[petsc-users] About hardware limitation/recommendation for PETSc (new user)

[Barry Smith]

   This is a very complex question. But most PETSc simulations are memory bandwidth limited so to first order you want to purchase something that delivers the highest possible memory bandwidth for your price tag. For a pure CPU system it is the cumulative memory bandwidth you can utilize over multiple cores that matters, https://petsc.org/release/faq/#what-kind-of-parallel-computers-or-clusters-are-needed-to-use-petsc-or-why-do-i-get-little-speedup <https://petsc.org/release/faq/#what-kind-of-parallel-computers-or-clusters-are-needed-to-use-petsc-or-why-do-i-get-little-speedup>

   GPUs are designed to be high memory bandwidth and hence can offer for large enough problems an higher overall bandwidth than a CPU can alone, but it may be difficult to take advantage of the bandwidth. In turns of mature software support in PETSc for GPUs the best choice is NVIDIA.

   If you have a good test case for your class of problems coded you might consider using some cloud services to benchmark some hardware choices for your problem before purchasing anything.

  Barry


  

> On Oct 13, 2021, at 12:18 PM, Tianchi Li <celestechevali at gmail.com> wrote:
> 
> Hello,
>  
> I’m planning to parallelize my C++ simulation code with PETSc. 
>  
> And I wish to buy a local workstation to perform the parallel code development and later productive runs.
>  
> Before purchasing the workstation, I wish to know if there is any limitation or recommendation for PETSc implementation on the hardware side ?
>  
> For example, is there any hardware limitation about the MPI or GPU parallelism ? (e.g. Nvidia or AMD graphics cards, Intel or AMD CPUs)
>  
> Concerning cost-effectiveness, does a hybrid CPU-GPU machine have more advantages than an all-CPU one ? (if using GPU acceleration) 
>  
> Thank you so much in advance.
> 
> I appreciate any advice that you provide.
> 
>  
>  

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20211013/46db37ff/attachment.html>