65 Watt TDP 12-core Ryzen 9 3900 @ 3.1 GHz Coming?

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Sep 26, 2019.

  1. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    This is false. 3900X has 2 chiplets that have same dimensions as 3800X while stock power draw is only bit higher than 3800X.
    3900X may produce more heat under full load, but it has easier time to dissipate it.

    I am using NT-H1 as you are. But in case I'll ever get 3900X, I already have CM Master Gel Maker in shopping cart. (Just not to forget.)
    I think that they need CPU with same feature set and similar to target performance. (Which can be tuned.)
    And then they'll need GPU capable to do same. If not, then they need those instructions emulated in API to enable development of visual effects. Then do final performance tuning as proper device gets to them.

    I am pretty sure that CPU wise even 1700 would do. GPU wise, that's hard part as there is real development of new instructions with magnitude higher performance on given workloads.
     
    Last edited: Sep 29, 2019
  2. Evildead666

    Evildead666 Guest

    Messages:
    1,309
    Likes Received:
    277
    GPU:
    Vega64/EKWB/Noctua
    They have had the Devkits for some time now.
    Probably based on Navi, but with the RDNA2 drivers.
    iirc Navi was a bit of a hybrid between GCN and RDNA ?

    I agree, the CPU could be almost anything Ryzen based.
     
    Deleted member 213629 likes this.
  3. Evildead666

    Evildead666 Guest

    Messages:
    1,309
    Likes Received:
    277
    GPU:
    Vega64/EKWB/Noctua
    Basically, what i'm saying is that when we overclocked, the Wall would come progressively, back in the day.
    As time goes on, and the fab process tech has evolved, the wall comes faster and faster.
    You just get a little OC room, and the the voltages just have to go straight up a cliff, whereas back in the day, it was more like a light hill, to moderate incline, until you hit the cliff.
    (Member Phase change coolers ? :))
    I remember drooling over those Athlon systems running at insane speeds....;)

    Now its like a hockey stick, and the manufacturers are clocking all the way up to those speeds, because they can reliably.
    I suppose they've just got really good at fabbing now, and have dialled it in that way for profits, because it certainly makes the chips a lot more equal.
     
    Deleted member 213629 likes this.
  4. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,780
    Likes Received:
    1,393
    GPU:
    黃仁勳 stole my 4090
    So what is false then? But your idea that it's easier to dissipate is wrong, unless you mean easier than if it was a single monolithic monstrosity.
     
    Deleted member 213629 likes this.

  5. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    Two Chiplets => Double Area. Let it sink in.
     
    Deleted member 213629 likes this.
  6. wavetrex

    wavetrex Ancient Guru

    Messages:
    2,450
    Likes Received:
    2,543
    GPU:
    TUF 6800XT OC
    Deleted member 213629 likes this.
  7. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,780
    Likes Received:
    1,393
    GPU:
    黃仁勳 stole my 4090
    You're thinking of it as if it's the same amount heat, from the same object, with double surface area making it easier to dissipate. No, it's actually more challenging to cool. Double the chip surface area under the same heat spreader, with more than double the heat being produced. And overclocking them is an entirely different world, it's not an apples to apples comparison and they don't remotely scale linearly compared to each other. It's not comparable. Spoiler alert: The one with more crap underneath is a bigger pain and hotter.

    Regardless of that, I don't know WTF about my post you were saying was false since that's not what it was talking about.

    The fact remains that a 3900X with a drop too much voltage is a procreating nuclear bomb going off. I have one, I don't need you telling me that I'm imagining the glass crater it caused.

    Edit: I forgot to mention, you should forget about CM's Master Gel, look at Hilbert's benchmarks. It won't give you any real gain over NT-H1.
     
    Last edited: Oct 1, 2019
  8. Things could change but Zen 3 has been slated for 2020 - according to multiple road-map slides I've seen and articles I've noticed. The design phase is finished. Issuing developer consoles (dev-kits) was done in the past with the Xbox etc and wouldn't be unusual to have done again. My comment was mere speculation. If AMD wanted to push Zen 3 in there - it could be done for 2020 (Issuing Engineering samples in said dev-kits). Developers often don't even release finalized products lol. Don't get me wrong I liked your post; I chuckled a bit because previous console gens did have pre-release SDKs etc that didn't reflect the final console release.:)

    [​IMG]
     
    Last edited by a moderator: Oct 2, 2019
  9. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    Turn your NH-D15 by 90°, so same heatpipe goes over both chiplets.
    If your IHS is magical heat spreading device, temperature will be same. as if different heatpipes go over each of chiplets.

    False was statement that 3900X is running hot as a Sun. Fact is that when 3800X outputs 200W during some OC, it does it with one chiplet and 3900X has 2 of them. Think about current too, if you can.

    And you are wrong with Master Gel too. Especially since I stated "Maker" variant.
    Take delta T to ambient and ability to shave extra 2~3°C in situations when heat goes into unpleasant zone. That matters the most. And price difference vs. volume of TIM in syringe makes it simply good choice.
     
    Last edited: Oct 2, 2019
  10. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,780
    Likes Received:
    1,393
    GPU:
    黃仁勳 stole my 4090
    Thanks for pointing out the Master Gel/Maker difference. But the 3900X is as hot as the sun. When seeing how far I can push it, it had a spike to 120C and I immediately shut it down. A small voltage difference to go for 4.4GHz as opposed to 4.2GHz pushes it into the 90s towards 100C and failure in certain programs (especially anything Cinema4D related, or video encoding) on air. As opposed to like 70C with that tiny bit less voltage. A Noctua NH-D15 can't handle it, even with the fan at 100%. And I'd have to check to verify but I'm pretty sure the heat pipes are as close to the chiplets as they can be.

    More surface area does make a huge difference. Each CCX holds 3 cores so that's a big difference vs concentrating a 4th one in the same spot. I don't know where you heard this 200W number but l've never seen my chip go over 150W if RTSS is correct. Whatever the number is, it definitely does not come anywhere close to 400W for both chiplets. A modern motherboard has hard limits on the current and voltage spikes unless you disable or loosen them.
     
    Last edited: Oct 12, 2019

  11. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    I made given post to kindly hint you that you write nonsensical things.

    For children version:
    3800X is single Chiplet CPU which does not overheat even while pulling more current than your 3900X where current splits into 2 chiplets.
    I did not write anything about 400W, that's just your assumption.
    I wonder too why you even had opportunity to shutdown your system as CPU reached 120°C, it should have cut power hard way to protect CPU.
    150W you wrote as your power draw in nothing crazy for D15, I had no problem to cool more with D14. And my 2700X stayed under 90°C even at 220W.

    Your another assumption is that thing about hard limits on current and voltage on "modern" MBs. You set voltage, components have resistivity, current happens unless limited (which happens by default PBO or not). Heat happens thanks to current. Heat alters resistivity, that alters current. If I wanted and CPU would not melt, I could easily pull 400W via VRMs of my board.
    And manual PBO settings would likely enable me to do so.

    I am telling you whole time, that area vs. heat, you should not have any issues with cooling unless you do something wrongly.
     
  12. My system / with 3700X went past TJMaxx under stress-test & wouldn't self hard-throttle either. Had to manually terminate. There's an engineering flaw in the current microcode out there. (This was on a B450) anyways - I posted specifics, Anyone is saving money for a Matisse build??
     
  13. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    At 120°C, it is not about throttling. It is about immediate power cut off => hard shutdown. Same way as Graphics card does when it detects that GPU or VRMs reached critical temperature.

    Btw, default throttle limit temperature in my BIOS is 95°C, so jumping over it to 96°C can be called measurement error. Especially since it is rather common that even under no load there are measurement spikes of 10+ °C with Zen.
     
  14. Great got it - wrong terminology on my part - point being if it's measuring it wrong or not; it's still not happening. No hard shutdowns or even attempts to throttle from 4GHz down to standard clocks to avoid overheating. It's a problem. It's not good for long-term health of the chip either. There's no going back or forth on this. It just simply is.

    The measurement spikes were taken into account in previous AGESA updates & even prior to those being released AMD stated Ryzen Master shows *actual* - which is what I used to record the tests I ran and posted.
     
    Last edited by a moderator: Oct 14, 2019
  15. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,780
    Likes Received:
    1,393
    GPU:
    黃仁勳 stole my 4090
    - Okay?
    - You said 200W for a chiplet on a 3800X is no problem to cool. You do realize that it draws more current the more cores there are? Even with only having 3 per CCX, 6 per chiplet, as opposed to 4 per CCX, that's going to get hot when it's 12 total with SMT on. Yes, SMT makes a big difference in heat, when you have SMT on 12 cores it's more heat added than SMT enabled on an 8 core chip. My temps drop 35C+ by turning off SMT. In video encoding and Cinema4D tests, from 95C to 60C or lower.
    - I didn't change the default safety limit for temperature, it should have shut off when it went over 100C. It does work, I've had it shut off in other tests that were about to go over 100, I don't know why it didn't work that time. It's an X470 board.
    - I don't know why it's hard for you to understand; the 3900X is not comparable to an 8 core chip.
    I checked the heatsink. Get it through your head. This chip is hot as hell. There is nothing wrong with how I mounted it, the fans are going full blast, the case has really good air flow, the thermal compound is good, and the room is like 19C to begin with. I don't know why you keep writing back to me what basically amounts to "nuh uh". I have it, you don't. Why would I make up some random numbers and post them on a forum? Does Intel pay me?

    I hope I'm doing something wrong so I can drop the temps on this miniature supernova I have. What am I doing wrong if it's not the mounting/paste/fans/airflow, oh wise one?
     

  16. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    It is easy to understand. draw 150W via 3800X and then do it via 3900X. 3900X has double chiplet area to dissipate same amount of heat.
    If you can't operate 3900X at 150W, then nobody can operate 3800X at 150W under remotely similar conditions.
     

Share This Page