NVIDIA will save 11 billion watts by cooling its graphics cards with water
There is no doubt that the market is going towards more and more consumption in any chip. But if it is about graphics cards then the problem is greater because the margin is also higher than that of other components. But if, in addition to this, companies like NVIDIA shoot up performance and take consumption hand in hand, we find ourselves with a problem that in servers will only have one solution: water cool A100 GPUs and later their H100.
Unlike other generations where efficiency was higher, NVIDIA and AMD are pushing each other towards higher performance, both on PC and server. The problem is that the A100 consume a lot and at the price of energy, many companies are beginning to tremble due to the costs of the electricity bill, so server assemblers are reacting to this.
Equinix will water the NVIDIA A100
This company is a global service provider that manages more than 240 data centers around the world and given the energy requirements of the new NVIDIA hardware and the needs of its customers, the company has taken action on the matter:
“We have 10,000 customers who are counting on us to help them with this journey. They demand more data and more intelligence, often with AI, and they want it in a sustainable way”
Talking about sustainability is not only consuming less non-green energy, but also reducing the environmental impact and, of course, the electricity bill. In the end, what any company is looking for is the highest performance, with the lowest investment and the lowest maintenance cost, without further ado, and that is why Equinix ensures that liquid cooling is the next step for accelerated computing within GPUs. NVIDIA, where they ensure that you can achieve up to 20 times more efficient in AI and HPC inference work versus CPUs.
Efficiency vs temperature vs consumption
The company claims that leaving CPU servers behind for AI and HPC and switching to GPUs for water can save a whopping 11 billion watt hours, which is equivalent to the consumption of 1.5 million households per year. Therefore, the company in association with NVIDIA is going to offer the A100 GPUs with its 80 GB with custom liquid cooling for its data centers, being something pioneering for the company:
“This is the first liquid-cooled GPU to show up in our lab, and that’s exciting for us because our customers are hungry for sustainable ways to harness AI.”
The system they will use will be closed to minimize liquid losses and they will have controlled environments in temperature and humidity to minimize capillarity. The first data they offer is very promising, since at the same workload in the same conditions when facing air cooling versus water they have managed to get the GPU to use almost 30% less energy.
But not only that, but data centers with GPUs by water can achieve twice the computation in the same physical space due to the reduction of a water block compared to a heatsink (1-slot vs 2-slot). The Equinix option is so promising that nearly a dozen manufacturers will jump on the bandwagon for its products, including: ASUS, ASRock Rack, Foxconn Industrial Internet, GIGABYTE, H3C, Inspur, Inventec, Nettrix, QCT, Supermicro, Wiwynn, and xFusion.
But the road does not end here, but rather begins, since Equinix will start working with the NVIDIA H100 as soon as the testing of the A100 is finished, something that according to the company will be done in the foreseeable future.
