reanimator/bee
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity

Add staged NVIDIA burn ramp-up mode

Browse Source
This commit is contained in:
Mikhail Chusavitin
2026-04-09 15:21:14 +03:00
parent a78fdadd88
commit 9481ca2805
11 changed files with 226 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
audit/internal/webui/tasks.go
View File

@@ -118,6 +118,7 @@ type taskParams struct {
StressMode bool `json:"stress_mode,omitempty"`
GPUIndices []int `json:"gpu_indices,omitempty"`
ExcludeGPUIndices []int `json:"exclude_gpu_indices,omitempty"`
StaggerGPUStart bool `json:"stagger_gpu_start,omitempty"`
SizeMB int `json:"size_mb,omitempty"`
Passes int `json:"passes,omitempty"`
Loader string `json:"loader,omitempty"`
@@ -162,6 +163,13 @@ func resolveBurnPreset(profile string) burnPreset {
}
}
func boolToNvidiaStaggerSeconds(enabled bool, selected []int) int {
if enabled && len(selected) > 1 {
return 180
}
return 0
}
func resolvePlatformStressPreset(profile string) platform.PlatformStressOptions {
acceptanceCycles := []platform.PlatformStressCycle{
{LoadSec: 85, IdleSec: 5},
@@ -592,7 +600,7 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
RunNCCL: t.params.RunNCCL,
ParallelGPUs: t.params.ParallelGPUs,
}, j.append)
case "nvidia-compute":
case "nvidia-compute":
if a == nil {
err = fmt.Errorf("app not configured")
break
@@ -601,7 +609,11 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
if t.params.BurnProfile != "" && dur <= 0 {
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
}
archive, err = a.RunNvidiaOfficialComputePack(ctx, "", dur, t.params.GPUIndices, j.append)
staggerSec := boolToNvidiaStaggerSeconds(t.params.StaggerGPUStart, t.params.GPUIndices)
if staggerSec > 0 {
j.append(fmt.Sprintf("NVIDIA staggered ramp-up enabled: %ds per GPU", staggerSec))
}
archive, err = a.RunNvidiaOfficialComputePack(ctx, "", dur, t.params.GPUIndices, staggerSec, j.append)
case "nvidia-targeted-power":
if a == nil {
err = fmt.Errorf("app not configured")
@@ -651,12 +663,13 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
if t.params.BurnProfile != "" && dur <= 0 {
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
}
archive, err = runNvidiaStressPackCtx(a, ctx, "", platform.NvidiaStressOptions{
DurationSec: dur,
Loader: t.params.Loader,
GPUIndices: t.params.GPUIndices,
ExcludeGPUIndices: t.params.ExcludeGPUIndices,
}, j.append)
archive, err = runNvidiaStressPackCtx(a, ctx, "", platform.NvidiaStressOptions{
DurationSec: dur,
Loader: t.params.Loader,
GPUIndices: t.params.GPUIndices,
ExcludeGPUIndices: t.params.ExcludeGPUIndices,
StaggerSeconds: boolToNvidiaStaggerSeconds(t.params.StaggerGPUStart, t.params.GPUIndices),
}, j.append)
case "memory":
if a == nil {
err = fmt.Errorf("app not configured")