Clevo Overclocker's Lounge

Discussion in 'Sager/Clevo Reviews & Owners' Lounges' started by Spartan, Mar 4, 2016.

  1. jaybee83

    jaybee83 Biotech-Doc

    Reputations:
    4,100
    Messages:
    11,564
    Likes Received:
    9,126
    Trophy Points:
    931
    indeed. at those speeds the sticks themselves are not at fault, neither is the CPU imc. both are very capable, just our laptop mobo traces suck :p
     
    jc_denton and Scerate like this.
  2. Meaker@Sager

    Meaker@Sager Company Representative

    Reputations:
    8,651
    Messages:
    54,002
    Likes Received:
    15,645
    Trophy Points:
    931
    The fact they go as far as they do is impressive to me. Desktops have a far easier time and less noisy sockets to boot.
     
    Scerate and jaybee83 like this.
  3. jc_denton

    jc_denton V̖̟en̰g̻̼̰̩͙ea̲n̪c̭e̼ ͍̘̤͓̟̤Is̙͔ ̤Mi̻̭̣͎͍ͅn̰ḙ

    Reputations:
    5,040
    Messages:
    2,088
    Likes Received:
    3,138
    Trophy Points:
    281
    Also, I needed 1.2v VccSA/VccIO for 3600Mhz to boot after applying my baseline settings for 5Ghz (-185mV)
    Not sure if undervolting messes with it.

    It wasn't too long ago where I tried breaking 3400 with these exact sticks but failed. Only recently to find a workaround. I wonder, if in the future someone might figure out a way to even break 3600/3700Mhz?
     
    Last edited: Dec 12, 2019
    jaybee83 likes this.
  4. jc_denton

    jc_denton V̖̟en̰g̻̼̰̩͙ea̲n̪c̭e̼ ͍̘̤͓̟̤Is̙͔ ̤Mi̻̭̣͎͍ͅn̰ḙ

    Reputations:
    5,040
    Messages:
    2,088
    Likes Received:
    3,138
    Trophy Points:
    281
    Update on the 3600Mhz:

    Seems like this is the trace limit, be it to signal/noise or the trace quality for this speed. Testing done with cold ambient in order to keep the B-Die cool enough.

    3600Mhz only boots at:
    CL17-17-17-39 at 1.25v
    CL16-18-18-36 at 1.35v
    CL16-16-16-36 at 1.45v
    "auto" 24-25-25 at 1.25v


    Some of the timings pass Windows Memory Diagnostic (lol) however, all of them report errors when put under load with MemTest. Variations of the above with auto or lose timings, also puke errors. After days of dialing in the secondary/tertiary timings, it can go through without errors only to freeze at some point during MemTest.

    3460Mhz displays the same behavior, also being voltage sensitive. So I expect the same results as above and compromised stability.

    Comparing gains going from 3333Mhz to 3600Mhz at 5Ghz x47 cache:

    3333Mhz CL14-15-15-28 (276) vdimm: 1.35v / VCCSA/VCCIO: Stock

    Read: 50762
    Write: 52408
    Copy: 50335
    Latency: 39.1 ns
    https://i.imgur.com/8yq3ehP.png

    Theoretical max read bandwidth: 128*3333/8= 53328 mb/s
    50762 is within 95% of theoretical limit.

    3600Mhz CL16-16-16-36 (535) vdimm: 1.45v / VCCSA/VCCIO: 1.2v
    Read: 54481 (+7.33%)
    Write: 56114 (+7.07%)
    Copy: 53535 (+6.36%)
    Latency: 39.1 ns
    https://i.imgur.com/9y6i7mS.png

    Theoretical max read bandwidth: 128*3600/8= 57600 mb/s
    54481 is within 95% of theoretical limit.

    For benchmarks such as FireStrike/TimeSpy that 3600Mhz passed, scores were worse or close to that of 3333Mhz.
    Though it might not be stable in the end, only Clevo laptops have been able to reach speeds such as these

    Closing notes:
    After thinking about the stability and similarity of error's/freezes, it might be due to retraining the ram after the initial fn+d reset. Ideally you would set the timings/voltage and test the stability right away. Changing timings after the initial reset might cause incorrect training from that point forward, as some values have already been pre-populated. This is something that I need to test in the future and see if this is true.
     
    Last edited: Dec 12, 2019
    Scerate and jaybee83 like this.
  5. jaybee83

    jaybee83 Biotech-Doc

    Reputations:
    4,100
    Messages:
    11,564
    Likes Received:
    9,126
    Trophy Points:
    931
    maybe use tm5 for testing instead of memtest. especially the extended suite put together by 1usmus.

    Sent from my Huawei Mate 20 X EVR-AL00 using Tapatalk
     
    Scerate and jc_denton like this.
  6. Meaker@Sager

    Meaker@Sager Company Representative

    Reputations:
    8,651
    Messages:
    54,002
    Likes Received:
    15,645
    Trophy Points:
    931
    Each 100mhz will be exponentially harder than the last.
     
    jaybee83 and Scerate like this.
  7. jc_denton

    jc_denton V̖̟en̰g̻̼̰̩͙ea̲n̪c̭e̼ ͍̘̤͓̟̤Is̙͔ ̤Mi̻̭̣͎͍ͅn̰ḙ

    Reputations:
    5,040
    Messages:
    2,088
    Likes Received:
    3,138
    Trophy Points:
    281
    They passed Windows Memory Diagnostic but when any real work needs to be done they crap out. So even if they pass tm5, it wouldn't change the fact they are unstable for any real world use.

    Going from 3000-3333 was nothing, heck I even had my RipJaws running 3333Mhz in the p775tm1 memtest stable. But these +3400Mhz speeds sure have a temper.

    I've reached out to bro @Johnksss again, he mentioned that his 3800 sticks were happiest at 1.25v, going above that would just make them unstable. That confirms my own findings that voltage is the key, when trying to break the 3400mhz speed barrier.

    It becomes a balancing act to get 3600Mhz stable at around 1.25v
     
  8. tps3443

    tps3443 Notebook Deity

    Reputations:
    317
    Messages:
    1,112
    Likes Received:
    491
    Trophy Points:
    101
    I’m going to lap my IHS to a mirror polish this weekend. And I’ll finish lapping my heatsink too. I want them both level and even and making perfect contact.

    If I can knock off about 5C I’ll be good to run BFV at 5GHZ with no AVX offset. I’m so close. 4.8Ghz solid with 2 hours straight and temps run around the 80’s. And will peak in the low 90’s under those extreme temp spikes.
     
  9. jc_denton

    jc_denton V̖̟en̰g̻̼̰̩͙ea̲n̪c̭e̼ ͍̘̤͓̟̤Is̙͔ ̤Mi̻̭̣͎͍ͅn̰ḙ

    Reputations:
    5,040
    Messages:
    2,088
    Likes Received:
    3,138
    Trophy Points:
    281
    I'd consider getting a copper ihs instead, that way you can still retain warranty or resell the chip with the original intel markings.
    And if you're planning on using LM, you don't want a mirror polish as @Falkentyne mentioned, it's better to have a slightly scuffed surface for it to grip onto.

    Btw, try running BF:V without the bottom panel and see how it affects the temps. I'd expect 5-10c lower temps just by improving the airflow.
     
    Scerate, Papusan and Falkentyne like this.
  10. Falkentyne

    Falkentyne Notebook Prophet

    Reputations:
    7,765
    Messages:
    5,672
    Likes Received:
    7,984
    Trophy Points:
    681
    Yes, wiping the die, ihs or top of ihs, if LM is used there, all with 1500 grit dry sandpaper, then cleaning with 91% alcohol before applying LM will give outstanding results. Make sure there are no foreign particles or chunks of anything when spreading it evenly and after it is scuffed, spend some time doing long and even wipes to work the LM fully into the recesses of the scuffs fully. Just do not press hard at all on the applicator and make sure it is lint free, as you don't want micro copper or nickel particles coming off. If you feel any sort of resistance at all or roughness when wiping, there are particles of something and you need to clean and start over. Doing the optional LM oven IHS bake method is extra steps but you only need a lint free coffee liner to remove the LM after the first oven bake if you do that. Any sanding should be done before the oven bake of course.

    If you can do this perfectly you will get excellent results.
    BTW these applicators are just god tier.

    https://www.amazon.com/gp/product/B00MEC5084/
     
    Scerate, Papusan and jc_denton like this.
Loading...

Share This Page