Intel person thinking of AMD move. Please explain.

Discussion in 'AMD x86 CPUs and chipsets' started by vladtepes, Sep 9, 2019.

  1. Elyzion

    Elyzion Member

    Joined:
    Oct 27, 2004
    Messages:
    7,434
    Location:
    Singapore
    I wouldn't trust anything he has to say. He says what ever he needs to say to get sponsors/advertisers.
     
  2. OJR

    OJR Member

    Joined:
    Jan 19, 2013
    Messages:
    4,952
    Location:
    Melbourne
    I don't mind his videos on older hardware and retro builds.
     
  3. Claymen

    Claymen Member

    Joined:
    Jan 27, 2002
    Messages:
    103
    Location:
    Perth, WA
    Just did something similar myself. Been running on a water cooled 4770K for about 6 years. Delidded and all the usual stuff to get that bit more out of it, though couldn't really get past about 4.5ghz without stability issues. In any case, have picked up a 3950x, x570 Master, and 4x 16GB 3600 CL16 (16-16-16-36 B-die). I was honestly considering a 10900k but with the lack of stock, motherboard prices and potential heat loads, felt like the AMD was the better option overall. Especially as this is effectively my work PC, so overall performance is key.

    In any case, I've been very happy with the 3950x system, and the overall polish on the current AMD processors. Might at some stage have a fiddle with overclocking a bit more, but have been pretty time poor so it was a case of get it in, flick on XMP, enable PBO/AutoOC and call it a day to get on with work.
     
  4. di_entropy

    di_entropy Member

    Joined:
    Jun 27, 2001
    Messages:
    1,352
    Location:
    Melbourne
    Nice, that is a beastly setup. I thought my 32GB of RAM was overkill, what are you going to do with 64GB. :lol::thumbup:

    I set up my new rig over the weekend and have been overclocking and fine tuning since.

    Currently have the 3800X at 4.5GHz manual all core OC and RAM at 3800MHz CL16. Still trying to find optimal heat / stability though. Started crashing a bunch today while stress testing and just realised it was ever since I put the side panel on for the first time. Swapped my top radiator so now both front and top are intakes and it is benching nearly 10deg cooler. LOL. Had been pulling my hair out wondering what had changed until I realised.

    Keen to know how you go with your OC results, I was actually getting better results at stock than with PBO and auto OC enabled. Much better results with manual all core OC now though.
     
  5. Claymen

    Claymen Member

    Joined:
    Jan 27, 2002
    Messages:
    103
    Location:
    Perth, WA
    I had 32GB on my old machine, and found that when running a few test VM's it would at a point start to impact overall performance. I support multiple customer environments, so having local VM's I can quickly test against is super useful. Whilst I do have some dev VM's on Azure, there is still an advantage to having things local. Plus I can mix and match what hypervisor I need (or even hypervisor in hypervisor), and can test some more specific situations, as an example, creating latency/packet loss to test replacement VPN products to see how well they cope.

    I've played a bit with PBO and AutoOC, as well as the apparent "bug" where you set EDC to super low values to force the boost algorithm to behave differently. At the moment I have mostly settled on forcing EDC to 20, and setting the PPT to 300, and TDC to 230 and so far that seems to have been the best combo. Like what the thread here reported (https://www.overclock.net/forum/13-amd-general/1741052-edc-1-pbo-turbo-boost.html) it seems to force the boost to work correctly so I now see 4.2-4.3 all core boost, whereas at stock settings it seemed to only go as high as 4.1~ which didn't seem right. Another nice side effect is that single core boosts are consistently higher on CCD0 which seems to be the better binned one. Seeing those cores boost to 4.7 far more consistently (or at least more often that I can capture it with HWInfo due to poll times/intervals).

    I have also tested upping the BCLK slightly to edge out just a hair more and get it more consistent at 100 (as often it will slightly drop under), and that seemed to also improve things slightly, though I've not done enough in depth testing to really confirm if it is truly stable. I can say it doesn't take much before things get a little weird which I put down to push that a bit too far.

    4.8.PNG

    clocks.PNG

    That said, I went overboard with cooling, running the EK-Quantum Momentum block for both my CPU + VRM, and have 2x 360x60mm radiators + 1x 280x30mm radiator in the loop.

    Whilst not definitive, I've been able to get some reasonable scores with CPU-Z to compare against each run quickly.

    Roughly the current config
    https://valid.x86.fr/iaxhxt

    Initial stock config
    https://valid.x86.fr/490kr1
     
    Last edited: Jul 15, 2020
  6. di_entropy

    di_entropy Member

    Joined:
    Jun 27, 2001
    Messages:
    1,352
    Location:
    Melbourne
    Yeah I was guessing lot's of VM's with that much RAM!

    Be careful with the EDC "bug", I also tested this on my system (forcing it to 1), but even though my reported clocks were much higher, the "Effective Clock" for each core in HWiNFO were actually not reaching full speed and my Cinebench R20 scores were actually LOWER than my regular PBO (and now far superior manual OC) scores.

    With that super low EDC setting you should keep an eye on your Effective Clocks while benching and compare with normal/default EDC values.

    What sort of max CPU temps are you getting after 10 mins of OCCT?
     
  7. Claymen

    Claymen Member

    Joined:
    Jan 27, 2002
    Messages:
    103
    Location:
    Perth, WA
    My understanding was effective clock is just the samples over time. As the normal clock is based on BCLK x Multi at that specific sample time, whereas effective is the hardware reporting the clocks over a time period helping to better view the clocks given how quickly the boost changes can happen, though still ultimately an average. Though I thought effective also included when things are sleeping, so may not be easy to use for tracking say a single core boost or when threads are shifted around lots of cores.

    I'm running an EDC of about 20, and all core boost was significantly better, and most of the quick benchmarks all benched higher than before.

    Had a quick test running it in the background, mostly hovers around 65-68c. My day to day temps are about 40-50~ when lightly loaded, depends as always on what is actually running. And should note, fans on low, fan noise drives me nuts, 6 of the 120mm's run at 900 rpm +/- 5%, not sure on the two 140's as they aren't running off my corsair link, need to work out which header they correspond to on the board sensors. The profile is set to low noise/speed though. All fans are push config.
     
    Last edited: Jul 15, 2020
  8. di_entropy

    di_entropy Member

    Joined:
    Jun 27, 2001
    Messages:
    1,352
    Location:
    Melbourne
    Yes, exactly, effective clock is sampled over the polling time of HWiNFO which is default of 2000ms. So when you're benching / stress testing you should see a continuous effective sustained clock for that 2000ms sample time, if you're not it is because you could be hitting power limit or thermal limits and the Ryzen architecture is "pulling back" the clock frequency at what could be microsecond intervals due to hitting these limits, which won't be picked up on the traditonal instant / discrete clock values that are reported.

    So for example, when running a bench like Cinebench R20 which will sustain your cores at 100% load for much longer than the 2000ms polling time, you should see your "current" effective clocks values stable and also hitting the the same frequencies as your regular core clocks throughout the whole bench.

    [​IMG]



    When I tried to do the EDC limitation trick, my core cocks were hitting up to 4550MHz on some cores, but the highest sustained effective clocks throughout the Cinebench run were something like 3900MHz.


    Regarding temps.. I must have some issues I need to sort out, hitting up to 80deg after a quick 10 min run of OCCT. :confused::Paranoid:
     
  9. Claymen

    Claymen Member

    Joined:
    Jan 27, 2002
    Messages:
    103
    Location:
    Perth, WA
    Cool, so that aligns with how I expected it to work. Yer I've kept an eye on both, but my reference with the standard clocks was to more look at general boosting (or more so, lightly threaded boosting) which granted it still needs to catch in the act, but was mainly to confirm that it was at least boosting like it should. All core boost I would expect to use effective clock more for. But in any case, effective during all core was sitting just under 4.3ghz (values like 4.292, 4.294 etc), some odd rounding/numbers as I'd been messing with BCLK.

    Weird, I'm not seeing that myself, but to be honest just goes to show that PBO and AutoOC are just... kinda broken. It's super inconsistent.

    I went fairly nuts with my cooling setup though :) Below is from when I had the 4770k, but it's the same case, cooling setup, and GPU. I'm not one for bling bling cooling setups, it's functional, was easy to fit, and clean enough for my general use. It was cleaner at one point, but after having a couple swivel fittings leak, I got rid of every single swivel fitting and now just have straight connectors, and made up the loss of routing by running slightly longer tubes.
    IMG_20191102_175415.jpg
     
    Last edited: Jul 15, 2020
  10. di_entropy

    di_entropy Member

    Joined:
    Jun 27, 2001
    Messages:
    1,352
    Location:
    Melbourne
    Old school barbs.. function over form, I get it. :lol: Plenty of cooling there.

    I have 1x EX420 top and 1x RX360 front rads.. should be plenty enough cooling for an 8c/16t Ryzen I would have thought, but still not getting great temps. Going to try a remount this weekend.

    IQcstES.jpg
     
  11. Claymen

    Claymen Member

    Joined:
    Jan 27, 2002
    Messages:
    103
    Location:
    Perth, WA
    Nice! And yer man, barbs are cheap, reliable, and hard to get wrong really. If they can work for engines running way higher water coolant temps (and pressure) and not leak, then should work fine on a PC. I'd have gone clear tubing as well but this (I think it's EK) tubing was less than half the cost, and was also in stock at the time. Super flexible and easy to work with too.

    Nice PSU there too, I've been running the AX1200i for a while now, after I went full nuts water cooling, including the GPU (1080ti), the next biggest noise was my old 750W PSU, under load the fans would spin right up. Which makes sense, after swapping in the AX1200i and reading the power draw data from it, I was pulling 550-600W total draw under max GPU+CPU load lol. Think it was an EVGA 750W, worked a treat to be honest, but obviously I couldn't hear the noise of it till I made everything else quiet lol.

    I do like the thick radiators, I know they can cause a bit more noise, or more so you get less cooling at lower fan speeds, vs thinner ones, so you have to run the fans faster to force more air through. But then they kind of act like a larger overall res as well, so ends up with quite a bit of water in the system acting as a nice buffer to even out temp changes. I had at one point seriously considered getting an Alphacool NexXxoS XT45 1260mm (the big giant thing) and mounting that to my desk as an external rad. But.. it was nice to have it all in the one case.
     
    Last edited: Jul 15, 2020

Share This Page

Advertisement: