> "they deserve bankruptcy"
Chapter 7, please! I would like them to *liquidate* Smaug's GPU hoard; and I would like them to do that all at once, in a swift and efficient fire sale.
> "they deserve bankruptcy"
Chapter 7, please! I would like them to *liquidate* Smaug's GPU hoard; and I would like them to do that all at once, in a swift and efficient fire sale.
I'd expect them to give "trusted partners" first dibs. You know, firms that can warehouse the chips and keep prices high. Gotta protect NVDA at all costs. I'd love to be proven wrong.
A lot of those are datacenter GPU racks which will be difficult to operate for most people, even for smaller companies that don't have watercooling setups.
Sure, but reallocating those to their AI competitors would offset those competitors' demand for AI GPU's and HBM RAM. Which frees up the TSMC, Samsung, Micron fabs to shift back to consumer production. Displaced demand.
Sounds like a fun challenge with bragging rights if someone gets it working! Don't threaten me with a good time ;)
It’s not hard. It’s power intensive. It’s $15k/mo at 15 cents / kWh. If you want the experience there’s a massive A100 liquidation sale that happened recently https://www.ebay.com/itm/178522410497
These come around once in a while. Go get that seller’s Gaudi rig if you want to test the experience. It costs one-time as much as the monthly power cost on a B300 NVL72 and uses a fifth the power.
can you imagine? gamers flush with all that fancy hardware?
Likely with no place to run it. I considered an old (circa 2017) retired DGX-1 which has 8x V100 32GB GPUs as an alternative to an M5 Ultra Mac Studio. The problem is the power draw. The modern GPUs AI companies are sitting on draw vastly more power.