Cold plates shaped for the heat map of each chip.
Cold plates and manifolds from physics-driven generative design. In simulation, a third of the pressure drop of skived fins for the same peak temperature.
Get a demoChip power keeps rising. The cooling loop has to keep up.
Each GPU generation puts more heat into the same footprint. The cold plate, the manifolds and the coolant distribution behind them set how much of that heat can leave, and at what pumping cost.
Skived fins and fixed channel layouts are tuned once and reused on every chip, whatever its heat map.
Every channel placed where the heat is.
ToffeeX grows fins and channels from the chip’s power map and your coolant limits, with flow and heat solved at every step. The same physics-driven generative design covers the system around the chip: distribution manifolds and rack cooling.
- 01GPU and CPU cold plates
- 02Direct-to-chip cooling
- 03Rack cooling
- 04Distribution manifolds
- Better cooling than a skived-fin cold plate
- 12%+
- GPU cold plate, customer pilot.
- Pressure drop for the same 61 °C peak
- One third
- Simulated. ToffeeX at 1.0 m/s, against 100 µm skived fins at 0.3 m/s.
- Better thermal resistance
- 6.7%
- Simulated, against 100 µm skived fins. It costs about 36% more pressure drop.
- Lower pressure drop
- ~10%
- Simulated, against 80 µm skived fins. Heat transfer is better too.
Have a chip that needs more cooling?
Bring your challenges. We will partner with you to solve them.
