AMD GPU Out of Memory Issue for 2nd GPU

1

Hi again,
I am having this issue MultiGPU malloc_device fails for second GPU (Out of Memory) · Issue #16038 · intel/llvm · GitHub
but I remember not having this issue before, so I was wondering what am I missing with my build routine and do you have any recipe to follow to get everything up and running ?

Thank you in advance,

2

It also fails with 2025.0.0 with codeplay plugin with this error

<HIP>[ERROR]: 
UR HIP ERROR:
        Value:           2
        Name:            hipErrorOutOfMemory
        Description:     out of memory
        Function:        getNextTransferStream
        Source Location: /tmp/tmp.6nq6FrBCn9/intel-llvm-mirror/build/_deps/unified-runtime-src/source/adapters/hip/queue.cpp:106
3

Hi,
I was wondering if there is any update/progress with this issue ?

Thanks!

4

Hi @br-ko,

we haven’t been able to replicate the setup that you have to be able to try this. We will try to create the nearest-possible setup soon though we might not be able to get to it before the end of the year.

If you had a small reproducer that exhibited this issue, this would be very helpful for us to be able to recreate your problem.

Many thanks,
Duncan.

5

Hi Duncan,
Thank you for the reply and explanation.
Would the provided reproducer in the intel/llvm issue linked in the first message helpful ?

Best,

6

Thanks @br-ko , you’re right, that should do it. We will try to take a look soon.