NVIDIA H100 server board cooling
NVIDIA H100 Liquid Cold Plate 8-GPU Server Cooling Board
Copper liquid cooling board engineered for 8-GPU NVIDIA H100 server baseboards: eight 700 W GPU positions plus four 186 W NVSwitch positions on one brazed copper assembly, validated with CFD before tooling and leak tested in production.
Request an H100 cold plate quote
Download datasheet (PDF)
Key specifications
| Model | TC-NS-H100 — NVIDIA H100 liquid cooling board (GPU + NVSwitch) |
|---|---|
| Heat dissipation | 700 W × 8 GPU positions + 186 W × 4 NVSwitch positions |
| Material | Copper (red copper), CNC machined |
| Joining method | Transient liquid-phase (TLP) diffusion bonding |
| Test conditions | Pure water coolant, 45°C inlet temperature |
| Flow rate / pressure drop | 9 LPM at 40 kPa |
| Cold plate surface temperature | ≤80°C under rated load |
| Working pressure | 1 MPa |
| Dimensions | 472.7 × 331 × 130.6 mm |
| Thermal resistance (plate, rated point) | ≤0.050 °C/W per GPU position — calculated as (surface-temperature limit − coolant inlet) / heat load: (80 − 45)°C / 700 W |
| ΔP–flow curve | Measured ΔP–flow curve is supplied with the prototype test report and DFM review — request it with your drawing |
| Wetted materials | Copper (C1100 / T2) channel and port surfaces; fittings, tubing and quick disconnects per customer specification — confirmed against your coolant during DFM |
| CAD / drawings | STEP model and 2D drawing available under NDA — request with your RFQ |
| Validation | CFD flow and thermal simulation; helium leak and proof-pressure testing |
H100 liquid cold plate datasheet (PDF, 2 MB)
Full specification table, dimensions, coolant and test conditions, materials and joining method — ready to drop into your thermal review.
ToneCooling manufactures the NVIDIA H100 Liquid Cold Plate — 700 W × 8 GPU positions + 186 W × 4 NVSwitch positions, CFD-validated before tooling and 100% leak tested under ISO 9001 at our Guangdong technical center, with North American customers supported by ToneCooling Texas LLC.
Why ToneCooling for H100 cooling boards
Design, engineering, and manufacturing run in-house at the ToneCooling technical center in Guangdong — one accountable partner from CFD model to leak-tested hardware. TLP diffusion bonding gives the copper assembly durable joints under repeated thermal cycling, and every board can be shipped with the test documentation your qualification process needs. Prototype orders start at 5 pieces with a 6–8 weeks lead time, so an evaluation unit reaches your lab fast.
Routine production testing
Acceptance criteria are agreed in writing before quotation, and the inspection results ship with the parts:
Leak testing (100% of units) — helium leak test to ≤1×10−9 mbar·L/s or nitrogen pressure-decay, per the agreed criteria.
Proof-pressure testing — to the project requirement, with burst verification on qualification samples.
Flow-resistance check — verified against the rated pressure-drop window at the specified flow rate.
Dimensional and flatness inspection — at the TIM interface, to drawing tolerances.
Bond inspection — diffusion-bonded parts are ultrasonically C-scanned per ASTM E2375; thermal-cycling qualification per IEC 60068-2-14 on qualification samples.
CFD validation before tooling
Before any metal is cut, the cold plate design is simulated against your operating point. For an 8-GPU H100 board the CFD model answers three questions that decide whether the board survives qualification:
Flow distribution — coolant must split evenly across all 8 GPU positions and 4 NVSwitch positions; an unbalanced branch shows up as one hot GPU at full load.
Pressure drop — the channel network is tuned so the board stays inside your pump budget at the 9 LPM rated flow.
Surface temperature map — predicted case temperatures are checked against the ≤80°C limit at 45°C inlet before the design is frozen.
The simulation report ships with the DFM review, and the prototype is then verified on the bench: measured ΔP–flow curve and thermal resistance against the CFD prediction, plus helium leak and proof-pressure testing.
Applications
8-GPU H100 server baseboards in AI training clusters, HPC nodes, and hyperscale direct-to-chip liquid cooling loops. For H200-class boards, GB200/GB300 racks, or single-slot GPU plates, see the related pages below — the same RFQ process covers every platform.
NVIDIA H100 cold plate FAQ
Does one cold plate cover all eight H100 GPUs?
Yes. The board is a single brazed copper assembly with 8 GPU positions and 4 NVSwitch positions, sized 472.7 × 331 × 130.6 mm.
Does one assembly cool the full 8-GPU H100 board?
Yes — eight 700 W GPU positions and four 186 W NVSwitch positions run on one brazed copper assembly at 9 LPM / 40 kPa, with flow distribution balanced so every position sees the same supply. Final fit is confirmed against your baseboard drawing before tooling.
Can you supply the rack manifold with the H100 boards?
Yes — ToneCooling also manufactures single-row and double-row liquid cooling manifolds with quick connectors, so the cooling board and its coolant distribution come from one engineering team.
What coolant should be used?
Rated performance is tested with pure water at 45°C inlet. Data-center loops following ASHRAE W55 water quality are supported; EGW mixtures are reviewed during DFM.
What flow rate does it need?
9 LPM at a 40 kPa pressure drop under the rated 6,344 W total load. Your pump budget and manifold layout are checked against this during engineering review.
Can the design be customized for our server tray?
Yes. Port locations, hose routing, mounting pattern, and surface finish are customized from your drawings. CFD simulation validates the change before tooling.
What is the MOQ and lead time?
Prototype MOQ is 5 pieces with a 6–8 weeks lead time from approved drawing and PO; production runs 4–6 weeks.
How is quality verified before shipment?
Helium leak testing and proof-pressure testing on the production line, plus flatness and dimensional checks to the acceptance criteria agreed in writing before quotation.
Request a quote
Send your board drawing, heat map, coolant, and flow budget to info@tonecooling.com or use the quote form — the engineering team returns a manufacturable concept with the quotation. MOQ 5 pcs, prototype in 6–8 weeks.
Send drawings for engineering review or download the datasheet (PDF)
