Blender Cycles Rendering Support For Intel Arc Via oneAPI + SYCL Under Review

[phoronix]

Phoronix: Blender Cycles Rendering Support For Intel Arc Via oneAPI + SYCL Under Review

Opened up at the end of March is the work-in-progress Intel oneAPI back-end for Blender's Cycles renderer. This Intel GPU back-end focused for supporting the company's forthcoming Intel Arc graphics cards is targeting the open-source oneAPI Base Toolkit and making use of SYCL. There still is more code work needed, but it's good to see this coming together to complement Blender's NVIDIA CUDA and AMD HIP support...

https://www.phoronix.com/scan.php?page=news_item&px=Intel-Blender-Backend-Review

[tuxd3v]

On one end, its nice to know that support for intel Arc graphics come to linux..
On the other end, we see a fragmentation of APIs supported, in blender at least..

While this is maybe not a problem for the final user( for blender ),
If this trend goes on, our dream of a unified API for compute will vanish?

[cl333r]

On one end, its nice to know that support for intel Arc graphics come to linux..
On the other end, we see a fragmentation of APIs supported, in blender at least..

While this is maybe not a problem for the final user( for blender ),
If this trend goes on, our dream of a unified API for compute will vanish?

It was bound to happen and stay this way until Linux gets a CUDA competitor both in terms of performance and ease of use.

Isn't oneAPI itself a collection of APIs?

[MadeUpName]

Intel: Software with out hardware.
AMD: Hardware with out software.
NVidia: Has both but you have to give up what make Linux great to use it.

[cl333r]

Intel: Software with out hardware.
AMD: Hardware with out software.
NVidia: Has both but you have to give up what make Linux great to use it.

It's funny cause it's true, though "with out" is one word, UNLESS IT ISN'T???

[tuxd3v]

It was bound to happen and stay this way until Linux gets a CUDA competitor both in terms of performance and ease of use.

We have opencl( and a mesa driver for it - Clover ),
I am not an expert in it, but it works...it seems not so easy to use as cuda, but we have a solution..
I think its not good or bad, its so so..
Last time I tested Clover,
It was around 10% faster than AMD legacy driver( but I don't test it for a long long time..and there are no image support yet, but there are a collection of patches..in mesa for it.. with a big challenge to anyone interested in participate in a mission.. )

Isn't oneAPI itself a collection of APIs?

Well, ..a collection of APIs that are called "oneAPI"

[Quackdoc]

We have opencl( and a mesa driver for it - Clover ),
I am not an expert in it, but it works...it seems not so easy to use as cuda, but we have a solution..
I think its not good or bad, its so so..
Last time I tested Clover,
It was around 10% faster than AMD legacy driver( but I don't test it for a long long time..and there are no image support yet, but there are a collection of patches..in mesa for it.. with a big challenge to anyone interested in participate in a mission.. )

No. I would say for most people clover is just downright bad. I have hope for rusticl, but clover in it's current state is NOT a solution at all.

I see vulkan still being the viable solution going forward. but it will take a lot of work to get there.

[wsippel]

It was bound to happen and stay this way until Linux gets a CUDA competitor both in terms of performance and ease of use.

Isn't oneAPI itself a collection of APIs?

HIP is probably the closest thing, as it's open source and has backends for AMD and Nvidia GPUs as well as CPUs. Ironically, oneAPI itself has a HIP backend as far as I understand, so you could presumably run applications written for Intel oneAPI on Nvidia GPUs through AMD HIP. But that adds an extra layer.

[qarium]

On one end, its nice to know that support for intel Arc graphics come to linux..
On the other end, we see a fragmentation of APIs supported, in blender at least..
While this is maybe not a problem for the final user( for blender ),
If this trend goes on, our dream of a unified API for compute will vanish?

we already have this unified compute api it is called WebGPU...

[MadeUpName]

We have opencl( and a mesa driver for it - Clover ),
I am not an expert in it, but it works...it seems not so easy to use as cuda, but we have a solution..
I think its not good or bad, its so so..
Last time I tested Clover,
It was around 10% faster than AMD legacy driver( but I don't test it for a long long time..and there are no image support yet, but there are a collection of patches..in mesa for it.. with a big challenge to anyone interested in participate in a mission.. )


Well, ..a collection of APIs that are called "oneAPI"

I have MESA22 installed and Clover doesn't have image support. Saying it could have it some day is no different from AMD saying their drivers may have it some day.