AMDGPU.jl VS klipper-lb

For myself, I use Julia to write my own software (that is run on AMD supercomputer) on Fedora system, using 6800XT. For my experience, everything worked nicely. To install you need to install rocm-opencl package with dnf, AMD Julia package (AMDGPU.jl), add yourself to video group and you are good to go. Also, Julia's KernelAbstractions.jl is a good to have, when writing portable code.

The Julia package AMDGPU.jl provides a Julia interface for AMD GPU (ROCm) programming. As they say, the package is being developed for Julia 1.7, 1.9 and above, but not 1.8. Therefore I downloaded the Julia binary of version 1.7.3 from the older releases Julia page.

This is exciting news! What's also exciting is that it's not just C++ that can run on this supercomputer; there is also good (currently unofficial) support for programming those GPUs from Julia, via the AMDGPU.jl library (note: I am the author/maintainer of this library). Some of our users have been able to run AMDGPU.jl's testsuite on the Crusher test system (which is an attached testing system with the same hardware configuration as Frontier), as well as their own domain-specific programs that use AMDGPU.jl.

What's nice about programming GPUs in Julia is that you can write code once and execute it on multiple kinds of GPUs, with excellent performance. The KernelAbstractions.jl library makes this possible for compute kernels by acting as a frontend to AMDGPU.jl, CUDA.jl, and soon Metal.jl and oneAPI.jl, allowing a single piece of code to be portable to AMD, NVIDIA, Intel, and Apple GPUs, and also CPUs. Similarly, the GPUArrays.jl library allows the same behavior for idiomatic array operations, and will automatically dispatch calls to BLAS, FFT, RNG, linear solver, and DNN vendor-provided libraries when appropriate.

I'm personally looking forward to helping researchers get their Julia code up and running on Frontier so that we can push scientific computing to the max!

Library link: <https://github.com/JuliaGPU/AMDGPU.jl>

GitHub - JuliaGPU/AMDGPU.jl: AMD GPU (ROCm) programming in Julia

https://github.com/JuliaGPU/AMDGPU.jl

https://github.com/JuliaGPU/oneAPI.jl

These are both less mature than CUDA.jl, but are in active development.

Service By, formerly Klipper LB, could also be an option, this is also by default integrated in k3s: https://github.com/k3s-io/klipper-lb

IMO Klipper is super simple compared to other load balancers: You setup a load balancer saying "hey I want to access this service on port 8080" and then you can access it from the outside by using "AnyKubernetesNodeIp:8080".

klipper (servicelb) is a very basic implementation of a load balancer, reading thru the issues you'll soon discover there's many reasons why metallb is preferred. For example klipper only supports externalTrafficPolicy: Cluster which means the real source IP of the request is not passed thru to the pod. The pod only gets the LB IP.

GitHub - k3s-io/klipper-lb: Embedded service load balancer in Klipper