Yifan Sun
Yifan Sun
The smallest example can recreate the bug: `./bitonicsort -timing -verify -length=256` It is not related to unified multi-GPU system. And there is no error in the emulation mode.
Have you tried to upgrade your local version of Akita?
Supersedes #159
You can check the mesh connector implementation here https://github.com/sarchlab/akita/blob/v3/noc/networking/mesh/mesh.go. Creating a ring topology is basically creating a ring connector like the mesh connector, but it will be much simpler.
@jvesely Thank you very much. Finally got a chance to test. The system level atomic works on APUs but not dGPUs. Do you know if there is any way to...
@JunjoFor `-unified-gpus` and `-use-unified-memory` should be available to all the benchmarks.
@JunjoFor The current implementation of unified memory is buggy. We are currently working on a project that will entirely revamp the implementation of unified memory. Do you need this problem...
Cache sizes are very easy to change by giving a different parameters at lines like here https://github.com/sarchlab/mgpusim/blob/v3/samples/runner/shaderarray.go#L359. DRAM size can be a bit tricky to change. May I know why...
I would not say the distribution is about GPU-GPU communication but about how the memory is allocated to GPUs. May I know what you want to understand about GPU-GPU communication?...
It is on our radar, but I do not see it can be delivered within 2025. We have access to MI100 and MI200 series GPUs, but not MI300 series. So,...