4 ms·

For per core performance, it surely is.

by huytersd 3y ago

BugsJustFindMe 3y ago
Clock only translates to performance when combined with Instructions-Per-Clock.
- bobim 3y ago
  It used to be clocks per instruction, we are spoiled nowadays.
  - AA-BA-94-2A-56 3y ago
    You say that, but applications use up orders of magnitude more system resources these days due to being built upon several layers of abstractions. Some modern applications feel more sluggish than their feature-complete counterparts from back in the day.
    - bobim 3y ago
      Correct, we somehow lost grip on things, still we aspire at a simpler ecosystem. But I wouldn’t give up on python that helps me powering so much of my digital life, with only SciTe and a command prompt.
- alkonaut 3y ago
  There are no dramatic shifts in IPC recently and within the same family of processors like Intel core. That makes clock speeds a pretty decent proxy for (peak single core) perf. You can compare a 4GHz from a couple of years ago and know with some accuracy what to expect from 6Gz.
  - hwillis 3y ago
    That's tautological. Commenter above is saying that IPC may have gone down. You're asserting that it didn't without any evidence. There are also plenty of other reasons speed doesn't translate well to performance, like thermal limits and memory bottlenecking.
    - alkonaut 3y ago
      Since there are small yet positive IPC improvements each generation it’s a reasonable assumption/expectation that at least IPC is no worse in gen 14 than in the generations immediately before it. Sure, until there are benchmarks we can’t know this. If I knew thus I’d no doubt be under NDA or embargo. But I’d be willing to bet that IPC difference is pretty small. > There are also plenty of other reasons speed doesn't translate well to performance, like thermal limits and memory bottlenecking. Indeed, the more interesting unknown for a new CPU gen these days isn’t what’s happened to IPC but how long and how many cores can actually run at peak clock. In terms of real perf that’s probably a much bigger unknown factor in the performance. When I see 6GHz I wonder if that’s for 1 core for one second… But that’s exactly why the thing that’s reasonably easy to guess is exactly that: the peak/instantaneous/theoretical single core perf. And actual “real world” perf is much harder to make any guesses about. If you take the IPC of the last gen and the clock of the next gen you have a pretty good value if you want to guesstimate the peak single core perf.
    - PH95VuimJjqBqy 3y ago
      no it's not, it's practical. causality does not imply causation but it sure as shit points a gigantic neon sign at it.
seabrookmx 3y ago
Assuming you're comparing two chips with the same architecture. Intel Prescott chips hit just shy of 4ghz and their performance certainly doesn't hold up.

For per core performance, it surely is.

BugsJustFindMe 3y ago

Clock only translates to performance when combined with Instructions-Per-Clock.

bobim 3y ago
It used to be clocks per instruction, we are spoiled nowadays.
- AA-BA-94-2A-56 3y ago
  You say that, but applications use up orders of magnitude more system resources these days due to being built upon several layers of abstractions. Some modern applications feel more sluggish than their feature-complete counterparts from back in the day.
  - bobim 3y ago
    Correct, we somehow lost grip on things, still we aspire at a simpler ecosystem. But I wouldn’t give up on python that helps me powering so much of my digital life, with only SciTe and a command prompt.
alkonaut 3y ago
There are no dramatic shifts in IPC recently and within the same family of processors like Intel core. That makes clock speeds a pretty decent proxy for (peak single core) perf. You can compare a 4GHz from a couple of years ago and know with some accuracy what to expect from 6Gz.
- hwillis 3y ago
  That's tautological. Commenter above is saying that IPC may have gone down. You're asserting that it didn't without any evidence. There are also plenty of other reasons speed doesn't translate well to performance, like thermal limits and memory bottlenecking.
  - alkonaut 3y ago
    Since there are small yet positive IPC improvements each generation it’s a reasonable assumption/expectation that at least IPC is no worse in gen 14 than in the generations immediately before it. Sure, until there are benchmarks we can’t know this. If I knew thus I’d no doubt be under NDA or embargo. But I’d be willing to bet that IPC difference is pretty small. > There are also plenty of other reasons speed doesn't translate well to performance, like thermal limits and memory bottlenecking. Indeed, the more interesting unknown for a new CPU gen these days isn’t what’s happened to IPC but how long and how many cores can actually run at peak clock. In terms of real perf that’s probably a much bigger unknown factor in the performance. When I see 6GHz I wonder if that’s for 1 core for one second… But that’s exactly why the thing that’s reasonably easy to guess is exactly that: the peak/instantaneous/theoretical single core perf. And actual “real world” perf is much harder to make any guesses about. If you take the IPC of the last gen and the clock of the next gen you have a pretty good value if you want to guesstimate the peak single core perf.
  - PH95VuimJjqBqy 3y ago
    no it's not, it's practical. causality does not imply causation but it sure as shit points a gigantic neon sign at it.

seabrookmx 3y ago

Assuming you're comparing two chips with the same architecture. Intel Prescott chips hit just shy of 4ghz and their performance certainly doesn't hold up.