CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

cuda: fix vmm oom issue on NVIDIA AGX Orin (#4687)

Browse Source 
Signed-off-by: hydai <hydai@secondstate.io>
This commit is contained in:
hydai
2023-12-30 00:31:19 +08:00
committed by GitHub
parent 04ac0607e9
commit 91bb39cec7
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml-cuda.cu
View File

@@ -6662,7 +6662,7 @@ static void ggml_cuda_pool_free_leg(int device, void * ptr, size_t size) {
// pool with virtual memory // pool with virtual memory
static CUdeviceptr g_cuda_pool_addr[GGML_CUDA_MAX_DEVICES] = {0}; static CUdeviceptr g_cuda_pool_addr[GGML_CUDA_MAX_DEVICES] = {0};
static size_t g_cuda_pool_used[GGML_CUDA_MAX_DEVICES] = {0}; static size_t g_cuda_pool_used[GGML_CUDA_MAX_DEVICES] = {0};
static const size_t CUDA_POOL_VMM_MAX_SIZE = 1ull << 36; // 64 GB static const size_t CUDA_POOL_VMM_MAX_SIZE = 1ull << 35; // 32 GB
static void * ggml_cuda_pool_malloc_vmm(int device, size_t size, size_t * actual_size) { static void * ggml_cuda_pool_malloc_vmm(int device, size_t size, size_t * actual_size) {
scoped_spin_lock lock(g_cuda_pool_lock); scoped_spin_lock lock(g_cuda_pool_lock);