Franscobec/gpufetch
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[v0.22] Do not show tensor cores in TU116

Browse Source
This commit is contained in:
Dr-Noob
2022-01-12 19:34:11 +01:00
parent 49119ae7eb
commit d3aaf7cfe5
1 changed files with 10 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/cuda/cuda.cpp
View File

@@ -5,6 +5,7 @@
#include "uarch.hpp" #include "uarch.hpp"
#include "../common/pci.hpp" #include "../common/pci.hpp"
#include "../common/global.hpp" #include "../common/global.hpp"
#include "../common/uarch.hpp"
bool print_gpu_cuda(struct gpu_info* gpu) { bool print_gpu_cuda(struct gpu_info* gpu) {
char* cc = get_str_cc(gpu->arch); char* cc = get_str_cc(gpu->arch);
@@ -84,7 +85,15 @@ int64_t get_peak_performance_cuda(struct gpu_info* gpu) {
int64_t get_peak_performance_tcu(cudaDeviceProp prop, struct gpu_info* gpu) { int64_t get_peak_performance_tcu(cudaDeviceProp prop, struct gpu_info* gpu) {
// Volta / Turing tensor cores performs 4x4x4 FP16 matrix multiplication // Volta / Turing tensor cores performs 4x4x4 FP16 matrix multiplication
// Ampere tensor cores performs 8x4x8 FP16 matrix multiplicacion // Ampere tensor cores performs 8x4x8 FP16 matrix multiplicacion
if(prop.major == 7) return gpu->freq * 1000000 * 4 * 4 * 4 * 2 * gpu->topo_c->tensor_cores; if(prop.major == 7) {
// TU116 does not have tensor cores!
// https://www.anandtech.com/show/13973/nvidia-gtx-1660-ti-review-feat-evga-xc-gaming/2
if(gpu->arch->chip == CHIP_TU116 || gpu->arch->chip == CHIP_TU116BM ||
gpu->arch->chip == CHIP_TU116GL || gpu->arch->chip == CHIP_TU116M) {
return 0;
}
return gpu->freq * 1000000 * 4 * 4 * 4 * 2 * gpu->topo_c->tensor_cores;
}
else if(prop.major == 8) return gpu->freq * 1000000 * 8 * 4 * 8 * 2 * gpu->topo_c->tensor_cores; else if(prop.major == 8) return gpu->freq * 1000000 * 8 * 4 * 8 * 2 * gpu->topo_c->tensor_cores;
else return 0; else return 0;
} }