github-starred/nvitop
2
0
Fork 0
mirror of https://github.com/XuehaiPan/nvitop.git synced 2026-05-15 14:15:55 -06:00
Code Issues 29 Projects Packages Wiki Activity Actions

Fix incorrect memory reporting on coherent UMA platforms (GB10 / DGX Spark)

Browse source 
On GB10 / DGX Spark, nvmlDeviceGetMemoryInfo returns NVML_SUCCESS with
total == system MemTotal (~121GB). This causes nvitop to display full
system RAM as GPU memory instead of actually allocatable memory.

Fix: detect UMA by comparing NVML total against system virtual memory
total. If total >= 90% of system RAM, treat as unified memory and use
system virtual memory (MemAvailable) for display instead.

Preserves existing behavior for discrete GPUs.

Note: requires validation on GB10 / DGX Spark hardware. The fix has
not been independently validated on a coherent UMA system.
This commit is contained in:
Joe McLaren 2026-04-16 04:10:25 -04:00
parent a6761eb5c4
commit de09aeb9f0
1 changed files with 15 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

22
nvitop/api/device.py
View file

@ -985,13 +985,21 @@ class Device: # pylint: disable=too-many-instance-attributes,too-many-public-me
memory_info = NA
if libnvml.nvmlCheckReturn(memory_info):
if memory_info.total > 0:
return MemoryInfo(
total=memory_info.total,
free=memory_info.free,
used=memory_info.used,
reserved=getattr(memory_info, 'reserved', NA),
)
has_unified_memory = True
# Detect coherent UMA platforms (e.g. GB10 Grace Blackwell):
# nvmlDeviceGetMemoryInfo returns NVML_SUCCESS with total == system MemTotal (~121GB).
# If total >= 90% of system RAM, treat as unified memory and use MemAvailable instead.
vm = host.virtual_memory()
if vm.total > 0 and memory_info.total >= vm.total * 9 // 10:
has_unified_memory = True
else:
return MemoryInfo(
total=memory_info.total,
free=memory_info.free,
used=memory_info.used,
reserved=getattr(memory_info, 'reserved', NA),
)
else:
has_unified_memory = True
if has_unified_memory:
# Device with unified memory
# Use system virtual memory as these devices share host memory