Questions about A100 gpgpu-sim.config

[bbooxxx]

hi, i'm a newer in GPU simulator, i'm just wondering why some of the rodinia-2.0-ft benchmark tests failed when i change the parameter of L2 size in A100 gpgpu-sim.config? i change the set number and associativity, just like from

-gpgpu_cache:dl2 S:128:128:16,L:B:m:L:X,A:192:4,32:0,32

to

-gpgpu_cache:dl2 S:256:128:16,L:B:m:L:X,A:192:4,32:0,32

then the test named "nn-rodinia-2.0-ft" failed due to COMPLETE_ERR_FILE_HAS_CONTENTS.

i wonder why this happens and how can i correctly change the L2 size? thks!

[beneslami]

Based on my experience, when you see the word kill, most probably it's because of the amount of memory you have.
The amount of memory that accel-sim needs to run the executions exceed the total capacity.

Simply add memory modules to your machine.

[bbooxxx]

Based on my experience, when you see the word kill, most probably it's because of the amount of memory you have. The amount of memory that accel-sim needs to run the executions exceed the total capacity.

Simply add memory modules to your machine.

i think it is not due to the amount of memory because under certain config these tests all work properly. Actually, i tried to use the traces downloaded from ./get-accel-sim-traces.py instead of those generated by my 3050 Ti, the problem disappeared. I'm still really curious about why this happen. ^^

[beneslami]

When you say "under certain config", what config do you refer to?
I still can't think about anything other than memory capacity.
Can you share the config that you're referring to?

[bbooxxx]

When you say "under certain config", what config do you refer to? I still can't think about anything other than memory capacity. Can you share the config that you're referring to?

Sorry for the ambiguity in my reply. Here, "config" means the config file of gpgpu-sim (accel-sim-framework/gpu-simulator/gpgpu-sim/configs/tested-cfgs/SM80_A100/gpgpu-sim.config). Under default parameters it worked well, but failed when i changed the L2 parameter from -gpgpu_cache:dl2 S:128:128:16,L:B:m:L:X,A:192:4,32:0,32 to -gpgpu_cache:dl2 S:256:128:16,L:B:m:L:X,A:192:4,32:0,32.

[beneslami]

Can you change the bsize to 256 to see if it works?
change S:256:128:16,L:B:m:L:X,A:192:4,32:0,32 to S:256:256:16,L:B:m:L:X,A:192:4,32:0,32

[JRPan]

Can you check the err output file and review its contents?

The way you change the L2 config should be fine.