Hi Извънземен,

Back in the days, I had bought a Supermicro SYS-E300-9D-4CN8TP system – That system had proven to be a neat piece of kit, with plenty of on board Ethernet connections etc. Although clearly that pal was way too loud for my targeted desktop/room environment…

Hence, that system ended up in its original box and laid on a shelf for years… Until Ramageddon hit us all =] After empirical thermal testing, a case swap to Lian Li B4 dropped idle temps to 50°C, delivered total silence and cool operation under full load. Here’s the complete journey.

📖 Table of Contents#

  1. The Issue
  2. Experimental Methodology
  3. Thermal Data Trail
  4. The Solution: B4 Migration
  5. Bill of Materials
  6. Wiring & Assembly Notes
  7. Final Results
  8. Migration Notes

The Issue#

Hardware: Supermicro SYS-E300-9D-4CN8TP (X11SDV-4C-TP8F motherboard + Xeon D-2123IT 60W SoC)

Symptoms:

  • CPU temperatures at 81°C at ~30% CPU usage
  • PCH temperature at 86°C+
  • All fans screaming - FAN1 @5200 RPM / FAN2 @5300 RPM / FANB @8600 RPM – using the Power Utilization Effectiveness BIOS profile

Root Cause: The SCE300 1U-short chassis funnels air in a straight line past components without forcing it through the passive heatsink. Volume of airflow ≠ effective airflow through the heat exchanger. A view of the original system running with 3× Noctua A4x20 (which is a failure LID close) + 1x Noctua A4x20 on CPU heatsink for experiments (here blowing up for testings - avoid it).

Image alt

Experimental Methodology#

Rather than guessing, controlled experiments isolated variables:

Test Configuration CPU Temp (Load) Finding
1 Lid ON + 4× Noctua A4x20 92°C+ climbing ❌ Chassis airflow blocked heat removal
2 Lid OFF + fan on heatsink 63°C ✅ Direct airflow through fins = cooling
3 Lid OFF + no fan on sink 75°C ⚠️ Case openness helps, but heatsink needs directed airflow
4 Lid ON + fan misaimed 75°C ⚠️ Misdirected fan = decorative (faint thermal contribution)

Key Insight: A running fan that doesn’t push air through the fin stack contributes zero thermal benefit. Aim > volume.

Thermal Data Trail#

Phase Idle Load (4 cores) Fans @ Load Verdict
E300 (closed) 75°C 92°C+ 5000+ RPM all 🔴 Throttling imminent
E300 (open lid) 40°C 63°C 5000+ RPM 🟡 Usable, noisy
B4 (A4x20 stand-in) 43°C 75°C 1000–4600 RPM 🟢 Silent, cool
B4 (projected A6x25) 43°C ~65°C <1000 RPM 🟢 Perfect

The Solution: B4 Migration#

Case Swap Rationale:

  • Flex ATX board fits any mATX case
  • 162mm+ cooler clearance vs ~40mm stock sink = headroom
  • Unrestricted airflow path vs E300’s choked channel
  • 3× bottom intakes + 2× top exhausts vs 40mm screamers

Build Decision:

  • Keep original 150W AC adapter
  • Feed JPV1 (8-pin 12V DC) directly, bypassing ATX PSU
  • External PSU brick = zero internal heat dump

Early migration/transplanting operations… Image alt

Bill of Materials#

Item Qty Purpose
Lian Li B4-mATX (Wood Edition) 1 New chassis, unrestricted airflow
Noctua NF-A6x25 PWM 1 CPU heatsink fan on sink
Noctua NF-A12x25 PWM 3 Bottom intake fans (3× ~30)
Dupont 20cm M-F Jumpers 2×20pk JF1 front-panel wiring
CR2032 CMOS Battery 1 Fix VBAT failure

Wiring & Assembly Notes#

Front Panel (JF1) Conversion#

The B4’s 4-pin F_PANEL block doesn’t match Supermicro’s 20-pin JF1 header. I simply connected the Dupont jumpers straight from the B4’s 4-pin F_PANEL block to JF1 (pins 1–2). If I remember correctly, the B4’s 4-pin F_PANEL is only wired with two wires, the Power Button wires indeed.

Power Entry via PCIe Slot#

Rather than cutting the case:

  1. Route DC cable through empty PCIe slot opening (Mobo has only two PCI slots)
  2. Secure with tape/foil to seal gap
  3. Connect to JPV1 (requires removing top fan bracket for access)
  4. External brick stays outside → no internal heat

I kept the original external PSU brick, my wiring solution… Image alt

Fan Assignment#

Fan Position Hardware PWM Header
Bottom ×2 NF-A12x25 PWM FANA–FAN3
Rear ×1 NF-A12x25 PWM FAN2
Top ×2 (stock) Slim 120mm exhaust FAN1
CPU Sink NF-A4x20 PWM FAN1 (choose CPU zone header/blowing down)

Fan curve: IPMI “Fan to Standard Speed” mode or manual raw commands for RPM tuning.
Case Top Fans: I did invert the blowing direction of the 2nd TOP fan as recommended in the B4’s user manual.

Near final migration/transplanting operations… Image alt

Final Results#

Thermal Performance#

Metric Before (E300) After (B4) Delta
Idle CPU Temp 75°C 43°C –32°C
Load CPU Temp (8 cores) 92°C+ 58–61°C –31°C
Idle PCH Temp 79°C 47°C –32°C
Fan Noise @ Load 5000+ RPM ~1000–2600 RPM Inaudible

Operational Notes#

  • ✅ All sensors report “Normal” status
  • ✅ No thermal throttling under sustained load
  • ✅ Full 4-core stress test completed successfully

Stress test used (on PVE | pve-manager/9.2.20/49318c671b82f31e (running kernel: 7.0.14-19-pve)):

apt update && apt install stress-ng
stress-ng --cpu 4 --cpu-method matrixprod --metrics-brief --timeout 300s

Migration/transplanting operations done… Image alt

Migration Notes#

Third fan slot:

  • B4 bottom mount may conflict with DOM modules
  • Workaround: Use rear 2.5" mount or the front fan available slot (thinner)

USB front panel:

  • B4 USB 3.0 cable has no matching header on TP8F (ports are rear I/O only)
  • Optional: Add PCIe USB 3.0 controller with internal header
  • Verdict: Skip for headless PVE hosts

Power LED:

  • B4 has no power LED
  • IPMI + network/mobo LEDs provide all necessary visibility for headless ops

Takeaways#

  1. Direct airflow through the fin stack beats ambient washing. Aim matters more than quantity.
  2. Small cases can run quiet if airflow paths aren’t choked. The B4’s unrestricted intake/exhaust solved the E300’s thermal chokehold.
  3. External power adapters are underrated. Zero waste heat dumped into a 21L case.
  4. Experimental rigor pays off. Control groups, isolate variables, document deltas — the data tells the truth.
  5. Vintage server boards deserve second lives. A 60W Xeon D + 125 GiB RAM for roughly CHF ~200 total is incredible homelab value.

What’s Next#

Expansion options:

  • Add 2.5" SATA drive via JPH1 + floppy→SATA adapter
  • Install Samsung 990 EVO Plus 2TB NVMe (PCIe 3.0-limited, but more than adequate)
  • re-seat previously hosted PCI expansion board (not really needed though better in there then on some shelves)
  • Opportunistically hunt for the I/O shield for the X11SDV-4C-TP8F - part number = MCP-260-00098-0N

Build completed September 2026. D-2123IT idle: 43°C. Load: 58–61°C. Silence achieved.