Single-core performance of Intel's Sunny Cove chips Surface - Shows Big IPC Increase

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Jun 17, 2019.

  1. BlackZero

    BlackZero Guest

    Yes, I did already touch on that subject.

    I don't think any body should expect different versions to show the same results, and I would expect most people know they don't.

    They are creating a utility that gives general results based on many different factors, and have to take in to account all the differences in instructions and processors to create a single utility that can give reasonable results that provide clarity in estimates. In all honesty, if making changes to scale, for e.g., provides better results, then that's a good change.

    I mean, it's got a data base included with each revision, and the chosen processors are clearly there for a very specific and well thought out reason.
     
  2. user1

    user1 Ancient Guru

    Messages:
    2,746
    Likes Received:
    1,279
    GPU:
    Mi25/IGP
    the thing is cpuid lists the benchmark version separately from the cpuz version, there is an expectation that it is the same test. they dont even list the cpuz version on their own charts, only the benchmark version.

    its fine for testing overclocks, but between systems is a no go unless tighly controlled. so again it should not be used as a general reference between systems/platforms.
     
  3. BlackZero

    BlackZero Guest

    As long as you use the results included with each revision, I don't see any issues.

    I also think CPU-Z is probably one of the most trusted utilities on the net.
     
  4. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    I do not think there is much reason to give ear to those results above measured score and then dereferencing it from clock.
    For example:
    Their software lists ST score for 2700X as 488, but does not show at what clock that single core did run.
    In this compiled list same score is accompanied by clock 4.35GHz. My chip scores that at around 4.25GHz. But that's not of importance.
    What's important is that it is as far as most of 2700X chips can go. While their listed 3800X @4.7GHz shows ST score of 635... That's good 30% better result even if we ignore clock they shown as that's not to be trusted.

    On intel's side CPUz shows:
    9900K @3.6GHz ST score 580. But lists shows 9900KS @ 5GHz as 600 only? And 9900K @5.4GHz as 660 only? That's because they mix boost clock with base clocks.
    Same way as CPUz shows 7700K @4.2GHz delivering ST score 492.

    They simply mix turbo clock score with base clock. Sometimes they show turbo clock score with turbo clock. And here they shown turbo clock score with base clock again.

    Question is: "What was the real clock at which Sunny Cove achieved 630~639 ST score?"
    Does anyone here really thinks it was at 3.6GHz? Because then its ST score would be ~880 @5GHz. Would it mean intel did miracle work? Or that they slowly developed much better tech, but did not release it till AMD finaly delivered competitive CPU... while milking everyone with mediocre improvements per generation.
    I am not into miracles, and I think intel was pretty lazy and felt safe as 1st Ryzen had still lower IPC and big trouble with clocks. That's most simple explanation.
    Issue with CPUz is in being quite small. It does not have wide variety of synthetic tests which would run for long enough time to test many different operations reliably. This smaller set of instructions tested makes it prone to be unfair. As was mentioned before when they changed benchmark because they realized that AMD's otherwise slower CPU scored too good at thing they were basing their test on.
    (Btw, they are changing their benchmark again, having AVX2 version. I wonder if that's still there as Zen2 comes around or they'll consider it unfair due to improvements made on AMD's platform.)
    Then you have that other issue and it is that they either change compiler flags, compiler itself (update or otherwise) and that's reason for different results in between version. (Even while I can't confirm as large difference in between versions as shown above.)
     
    Last edited: Jun 18, 2019
    Kool64 likes this.

  5. user1

    user1 Ancient Guru

    Messages:
    2,746
    Likes Received:
    1,279
    GPU:
    Mi25/IGP
    all I said was that they had a methodology problem, I stand by that, I have used cpuz for more than 10 years myself, its a useful piece of software.
    trusted or not, the benchmark part of it, the way it is recorded, presented and managed is highly flawed.
     
  6. BlackZero

    BlackZero Guest

    You're creating an issue that doesn't exist.

    They provide quick estimates, based on an algorithm that must look at factors beyond a single instruction set such as AVX2. Also, at no point is it implied that the results are any thing other than a general estimate, and it is pretty clear that they would have to make specific changes to that algorithm with new processors.

    Opinion noted.

    That sad, if they do present the results in a misleading way on their site, then that should absolutely be fixed.
     
  7. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    Issue at hand was, that they made changes because Zen was better at something. And it really was better at that thing they tested. But they wanted to show some other comparison instead as this became apparent.

    And unless you know exactly what they want to show and towards what goal they alter their benchmark, you are being manipulated. That's why I like Geekbench. While it does not really show if one CPU is really better than another. It names their sub tests well enough to know where score came from.
     
  8. BlackZero

    BlackZero Guest

    I suppose in life, who knows what.

    My thoughts were just that why couldn't it be a case of them having to make changes because the results were not as expected, based on what they think is correct in their own testing. In addition, based on all the various tests and results from other sources I have seen, and this of course includes lots of putting the numbers in one place, the differences they show seem to be reasonably accurate. So, I just find they provide pretty good results based on my own subjective experience, and I agree that I simply can not know exactly what goes on in their office.

    Any way, compared to other tests like Time Spy, which focusses unrealistically on specific tests based on AVX, for e.g., I feel CPU-Z gets a much better balance. Again, all subjective based on personal experience, but then there isn't much else one can do. :)
     
    Last edited by a moderator: Jun 18, 2019
  9. oxidized

    oxidized Master Guru

    Messages:
    234
    Likes Received:
    35
    GPU:
    GTX 1060 6G
    Fair.
     
  10. Aura89

    Aura89 Ancient Guru

    Messages:
    8,413
    Likes Received:
    1,483
    GPU:
    -
    Where?
     

  11. D3M1G0D

    D3M1G0D Guest

    Messages:
    2,068
    Likes Received:
    1,341
    GPU:
    2 x GeForce 1080 Ti
    And you read that as 15% across the board in all situations and configurations? Wasn't it obvious that the figure was a rough estimate or an upper limit?
     
  12. Aura89

    Aura89 Ancient Guru

    Messages:
    8,413
    Likes Received:
    1,483
    GPU:
    -
    And where was the lie?

    Or do you think there's some sort of universal % IPC gain that can be claimed by any company?

    If you do apparently you have zero understanding on how this stuff works.

    Did they lie because there's 30% IPC gain in certain scenarios?

    Or did they lie because there's 5% IPC gain in certain scenarios?

    Would they have lied if they said 30%? Or 5%? How could they lie, if both would be true?

    And where do you get the information that 15% is a lie in the first place when Zen2 hasn't been released? Leaks? So wait you're trusting limited leaks that could not be true to determine if a company has lied on a unreleased product?

    Just so we're clear, i'm not saying there's any scenario where 30% IPC gain is a thing, only that since they stated 15%, i doubt 15% is the highest IPC gain, it's likely higher, but obviously, just like with ANY company in existence, there is no "percentage of IPC that is universally better"

    So many things wrong with whatever methodology you are thinking with.
     
    Neo Cyrus and Fox2232 like this.
  13. D3M1G0D

    D3M1G0D Guest

    Messages:
    2,068
    Likes Received:
    1,341
    GPU:
    2 x GeForce 1080 Ti
    There was no need to. Such figures are always an "up to" figure or some kind of favorable average - my immediate thought on seeing that was that it was a best-case scenario (possibly due to increased AVX support) and it would range anywhere from 0% - 15% depending on the application and task. At no point did I assume that it was a flat 15% increase in all scenarios - that goes against everything we know about the way CPUs work.
     
    carnivore, las and Aura89 like this.
  14. Astyanax

    Astyanax Ancient Guru

    Messages:
    17,011
    Likes Received:
    7,352
    GPU:
    GTX 1080ti
    compared clock for clock, the gain from zen+ to zen2 is 15%
     
  15. Aura89

    Aura89 Ancient Guru

    Messages:
    8,413
    Likes Received:
    1,483
    GPU:
    -
    Who does that and how would that be relevant?

    The best case scenario is they should give an AVERAGE IPC increase percentage and then tell us based off of what, as any ONE workload is completely pointless to claim as your IPC gain.

    Now i don't know if it's an actual average they have given us based off their own internal tests, or, if it's a nice number that technically is not incorrect as their IPC gain, but the whole point of this conversation you brought up is that they were "lying" about the 15% IPC gain, and i ask again: Where?
     

  16. Exodite

    Exodite Guest

    Messages:
    2,087
    Likes Received:
    276
    GPU:
    Sapphire Vega 56
    My understanding, from the vast majority of reporting I've read on the subject, was that the stated IPC gain were vs. original Zen and not Zen +. Did we get other information from AMD at E3, or even Computex? I did watch the streams but I confess I didn't pay enough attention to catch any such details.
     
  17. Astyanax

    Astyanax Ancient Guru

    Messages:
    17,011
    Likes Received:
    7,352
    GPU:
    GTX 1080ti
    im looking at numbers from a 2600 to 3600 comparison.
     
  18. sverek

    sverek Guest

    Messages:
    6,069
    Likes Received:
    2,975
    GPU:
    NOVIDIA -0.5GB
    If it did OC much better, we would already see hype how Matisse reached 5GHz on air/AIO. AMD themselves would proudly mention it.

    Basically, AMD already overclocked the CPU for us, so there no further overclocking without adding lots of voltage, which will require LN2, which means nothing for us.

    Ryzen 9 3950X having 4.7GHz on Precision Boost 2, meaning higher end CPUs getting all the sweet binned cores.

    Good luck reaching 4.7GHz on ALL cores with Ryzen 7 3800X. Not happpening IMO.
    Community built too much hype for Zen2.
     
    fantaskarsef likes this.
  19. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    How did MSI debunked it?
     
  20. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    Elaborate on that.
     

Share This Page