- Cooling costs rise as Nvidia pushes power limits in successive rack generations
- Compute trays dominate expenses due to increasing cold plate requirements and thermal density.
- Savings on switching trays can’t offset growing demands for high-power GPU trays
The cost of keeping Nvidia’s high-end rack systems cool continues to rise as each generation delves into extreme power levels.
A Morgan Stanley report obtained by @Jukanlosreve reveals that the liquid cooling hardware inside the GB300 NVL72 costs $49,860, roughly enough to buy a new Tesla Model Y.
The report further estimates that the liquid cooling system required for the new Vera Rubin NVL144 configuration will approach $55,710, an increase of 17%.
The economics of tray level cooling
This platform relies on more powerful Rubin GPUs with power up to 1,800W per unit, along with next-generation NVSwitch 6.0 components.
The cost of cooling this system is tied to the individual compute trays, and each compute tray will need higher capacity cold plates.
The cost per compute tray is expected to increase 18% to approximately $2,660, and since the Vera Rubin NVL144 system has 18 trays, the total cooling expense on the computer side reaches approximately $47,880.
The increase comes from higher capacity cold plates, which rise to $400 per unit as CPUs and GPUs push thermal limits.
Switching tray cooling, meanwhile, appears less onerous, dropping to $870 per tray and totaling $7,830 per rack.
However, this The reduction is dwarfed by the much larger jump on the compute side, as the cost trajectory follows a pattern: The GB200 NVL72 transition saw a 20% increase in cooling demands for the GB300 NVL72.
Likewise, moving from the GB300 NVL72 to the Vera Rubin NVL144 adds another 17%. Power levels explain this trend.
Each GPU in the Blackwell Ultra data center consumes 1400 W, a Grace CPU consumes 300 W, and memory contributes 200 W per socket.
As workloads increase, the value of precision cooling grows just as quickly, but future systems will further exacerbate this. Nvidia plans a move to Rubin Ultra GPUs that can reach a thermal design power of 3,600W per package, and meeting that requirement may force new types of cold plates or more aggressive cooling techniques.
Nvidia is also preparing the liquid-cooled NVL576 “Kyber” system, which will include 144 GPU packages and offer higher performance than the Vera Rubin NVL144, while having an even higher cooling bill.
Although the final dollar amount is not confirmed, high-capacity plates capable of removing 3.6 kW of heat will clearly exceed the current $400 per unit.
This sends a signal that future data center installations will face even higher thermal expenses.
Via Toms Hardware
Follow TechRadar on Google News and add us as a preferred source to receive news, reviews and opinions from our experts in your feeds. Make sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form and receive regular updates from us on WhatsApp also.
