Game On? AI’s Journey Ahead in Video Games
Generative AI for video games is still far from being real. The main challenges are latency of cloud gaming and the very high cost of
Direct-to-chip cooling system
Traditionally used to cool large supercomputers for decades, direct-to-chip cooling is becoming a more mainstream reality in the data center due to the introduction of AI. The introduction of thousands of GPUs into data centers to run parallelized AI workloads requires much more cooling and refrigeration, and air cooling is simply not enough.
AI workloads consume more power than traditional data center workloads, driving up server density. While traditional data center racks (no accelerated computing) used to have densities of 5-10kW, racks used in AI workloads can have densities of 60-100kW, up to 20 times more. With denser servers there is more heat per rack, and traditional air cooling equipment like raised floors or computer room air handlers (CRAHs) is not enough.
In an average data center air cooling systems are normally sufficient: they are technically simpler, easier to install and require less capital expenditures. But they are not enough for AI workloads, which involve massive computations from GPUs, also supported by CPUs and high-bandwidth-memory, among others.
In direct-to-chip cooling, also known as cold plate cooling, water flows through a conductive plate where the electronic components are placed, dissipating heat at a much faster pace than air cooling. Some data center IT equipment can be highly sensitive and cost thousands of dollars, so it is paramount to maintain it at a stable temperature to maximize its performance and avoid any issues. Industry leaders like Meta, Google or Equinix are already adopting this approach in their data centers. Direct-to-chip liquid cooling can also be complemented with air cooling for less sensitive components that may not come into direct contact with the cold plate.
John Niemann, Vertiv’s Senior Vice President, mentioned in a recent earnings call “direct-to-chip cooling is where we are seeing scale being deployed”, implying this will be the default solution for cooling AI data centers. Vertiv, which manufactures equipment and cooling systems for data centers, has seen its sales increase to $5 billion in the first nine months of 2023, a 24% year-over-year increase. This is, in part explained, by higher demand for direct to chip cooling systems.
Generative AI for video games is still far from being real. The main challenges are latency of cloud gaming and the very high cost of
RISC-V, an open standard instruction set architecture (ISA), could have the potential to mirror the success of Linux two decades later, but in the hardware domain.
GPU vs CPU is a parallelization vs complexity dilemma. While GPUs can manage very large parallel calculations, they struggle with linear, more heterogeneous tasks, where CPUs excel.
AI workloads require more server density and generate more heat. Direct-to-chip liquid cooling is emerging as the preferred cooling solution for AI with leaders like Meta, Google or Equinix adopting it.
Smartphone on-device AI is expected by 2024, and will likely be combined with cloud AI. Smartphone makers will capitalize on it by increasing prices.
Logic chips are the “brains” of the electronic device. They process information and perform calculations. Memory chips store information and data.
The information provided on this site is for informational purposes only. The content is based on the authors´ knowledge and research at the time of writing. It is not intended as professional advice or a substitute for professional consultation. Readers are encouraged to conduct their own research and consult with appropriate experts before making any decisions based on the information provided. The blog may also include opinions that do not necessarily reflect the views of the blog owners or affiliated individuals. The blog owners and authors are not responsible for any errors or omissions in the content or for any actions taken in reliance on the information provided. Additionally, as technology is rapidly evolving, the information presented may become outdated, and the blog owners and authors make no commitment to update the content accordingly.