Re:AMD EPYC Milan Review Part 2: Testing 8 to 64 Cores in a ...
by tyingq on Wednesday June 30, 2021 @09:02AM
(#27687959)
Benchmarks and price/performance are going to start getting confusing now that cloud providers are selling a mix of vCPUs where some are real cores, and some are hyperthreads. Amazon's Graviton instances are full cores. I expected that since ARM doesn't have hyperthreads. But now Google is selling EPYC full cores as vCPUs with Tau. I would guess that forces the other providers to follow suit.
[ Parent
| on HN ]
Re:AMD EPYC Milan Review Part 2: Testing 8 to 64 Cores in a ...
by MR4D on Wednesday June 30, 2021 @09:53AM
(#27688532)
If we’re lucky, we’ll end up where we need to be - performance per dollar.
But that’s probably a big “if”.
Re:AMD EPYC Milan Review Part 2: Testing 8 to 64 Cores in a ...
by nine_k on Wednesday June 30, 2021 @07:04PM
(#27694567)
Performance is hard to measure as a scalar.
Re:AMD EPYC Milan Review Part 2: Testing 8 to 64 Cores in a ...
by fulafel on Wednesday June 30, 2021 @02:43PM
(#27692073)
Terminology nit: s/hyperthread/SMT thread/ - HyperThreading is an Intel marketing name for their implementation of SMT.
Re:AMD EPYC Milan Review Part 2: Testing 8 to 64 Cores in a ...
by mhh__ on Wednesday June 30, 2021 @10:34AM
(#27689049)
There are a few ARM CPUs that do SMT
Re:AMD EPYC Milan Review Part 2: Testing 8 to 64 Cores in a ...
by rbanffy on Wednesday June 30, 2021 @04:19PM
(#27692999)
True, but I suspect that the simpler instruction format makes it much easier to extract parallelism from a single thread than it is with x86's.
Re:AMD EPYC Milan Review Part 2: Testing 8 to 64 Cores in a ...
by bserge on Wednesday June 30, 2021 @11:37AM
(#27689816)
I was under the impression that was already the case with all Intel processors.