Intel ARC GPUs, Xe Architecture for dGPUs [2018-2022]

Status
Not open for further replies.
Starting with the hardware requirements, you'll need a 400/500/600-series motherboard paired with a 10th/11th/12th Gen Intel Core CPU. Moreover, the motherboard requires a full-size PCIe 3.0 (or newer) x16 slot and Resizable BAR/AMD Smart Access Memory support. However, considering Intel states the latter requirement is to ensure “optimal performance in all applications using Intel Arc A-Series graphics,” it may be just a recommendation rather than a strict requirement.

Operating system requirements include:
  • Windows 10 64-bit 20H2 or newer
  • Windows 11 64-bit
  • Confirm the Operating System is using the GPT partition type:
    • For Windows 11, this mode is configured by default.
    • For Windows 10, the partition type can be converted if installed with an MBR partition type.
 

Nice I haven’t seen a synthetic triangle throughput test in a long time. It’s just a single data point but the authors seem convinced that mesh shaders are too little too late and compute based raster is already the best option today.

This is my understanding of the tests. Please correct if wrong.

DIP - Draw Indexed Primitive. DX9 style draw call for one instance at a time. No meshlets. Hardware rasterizer.

Mesh Shader - DX12 mesh Shading of meshlets. Hardware rasterizer.

MDI - MultiDrawIndirect. DX12 style draw calls using vertex shading of meshlets. Hardware rasterizer.

Compute Shader - Custom compute shader for geometry processing and rasterization.

There are a few interesting nuggets in the triangle crunching data. RDNA 2 is much faster at DIP than MDI or mesh shading. Seems broken. Mesh shading seems broken on Intel as well. Nvidia and Intel are just as fast when running MDI or compute which makes more sense.

The 6700xt and 2080Ti are 6x faster than A370m. Wowza.

RT has some interesting results too. RDNA 2 is over 2x as fast as Arc on fast BVH builds but it’s a tie with full rebuilds.

The relative performance of static BVH and fast builds is consistent across all 3 architectures. However Nvidia and AMD lose ground to Intel with full builds. Maybe seeing some of Intel’s secret RT sauce at work.
 

With 1 exception confirming the rule, A380 is faster with i5-12400 than R5 5600
 
We need Intel to compete with NV, since AMD isn't really doing that. Competition will enable for lower prices and probably more performance.
 
Performance-wise there will likely be some increase due to future driver improvements. Though not sure how much can be gained based on Hitman 3 results (which likely had driver optimizations).

Intel-A380-1080p-Gaming-Benchmarks.jpg
 
As expected, it's a complete shit show.
 

Oh dear, that’s not very promising. If Arc is struggling to outmuscle lower powered Ampere parts it’ll be coming to market with a whimper.
 
Double the power of the nvidia chips they compared to being 15% faster? Probably won't even be faster when third party tests come out.
 
Status
Not open for further replies.
Back
Top