# RTX 5090 FE — Thermal & Power Baseline
25 August 2026
## Test configuration
| | |
|---|---|
| GPU | RTX 5090 FE |
| VBIOS | 98.02.2e.00.03 |
| Driver | 610.88 |
| Power cap | 575 W (600 W max) |
| Display | AW3225QF 4K @ 239 Hz |
| Tuning | core undervolt, custom V/F curve (memory + power limit stock) |
## Headline results
| Metric | Value |
|---|---|
| OCCT errors | 0 (all tests) |
| Steel Nomad — best loop | 12,955 |
| Steel Nomad — worst loop | 12,656 |
| 20-loop stability | 97.7% (threshold 97%) |
| Pin spread @ 568 W | 4.9% |
| Pin spread @ 490 W | 4.8% |
## Peak values vs limits (Steel Nomad)
| Metric | Peak | Limit | % of limit |
|---|---|---|---|
| GPU power | 567.8 W | 575 W | 99% |
| Per-pin current | 8.69 A | 9.5 A | 91% |
| Memory junction | 86 °C | 105 °C | 82% |
| Hot spot | 85 °C | 105 °C | 81% |
| GPU core | 74.9 °C | 83 °C | 90% |
## Load generator comparison
| Metric | OCCT 3D Adaptive | Steel Nomad | Delta |
|---|---|---|---|
| GPU power (mean) | 484.4 W | 556.3 W | +71.9 W |
| GPU power (peak) | 490.6 W | 567.8 W | +77.2 W |
| Power cap asserted | never | 91.7% of run | — |
| Hot spot (peak) | 79.8 °C | 85.0 °C | +5.2 °C |
| Memory junction (peak) | 80 °C | 86 °C | +6 °C |
| Peak pin current | 7.55 A | 8.69 A | +1.14 A |
| SM clock (peak) | 2745 MHz | 2865 MHz | +120 MHz |
## Phase results (steady-state windows)
| Phase | Core | Hot spot | Mem jct | GPU power | Fan |
|---|---|---|---|---|---|
| Idle baseline | 32.1 °C | 34.9 °C | 44.0 °C | 64.7 W | 1607 rpm |
| 3D Adaptive | 68.7 °C | 77.8 °C | 78.1 °C | 484.4 W | 2297 rpm |
| Cooldown | 39.8 °C | 43.1 °C | 54.1 °C | 64.3 W | 1637 rpm |
| VRAM | 37.7 °C | 40.1 °C | 52.0 °C | 109.2 W | 1607 rpm |
| Power (CPU+GPU) | 67.7 °C | 76.5 °C | 75.1 °C | 451.9 W | 2145 rpm |
| Steel Nomad stress | 72.4 °C | 81.4 °C | 83.3 °C | 556.3 W | 2324 rpm |
| VRAM hot (combined) | 62.6 °C | 69.9 °C | 70.5 °C | 377.2 W | 1945 rpm |
## 12V-2x6 per-pin current — WireView Pro 2
Averaged over Steel Nomad active window, card at power cap.
| Pin | Current | |
|---|---|---|
| 1 | 7.89 A | |
| 2 | 7.94 A | |
| 3 | 7.95 A | |
| 4 | 7.75 A | lowest, every phase |
| 5 | 7.94 A | |
| 6 | 8.14 A | highest, every phase |
| | |
|---|---|
| Mean | 7.94 A |
| Range | 7.33 – 8.44 A |
| Per-pin rating | 9.5 A (84% of rating) |
| Spec @ 600 W | 8.33 A per pin |
| Total draw | 48.5 A = 8.09 A/pin avg |
## Rail voltage
| | |
|---|---|
| Idle | 11.94 V |
| @ ~40 A | 11.71 V |
| Path resistance | ~6 mΩ |
## Additional measured
| Metric | Value |
|---|---|
| Thermal plateau reached | 137 s (within 1 °C of peak) |
| Core-to-hotspot delta | ~9 °C |
| Cooldown decay | 53 → 35 °C in 180 s |
| Fan pair tracking | within 3 rpm, all phases |
| Max fan speed observed | 2386 rpm |
| VRAM combined test | 23.4 GB under test, 0 errors |
| VRAM combined thermal | >70 °C for 750 of 899 s, peak 78 °C |
| VRAM log note | 70 s idle gap (335–405 s), test stopped/restarted; active-load power avg 117.1 W excluding gap |
## Telemetry
- nvidia-smi @ 1 Hz, 7 phases, 4,141 samples
- HWiNFO64 @ 1 Hz, 4,781 samples, 174 sensors
- Merged on wall-clock timestamp
- Hot spot, memory junction, and per-pin values from HWiNFO (driver 610.88 does not expose them via NVML)
- Stability parsed from result.3dmark-result