Skip to content

Fix NVML memory reporting regression on coherent UMA platforms (Fixes… #463

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
parallelArchitect wants to merge 1 commit into from
Closed

Fix NVML memory reporting regression on coherent UMA platforms (Fixes… #463

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 17 additions & 2 deletions src/extract_gpuinfo_nvidia.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -655,9 +655,24 @@ static void gpuinfo_nvidia_refresh_dynamic_info(struct gpu_info *_gpu_info) {
memory_info.version = 2;
last_nvml_return_status = nvmlDeviceGetMemoryInfo_v2(device, &memory_info);
if (last_nvml_return_status == NVML_SUCCESS) {
// Check if this is a unified memory GPU (total == 0 indicates unified memory)
got_meminfo = true;
if (memory_info.total == 0) {
// Detect coherent UMA platforms (e.g. GB10 Grace Blackwell):
// NVML returns NVML_SUCCESS but total == full system MemTotal.
// This is not usable VRAM — treat as unified memory and use MemAvailable.
unsigned long long sys_mem_total = 0;
FILE *mf = fopen("/proc/meminfo", "r");
if (mf) {
char ml[256];
while (fgets(ml, sizeof(ml), mf)) {
if (sscanf(ml, "MemTotal: %llu kB", &sys_mem_total) == 1) {
sys_mem_total *= 1024;
break;
}
}
fclose(mf);
}
if (memory_info.total == 0 ||
(sys_mem_total > 0 && memory_info.total >= sys_mem_total * 9 / 10)) {
has_unified_memory = true;
} else {
SET_GPUINFO_DYNAMIC(dynamic_info, total_memory, memory_info.total);
Expand Down
Loading