NVIDIA Watch posted an update
Nvidia’s Tesla V100, a 2017 accelerator with 32GB of HBM2 memory, is finding new buyers for local AI inference, eTeknix reports. The site says second-hand prices have risen by about $200 in recent months.
Why it mattersThe card’s 32GB of memory is the draw for running local models, and eTeknix says a V100-based system could be built for around $2,000. That is a reported estimate, not a ready-made shopping list. There are trade-offs: the V100 has a 250W power draw, lacks newer precision formats such as FP8, and the SXM2 version needs special cooling or adapters. Cheap capacity is appealing; cheap capacity that fits your setup is the more useful calculation.
Discuss: Would you choose an older 32GB accelerator for local AI, or pay more for a newer card with lower power use and newer formats?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.