Does anyone know what the cache here refers to?

[user110864]

[kayccc]

May I Know where is this table from？

[user110864]

I found this table at CUDA for Tegra

[AastaLLL]

Hi,

The cache in the CPU column indicates the CPU cache.
The one in the iGPU column indicates the GPU cache.

Thanks.

[user110864]

Thank you for your response, but I still have doubts. Does this mean that when a function uses Pinned host memory and is accessed by the iGPU, it doesn’t utilize the L2 or L1 cache?

[AastaLLL]

Hi,

Yes.
Pinned is a page-lock CPU memory.
It doesn’t have a copy on GPU so it won’t be cached when GPU tries to access it.

Thanks.

[user110864]

Thank you, but where does the unified memory exist in the igpu, on the cache? I think the cache is too small to hold all the content, or the cache is only part of it.

[AastaLLL]

Hi,

No. All the data is stored in the memory.
But cached indicates it can use CPU/GPU cache to allow fast access.
GPU cache mechanism is similar to the CPU version.

Thanks.

[user110864]

Thanks, i get it。