might still be testing a new card internally. Spotted over CompuBench website, there is an Arc GPU running test driver (9999) with 256 Compute Units and clock speed varying from 2400 to 2450 MHz. Needless to say this card is not the mobile A550M, but an unreleased desktop variant.
With 256 Compute Units reported, this means a configuration of 16 Xe-Cores. However, the A580 desktop model is said to feature 24 cores. It is therefore a change of specs for the A580 or this is yet another variant that we do not know yet. In theory, it could also be one of the upcoming Mainstream/Performance GPUs that were leaked a few weeks ago. Such cards are to launch by the end of this quarter.
Intel is reportedly working on a third Arc GPU called ACM-G12. Such a GPU would feature exactly 16 Xe-Cores and in theory it would be a perfect fit for this desktop model.
CompuBench is quite possibly the last benchmark a gamer would have chosen for comparison, yet this is all we have got for now. A quick check on A770 performance tells us that on average, it is 42% slower than A770. Technically speaking, it should be half of the speed, considering only 16 Xe-Cores will be enabled on this GPU.
|Intel Arc A-Series in CompuBench OpenCL|
|VideoCardz.com||Arc A???||Arc A770||A??? vs. A770|
|Level Set Segmentation – 128||7923.273 mVoxels/s||21483.02 mVoxels/s||-63%|
|Level Set Segmentation – 256||8235.018 mVoxels/s||14821.874 mVoxels/s||-44%|
|Ocean Surface Simulation||3285.159 Iterations/s||4457.6 Iterations/s||-26%|
|4457.6 Iterations/s||132.738 mTriangles/s||176.837 mTriangles/s||-25%|
|Catmull-Clark Subdivision Level 5||283.535 mTriangles/s||511.032 mTriangles/s||-45%|
|Vertex Connection and Merging||14.884 mPixels/s||22.693 mPixels/s||-34%|
|Subsurface Scattering||7360.095 mSample/s||15300.483 mSample/s||-52%|
|Subsurface Scattering multiple view||7691.361 mSample/s||14431.641 mSample/s||-47%|
|TV-L1 Optical Flow||50.924mPixels/s||93.356 mPixels/s||-45%|
Source: CompuBench via KOMACHI_ENSAKA