Question on Customizing SQC and TCP Cache Settings in apu_se.py Architecture #603
-
Hello authors, I am currently working with the gem5 simulator, specifically with the
Thank you for your assistance and looking forward to your suggestions. |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments
-
GPU_VIPER.py has a number of arguments to adjust cache size/latency/associativity/etc., starting here: https://github.com/gem5/gem5/blob/develop/configs/ruby/GPU_VIPER.py#L413. You can either use these on the command line or change the defaults in the Python file per your preference. Note that this will change once the GPU has a board like the CPUs though. Regarding your second point, I'm assuming what you really mean is you want the main memory to be DDR3 instead of HBM? Because making the caches use DDR3 instead of SRAM I have no idea how to do. In terms of the main memory, you can use the --mem-type parameter on the command line or in the Python file: https://github.com/gem5/gem5/blob/develop/configs/common/MemConfig.py#L127. In terms of what memories are possible to use here, I believe all of them here are possible (including DDR3): https://github.com/gem5/gem5/blob/develop/src/mem/DRAMInterface.py. Note that this will also change once the GPU gets a board. Beyond this, it is up to you to modify to your specifications and accuracy requirements. Hope this helps, |
Beta Was this translation helpful? Give feedback.
-
Hi @EricChenSudo , I agree with everything that Prof. @ mattsinc mentioned. Those files seem like they will allow you to modify the parameters you are looking to change. Just as a nitpicky clarification to the original post, SQC / TCP actually decode to SeQencer Cache and Texture Cache per Pipeline (CU) - Want to make sure it's clear TCP is a cache and not a processor - Computer architects use a lot of overloaded acronyms and these are some of them |
Beta Was this translation helpful? Give feedback.
-
I am converting this issue to a Q/A discussion. |
Beta Was this translation helpful? Give feedback.
GPU_VIPER.py has a number of arguments to adjust cache size/latency/associativity/etc., starting here: https://github.com/gem5/gem5/blob/develop/configs/ruby/GPU_VIPER.py#L413. You can either use these on the command line or change the defaults in the Python file per your preference. Note that this will change once the GPU has a board like the CPUs though.
Regarding your second point, I'm assuming what you really mean is you want the main memory to be DDR3 instead of HBM? Because making the caches use DDR3 instead of SRAM I have no idea how to do. In terms of the main memory, you can use the --mem-type parameter on the command line or in the Python file: https://github.com/gem5/gem5/blob/…