RunNvidiaPowerBench already performs a full internal ramp from 1 to N
GPUs in Phase 2. Spawning N tasks with growing GPU subsets meant task K
repeated all steps 1..K-1 already done by tasks 1..K-1 — O(N²) work
instead of O(N). Replace with a single task using all selected GPUs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Benchmark page now shows two result sections: Performance (scores) and
Power / Thermal Fit (slot table). After any benchmark task completes
the results section auto-refreshes via GET /api/benchmark/results
without a full page reload.
- Power results table shows each GPU slot with nominal TDP, achieved
stable power limit, and P95 observed power. Rows with derated cards
are highlighted amber so under-performing slots stand out at a glance.
Older runs are collapsed in a <details> summary.
- memtester is now wrapped with timeout(1) so a stuck memory controller
cannot cause Validate Memory to hang indefinitely. Wall-clock limit is
~2.5 min per 100 MB per pass plus a 2-minute buffer.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Phase 1 now calibrates each GPU individually (sequentially) so that
PowerRealizationPct reflects real degradation from neighbour thermals and
shared power rails. Previously the baseline came from an all-GPU-together
run, making realization always ≈100% at the final ramp step.
Ramp step 1 reuses single-card calibration results (no extra run); steps
2..N run targeted_power on the growing GPU subset with derating active.
Remove OccupiedSlots/OccupiedSlotsNote fields and occupiedSlots() helper —
they were compensation for the old all-GPU calibration approach.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously each GPU was calibrated sequentially (one card fully done
before the next started), producing the staircase temperature pattern
seen on the graph.
Now all GPUs run together in a single dcgmi diag -r targeted_power
session per attempt. This means:
- All cards are under realistic thermal load at the same time.
- A single DCGM session handles the run — no resource-busy contention
from concurrent dcgmi processes.
- Binary search state (lo/hi) is tracked independently per GPU; each
card converges to its own highest stable power limit.
- Throttle counter polling covers all active GPUs in the shared ticker.
- Resource-busy exponential back-off is shared (one DCGM session).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove telemetry-guided initial candidate; use strict binary search
midpoint at every step. Clean and predictable convergence in O(log N)
attempts within the allowed power range [minLimitW, startingLimitW].
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Power calibration previously stepped down 25 W at a time (linear),
requiring up to 6 attempts to find a stable limit within 150 W range.
New strategy:
- Binary search between minLimitW (lo, assumed stable floor) and the
starting/failed limit (hi, confirmed unstable), converging within a
10 W tolerance in ~4 attempts.
- For thermal throttle: the first-quarter telemetry rows estimate the
GPU's pre-throttle power draw. nextLimit = round5W(onset - 10 W) is
used as the initial candidate instead of the binary midpoint, landing
much closer to the true limit on the first step.
- On success: lo is updated and a higher level is tried (binary search
upward) until hi-lo ≤ tolerance, ensuring the highest stable limit is
found rather than the first stable one.
- Let targeted_power run to natural completion on throttle (no mid-run
SIGKILL) so nv-hostengine releases its diagnostic slot cleanly before
the next attempt.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
During power calibration: if a thermal throttle (sw_thermal/hw_thermal)
causes ≥20% clock drop while server fans are below 98% P95 duty cycle,
record a CoolingWarning on the GPU result and emit an actionable finding
telling the operator to rerun with fans manually fixed at 100%.
During steady-state benchmark: same signal enriches the existing
thermal_limited finding with fan duty cycle and clock drift values.
Covers both the main benchmark (buildBenchmarkFindings) and the power
bench (NvidiaPowerBenchResult.Findings).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When a targeted_power attempt is cancelled (e.g. after sw_thermal
throttle), nv-hostengine holds the diagnostic slot asynchronously.
The next attempt immediately received DCGM_ST_IN_USE (exit 222)
and incorrectly derated the power limit.
Now: exit 222 is detected via isDCGMResourceBusy and triggers an
exponential back-off retry at the same power limit (1s, 2s, 4s, …
up to 256s). Once the back-off delay would exceed 300s the
calibration fails, indicating the slot is persistently held.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
USB Export Drive:
lsblk reports TRAN only for whole disks, not partitions (/dev/sdc1).
Strip trailing partition digits to get parent disk before transport check.
LiveCD in RAM:
When RunInstallToRAM copies squashfs to /dev/shm/bee-live/ but bind-mount
of /run/live/medium fails (CD-ROM boots), /run/live/medium still shows the
CD-ROM fstype. Add fallback: if /dev/shm/bee-live/*.squashfs exists, the
data is in RAM — report status OK.
Dashboard Hardware Summary:
Show server Manufacturer + ProductName as heading and S/N as subline above
the component table, sourced from hw.Board (dmidecode system-type data).
Validate:
Remove Cycles input — always run once. cycles=1 hardcoded in runAllSAT().
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
NCCL all_reduce is always attempted when 2+ GPUs are selected; a failure
leaves InterconnectScore=0 (no bonus, no penalty) and OverallStatus
unaffected. Exposing the checkbox implied NCCL is optional and made a
failed run look like a deliberate skip.
- Remove benchmark-run-nccl checkbox and its change listener from pages.go
- Client sends run_nccl: selected.length > 1 (automatic)
- api.go default runNCCL=true is unchanged
- Selection note now mentions NCCL automatically for multi-GPU runs
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add RampStep/RampTotal/RampRunID to NvidiaBenchmarkOptions, taskParams, and
NvidiaBenchmarkResult so ramp-up steps can be correlated across result.json files
- Add ScalabilityScore field to NvidiaBenchmarkResult (placeholder; computed externally
by comparing ramp-up step results sharing the same ramp_run_id)
- Propagate ramp fields through api.go (generates shared ramp_run_id at spawn time),
tasks.go handler, and benchmark.go result population
- Apply ServerPower penalty to CompositeScore when IPMI reporting_ratio < 0.75:
factor = ratio/0.75, applied per-GPU with a note explaining the reduction
- Add finding when server power delta exceeds GPU-reported sum by >25% (non-GPU draw)
- Report header now shows ramp step N/M and run ID instead of "parallel" when in ramp mode;
shows scalability_score when non-zero
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When --query-gpu CSV fields fail (exit status 2 on some Blackwell +
driver combos), enrichGPUInfoWithMaxClocks now also parses from the
verbose nvidia-smi -q output already collected at benchmark start:
- Default Power Limit → DefaultPowerLimitW
- Current Power Limit → PowerLimitW (fallback)
- Multiprocessor Count → MultiprocessorCount
Fixes PowerSustainScore=0 on systems where all three CSV query
variants fail but nvidia-smi -q succeeds (confirmed on RTX PRO 6000
Blackwell + driver 590.48.01).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Before the per-GPU compute phases, run `dcgmi diag -r targeted_power`
for 45 s while collecting nvidia-smi power metrics in parallel.
The p95 power per GPU is stored as calibrated_peak_power_w and used
as the denominator for PowerSustainScore instead of the hardware default
limit, which bee-gpu-burn cannot reach because it is compute-only.
Fallback chain: calibrated peak → default limit → enforced limit.
If dcgmi is absent or the run fails, calibration is skipped silently.
Adjust composite score weights to match the new honest power reference:
base 0.35, thermal 0.25, stability 0.25, power 0.15, NCCL bonus 0.10.
Power weight reduced (0.20→0.15) because even with a calibrated reference
bee-gpu-burn reaches ~60-75% of TDP by design (no concurrent mem stress).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- BenchmarkHostConfig captures CPU model, sockets, cores, threads, and
total RAM from /proc/cpuinfo and /proc/meminfo at benchmark start.
- BenchmarkCPULoad samples host CPU utilisation every 10 s throughout
the GPU steady-state phase (sequential and parallel paths).
- Summarises avg/max/p95 and classifies status as ok / high / unstable.
- Adds a finding when CPU load is elevated (avg >20% or max >40%) or
erratic (stddev >12%), with a plain-English description in the report.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds a new checkbox (enabled by default) in the benchmark section.
In ramp-up mode N tasks are spawned simultaneously: 1 GPU, then 2,
then 3, up to all selected GPUs — each step runs its GPUs in parallel.
NCCL runs only on the final step.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
clocks.max.graphics / clocks.max.memory CSV fields return exit status 2 on
RTX PRO 6000 Blackwell (driver 98.x), causing the entire gpu inventory query
to fail and clock lock to be skipped → normalization: partial.
Fix:
- Add minimal fallback query (index,uuid,name,pci.bus_id,vbios_version,
power.limit) that succeeds even without clock fields
- Add enrichGPUInfoWithMaxClocks: parses "Max Clocks" section of
nvidia-smi -q verbose output to fill MaxGraphicsClockMHz /
MaxMemoryClockMHz when CSV fields fail
- Move nvidia-smi -q execution before queryBenchmarkGPUInfo so its output
is available for clock enrichment immediately after
- Tests: cover enrichment and skip-if-populated cases
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- collector/pcie: add applyPCIeLinkSpeedWarning that sets status=Warning
and ErrorDescription when current link speed is below maximum negotiated
speed (e.g. Gen1 running on a Gen5 slot)
- collector/pcie: add pcieLinkSpeedRank helper for Gen string comparison
- collector/pcie_filter_test: cover degraded and healthy link speed cases
- platform/techdump: collect lspci -vvv → lspci-vvv.txt for LnkCap/LnkSta
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- schema: add ToRAMStatus and USBExportPath fields to RuntimeHealth
- platform/runtime.go: collectToRAMHealth (ok/warning/failed based on
IsLiveMediaInRAM + toramActive) and collectUSBExportHealth (scans
/proc/mounts + lsblk for writable USB-backed filesystems)
- pages.go: add USB Export Drive and LiveCD in RAM rows to the health table
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- nvidia.go: add Name field to nvidiaGPUInfo, include model name in
nvidia-smi query, set dev.Model in enrichPCIeWithNVIDIAData
- pages.go: fix duplicate GPU count in validate card summary (4 GPU: 4 x …
→ 4 x … GPU); fix PSU UNKNOWN fallback from hw.PowerSupplies; treat
activating/deactivating/reloading service states as OK in Runtime Health
- support_bundle.go: use "150405" time format (no colons) for exFAT compat
- sat.go / benchmark.go / platform_stress.go / sat_fan_stress.go: remove
.tar.gz archive creation from export dirs — export packs everything itself
- charts_svg.go: add min-max downsampling (1400 pt cap) for SVG chart perf
- benchmark_report.go / sat.go: normalize GPU fallback to "Unknown GPU"
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two issues:
1. BMC/management VGA chips (e.g. Huawei iBMC Hi171x, ASPEED) were included
in GPU inventory because shouldIncludePCIeDevice only checked the PCI class,
not the device name. Added a name-based filter for known BMC/management
patterns when the class is VGA/display/3d.
2. New NVIDIA GPUs (e.g. RTX PRO 6000 Blackwell, device ID 2bb5) showed as
"Device 2bb5" because lspci's database lags behind. Added "name" to the
nvidia-smi query and use it to override dev.Model during enrichment.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
hpl was listed in baseTargets and stressOnlyTargets but /api/sat/hpl/run
was never registered, causing a 405 Method Not Allowed (not valid JSON)
error when Validate one by one was triggered in stress mode.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Runtime Health now shows only LiveCD system status (services, tools,
drivers, network, CUDA/ROCm) — hardware component rows removed
- Hardware Summary now shows server components with readable descriptions
(model, count×size) and component-status.json health badges
- Add Network Adapters row to Hardware Summary
- SFP module static info (vendor, PN, SN, connector, type, wavelength)
now collected via ethtool -m regardless of carrier state
- PSU statuses from IPMI audit written to component-status.json so PSU
badge shows actual status after first audit instead of UNKNOWN
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Bootloader: GRUB fallback text colors → yellow/brown (amber tone)
- CLI charts: all GPU metric series use single amber color (xterm-256 #214)
- Wallpaper: logo width scaled to 400 px dynamically, shadow scales with font size
- Support bundle: renamed to YYYY-MM-DD (BEE-SP vX.X) SRV_MODEL SRV_SN ToD.tar.gz
using dmidecode for server model (spaces→underscores) and serial number
- Remove display resolution feature (UI card, API routes, handlers, tests)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
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>
- Replace diag level 1-4 dropdown with Validate/Stress radio buttons
- Validate: dcgmi L2, 60s CPU, 256MB/1p memtester, SMART short
- Stress: dcgmi L3 + targeted_stress in Run All, 30min CPU, 1GB/3p memtester, SMART long/NVMe extended
- Parallel GPU mode: spawn single task for all GPUs instead of splitting per model
- Benchmark table: per-GPU columns for sequential runs, server-wide column for parallel
- Benchmark report converted to Markdown with server model, GPU model, version in header; only steady-state charts
- Fix IPMI power parsing in benchmark (was looking for 'Current Power', correct field is 'Instantaneous power reading')
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds a "Multi-GPU tests — use all GPUs" checkbox to the NVIDIA GPU
selector (checked by default). When enabled, PSU Pulse, NCCL, and
NVBandwidth tests run on every GPU in the system regardless of the
per-GPU selection above — which is required for correct PSU stress
testing (synchronous pulses across all GPUs create worst-case
transients). When unchecked, only the manually selected GPUs are used.
The same logic applies both to Run All (expandSATTarget) and to the
individual Run button on each multi-GPU test card.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
pulse_test is a PSU/power-delivery test, not a per-GPU compute test.
Its purpose is to synchronously pulse all GPUs between idle and full
load to create worst-case transient spikes on the power supply.
Running it one GPU at a time would produce a fraction of the PSU load
and miss any PSU-level failures.
- Move nvidia-pulse from nvidiaPerGPUTargets to nvidiaAllGPUTargets
(same dispatch path as NCCL and NVBandwidth)
- Change card onclick to runNvidiaFabricValidate (all selected GPUs at once)
- Update card title to "NVIDIA PSU Pulse Test" and description to
explain why synchronous multi-GPU execution is required
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
nvidia-interconnect (NCCL all_reduce_perf) and nvidia-bandwidth
(NVBandwidth) verify fabric connectivity and bandwidth — they are
not sustained burn loads. Move both from the Burn section to the
Validate section under the stress-mode toggle, alongside the other
DCGM diagnostic tests moved in the previous commit.
- Add sat-card-nvidia-interconnect and sat-card-nvidia-bandwidth
validate cards (stress-only, all selected GPUs at once)
- Add runNvidiaFabricValidate() for all-GPU-at-once dispatch
- Add nvidiaAllGPUTargets handling in expandSATTarget/runAllSAT
- Remove Interconnect / Bandwidth card from Burn section
- Remove nvidia-interconnect and nvidia-bandwidth from runAllBurnTasks
and the gpu/tools availability map
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- bee-gpu-stress.c: remove per-wave cuCtxSynchronize barrier in both
cuBLASLt and PTX hot loops; sync at most once/sec so the GPU queue
stays continuously full — eliminates the CPU↔GPU ping-pong that
prevented reaching full TDP
- sat_fan_stress.go: default SizeMB 0 (auto = 95% VRAM) instead of
hardcoded 64 MB; tiny matrices caused <0.1 ms kernels where CPU
re-queue overhead dominated
- pages.go: move nvidia-targeted-power and nvidia-pulse from Burn →
Validate stress section alongside nvidia-targeted-stress; these are
DCGM pass/fail diagnostics, not sustained burn loads; remove the
Power Delivery / Power Budget card from Burn entirely
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
