sarchlab / mgpusim

A highly-flexible GPU simulator for AMD GPUs.

sarchlab/mgpusim Issues

total_kernel_time = 0 when using unified-gpus
Updated 13 days ago1
vgg16 benchmark
Updated 2 months ago
DRAM Latency is not recorded in `metrics.csv`
Updated 2 months ago
Failed Execution for Program Compiled with ROCm 1.7.1
Updated 2 months ago2
Support HIP, as well as OpenCL kernels compiled by new ROCm versions
Updated 3 months ago3
Rewrite the DNN in a way that all the memory are passed from outside
Updated 3 months ago
Update instruction tracing so that all instructions show in one interface
Updated 3 months ago
adding perfRegister
Updated 4 months ago
Perf analyzer cannot be used in mgpusim
Closed 4 months ago
MMU is not being traced for visualization
Updated 4 months ago
Larger simulations are not deterministic
Updated 4 months ago11
Parallel simulation results is very different from serial simulation
Updated 4 months ago1
where is private segment memory allocation for kernel load ?
Closed 4 months ago5
We need a GMMU implementation that can walk the page table locally on each GPU
Updated 5 months ago
Add a flag `--no-monitor` to disable AkitaRTM
Updated 5 months ago
NVIDIA: Support block dispatching that considers SM capacity.
Updated 6 months ago
Add units to metrics.csv report
Updated 6 months ago
Report WF count per GPU
Updated 6 months ago
Deterministic test is not working
Closed 7 months ago
Report RDMA Trace
Closed 7 months ago1
While VALU should take 4 cycles to execute, it should not be a pipeline.
Updated 7 months ago
VALU instructions should take 4 cycles rather than 5 cycles.
Updated 7 months ago
DRAM latency is reported as 0
Closed 8 months ago
Show progress bar for memory copy
Updated 8 months ago
Analyzer is not working
Updated 9 months ago1
Merge `gitlab.com/akita/dnn` into `github.com/sarchlab/mgpusim`
Closed 9 months ago
RDMA itself should not be a performance bottleneck
Closed 10 months ago
RMDA should have a builder
Closed 10 months ago
ISADebugger: Tracer is only tracing a specific WF. The condition should remove.
Updated 10 months ago
The design of the scratchpad is impacting the performance
Updated a year ago
MultiNode Simulation
Closed a year ago13