DO and friends wouldn't exist in the CPU side if that was true, so a spot economy in GPU side makes sense, and even more so bc they are higher-end time shares
Even if we didn't have an open ticket with Azure since Christmas for more GPUs, the ability to burst non-sensitive tasks on say 20 cheaper GPUs for 5-30 min is attractive.
That stuff adds up fast, and lowering cost both opens accessibility for low-end users and helps scale what power users can do.
My bigger surprise is vast has been around awhile and only ~20 available servers, so I'm guessing < 100 total. Is this a friction issue?
Even if we didn't have an open ticket with Azure since Christmas for more GPUs, the ability to burst non-sensitive tasks on say 20 cheaper GPUs for 5-30 min is attractive.
That stuff adds up fast, and lowering cost both opens accessibility for low-end users and helps scale what power users can do.
My bigger surprise is vast has been around awhile and only ~20 available servers, so I'm guessing < 100 total. Is this a friction issue?