Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
1–10 of 28 posts
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#2If I recall correctly the AMD hardware is competitive or better in raw compute than their nVidia counterparts but Cuda makes developing software so easy so I’m curious what ORNL’s take on that is.
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#3If I recall correctly the AMD hardware is competitive or better in raw compute than their nVidia counterparts but Cuda makes developing software so easy so I’m curious what ORNL’s take on that is.
I'm wondering the same. I'd guess that 99% of science code that uses the GPU now depends on CUDA. The other 1% use other third-party GPU frontends which might also be tuned for Nvidia GPUs. I feel that it's good to use the AMD ecosystem to mix up the competition a bit, and the majority of other clusters will still use Nvidia if your application requires it, but I'm just not aware of any momentum at all (in the form of software ecosystems) that target AMD, so library-writers will have to basically start from scratch if this AMD cluster is confirmed.
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#4If I recall correctly the AMD hardware is competitive or better in raw compute than their nVidia counterparts but Cuda makes developing software so easy so I’m curious what ORNL’s take on that is.
I'm wondering the same. I'd guess that 99% of science code that uses the GPU now depends on CUDA. The other 1% use other third-party GPU frontends which might also be tuned for Nvidia GPUs. I feel that it's good to use the AMD ecosystem to mix up the competition a bit, and the majority of other clusters will still use Nvidia if your application requires it, but I'm just not aware of any momentum at all (in the form o…
>... so library-writers will have to basically start from scratch if this AMD cluster is confirmed.
I doubt library writers will care beyond anyone involved in the project. That said, it's possible to port CUDA code already with a bit of work, so not quite from scratch in any case:
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#5If I recall correctly the AMD hardware is competitive or better in raw compute than their nVidia counterparts but Cuda makes developing software so easy so I’m curious what ORNL’s take on that is.
I’m wondering the same thing. With the size of this lab, maybe this will drive the adoption of open alternatives to CUDA. Serious amount of software will be written for this supercomputer. In Canada, however, all public GPU clusters use nvidia...
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#6If I recall correctly the AMD hardware is competitive or better in raw compute than their nVidia counterparts but Cuda makes developing software so easy so I’m curious what ORNL’s take on that is.
I don't work for ORNL and don't know what their take in this is, but a lot of codes that the DoE uses are being ported to the Kokkos framework out of the Sandia National Lab. The application developer at that point basically doesn't have to care if the code runs on KNL, on Nvidia cards using Cuda or on AMD cards using ROCm. The work of tuning Kokkos well on AMD cards is cheap compared to getting stuck on a single vendor.
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#7If I recall correctly the AMD hardware is competitive or better in raw compute than their nVidia counterparts but Cuda makes developing software so easy so I’m curious what ORNL’s take on that is.
Presumably they intend to expand CAAR[1] to help with this.
They're going to need to, if they want this to succeed.
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#8If I recall correctly the AMD hardware is competitive or better in raw compute than their nVidia counterparts but Cuda makes developing software so easy so I’m curious what ORNL’s take on that is.
I don't work for ORNL and don't know what their take in this is, but a lot of codes that the DoE uses are being ported to the Kokkos framework out of the Sandia National Lab. The application developer at that point basically doesn't have to care if the code runs on KNL, on Nvidia cards using Cuda or on AMD cards using ROCm. The work of tuning Kokkos well on AMD cards is cheap compared to getting stuck on a single ven…
Is it open source? It’d be huge to have something compete with cuda so that the best hardware may win
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#9Earlier quoted context omitted.
I don't work for ORNL and don't know what their take in this is, but a lot of codes that the DoE uses are being ported to the Kokkos framework out of the Sandia National Lab. The application developer at that point basically doesn't have to care if the code runs on KNL, on Nvidia cards using Cuda or on AMD cards using ROCm. The work of tuning Kokkos well on AMD cards is cheap compared to getting stuck on a single ven…
Is it open source? It’d be huge to have something compete with cuda so that the best hardware may win
Kokkos (as in the C++ template frame work and a lot of tools to get insights into bottle necks) is opens source. See https://github.com/kokkos/kokkos and https://github.com/kokkos/kokkos-tools . They take pull requests and react to bug reports. They even plan to upstream some things from kokkos (such as multi-dimensional arrays) into future C++ standards.
The mission related codes based on kokkos are very much not open source.
Re: Frontier: ORNL's 2021 exascale supercomputer will run on AMD CPUs and GPUs
#10Here is a recent training from AMD for Frontier:
Link: https://www.exascaleproject.org/event/amd-gpuprogramming-hip...
Video: https://youtu.be/3ZXbRJVvgJs
Slides: https://www.exascaleproject.org/wp-content/uploads/2017/05/O...
It will be very interesting to see what AMD came up with to convince ORNL to switch from 7 years (or more?) of NVIDIA to something else. I don't think it's just a lower price. Perhaps AMD is doing a closer tie-in between CPU and GPU. Since they own both.