Add HPL (LINPACK) benchmark as validate/stress task

HPL 2.3 from netlib compiled against OpenBLAS with a minimal
single-process MPI stub — no MPI package required in the ISO.
Matrix size is auto-sized to 80% of total RAM at runtime.

Build:
- VERSIONS: HPL_VERSION=2.3, HPL_SHA256=32c5c17d…
- build-hpl.sh: downloads HPL + OpenBLAS from Debian 12 repo,
  compiles xhpl with a self-contained mpi_stub.c
- build.sh: step 80-hpl, injects xhpl + libopenblas into overlay

Runtime:
- bee-hpl: generates HPL.dat (N auto from /proc/meminfo, NB=256,
  P=1 Q=1), runs xhpl, prints standard WR... Gflops output
- platform/hpl.go: RunHPL(), parses WR line → GFlops + PASSED/FAILED
- tasks.go: target "hpl"
- pages.go: LINPACK (HPL) card in validate/stress grid (stress-only)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-04-08 07:08:18 +03:00
parent b2f8626fee
commit 16e7ae00e7
10 changed files with 541 additions and 6 deletions

View File

@@ -139,6 +139,7 @@ type satRunner interface {
RunFanStressTest(ctx context.Context, baseDir string, opts platform.FanStressOptions) (string, error)
RunPlatformStress(ctx context.Context, baseDir string, opts platform.PlatformStressOptions, logFunc func(string)) (string, error)
RunNCCLTests(ctx context.Context, baseDir string, logFunc func(string)) (string, error)
RunHPL(ctx context.Context, baseDir string, opts platform.HPLOptions, logFunc func(string)) (string, *platform.HPLResult, error)
}
type runtimeChecker interface {
@@ -737,6 +738,13 @@ func (a *App) RunNCCLTestsResult(ctx context.Context) (ActionResult, error) {
return ActionResult{Title: "NCCL bandwidth test", Body: body}, err
}
func (a *App) RunHPL(ctx context.Context, baseDir string, opts platform.HPLOptions, logFunc func(string)) (string, *platform.HPLResult, error) {
if a == nil {
return "", nil, fmt.Errorf("app not configured")
}
return a.sat.RunHPL(ctx, baseDir, opts, logFunc)
}
func (a *App) RunFanStressTestResult(ctx context.Context, opts platform.FanStressOptions) (ActionResult, error) {
path, err := a.RunFanStressTest(ctx, "", opts)
body := formatFanStressResult(path)

View File

@@ -282,6 +282,9 @@ func (f fakeSAT) RunPlatformStress(_ context.Context, _ string, _ platform.Platf
func (f fakeSAT) RunNCCLTests(_ context.Context, _ string, _ func(string)) (string, error) {
return "", nil
}
func (f fakeSAT) RunHPL(_ context.Context, _ string, _ platform.HPLOptions, _ func(string)) (string, *platform.HPLResult, error) {
return "", nil, nil
}
func TestNetworkStatusFormatsInterfacesAndRoute(t *testing.T) {
t.Parallel()

View File

@@ -0,0 +1,142 @@
package platform
import (
"context"
"fmt"
"os"
"os/exec"
"path/filepath"
"strconv"
"strings"
"time"
)
// HPLOptions configures the HPL (LINPACK) benchmark run.
type HPLOptions struct {
MemFraction float64 // fraction of RAM to use (default 0.80)
NB int // block size (default 256)
}
// HPLResult holds the parsed result of an HPL run.
type HPLResult struct {
N int // matrix dimension
NB int // block size
P int // process grid rows
Q int // process grid cols
TimeSec float64 // wall time in seconds
GFlops float64 // achieved performance
Residual float64 // backward error residual (from HPL verification line)
Status string // "PASSED" or "FAILED"
RawOutput string // full xhpl output
}
func applyHPLDefaults(opts *HPLOptions) {
if opts.MemFraction <= 0 || opts.MemFraction > 1 {
opts.MemFraction = 0.80
}
if opts.NB <= 0 {
opts.NB = 256
}
}
// RunHPL runs bee-hpl and returns parsed results plus a tar.gz artifact path.
func (s *System) RunHPL(ctx context.Context, baseDir string, opts HPLOptions, logFunc func(string)) (string, *HPLResult, error) {
applyHPLDefaults(&opts)
if baseDir == "" {
baseDir = "/var/log/bee-sat"
}
ts := time.Now().UTC().Format("20060102-150405")
runDir := filepath.Join(baseDir, "hpl-"+ts)
if err := os.MkdirAll(runDir, 0755); err != nil {
return "", nil, fmt.Errorf("mkdir %s: %w", runDir, err)
}
logPath := filepath.Join(runDir, "hpl.log")
cmd := []string{
"bee-hpl",
"--mem-fraction", strconv.FormatFloat(opts.MemFraction, 'f', 2, 64),
"--nb", strconv.Itoa(opts.NB),
}
if logFunc != nil {
logFunc(fmt.Sprintf("HPL: N will be auto-sized to %.0f%% of RAM, NB=%d", opts.MemFraction*100, opts.NB))
}
out, err := runSATCommandCtx(ctx, "", "hpl", cmd, nil, logFunc)
_ = os.WriteFile(logPath, out, 0644)
result := parseHPLOutput(string(out))
result.RawOutput = string(out)
if err != nil && err != context.Canceled {
return "", result, fmt.Errorf("bee-hpl failed: %w", err)
}
if err == nil && result.GFlops <= 0 {
return "", result, fmt.Errorf("HPL completed but no Gflops result found in output")
}
// Write summary
summary := fmt.Sprintf("N=%d NB=%d time=%.2fs gflops=%.3f status=%s\n",
result.N, result.NB, result.TimeSec, result.GFlops, result.Status)
_ = os.WriteFile(filepath.Join(runDir, "summary.txt"), []byte(summary), 0644)
if logFunc != nil {
logFunc(fmt.Sprintf("HPL result: N=%d NB=%d %.2fs %.3f Gflops %s",
result.N, result.NB, result.TimeSec, result.GFlops, result.Status))
}
ts2 := time.Now().UTC().Format("20060102-150405")
archive := filepath.Join(baseDir, "hpl-"+ts2+".tar.gz")
if archErr := createTarGz(archive, runDir); archErr != nil {
return runDir, result, err
}
return archive, result, err
}
// parseHPLOutput extracts N, NB, time, and Gflops from standard HPL output.
//
// HPL prints a result line of the form:
//
// WR00L2L2 45312 256 1 1 1234.56 5.678e+01
// T/V N NB P Q Time Gflops
func parseHPLOutput(output string) *HPLResult {
result := &HPLResult{Status: "FAILED"}
for _, line := range strings.Split(output, "\n") {
line = strings.TrimSpace(line)
// Result line starts with WR
if strings.HasPrefix(line, "WR") {
fields := strings.Fields(line)
// WR00L2L2 N NB P Q Time Gflops
if len(fields) >= 7 {
result.N, _ = strconv.Atoi(fields[1])
result.NB, _ = strconv.Atoi(fields[2])
result.P, _ = strconv.Atoi(fields[3])
result.Q, _ = strconv.Atoi(fields[4])
result.TimeSec, _ = strconv.ParseFloat(fields[5], 64)
result.GFlops, _ = strconv.ParseFloat(fields[6], 64)
}
}
// Verification line: "||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= ... PASSED"
if strings.Contains(line, "PASSED") {
result.Status = "PASSED"
fields := strings.Fields(line)
for i, f := range fields {
if f == "PASSED" && i > 0 {
result.Residual, _ = strconv.ParseFloat(fields[i-1], 64)
}
}
}
}
return result
}
// hplAvailable returns true if bee-hpl and xhpl are present and executable.
func hplAvailable() bool {
if _, err := exec.LookPath("bee-hpl"); err != nil {
return false
}
_, err := os.Stat("/usr/local/lib/bee/xhpl")
return err == nil
}

View File

@@ -1143,6 +1143,16 @@ func renderValidate(opts HandlerOptions) string {
`</div>` +
`</div>
<div class="grid3" style="margin-top:16px">
` + `<div id="sat-card-hpl">` +
renderSATCard("hpl", "LINPACK (HPL)", "runSAT('hpl')", "", renderValidateCardBody(
``,
`Standard High Performance LINPACK benchmark. Measures sustained FP64 GFLOPS and memory bandwidth of the CPU subsystem. Uses 80% of available RAM. Pass/fail based on HPL residual check.`,
`<code>xhpl</code> (HPL 2.3, OpenBLAS)`,
`Skipped in Validate mode. Runs in Stress mode only. Runtime scales with RAM — expect 530 min.<p id="sat-hpl-mode-hint" style="color:var(--warn-fg);font-size:12px;margin:8px 0 0">Only runs in Stress mode. Switch mode above to enable in Run All.</p>`,
)) +
`</div>` +
`</div>
<div class="grid3" style="margin-top:16px">
` + renderSATCard("amd", "AMD GPU", "runAMDValidateSet()", "", renderValidateCardBody(
inv.AMD,
`Runs the selected AMD checks only. GPU Validate collects inventory; MEM Integrity uses the RVS MEM module; MEM Bandwidth uses rocm-bandwidth-test and the RVS BABEL module.`,
@@ -1178,6 +1188,7 @@ function satModeChanged() {
{card: 'sat-card-nvidia-pulse', hint: 'sat-pt-mode-hint'},
{card: 'sat-card-nvidia-interconnect', hint: 'sat-ni-mode-hint'},
{card: 'sat-card-nvidia-bandwidth', hint: 'sat-nb-mode-hint'},
{card: 'sat-card-hpl', hint: 'sat-hpl-mode-hint'},
].forEach(function(item) {
const card = document.getElementById(item.card);
if (card) {
@@ -1188,7 +1199,7 @@ function satModeChanged() {
});
}
function satLabels() {
return {nvidia:'Validate GPU', 'nvidia-targeted-stress':'NVIDIA Targeted Stress (dcgmi diag targeted_stress)', 'nvidia-targeted-power':'NVIDIA Targeted Power (dcgmi diag targeted_power)', 'nvidia-pulse':'NVIDIA Pulse Test (dcgmi diag pulse_test)', 'nvidia-interconnect':'NVIDIA Interconnect (NCCL all_reduce_perf)', 'nvidia-bandwidth':'NVIDIA Bandwidth (NVBandwidth)', memory:'Validate Memory', storage:'Validate Storage', cpu:'Validate CPU', amd:'Validate AMD GPU', 'amd-mem':'AMD GPU MEM Integrity', 'amd-bandwidth':'AMD GPU MEM Bandwidth'};
return {nvidia:'Validate GPU', 'nvidia-targeted-stress':'NVIDIA Targeted Stress (dcgmi diag targeted_stress)', 'nvidia-targeted-power':'NVIDIA Targeted Power (dcgmi diag targeted_power)', 'nvidia-pulse':'NVIDIA PSU Pulse Test (dcgmi diag pulse_test)', 'nvidia-interconnect':'NVIDIA Interconnect (NCCL all_reduce_perf)', 'nvidia-bandwidth':'NVIDIA Bandwidth (NVBandwidth)', hpl:'LINPACK (HPL)', memory:'Validate Memory', storage:'Validate Storage', cpu:'Validate CPU', amd:'Validate AMD GPU', 'amd-mem':'AMD GPU MEM Integrity', 'amd-bandwidth':'AMD GPU MEM Bandwidth'};
}
let satNvidiaGPUsPromise = null;
function loadSatNvidiaGPUs() {
@@ -1437,8 +1448,8 @@ function runAllSAT() {
const cycles = Math.max(1, parseInt(document.getElementById('sat-cycles').value)||1);
const status = document.getElementById('sat-all-status');
status.textContent = 'Enqueuing...';
const stressOnlyTargets = ['nvidia-targeted-stress', 'nvidia-targeted-power', 'nvidia-pulse', 'nvidia-interconnect', 'nvidia-bandwidth'];
const baseTargets = ['nvidia','nvidia-targeted-stress','nvidia-targeted-power','nvidia-pulse','nvidia-interconnect','nvidia-bandwidth','memory','storage','cpu'].concat(selectedAMDValidateTargets());
const stressOnlyTargets = ['nvidia-targeted-stress', 'nvidia-targeted-power', 'nvidia-pulse', 'nvidia-interconnect', 'nvidia-bandwidth', 'hpl'];
const baseTargets = ['nvidia','nvidia-targeted-stress','nvidia-targeted-power','nvidia-pulse','nvidia-interconnect','nvidia-bandwidth','hpl','memory','storage','cpu'].concat(selectedAMDValidateTargets());
const activeTargets = baseTargets.filter(target => {
if (stressOnlyTargets.indexOf(target) >= 0 && !satStressMode()) return false;
const btn = document.getElementById('sat-btn-' + target);
@@ -2082,7 +2093,7 @@ func benchmarkHistoryParallelLabel(serverModel, gpuName string, count int) strin
func renderBurn() string {
return `<div class="alert alert-warn" style="margin-bottom:16px"><strong>&#9888; Warning:</strong> Stress tests on this page run hardware at high load. Repeated or prolonged use may reduce hardware lifespan. Use only when necessary.</div>
<div class="alert alert-info" style="margin-bottom:16px"><strong>Scope:</strong> DCGM diagnostics and ` + "targeted_stress" + ` remain in <a href="/validate">Validate</a>. Burn exposes official NVIDIA load recipes by test goal plus separate custom stress tools.</div>
<div class="alert alert-info" style="margin-bottom:16px"><strong>Scope:</strong> DCGM diagnostics (` + "targeted_stress, targeted_power, pulse_test" + `), NCCL, NVBandwidth, and LINPACK remain in <a href="/validate">Validate → Stress mode</a>. Burn exposes sustained GPU compute load recipes.</div>
<p style="color:var(--muted);font-size:13px;margin-bottom:16px">Tasks continue in the background — view progress in <a href="/tasks">Tasks</a>.</p>
<div class="card" style="margin-bottom:16px">

View File

@@ -741,8 +741,8 @@ func TestBurnPageRendersGoalBasedNVIDIACards(t *testing.T) {
for _, needle := range []string{
`NVIDIA Max Compute Load`,
`dcgmproftester`,
`targeted_stress remain in <a href="/validate">Validate</a>`,
`NVIDIA Interconnect Test (NCCL all_reduce_perf)`,
`NCCL`,
`Validate → Stress mode`,
`id="burn-gpu-list"`,
} {
if !strings.Contains(body, needle) {

View File

@@ -39,6 +39,7 @@ var taskNames = map[string]string{
"nvidia-interconnect": "NVIDIA Interconnect Test (NCCL all_reduce_perf)",
"nvidia-bandwidth": "NVIDIA Bandwidth Test (NVBandwidth)",
"nvidia-stress": "NVIDIA GPU Stress",
"hpl": "LINPACK (HPL)",
"memory": "Memory SAT",
"storage": "Storage SAT",
"cpu": "CPU SAT",
@@ -739,6 +740,19 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
}
archive, err = runSATStressPackCtx(a, ctx, "", dur, j.append)
case "hpl":
if a == nil {
err = fmt.Errorf("app not configured")
break
}
opts := platform.HPLOptions{
MemFraction: 0.80,
NB: 256,
}
archive, err = func() (string, error) {
path, _, runErr := a.RunHPL(ctx, "", opts, j.append)
return path, runErr
}()
case "platform-stress":
if a == nil {
err = fmt.Errorf("app not configured")