Has AMD passed Intel or just pulled even or still slightly behind

Discussion in 'AMD Processors' started by Wolverine2349, Nov 29, 2019.

  1. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,347
    Joined:
    Mar 13, 2017
    The 3900X is held back by the "shitlet". One good chiplet for high boost clocks, and one shitlet that's probably no better than a 3600 or 3600X to get that core count up. The shitlet holds back all core OC.

    From what Gamers Nexus was achieving on their 3950X (4.4 stable and they said 4.5 looked possible as stable with some more tweaking - they said it was ALMOST stable at 4.5), it's possible both chiplets on that are pretty golden. Might get better OC results. Then again, perhaps not - perhaps GN just got lucky.
     
    N4CR and Brackle like this.
  2. Mega6

    Mega6 2[H]4U

    Messages:
    2,173
    Joined:
    Aug 13, 2017
    Chiplet and Shitlet. I was wondering what AMD used to build these things.
     
    mikeo, Verado and DuronBurgerMan like this.
  3. Dan_D

    Dan_D [H]ard as it Gets

    Messages:
    55,035
    Joined:
    Feb 9, 2002
    GN got a sample from AMD. The samples they send out are likely cherry picked examples of the CPU in question. Review 3900X's were topping out at 4.3GHz. However retail CPU's often were reportedly only getting 4.1GHz or 4.2GHz all core overclocks. Silicon lottery reported pretty awful results on 3900X's as a whole saying few did 4.1GHz all core. The test CPU AMD sent me will do 4.3GHz all core and that's it. Why? A chiplet and a shitlet. One good die and one that's mediocre as hell. I'm getting a retail 3950X and not a cherry picked sample so we'll see how it goes but I'm not optimistic about it.
     
    IdiotInCharge likes this.
  4. noko

    noko [H]ardness Supreme

    Messages:
    4,417
    Joined:
    Apr 14, 2010
    What one gets from the store maybe more random, anyways my Bestbuy 3900x will do over 4.5ghz on one Chiplet and 4.3ghz+ on the Shitlet testing with all cores. Anyways if I bothered to run games on the fastest chiplet I am sure performance numbers would look better -> I just tend to think that the gaming experience difference would not really matter in the end. I did test individual cores and on the good chiplet they would do 4.6ghz on each one and one slightly faster but anything higher with the voltage I was using failed - pushing 4.6ghz out of one chiplet maybe possible with a very good sample and some severely high voltages and probably a chiller -> doubt that it would be remotely worth it though.

    https://hardforum.com/threads/manually-overclocking-ryzen-2-3900x.1986951/
     
    N4CR likes this.
  5. Dan_D

    Dan_D [H]ard as it Gets

    Messages:
    55,035
    Joined:
    Feb 9, 2002
    Mine was the same way. Individual cores can do 4.5GHz or more on the chiplet but the shitlet is tapped out at 4.3GHz. Cooling doesn't seem to really be the issue either.
     
    N4CR and DuronBurgerMan like this.
  6. Brackle

    Brackle Old Timer

    Messages:
    7,264
    Joined:
    Jun 19, 2003
    From what they said, they received a retail sample. As did HUB Linus etc etc.
     
  7. Dan_D

    Dan_D [H]ard as it Gets

    Messages:
    55,035
    Joined:
    Feb 9, 2002
    Negative. AMD doesn't send retail samples. It sends retail packages which are marked as "not for resale" but are otherwise indistinguishable from retail packages. The CPU's themselves do not have any identifier which suggests they are different from retail chips. In contrast, Intel's CPU's always identify as "Engineering or ES" in CPU-Z. Thus, have a modified identifier. However, I have long suspected that these are cherry picked examples. I had a number of these CPU's and some retail CPU's for friends wanting Ryzen systems on hand. In every case, the CPU's I got from AMD were superior overclockers. They were also able to hit their advertised boost clocks, or damn close to them on some motherboard on day one. This was not the case with the retail CPU's which all acted the same on every board. In my 3700X review, I talked about this experience showcasing the difference between these chips.

    Now, I will admit that this is circumstantial evidence at best. Maybe I just got lucky with my samples, but so far, my experience is that actual retail CPU's do not clock as well as the "retail samples" you get from AMD. This has been the case with the 3700X and 3600X's I've received / worked with. By most of the accounts I've read, the 3900X sample is better than the bulk of retail chips as well. That's three for three on that. I don't think I'm all that lucky either as I've had some pretty dismal overclockers over the years. I'm the only person I know of to buy a Celeron 300A that wouldn't hit 450MHz. My second one would, but my first one didn't.

    Like I said, I have no evidence this is the case, but so far that's what my experiences point to.
     
    IdiotInCharge likes this.
  8. Gideon

    Gideon 2[H]4U

    Messages:
    2,360
    Joined:
    Apr 13, 2006
    My retail 3900X will go to 4.4 but voltage has to be 1.4 volts which is not where I want to run it at all the time but easily does 4.3, however I just leave it alone as I prefer the way it runs when the chip is in control of the clocks. I wish I could just tell PBO what clocks I want to run at for however many cores are utilized then I would be happy tinkering with that.
     
  9. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,347
    Joined:
    Mar 13, 2017
    I don't even bother with PBO. Just leave it stock and decently cooled, and it's fine. Overclocking these just isn't worth it. Kinda takes some of the fun out of it, TBH.
     
  10. bobzdar

    bobzdar [H]ard|Gawd

    Messages:
    1,662
    Joined:
    Jun 6, 2003
    Well - would disabling the shitlet for gaming and then turning it back on for productivity provide a boost? Even if a minor one, it'd mean a true no compromise setup. Thinking about trying that with my 3950 as there are a whopping zero games that use more than 8 cores. So run the golden chiplet at 4.5-4.6 for games and then turn the shitlet back on when it's time to do video editing and photogrammetry.
     
  11. Gideon

    Gideon 2[H]4U

    Messages:
    2,360
    Joined:
    Apr 13, 2006
    I dont think the 3950 is going to have a shitlet, both chiplets should be of good quality.
     
  12. dave343

    dave343 [H]ard|Gawd

    Messages:
    1,631
    Joined:
    Oct 17, 2000
    no... definitely not the only person ha, I had a 300A that wouldn't do 450 either, which at the time with how hot these CPU's were, and the almost 100% promise they all did 450 was a surprise. But then I got lucky and had a Celeron 600 that did over 1ghz but I remember having to push 2V :ROFLMAO: (mind you I ran a peltier at the time)
     
  13. kac77

    kac77 2[H]4U

    Messages:
    2,290
    Joined:
    Dec 13, 2008
    The other problem with the Intel HEDT is leaving out ECC for the chips recently released. For all the talk of AMD raising prices, the truth is if you want a full-on system that gives you the most features AMD still gives the end-user a bang for the buck.

    You want ECC on the Intel HEDT? Then you will spend an extra $1000 to $2,000 for the chip. You can't do scientific calculations with out it. gaming is wonderful and all but that is not the only metric that these systems are built for. Hell even in the consumer space, gaming isn't the only thing these systems are built for. They call it general computing for a reason.

    Running AVX 512 without ECC seems pretty counter intuitive. And ECC is on every AMD chip from the top to the bottom.
     
  14. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,394
    Joined:
    Jun 13, 2003

    Not quite ;)
     
  15. kac77

    kac77 2[H]4U

    Messages:
    2,290
    Joined:
    Dec 13, 2008
    Which are missing it? Maybe the 3000G but that's like rock bottom
     
  16. somebrains

    somebrains Gawd

    Messages:
    821
    Joined:
    Nov 10, 2013
    My buddies were all building Celeron A back then.
    Issues clocking were using free Fry's mobos like Soyo or ECS.

    CS was my thing so I built a Duron box with a golden orb. StarCraft guys all used Celerons.

    On that note AMD might be in another K6-2/Duron-Athlon-Athlon64.
    Eventually Intel will hit their arc, which historically was longer then AMDs.
    AMD consumers like sticking to a single socket to define cpu arc, AMD is going to have to change those assumptions in the consumer segment like they're doing with Threadripper.

    The manner in which Asmedia executes the move for AMD motherboard chipsets will be interesting to watch.
     
    Last edited: Dec 6, 2019 at 2:09 PM
  17. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,394
    Joined:
    Jun 13, 2003
    The APU lineup. They're perfect for small servers, you know, except...
     
  18. kac77

    kac77 2[H]4U

    Messages:
    2,290
    Joined:
    Dec 13, 2008
    Yeah I figured the APU's. Those are like 60 bucks. Actually the 3000G is 50 bucks. But spending $119 gets me the ECC support and two extra cores. I have tons of extra video cards laying about so and extra $70 bucks goes a long way.
     
  19. OFaceSIG

    OFaceSIG 2[H]4U

    Messages:
    2,108
    Joined:
    Aug 31, 2009
    How would you specifically only disable the shitlet? Just turn off the second half of the core count? Would it line up directly that way?

    That's assuming the bad one is the second one.
     
  20. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,394
    Joined:
    Jun 13, 2003
    This assumes that you want a GPU at all -- that's kind of the point of an APU.
     
  21. kac77

    kac77 2[H]4U

    Messages:
    2,290
    Joined:
    Dec 13, 2008
    It's not really necessary for a headless server though.
     
  22. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,394
    Joined:
    Jun 13, 2003
    They why bother with a GPU at all?
     
  23. somebrains

    somebrains Gawd

    Messages:
    821
    Joined:
    Nov 10, 2013
    You need something if you are going to work locally on headless gear.

    I'm of the "snapshot it's local state for further examination, if all hardware diagnostics show green redeploy from golden image or last known clean updated state.". But that's at work where I'm dealing with rows of compute.
     
  24. kac77

    kac77 2[H]4U

    Messages:
    2,290
    Joined:
    Dec 13, 2008
    The GPU is needed just to configure the server. After that it can sit in a dust bin.

    Full server motherboards have the GPU on the motherboard. But if you need a server in a pinch for considerably less money then AMD fits the bill quite easily.
     
  25. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,394
    Joined:
    Jun 13, 2003
    At that point, so does Intel.
     
  26. bobzdar

    bobzdar [H]ard|Gawd

    Messages:
    1,662
    Joined:
    Jun 6, 2003
    I suppose I'd clock it low and tell core prio to not use it so it'd stay idle and not generate heat.
     
  27. kac77

    kac77 2[H]4U

    Messages:
    2,290
    Joined:
    Dec 13, 2008
    you need ECC. That's required for storage server and many other things.
     
    N4CR likes this.
  28. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,394
    Joined:
    Jun 13, 2003
    And I can get an i3 or quad-core Xeon with that, lol.
     
  29. tweaked

    tweaked n00b

    Messages:
    53
    Joined:
    Sep 28, 2004
    I love how you morons are just blindly running with this shitlet narrative.

    Both my chiplets clock to 4.45 at 1.33 v and boost to 4.6 when using PBO.

    I've yet to see anyone complain about a "shit" second core.
     
    IdiotInCharge likes this.
  30. kac77

    kac77 2[H]4U

    Messages:
    2,290
    Joined:
    Dec 13, 2008
    Yup and both of them are slower.
     
    N4CR likes this.
  31. somebrains

    somebrains Gawd

    Messages:
    821
    Joined:
    Nov 10, 2013
    Hence if you look around a cage you should find a couple basic gpus for host troubleshooting.
    Granted thats for out of service contract gear we keep for test or experiments.
     
  32. Nenu

    Nenu [H]ardened

    Messages:
    19,038
    Joined:
    Apr 28, 2007
    Strange, I recall reading about it.
    Now where was that ...
     
  33. N4CR

    N4CR 2[H]4U

    Messages:
    4,092
    Joined:
    Oct 17, 2011
    Late stepping 3600xs are getting 4.4 more often now So its likely later 3900 and 3950s will be similar soon.
     
    IdiotInCharge likes this.
  34. N4CR

    N4CR 2[H]4U

    Messages:
    4,092
    Joined:
    Oct 17, 2011
    It does have a shitlet just like the 3900x but better on average.
    As tweaked said not all are that way now especially with later sillicon...
    Now you see one reason why AMD delayed the release..
     
    Last edited: Dec 7, 2019 at 4:37 AM
  35. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,394
    Joined:
    Jun 13, 2003
    Than an AMD APU? Hardly.
     
    drescherjm likes this.
  36. FlawleZ

    FlawleZ Gawd

    Messages:
    826
    Joined:
    Oct 20, 2010
    I think on average yields will improve in quality over time and the average stable OC we see today will rise some within the next year or so.