NVIDIA GH200 Grace CPU vs. AMD EPYC 9005 Turin CPU Performance

[coder]

But beyond that most cloud providers offer very limited access to free/gratis instances for benchmarking, especially after launch of any new instance type. And not within budget to do all that extra cloud benchmark runs when not provided by the CSP.

Have you ever tried running benchmarks on spot instances? Or do they go offline too quickly to be a practical option?

Edit: and with the CSP comparisons also worth mentioning the lack of CPU power monitoring access.

It's obviously imperfect, but here's where I look to pricing as a proxy. Especially with spot pricing, you can be sure they're not going to price it lower than the marginal costs are. I'll bet most people using spot instances are doing so for compute-heavy work, since that's about all they're good for, so the prices should account for most of the cores running flat-out, most of the time.

That's said, there's obviously some supply/demand aspect of spot pricing at AWS, since like the R7g.12xlarge you just tested currently has a spot price of $0.4759, whereas an identical instance with half the RAM has a spot price of only $0.3176. However, cutting the RAM in half again saves nothing, with the C7G.12xlarge having a spot price $0.0005 higher.

BTW, how much consistency in benchmark scores do you see from run-to-run, on these cloud instances?