3 ms·

> Considering hyperthreading as a way of obtaining an extra "real core" is a gross misunderstanding of what hyperthreading is meant to achieve. That’s cool but

> Considering hyperthreading as a way of obtaining an extra "real core" is a gross misunderstanding of what hyperthreading is meant to achieve. That’s cool but when a vcpu is a hyper thread (rather than a physical core with two hyper threads) that’s the reality of your cloud experience.

Dalewyn 4y ago
If you really need or want an entire physical CPU core, you pay extra for physical hardware instead of a virtualization. Yes, it's expensive, but the performance of your server is a simple question of how much money you're willing to part with.
- dontlaugh 4y ago
  The point is that you can get that with graviton too, which is a much smaller transition.
- Symmetry 4y ago
  If you're got a workload that isn't characterized by missing cache and waiting around for main memory fairly often then usually SMT is going to be pretty inefficient. As a crude rule of thumb, the single thread performance you get is going to scale as something like the square root of the number of transistors you throw at the problem. So if you want two threads of a certain performance you can either use one bit core with SMT-2 or two cores each a third or so the size. And the two cores will tend to be more power efficient two with less data moving long distances through the caches. Now, if you are hitting main memory often it makes sense to go wild with thread and use SMT-8 like IBM's POWER cores or Sun's SPARC cores did. And if you mostly just care about maxing out single threaded performance for user responsiveness then you do indeed have lots of unused resources most of the time and you might as well add SMT for when you're more throughput bound. But while the design and transistor costs of adding SMT to a design are very modest, everything I've heard about the test and verification of it seems pretty hairy.

3 ms·

> Considering hyperthreading as a way of obtaining an extra "real core" is a gross misunderstanding of what hyperthreading is meant to achieve. That’s cool but

by masklinn 4y ago

Dalewyn 4y ago
If you really need or want an entire physical CPU core, you pay extra for physical hardware instead of a virtualization. Yes, it's expensive, but the performance of your server is a simple question of how much money you're willing to part with.
- dontlaugh 4y ago
  The point is that you can get that with graviton too, which is a much smaller transition.
- Symmetry 4y ago
  If you're got a workload that isn't characterized by missing cache and waiting around for main memory fairly often then usually SMT is going to be pretty inefficient. As a crude rule of thumb, the single thread performance you get is going to scale as something like the square root of the number of transistors you throw at the problem. So if you want two threads of a certain performance you can either use one bit core with SMT-2 or two cores each a third or so the size. And the two cores will tend to be more power efficient two with less data moving long distances through the caches. Now, if you are hitting main memory often it makes sense to go wild with thread and use SMT-8 like IBM's POWER cores or Sun's SPARC cores did. And if you mostly just care about maxing out single threaded performance for user responsiveness then you do indeed have lots of unused resources most of the time and you might as well add SMT for when you're more throughput bound. But while the design and transistor costs of adding SMT to a design are very modest, everything I've heard about the test and verification of it seems pretty hairy.