How much of each GPU works, beside whether anything does
The busy figure is nvidia-smi's utilization.gpu: the share of time any kernel ran, which a kernel on one SM fills. On cuda01 decode read 92% busy while the SMs had work 69% of the time and issued on 18% of their cycles. Busy is now said to be that, and beside it: each card's draw against its power limit (efficiency table, fleet and node charts, the console's GPU work chart), and where the profiling counters can be read, SMs active, SM occupancy, tensor cores and DRAM (llmgw_gpu_*_active_percent), from NVML's GPM on Hopper and later or a dcgm-exporter on the node (discovery.dcgm_exporter). The GPU map colours by work by default and keeps busy as a mode. A MIG instance is never published with its card's profile. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
7bfe70c8df
commit
c866b3a8ee
34 files changed
+954
-52
No files matched your search
@@ -64,11 +64,13 @@
|
||||
{{/* Per card, because a model on eight cards making twice the tokens of one
|
||||
on two is the worse of the two; per watt-hour, because that is what the
|
||||
room is paying. */}}
|
||||
<table class="rows"><tr><th>{{t "dash.col.model"}}</th><th class="num">GPUs</th><th class="num">{{t "dash.eff.tps"}}</th><th class="num">{{t "dash.eff.per_gpu"}}</th><th class="num">{{t "dash.eff.busy"}}</th><th class="num">{{t "dash.eff.draw"}}</th><th class="num">{{t "dash.eff.per_wh"}}</th></tr>
|
||||
<table class="rows"><tr><th>{{t "dash.col.model"}}</th><th class="num">GPUs</th><th class="num">{{t "dash.eff.tps"}}</th><th class="num">{{t "dash.eff.per_gpu"}}</th><th class="num">{{t "dash.eff.busy"}}</th><th class="num">{{t "dash.eff.power_load"}}</th><th class="num">{{t "dash.eff.sm_active"}}</th><th class="num">{{t "dash.eff.draw"}}</th><th class="num">{{t "dash.eff.per_wh"}}</th></tr>
|
||||
{{range .}}<tr><td>{{.Model}}</td><td class="num">{{num .GPUs 0}}</td><td class="num">{{num .TokensPerSec 1}}</td>
|
||||
<td class="num">{{num .PerGPU 1}}</td><td class="num">{{pc .Util 0}}</td>
|
||||
<td class="num">{{if ge .PowerLoad 0.0}}{{pc .PowerLoad 0}}{{else}}—{{end}}</td><td class="num">{{if ge .SMActive 0.0}}{{pc .SMActive 0}}{{else}}—{{end}}</td>
|
||||
<td class="num">{{if .Watts}}{{num .Watts 0}} W{{else}}—{{end}}</td><td class="num">{{if .PerWh}}{{num .PerWh 0}}{{else}}—{{end}}</td></tr>
|
||||
{{end}}</table>
|
||||
<p class="note">{{t "dash.eff.note"}}</p>
|
||||
{{end}}
|
||||
{{with .IdleGPUs}}
|
||||
<h3>{{t "dash.idle.title"}}</h3>
|
||||
@@ -282,6 +284,10 @@ document.querySelectorAll("form.build-choice select").forEach(function (s) {
|
||||
{{template "tschart" dict "t" .GPUBusy "title" (t "dash.fleet.gpu_busy") "note" (t "dash.fleet.gpu_busy_note")}}
|
||||
{{template "tschart" dict "t" .GPUMemory "title" (t "dash.fleet.gpu_memory") "note" (t "dash.fleet.gpu_memory_note")}}
|
||||
</div>
|
||||
<div class="chart-row">
|
||||
{{template "tschart" dict "t" .GPUPowerLoad "title" (t "dash.fleet.gpu_power_load") "note" (t "dash.fleet.gpu_power_load_note")}}
|
||||
{{if not .GPUSMActive.Empty}}{{template "tschart" dict "t" .GPUSMActive "title" (t "dash.fleet.gpu_sm_active") "note" (t "dash.fleet.gpu_sm_active_note")}}{{end}}
|
||||
</div>
|
||||
{{end}}
|
||||
<div class="chart-row">
|
||||
{{template "tschart" dict "t" .Queue "title" (t "dash.fleet.queue") "note" (t "dash.fleet.queue_note")}}
|
||||
|
||||
Reference in new issue
Block a user