Musk’s model-building outfit xAI – which makes its generative Grok models mainly available through the billionaire’s X – has a combination Dell and Supermicro systems installed in its 100,000 GPU Colossus supercomputer in Memphis, Tennessee.
The system will be upgraded to include an additional 100,000 GPUs with plans to expand it to a million accelerators at a later date. These HPE systems, according to reports, will either end up at X for AI inference and learning work, or xAI to do the same. X and xAI are quite tangled; as we said, xAI provides a Claude/ChatGPT/Gemini-like model called Grok that netizens interact with mainly from X. Musk has previously redirected 12,000 Nvidia H100 GPUs originally bound for Tesla to xAI, arguing the parts would have sat unused otherwise.
Last week, X made Grok’s image-generation capabilities available to all X users. Previously, this was only possible for X Premium subscribers. This month, xAI launched a standalone Grok iOS app. The availability of these services will likely require additional computing resources, especially as future Grok models grow in size. HPE is no stranger to accelerated computation. Its Cray division builds some of the largest computers in the world. These include the El Capitan supercomputer, the Frontier supercomputer, and the Aurora supercomputer, which are among the top three systems on the publically known global top 500 list. HPE announced that it has developed a new series of liquid-cooled GPUs with your choice of Nvidia GPUs, AMD GPUs, or Intel AI accelerators. The X company has not yet revealed which of the three it has chosen, but we can note that Musk’s companies have used Nvidia accelerators in most cases. Nvidia retaliates to Biden’s alleged ‘innovation-killing AI chip export restrictions’