Compact small-form-factor gaming PC built following the Eiga Works FormD T1 5090 Air guide.

| Component | Part |
|---|---|
| CPU | AMD Ryzen 7 9800X3D |
| CPU Cooler | Thermalright AXP90-X47 Full Copper |
| Cooler Fan | Noctua NF-A9x14 HS-PWM chromax.black.swap |
| Motherboard | ASUS ROG STRIX B850-I GAMING WIFI (Mini ITX, AM5) |
| RAM | Patriot Venom 64 GB (2x32 GB) DDR5-6000 CL30 |
| Storage | Samsung 990 Pro 2 TB (M.2 PCIe 4.0 x4 NVMe) |
| GPU | NVIDIA GeForce RTX 5080 Founders Edition 16 GB |
| Case | FormD T1 V2.1 |
| Case Add-ons | T1 GPU Travel Kit, T1 PCIe SS200 Gen5 Riser, T1 Essentials Pack |
| PSU | Corsair SF1000 (2024) 80+ Platinum, SFX |
| Exhaust Fans | 2x Phanteks T30-120 |
| Contact Frame | Thermal Grizzly AM5 Contact Frame |
| Backplate | Thermal Grizzly AM5 Short Backplate |
| Thermal Paste | Thermal Grizzly Duronaut |
| Cables | Custom Black Teflon cables from CablesterCustom (24P 170mm, CPU 8P 290mm, GPU 12V2X6 300mm) |
| Screws | 12x M3 flat head 4mm, 16x M3 countersunk 5mm, 11x M3 countersunk 8mm |
This build follows the Eiga guide which targets the RTX 5090 FE, but uses an RTX 5080 FE instead. The assembly process is identical since both cards share the same Founders Edition dual-slot cooler design. Same screws as the guide.
Using MSI Afterburner (Beta with RTX 50 series support).
Method: Set Core Clock offset to -250 MHz, open Curve Editor (Ctrl+F), drag the 950mV point to target frequency, flatten everything to the right with Shift + click-drag in empty space then Shift+Enter twice.
| Stock | Undervolted (950mV/2900 + 1000mem) | |
|---|---|---|
| Core Clock | 2742 MHz | 2830 MHz |
| Memory | 1875 MHz | 2000 MHz |
| Voltage | 1.004V | 0.934V |
| Power | 251W | 251W |
| GPU Temp | 62.3°C | 61.4°C |
| VRAM Temp | 72.3°C | 72.4°C |
+88 MHz over stock at the same power draw, 1°C cooler. For reference, the RTX 5080 FE in open-air reviews (GamersNexus) runs 65-67°C -- our T1 build undervolted runs 61-63°C.
Battlefield 6 (128 players): CPU averaging 97°C at 108W, hitting the 95°C thermal throttle. EXPO auto-voltages dangerously high: VSOC 1.272V, VDDIO 1.376V.
Use the AMD Overclocking path, NOT Ai Tweaker for PBO/CO. The same PBO options also exist under Ai Tweaker -> Precision Boost Overdrive; keep those on Auto so limits are only set in one place.
Advanced -> AMD Overclocking -> [Accept Disclaimer] -> Precision Boost Overdrive
| Setting | Stock | Optimized |
|---|---|---|
| PBO | Auto | Advanced |
| PBO Limits | Auto | Manual |
| PPT Limit | 162W | 90000 (90W) |
| TDC Limit | 120A | 75000 (75A) |
| EDC Limit | 180A | 110000 (110A) |
| PBO Scalar | Auto | Manual, 1X |
| Platform Thermal Throttle Limit | 95°C | 85°C |
| Curve Optimizer | Disabled | All Cores, Negative, 30 |
Ai Tweaker
| Setting | Stock | Optimized |
|---|---|---|
| CPU SOC Voltage (VDDSOC Voltage Override) | Auto (1.272V) | Manual, 1.200V |
| CPU VDDIO / MC Voltage | Auto (1.376V) | 1.300V |
| DIGI+ Power Control -> CPU Load-line Calibration | Auto | Level 1 |
Advanced -> AMD CBS -> CPU Common Options
| Setting | Stock | Optimized |
|---|---|---|
| Global C-state Control | Auto | Enabled |
Notes:
CPU PPT Limit [%] to confirm which limit is actually active.Tool -> ASUS User Profile); the BIOS rejects NTFS.Tool: CoreCycler (default config, Prime95, ~6 min per core). Loads one core at a time at max boost, which is where an overly aggressive CO fails. A core that errors is skipped early, so a full 6 min on every core means no errors. WHEA errors checked in Event Viewer -> Windows Logs -> System, filtered on source WHEA-Logger (Event ID 19 = corrected, 18 = fatal; core = Processor APIC ID / 2).
| CO -25 | CO -30 | |
|---|---|---|
| Iterations | 2 | 1.6 |
| CoreCycler errors | 0 | 0 |
| WHEA errors | 0 | 0 |
| Tested core clock | 5225 MHz | 5225 MHz |
| Avg VID | 1.20V | 1.17V |
| Package Power avg | 68.9W | 66.1W |
| CPU Temp avg / max | 81.3°C / 86.0°C | 80.4°C / 85.0°C |
| Thermal throttling (HTC) | No | No |
Both runs were logged with Windows CPU idle disabled (all cores held in C0), which inflated package power and temps by ~30W during a single-core test. See below.
Status: -30 passed short runs. Still to do: one overnight CoreCycler run (6-8h), a WHEA check the next morning, then a week of normal use (watch for crashes at idle, on wake from sleep, on loading screens). Fallback if anything fails: -25 (validated), or Per Core with only the failing core at 25.
The active plan was optimizerDuck PowerPlan (installed by an optimization tool), which sets Processor idle disable on AC. Every core stayed in C0 at full clock 100% of the time (0% C1/C6 residency), even when idle.
Check current state (admin terminal):
powercfg /qh SCHEME_CURRENT SUB_PROCESSOR IDLEDISABLE
Current AC Power Setting Index: 0x00000001 = idle disabled, 0x00000000 = idle enabled.
Toggle (applies instantly, no reboot):
:: Enable idle (kept)
powercfg /setacvalueindex SCHEME_CURRENT SUB_PROCESSOR IDLEDISABLE 0
powercfg /setactive SCHEME_CURRENT
:: Disable idle
powercfg /setacvalueindex SCHEME_CURRENT SUB_PROCESSOR IDLEDISABLE 1
powercfg /setactive SCHEME_CURRENT
To show the setting in the classic Advanced power settings UI: powercfg -attributes SUB_PROCESSOR IDLEDISABLE -ATTRIB_HIDE
Same scene, game restarted between runs, HWiNFO + PresentMon logging. Menus, loading screens and the first minute excluded (~5 min and ~7 min of gameplay). One run each; the idle-enabled run was done second in an already warm case (motherboard 56°C vs 52°C).
| Idle disabled (C0) | Idle enabled | Change | |
|---|---|---|---|
| FPS avg | 195 | 191 | -2% (run-to-run noise) |
| 1% low | 139.8 | 140.4 | same |
| 0.1% low | 135.3 | 136.5 | same |
| Frame time 99th pct | 6.99 ms | 6.92 ms | same |
| Worst spike (0.1% high max) | 56.6 ms | 14.1 ms | better with idle |
| CPU Package Power | 87.1W | 84.4W | -2.7W |
| CPU Temp avg | 85.9°C | 85.4°C | -0.5°C |
| Time at thermal limit | 92% | 80% | -12 pts |
Decision: idle enabled. No smoothness benefit from disabling idle on this system. BF6 keeps all 8 cores busy anyway (C1 ~25%, no C6), so the in-game saving is small; the big saving (~30W) is at the desktop and in light loads.
Optimization tools can re-apply their plan on update or reinstall. If temps rise for no reason, re-check with powercfg /qh.
Adapted from Eiga's FanControl config (originally for 5090 FE), swapping GPU identifiers from GB202-A to GB203-A.
CPU Fan (Noctua NF-A9x14, driven by CPU Tctl/Tdie):
| Temp | Fan % |
|---|---|
| 50-60°C | 30% |
| 60-70°C | 35% |
| 70-80°C | 40% |
| 80-92°C | 45% |
| 92°C+ | 50% |
Exhaust Fans (2x Phanteks T30, driven by a Mix sensor -- max of GPU temp and CPU temp with -15°C offset):
| Temp | Fan % |
|---|---|
| 50-60°C | 40% |
| 60-65°C | 50% |
| 65-75°C | 60% |
| 75°C+ | 70% |
GPU fans left on Auto (firmware control). FanControl and MSI Afterburner both set to start with Windows minimized.
Observation: this curve is silence-first. The CPU fan tops out around 1450 RPM, and in BF6 the CPU sits on the 85°C limit 80-92% of the time, averaging ~4980 MHz instead of 5225 MHz. Performance impact is small (GPU ~94% loaded), but if lower CPU temps are wanted, raise only the top steps, e.g.:
| Temp | Eiga | Candidate |
|---|---|---|
| 70-80°C | 40% | 45% |
| 80-85°C | 45% | 60% |
| 85°C+ | 50% | 75% |
Adding a few seconds of response time and 3-5°C hysteresis also stops the fan chasing X3D temperature spikes.
Optimized column: CO -20, 90W PPT cap.
| Stock | Optimized | Change | |
|---|---|---|---|
| Score | ~20,000 | 19,650 | -1.7% |
| CPU Temp avg | 95.0°C | 81.0°C | -14°C |
| CPU Temp max | 95.6°C | 83.2°C | -12.4°C |
| Package Power avg | 110.3W | 89.4W | -20.9W |
| Package Power max | 138.6W | 90.0W | -48.6W |
| SoC Power | 16.2W | 12.7W | -3.5W |
| VSOC | 1.272V | 1.20V | -72mV |
| VDDIO | 1.376V | 1.30V | -76mV |
| Bottleneck | Thermal (95°C wall) | PPT (99.4%) | -- |
Optimized column: CO -20, no PPT cap (PPT % is relative to the stock ~162W limit), idle disabled.
| Stock | Optimized | Change | |
|---|---|---|---|
| CPU Temp avg | 97°C | 82°C | -15°C |
| CPU Temp max | 97°C | 86.9°C | -10°C |
| Package Power avg | 108W | 84.7W | -23.3W |
| Core Clock avg | -- | 5064 MHz | Boosting freely |
| PPT Used | -- | 52% | Nowhere near limit |
Current config (CO -30, 90W PPT, idle enabled) -- see the idle test above:
| Metric | Value |
|---|---|
| FPS avg / 1% low / 0.1% low | 191 / 140.4 / 136.5 |
| CPU Temp avg / max | 85.4°C / 86.5°C |
| Package Power avg | 84.4W (94% of PPT) |
| Core Clock avg | 4990 MHz |
| Time at 85°C thermal limit | 80% |
| GPU Load avg | 94% |
Run details: 3dmark
| Metric | Score |
|---|---|
| Overall | 27,511 |
| Graphics | 31,888 |
| CPU | 15,476 |
GPU (950mV undervolt + 1000 memory):
| Metric | Avg | Max |
|---|---|---|
| GPU Clock | 2727 MHz | 2805 MHz |
| GPU Temp | 61.4°C | 67.7°C |
| VRAM Temp | 64.5°C | 72.0°C |
| GPU Power | 274.6W | 326.4W |
| TDP Used | 76.8% | 91.0% |
| Fan Speed | 40% | 43% |
CPU (CO -20, 85°C thermal limit, no PPT cap):
| Metric | Avg | Max |
|---|---|---|
| Core Clock | 4678 MHz | 5225 MHz |
| CPU Temp | 69.5°C | 86.5°C |
| Package Power | 60.5W | 110.4W |
| PPT Used | 37.3% | 68.2% |
Graphics score of 31,888 is within the normal range for a stock RTX 5080 FE (30,229-32,642 per reviews), confirming the 950mV undervolt has no performance penalty in 3DMark. CPU barely breaks a sweat at 69.5°C average.
Created 2026-02-07T23:51:13+01:00, updated 2026-10-07T20:42:38+02:00 · History · Edit