Simulating a much larger (imaginary) GPU

[rvnonsec]

Hi,

Can someone help with the considerations to extend the config to represent a much larger machine than ampere (or even blackwell). For eg, say 256 to 512SMs and scaling up cache/memory bw similarly?

Specifically, outside of obvious things like SM count and memory, what other config knobs are likely to have to be reworked for such a config

Thanks!

[JRPan]

We have created a config for A100. However, please note it has not been tested yet. Some Ampere-specific features are under dev.
Both trace and execution mode uses the same backend. However, some API calls/instructions may not be available in the PTX execution-driven mode.

[JRPan]

#344
accel-sim/gpgpu-sim_distribution#80