3 ms·
The finer grained performance per watt vs silicon cost set in the context of use case is just lost in HN hardware conversations like this that ignore total cost
by simfree 2y ago
The finer grained performance per watt vs silicon cost set in the context of use case is just lost in HN hardware conversations like this that ignore total cost of ownership, vendor politics and such.
Everyone is buying the tool that does the job, or building that tool if they want to make that large investment...
- llm_trw 2y agoThere's a difference in cost per flop at home and in a data center. I'm updating my wiring and air conditioning for a 7x5090 workstation because having that power for experiments under the desk is worth the cost (and fire hazard). If I had to build 10,000 of those I'd be banned by NVidia from ever buying their hardware again.
- maayank 2y agoWhat are you doing with the workstation that it’s worth the cost?
- llm_trw 2y agoHigh dimensional computations for time series data in finance.
- maayank 2y agoPrivate algotrading? Nice. May the odds be always at your favor!
- DeathArrow 2y agoIf I'd have those GPUs, I'd use them to run some AI models locally. But since I am not rich and I don't make money from that, I won't buy those GPUs.