I want to use the L1 size of a TitanV (volta) device in my program and I am confused with the actual size. According to [1] :
In Volta the L1 cache, texture cache, and shared memory are backed by a combined 128 KB data cache.
…
Volta increases the maximum capacity of the L1 cache to 128 KB
So should I use the total 128KB size or exclude texture and shared capacities? Shared size is also variable.
On the other hand, in TechPowerup [2], I see that L1 size is 96KB.
The same question exists for 2080Ti (Turing). According to [3]
The total size of the unified L1 / Shared Memory cache in Turing is 96 KB.
…
Turing supports two carveout configurations, either with 64 KB of shared memory and 32 KB of L1, or with 32 KB of shared memory and 64 KB of L1.
But TechPowerup says it is 64KB [4].
What are the actual sizes?
[1] https://docs.nvidia.com/cuda/volta-tuning-guide/index.html#l1-cache
[2] NVIDIA TITAN V Specs | TechPowerUp GPU Database
[3] https://docs.nvidia.com/cuda/turing-tuning-guide/index.html#l1-cache
[4] NVIDIA GeForce RTX 2080 Ti Specs | TechPowerUp GPU Database