Ryzen 3 3700x vs 3800x

Discussion in 'AMD Processors' started by dexvx, Jun 18, 2019.

  1. dexvx

    dexvx [H]ard|Gawd

    Messages:
    1,091
    Joined:
    Aug 14, 2002
    They are both 8C/16T unlocked. Is there any point getting the 3800x? The mystical claims of better binning?
     
  2. Randall Stephens

    Randall Stephens Limp Gawd

    Messages:
    408
    Joined:
    Mar 3, 2017
    Just buy it. Them. Buy both of them.
     
  3. sirmonkey1985

    sirmonkey1985 [H]ard|DCer of the Month - July 2010

    Messages:
    21,247
    Joined:
    Sep 13, 2008
    probably not even binned since it's a 105w part vs a 65w part. it's basically like the 1800x where it's slightly overclocked for those that don't want to overclock themselves. won't know for sure though til the 7th.
     
  4. ChadD

    ChadD 2[H]4U

    Messages:
    3,948
    Joined:
    Feb 8, 2016
    If I had to guess as these are chiplet based chips. The 3700 has one chiplet with 8 cores.... and as its only powering one its a 65w part. While the 3800 is 2 chiplets with 4 cores each.

    I could be wrong... but that is all I could think of as to why there is such a drastic power difference. Seems logical they would be able to squeeze a few more MHZ out of the part where they can basically disable the weakest cores.

    I am looking at both but will be waiting for a few good in depth reviews. The extra power draw might be worth it... or not. :)
     
  5. ChadD

    ChadD 2[H]4U

    Messages:
    3,948
    Joined:
    Feb 8, 2016
    Also if it turns out the 3800 is a dual chiplet and has some nice performance advantage over the 3700... well at that point your so close to the price of a 3900. ;) lol
     
  6. bizzmeister

    bizzmeister [H]ard|Gawd

    Messages:
    1,487
    Joined:
    Apr 26, 2010
    Please just be an amazing cpu at an amazing price ( for gaming ) don’t give a shit about anything else personally.
     
    gigatexal likes this.
  7. dexvx

    dexvx [H]ard|Gawd

    Messages:
    1,091
    Joined:
    Aug 14, 2002
    Why would two chiplets have any advantage over 1 chiplet?
     
  8. drescherjm

    drescherjm [H]ardForum Junkie

    Messages:
    14,371
    Joined:
    Nov 19, 2008
    Maybe cooling and possibly L3 cache if its not disabled.
     
  9. Randall Stephens

    Randall Stephens Limp Gawd

    Messages:
    408
    Joined:
    Mar 3, 2017
    the L3 would be the big difference, I feel. But based on specs either AMD is lopping off the L3 of the 2nd chiplet, or just going with a single die. the 12 core is the big L3 bump per their specs.
     
  10. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    8,819
    Joined:
    Apr 22, 2006
    That's a virtual certainty.

    It's just like 2700 vs 2700x. The top part is max clock speed part.

    Auto-overclocking for those who don't want to bother overclocking.
     
  11. BrotherMichigan

    BrotherMichigan [H]Lite

    Messages:
    103
    Joined:
    Apr 8, 2016
    The L3 is a victim cache distributed on a per-CCX basis, but the 3800X is certainly going to be a single chiplet design if only because the performance penalty of going off-die will be too great compared to having two CCXs on one die.

    (Ninja'd by Snowdog on the rest of what I was going to say.)
     
  12. ChadD

    ChadD 2[H]4U

    Messages:
    3,948
    Joined:
    Feb 8, 2016
    So you believe that a 200mhz difference requires almost twice as much power ? Perhaps. Hope launch day brings a few decent reviews.
     
  13. ChadD

    ChadD 2[H]4U

    Messages:
    3,948
    Joined:
    Feb 8, 2016
    Unknown right now. Could be an advantage. Could be a disadvantage as well... as I don't believe chiplets can share cache with each other. Its possible if the 3700 is actually just one chiplet it has the advantage. Or there could be none. We need a good [H] type review.
     
    travanx and IdiotInCharge like this.
  14. BrotherMichigan

    BrotherMichigan [H]Lite

    Messages:
    103
    Joined:
    Apr 8, 2016
    No, just like how the 1700X didn't consume 30 more watts than the 1700 and the 2700X didn't consume 40 more watts than the 2700 because TDP != power consumption. The difference comes about because of the 300 MHz higher base clock of the 3800X and the extra headroom for Precision Boost. See the following plot from Tom's for an example of this difference:

    aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9FLzIvNzcwMzMwL29yaWdpbmFsL2ltYWdlMDA5LnBuZw==.png

    The 2700X maintains significantly higher clockspeeds on average (much higher than the 100 - 300 MHz difference in boost or base clocks) because of the greater thermal headroom allowed by its higher TDP.
     
    Snowdog and ChadD like this.
  15. BrotherMichigan

    BrotherMichigan [H]Lite

    Messages:
    103
    Joined:
    Apr 8, 2016
    It would certainly be a disadvantage. Even with the improvements AMD has made to Infinity Fabric, you will still have much greater bandwidth (and MUCH lower latency) between CCXs on the same die than you would between CCXs on difference dice with a hop across the IO die in between.
     
    kirbyrj and ChadD like this.
  16. dexvx

    dexvx [H]ard|Gawd

    Messages:
    1,091
    Joined:
    Aug 14, 2002
    IMO, back when Ryzen launched, there was little reason to get the 1800x over the 1700x.

    To me, the 3800x feels somewhat strange on the product stack, given where the 3700x is.
     
    Mav451 and ChadD like this.
  17. ChadD

    ChadD 2[H]4U

    Messages:
    3,948
    Joined:
    Feb 8, 2016
    I agree that is why I guess I'm expecting some sort of catch... like 2 chiplets vs 1 or something.

    BrotherM could well be correct. I am planning to go one or the other unless the 3900 just calls my wallet. lol I will for sure be waiting to read a few decent reviews first though.
     
  18. lightsout

    lightsout Gawd

    Messages:
    870
    Joined:
    Mar 15, 2014
    https://hardforum.com/threads/ryzen-3000-ccx-and-core-layout.1982784/#post-1044229047

    Seems to not be the case. But this doesn't specify between multiple 8 core chips.

    I have wondered this same thing myself. Looking at Zen+ it would make more sense if the cheaper 8 core was a non-x. With current gen ryzen X chips are imo a good amount better. The auto overclocking is so good it makes manual clocking almost pointless.

    But with them both being X models its hard to know without them releasing what if any real difference there will be.
     
    ChadD likes this.
  19. RamonGTP

    RamonGTP [H]ardness Supreme

    Messages:
    7,577
    Joined:
    Nov 9, 2005
    The L3 is there to minimize the latency that's inherent with multi chipset designs. Everything being equal, (clock speed, core/thread count) a dual chiplet design will be slower than a single chiplet in most scenarios even with larger cache. Not that it matters, both the chips in question are single 8 core chiplets. The 3800x is not a 4x2 configuration.

    I should preface this by "at least according to Linus"

    https://linustechtips.com/main/topi...md-3000-specs-47-ghz-r9-3950x-r7-3700x-3800x/

    He mentions the chiplet configuration in the 12 and 16 core parts but no mention of it in the 3700x and 3800x. This leads me to believe both are single chiplets.

    Further confirmation at anandtech

    https://www.anandtech.com/show/1440...-cores-for-499-up-to-46-ghz-pcie-40-coming-77

    The Ryzen 3000 series will debut a new product tier for AMD: Ryzen 9. In this case, the Ryzen 9 3900X will be AMD’s first mainstream desktop 12-core processor. The processor is the only one of the group that uses two chiplets, in a 6+6 configuration.
     
    Last edited: Jun 18, 2019
    ChadD likes this.
  20. dexvx

    dexvx [H]ard|Gawd

    Messages:
    1,091
    Joined:
    Aug 14, 2002
  21. Randall Stephens

    Randall Stephens Limp Gawd

    Messages:
    408
    Joined:
    Mar 3, 2017
    Don't they say that the lower tdp parts are worse when run at the higher frequencies? I.e. they leak more, so they limit their vcore to where their efficiency falls in line?
     
  22. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    10,406
    Joined:
    Jun 13, 2003
    This is going to architecture, process, and even bin dependent.

    We'll see ;)
     
  23. chameleoneel

    chameleoneel 2[H]4U

    Messages:
    2,920
    Joined:
    Aug 15, 2005
    As with the last Ryzens, it probably needs the extra voltage to get over the hump, for the extra MHZ.
     
  24. BrotherMichigan

    BrotherMichigan [H]Lite

    Messages:
    103
    Joined:
    Apr 8, 2016
    Please read my posts above regarding the higher TDP of the 3800X vs the 3700X. It has nothing to do with binning or voltages required for clock bumps, it is simply greater thermal headroom for Precision Boost.
     
  25. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    8,819
    Joined:
    Apr 22, 2006
    No. See post #14 above yours, and compare the clock speed lines 2700 vs 2700X. That is the difference.

    AMD doesn't seem to do a lot of binning on consumer parts, but if there is any here, it will be on the 3800X to ensure it can sustain much better all core clock speeds.
     
  26. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Mar 13, 2017
    Both the 3700X and 3800X are single chiplet designs. not dual like the 3900X and 3950X.

    If there is any binning at all, it will just be for those which can sustain higher clocks given the extra headroom. In some ways, I expect the 3800X to actually be inferior ( like for instance with efficiency at lower clockspeeds). Just like the 2700 vs 2700X.

    I suspect - and this is supposition, but it makes sense - that the 3700X couldn't quite compete with the 9900k, and so AMD also gives us the 3800X option which would - albeit with a little extra power and thermal headroom. If the 3700X more or less tied the 9900k in Cinebench in the CES demo, which favors Ryzen, then we can expect the 3700X to be a bit slower. AMD probably wants to break this image of them being slower at equivalent core counts, so we get a 3800X with enough extra juice to trade blows with the 9900k. But I expect (and AMD might, too) that the 3700X would be the volume seller.
     
  27. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    8,819
    Joined:
    Apr 22, 2006
    Where is there any evidence of this?
     
  28. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Mar 13, 2017
    The_Stilt discusses it here: https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-72

    The money quote regarding leakage/power consumption:

    "Because of that, the 2700X SKUs most likely uses silicon which features higher SIDD (static leakage) than the average silicon used in other lower clocked SKUs. Silicon which features higher SIDD will most of the time require less voltage to reach certain frequency, compared to silicon with lower leakage characteristics. However, despite the lower voltage on the higher SIDD silicon, the power consumption is usually slightly higher due to the higher current draw and temperatures (and therefore lower overall efficiency)."

    Translation: some silicon can hit higher clockspeeds at lower voltage, but that silicon tends to draw higher power (than other silicon) at lower frequencies. A 2700 won't clock as high as a 2700X at a given voltage, but it will be more efficient at lower clockspeeds. AMD can bin for this with Zen 2, as well - which is what I suspect the difference is between the 3700X and 3800X.
     
  29. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    8,819
    Joined:
    Apr 22, 2006
    Seriously? Using speculation of someone on another forum as "evidence". He didn't even test a 2700, he was just guessing. :rolleyes:

    In reality, typically the higher binned parts, tend to run higher clocks at lower voltage and less power, those same parts tend to exhibit that same behavior when clocked lower, IOW they are better everywhere.

    What I think both of you are confusing, is when you have a specific "low power process(LPP)", that uses less power at lower frequencies than the "high power process(HPP)". But LPP falls apart at higher clock speeds. Likewise HPP tends to be inefficient at lower clock speeds.

    But this doesn't translate to binned parts in the same process, top binned parts just tend to be "best of" process examples and are better everywhere.

    3700X = 2700. A lower clocked part, clamped to a lower power. If you manually overclock they would be about the same as their higher clocked counterpart, but then lacking the features that reduce clocks and power under lighter loads.

    Just like for 2700/2700X, with 3700X/3800X you get the 3700X if you want to manually overclock, though you will end up using more power (require more cooling) than what the 3800X will give you automatically.
     
  30. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Mar 13, 2017
    Reading comprehension, mang. This is regarding CPUs on the same process, so it's not a confusion of LPP vs HPP.

    He did test the 1700 vs the 1800X earlier in that thread, and those had the characteristics he described. He also tested earlier Bulldozer binnings and those had the characteristics too. He says that in the link I sent you. AMD has a history of binning this way. He extrapolated re: 2700 vs 2700X. The Stilt has a history of being right on these things, too. He wrote the book on Ryzen memory timings and speeds. In an earlier thread you accused me of claiming I was an expert. I wasn't, and am not. But The Stilt is.

    The part with lower SIDD will not clock as high (unless you give it extra voltage vs. the part with higher SIDD), but will be more efficient at lower clocks. It's not really a hard concept to grasp. Binning is more complicated than just "this one is better in all the things."

    Edit: Hell, even Juanrga, Intel shill that he is, generally accepts The Stilt's conclusions on these matters.
     
    Last edited: Jun 26, 2019
  31. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    8,819
    Joined:
    Apr 22, 2006
    This is guessing based on a belief that the same process work like the difference between LPP and HPP.

    Testing one or two samples says nothing. Especially 1700 vs 1800x. Anyone paying attention in those days knows there wasn't much binning at all. 1700 often over-clocked higher than 1800x. It was very much a silicon lottery.

    Evidence would require a demonstrated consistent pattern, and guessing based on the result testing a couple of 1700/1800x parts is laughable. :D
     
  32. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Mar 13, 2017
    Well, you're the expert, right?

    See, at the end of your post, you say "a consistent pattern" involving many samples would be required to make a conclusion. And then you make a conclusion re: not much binning on the basis of no evidence of that kind. As it so happens, this is easier to look up than power efficiency at low clockspeeds. HWBot has this data: https://hwbot.org/compare/processors#5395_1,5392_1-3,14.

    Average OC of the 1800X is noticeably higher than the 1700. Conclusion: they were binned.

    You are confusing "proof" with "evidence". One sample is evidence, albeit not necessarily very convincing evidence. One sample is not proof. As it so happens, he tested a few Ryzen CPUs, and also a few Bulldozer CPUs, and noticed the same pattern in all of his samples. He also supplied a logical technical explanation for the results he saw. Believe or disbelieve if you wish. But he is an expert on this shit, and he's been spot-on with everything else. If you require a test of, say, a few hundred underclocked CPUs measured for power efficiency, so far as I know nobody has done that. Feel free to do it yourself if you want, though.
     
    TheFlayedMan likes this.
  33. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    8,819
    Joined:
    Apr 22, 2006
    Then show that actual information. One sample is weak evidence at best.

    But you didn't even provide that. You provided a link to speculation, not the actual weak "evidence".
     
  34. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Mar 13, 2017
    It's all in that thread. He dissects Zen's characteristics pretty thoroughly there, and has charts/data for all of his various tests, including the power curves. Read it, or don't.

    You demand that I adhere to a standard of evidence you yourself aren't willing to provide for your own assertions, and then don't even want to read what I link you to. ¯\_(ツ)_/¯
     
  35. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    8,819
    Joined:
    Apr 22, 2006
    Gee, why wouldn't I want to read 2000+ post haystack in search of needle of thin evidence? :rolleyes:

    If you ever actually find the appropriate post, I will be happy to read it.
     
  36. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Mar 13, 2017
    Filter by "The Stilt". Everything else in that thread is just noise/comments. I found you the original I was referencing when you asked for it, if you want more, I'm not your personal search engine. But it's in there.

    Also if you want to read more on the concept he is referencing, you can familiarize yourself with static leakage here: http://www.ruf.rice.edu/~mobile/elec518/readings/DevicesAndCircuits/kim03leakage.pdf

    It's been a big problem for a while now and continues to get more difficult to manage with smaller nodes.
     
  37. dexvx

    dexvx [H]ard|Gawd

    Messages:
    1,091
    Joined:
    Aug 14, 2002
    So going by hwbot's numbers, the 2700x's average air overclock is about 220 MHz more than the 2700 which is about the same as the 1700 vs 1800x.

    Below chart indicates that 1700 is *worse* at efficiency than the 1800x at the same clock speeds (unless I'm reading it wrong). Of course that's limited samples.
    https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-73
     
    DuronBurgerMan likes this.
  38. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Mar 13, 2017
    Correct. At high clock speeds, the 1700 will show worse efficiency than the 1800X. That's part of the binning process for the 1800X. The dispute is over what happens at low clock speeds. For that, we would have to underclock 1800Xs to behave like 1700s, and then measure power draw accordingly.

    This isn't the first time this topic has come up in these forums, either: https://hardforum.com/threads/amd-launches-zen-12nm-ryzen-and-x470-motherboards.1952044/page-15
     
  39. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    8,819
    Joined:
    Apr 22, 2006
    Filtering in a thread doesn't appear available to non-members. I can see it here at [H], but not at Anand where I'm not a member.


    I know about static leakage, it was a huge issue around 28nm transition until FinFets were used to mitigate to tolerable levels.

    But nothing says that a processor on the same process, that uses less power at high frequency (when compared to another) would reverse that trend at low power.

    Top binned parts, tend to be winners in the silicon lottery that run at lower voltage across the board.

    So if you want to compare power at lower speed, both parts should be set to their minimum stable voltage for that clock speed, which should be lower for the better binned part, which should equate lower power. Lower supply voltage leads to lower static power leakage, and lower operating power.
     
  40. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Mar 13, 2017
    Good.

    Remember when discussing power draw, we're not talking about just voltage. We are also talking about current and thermals, and thus implicitly resistance as well (remember Ohm's Law).

    Two tests would need to be conducted, at both extremely high clocks/power, and extremely low clocks/power - and thus it's usually only extreme overclockers (The_Stilt used to hold the AMD record - don't know if he still does) who care overmuch about this and would have this data - well, aside from whoever does the binning at AMD and Intel (and presumably Nvidia as well - this applies to all silicon at competitive nodes these days, not just CPUs).

    Test 1 would be to compare two SKUs on the same process at extremely high clocks with a consistent level of stability, and observe power draw at the wall.
    Test 2 would be to compare the same two SKUs at extremely low clocks, with minimized power, with a consistent level of stability, and again, observe power draw at the wall.

    This needs to be done with wall power as your measurement criteria, because voltage alone may be bullshitting you because of how static leakage works.

    The idea being, of course, to create as much spread as possible to exaggerate the difference in order to get your results outside the margin of error.