Intel's Open-Source Linux Compute Stack Maturing Very Well For Arc Graphics

Written by Michael Larabel in Display Drivers on 13 March 2023 at 12:00 PM EDT. Page 2 of 8. 15 Comments.

Blender benchmark with settings of Blend File: BMW27, Compute: Intel oneAPI. Intel Arc A770 was the fastest.

With this latest Compute-Runtime stack, the Blender oneAPI support is in great shape for testing! Blender with oneAPI has worked for me going back to Blender 3.3 when using the Blender interface and enabling it from the preferences dialog, etc. But when activating oneAPI via the Blender Python API and launching Blender in the headless mode for the scripted benchmarking, using the oneAPI back-end had previously just yielded a segmentation fault. Thankfully with the latest Compute Runtime stack this is no longer happening so I am able to test the oneAPI back-end just like with the NVIDIA CUDA/OptiX and Radeon ROCm back-ends. Smooth sailing.

Blender benchmark with settings of Blend File: BMW27, Compute: NVIDIA CUDA. RTX 3060 was the fastest.

The Intel oneAPI results relative to NVIDIA CUDA were fairly promising for the Arc Graphics A750/A770 compared to the mature NVIDIA CUDA back-end with Blender with the cards ahead of the RTX 3060! Meanwhile no Blender benchmarks on ROCm in this article since ROCm was having issues with the install on Ubuntu 23.04 + Linux 6.2 though the ROCm OpenCL stack was working out fine.

Blender benchmark with settings of Blend File: BMW27, Compute: NVIDIA OptiX. RTX 3060 was the fastest.

But with the NVIDIA OptiX back-end is where the NVIDIA GPU performance really shines with Blender... The Blender RTX 2060 and RTX 3060 performance was much faster than what was provided by Arc Graphics A750/A770. Though it will be interesting to see how the Intel Arc Graphics ray-tracing performance is once there is support with Blender.

Blender benchmark with settings of Blend File: Classroom, Compute: Intel oneAPI. Intel Arc A770 was the fastest.

Blender benchmark with settings of Blend File: Classroom, Compute: NVIDIA CUDA. RTX 3060 was the fastest.

Blender benchmark with settings of Blend File: Classroom, Compute: NVIDIA OptiX. RTX 3060 was the fastest.

With the more demanding Classroom scene, the Intel oneAPI performance with the Arc Graphics A750/A770 was competing with the RTX 2060 and RTX 3060 when using the CUDA back-end but with the speedy OptiX ray-tracing is where NVIDIA easily pulls ahead.

Blender benchmark with settings of Blend File: Barbershop, Compute: Intel oneAPI. Intel Arc A770 was the fastest.

Blender benchmark with settings of Blend File: Barbershop, Compute: NVIDIA CUDA. RTX 3060 was the fastest.

Blender benchmark with settings of Blend File: Barbershop, Compute: NVIDIA OptiX. RTX 3060 was the fastest.

With the Barbershop scene, the Arc Graphics A750/A770 were even faster than the RTX 3060 performance using the CUDA back-end.

But the Arc Graphics A750/A770 were consuming much more power here than with the NVIDIA GPUs.

15 Comments - Next Page

Page: 1 2 3 4 5 6 7 8 Next Page