AMD ROCm Open-Source Stack Coming To Xilinx FPGAs

[phoronix]

Phoronix: AMD ROCm Open-Source Stack Coming To Xilinx FPGAs

Now here is some darn interesting software news from SC20... AMD, which is in the process of acquiring Xilinx, is bringing the Radeon Open eCosystem "ROCm" stack to Xilinx hardware...

http://www.phoronix.com/scan.php?page=news_item&px=AMD-ROCm-Xilinx

[CochainComplex]

AMD is really pushing the OpenSource train. I hope they do not fall for an AMD owned CUDA thing

[ms178]

And I thought they would go the SYCL 2020+ route with bringing both to fruition. Also how would AMD + Xilinx work on Windows? As ROCm is Linux-specific I wonder how that take over would translate into the Windows/Desktop world.

[baryluk]

This is awesome. Would really streamline a lot of integrations.

I wonder how managing memory, and writing kernels for FPGA with HIP would work, and how involved will it be.

[uid313]

Whatever, can we have open source OpenCL 3.0 support?

[ivan.cwb]

Awesome!

And I thought they would go the SYCL 2020+ route with bringing both to fruition. Also how would AMD + Xilinx work on Windows? As ROCm is Linux-specific I wonder how that take over would translate into the Windows/Desktop world.

It looks like that Linux is the dominant platform for high computing. Microsoft has a lot of money to make it working on Windows, but they would probably invest in a solution for WSL, like they did recently with CUDA (https://developer.nvidia.com/cuda/wsl).

I think the question here is when AMD will support ROCm in gaming gpus in a more serious way. That's the catch from NVIDIA: developers who have a gaming GPU can learn how to develop for CUDA without spending some money just for that.

[bridgman]

I think the question here is when AMD will support ROCm in gaming gpus in a more serious way. That's the catch from NVIDIA: developers who have a gaming GPU can learn how to develop for CUDA without spending some money just for that.

We have been working from the bottom up - first step was making sure we had KFD/ROCR running on all new hardware starting with Picasso and Navi1x. We are just rolling out the next step now, which is making KFD/ROCR the default OpenCL back end for all Vega and Navi parts on Linux.

Michael's OpenCL testing for the 6800/XT used the ROCm back end.

[Spacefish]

And I thought they would go the SYCL 2020+ route with bringing both to fruition. Also how would AMD + Xilinx work on Windows? As ROCm is Linux-specific I wonder how that take over would translate into the Windows/Desktop world.

is there even a market for that? i would doubt it.. HPC isn´t typically windows based. And you typically don´t have large FPGAs in consumer hardware, as the hardware typically is fixed function / it makes sense to create an ASIC if you produce your product in volume..

[ms178]

is there even a market for that? i would doubt it.. HPC isn´t typically windows based. And you typically don´t have large FPGAs in consumer hardware, as the hardware typically is fixed function / it makes sense to create an ASIC if you produce your product in volume..

I wasn't thinking of HPC-focused products from today, but what a more integrated solution of AMD/Xilinx on a single package for consumers in the future might look like for which would be use in a Windows environment and which needs to work there seemlessly, too.

[oleid]

I wasn't thinking of HPC-focused products from today, but what a more integrated solution of AMD/Xilinx on a single package for consumers in the future might look like for which would be use in a Windows environment and which needs to work there seemlessly, too.

Parts of ROCm will work / are working on Windows. DirectML for example is tied into ROCm's MiOpen.