A former Twitter engineer who worked at the company during Elon Musk’s takeover now recalls discovering a cluster of 700 Nvidia V100 GPUs in mid-2022. Tim Zaman discovered the cluster in a Twitter data center — running but idle.
Weird GPU Discovery on Twitter
Zaman described his bizarre discovery in an X post as “the forgotten remnants of an honest attempt to create a Twitter 1.0 cluster.” Engineer Zaman got the idea to report his discovery after reading about xAI’s Memphis Supercluster. This trains the Grok 3 chatbot and is powered by 100,000 liquid-cooled Nvidia H100 GPUs..
“How times have changed!” Zaman declared. After all, the former Twitter company had 700 of the world’s most powerful GPUs running for years—and it was completely useless. Another curiosity of Zaman was to discover that the 700 Nvidia V100 were PCIe GPUs and not the NVLink variant in the SXM2 form factor.
Zaman’s tweet also reflected on Musk’s new “Gigafactory of Compute.” “Running 100,000 GPUs on the same fabric should be an epic challenge,” the engineer said. “At this scale, the only guarantee is failure, and it’s about managing failure properly.”
what's your opinion about this subject? Use the comment function and let us know what you think. To comment you must be logged in to PCGH.de or the Extreme forum. If you do not have an account yet, you can consider registering without obligation, which has many advantages. Please follow the forum rules when commenting.
