Only 11 percent of GPUs at Elon Musk's xAI are operational
Elon Musk's xAI company is facing significant challenges in expanding its computing power, with only 11% of its installed GPUs currently operational. The company has set up nearly 550,000 NVIDIA H100 and H200 graphics accelerators across its Memphis and Colossus clusters, but issues with managing such a vast system have led to inefficiencies. Unlike major companies like Meta and Google, which achieve utilization rates of 43% and 46% respectively, xAI aims to increase its GPU load to 50%. The inefficiencies stem from delays in data transmission and synchronization among the GPUs, highlighting the need for improvements in xAI's software and distributed architecture.
xAI, the company owned by Elon Musk that is developing the Grok artificial intelligence, has encountered a serious problem in expanding its computing power. According to The Information, the company has installed approximately 550,000 NVIDIA H100 and H200 graphics accelerators in its Memphis and Colossus clusters. This was reported by Ixbt.com.
However, the current utilization of this infrastructure stands at only 11 percent.
This means that only the equivalent of 60,000 GPUs is operating effectively from the massive fleet of accelerators. The problem lies not in a shortage of devices, but in the difficulty of managing such a vast system.
On the scale of hundreds of thousands of graphics processors, delays between data transmission and synchronization stages accumulate rapidly.
According to the source, xAI's software and distributed architecture are not yet sufficiently advanced. As a result, the GPUs spend a significant amount of time waiting for data or tasks, which reduces overall efficiency.
For comparison, major companies such as Meta and Google are achieving utilization rates of up to 43 and 46 percent, respectively. xAI has set a goal to reach 50 percent utilization.
Comments 0
…