Clevo PB50RF 9750H w/ RTX 2070 - OC and UV benchmarks and thermals

Discussion in 'Sager and Clevo' started by NomBidon, Oct 10, 2019.

  1. NomBidon

    NomBidon Newbie

    Reputations:
    0
    Messages:
    2
    Likes Received:
    0
    Trophy Points:
    5
    Hi guys,


    this is a thread about my Clevo PB50RF with 9750H CPU and RTX2070 GPU and 32Gb RAM (2666MHz). I bought it from PCSpecialist (Vortex IX) as they were significantly cheaper than their competitors offering the same chassis. My buying experience went very smoothly and I can recommend them highly.

    As this is my first gaming laptop, I was quite excited and after running the first few bechmarks, I spent some time learning the basics of OC, UV and thermals.

    With all stock clocks and voltages and standard thermal paste, this laptop is quite severly thermally limited.

    3DMark Time Spy scores in all stock config were approx. 6980 pts aggregate (6993 GPU and 6888 CPU) with both CPU and GPU hitting 79-81°C (GPU is thermally throttled starting from approx. 78°C). CPU is maxed out at 3.99GHz on all cores.

    Please note that all tests were run with the back of the laptop raised by 1-2 inches to improve cooling. Also the fans were manually set to max. before the start of each benchmark run.

    Looking to improve performance, I then undervolted the CPU core by 250mV and the cache by 100mV and overclocked the GPU voltage/freq curve by a constant 175MHz. TimeSpy scores improved by 5% to 7311 pts (7411 GPU and 6800 CPU). No change in temperatures during the benchmark run. At this stage, Port Royal scores were 4426 pts.

    As a next step, I ran the MSI Afterburner OC scanner, which defined the optimal voltage/frequency curve for my GPU. The Time Spy scores improved again by 5% to 7694 pts (7850 GPU and 6900 CPU). At this point, the improvement over stock was a nice 10%.

    I decided to go one step further and replace the standard thermal paste with a much more efficient Thermal Grizzly Arctic Silver 5 paste. This brought an nice improvement bringing the Time Spy score to 7836 pts (8043 GPU and 6841 CPU). The temperatures, however, did not show much more than a 3-4°C improvement, and as a result, the GPU was still partially slowed down by thermal throttling.

    I then decided to use another type of thermal paste: Thermal Grizzly Kryonaut. As I took appart the cooling system, I noticed that the copper heatsink on the GPU was only making partial contact with the GPU die, leaving about 1/4 of the surface without contact with the copper plate. I decided to apply the new thermal paste more generously to fill the gap between die and heat-sink.

    After putting everything back together, it was time to run more benchmarks. This time, the thermals were much improved, allowing the GPU to run much higher clocks and not hit the thermal limit. As a consequence, I could manually tweak the voltage/frequency curve of the GPU a lot further. The UV on the CPU is now -250mV core and -115mV cache.

    Results (as of 10.10.19):

    3DMark Time Spy: 8295 (8612 GPU / 6895 CPU). 4th on 3DMark rankings. Temps: 74°C GPU / 75°C CPU (+19% over stock)

    3DMark Port Royal: 5001. 1st on 3DMark rankings. Temps: 67°C GPU / 67°C CPU

    3DMark Fire Strike: 18460 (22502 Graphics / 16178 Physics / 8399 Combined). 14th on 3DMark rankings. Temps: 70°C GPU / 73°C CPU

    UNIGINE Superposition:
    1080P Medium: 15879; 1080P High: 11707

    Forza Horizon 4 (in-game benchmark):
    1080P Ultra: 121 FPS

    Shadow of the Tomb Raider (in-game benchmark):
    1920x1080: Highest preset: AA SMAAT2x: 99 FPS
    1920x1080: Highest preset: AA TAA: 101 FPS

    2560x1140: Highest preset / AA SMAAT2x: 68 FPS
    2560x1140: Highest preset / AA TAA: 71 FPS

    Needless to say, I am very happy with these results, since for the SOTTR or FH4 benchmarks, this laptop RTX 2070 would place between an RTX 2070 and an RTX 2070 Super desktop GPU and above a GTX 1080 by some margin.

    I will try to post some screenshots of the benchmark results when time permits.

    If you reached better results with similar hardware, I would be interested to hear, as there might be some additional performance still to be achieved.



    David
     
  2. fipz

    fipz Newbie

    Reputations:
    0
    Messages:
    3
    Likes Received:
    0
    Trophy Points:
    5
    Hi David.

    Thank you for sharing your results.
    I own a PB50RF-G myself, and can't get anywhere near the GPU scores in benchmarks that you are getting - even with Thermal Grizzly Kryonaut on the GPU and Conductonaut on the CPU.
    Getting inspiration from other forum threads on this forum regarding a Afterburner Frequency/voltage-curve for the PB50RF, I'm struggling to get past a graphics score of 20.700 in Firestrike. Temperatures on the GPU maxes out at 66C on the GPU during benchmarks.

    In Firestrike, I'm consistently getting around a 17.000 physics score probably due to 2x16 G.Skill-ram (F4-3000C16D-32GRS) running at 3100 MHz (15-17-17-34-1T).
    I'd love to have the Corsair Vengeance SO-DIMM 2x16 GB-kit (16-18-18-39-1T) which consists of Samsung B-die chips but I don't want to splurge an extra 100€+ in my country to get around 1-2% real world performance after tweaking.

    Can you post a screenshot of your MSI Afterburner Frequency/voltage curve? Also, what is your memory clock speed on the GPU?
    I'm very curious to see what it looks like so I can hopefully squeeze out the last potentiel performance of the GPU in my system.

    Thank you.
     
  3. yrekabakery

    yrekabakery Notebook Virtuoso

    Reputations:
    804
    Messages:
    2,485
    Likes Received:
    2,342
    Trophy Points:
    181
    I’m using the single rank version of the 3000MHz Corsair B-die modules and am not able to tweak them any higher than you are. Trying to set command rate to 1T automatically reverts to 2T every time it reboots as well, not sure if this is a firmware or module limitation. Aside from the fact that there are different binning levels of B-die, trying to run at higher speeds (3000MHz+) in laptops is limited by mobo traces rather than IMC or the modules themselves, so buying uber expensive high speed SODIMMs intended for mini ITX boards is not worth it considering you might be able to eek out an extra 200MHz or so with significant time/effort and money investment.
     
  4. fipz

    fipz Newbie

    Reputations:
    0
    Messages:
    3
    Likes Received:
    0
    Trophy Points:
    5
    You can't change the command rate inside the UEFI of the PB5xRF.
    Mark J. Foster, another forum user (in this forum thread: http://forum.notebookreview.com/threads/sager-np8454-clevo-pb51rf-g-early-impression.829577/page-17 ), managed to get the kit I linked to 13-12-12-28-1T on the PB5xRF. That's the 2x16 GB Dual-rank version of your kit.
    It's possible that the IMC in your i5-8600k is weaker than the I7-9750h or as you say, the mobos are too weak to handle tighter timings at higher than 3100 MHz. I was not able to get 3200 MHz stable at 1.35V even with very loose timings and subtimings on the G.skill-kit. I'm pretty sure it's Hynix RAM chips sitting on my kit.
    As always though, it's hard to make any final conclusions on anything with a sample size of 1.
     
  5. yrekabakery

    yrekabakery Notebook Virtuoso

    Reputations:
    804
    Messages:
    2,485
    Likes Received:
    2,342
    Trophy Points:
    181
    Well I’m not sure how he was able to change command rate if your laptop’s BIOS doesn’t have that option, or make 1T stick. XTU maybe? Also I’m not sure how reliable that spreadsheet is as it contains some errors. For example he lists the 3000MHz dual rank Corsair modules has being 1T and 1.35V in the XMP-3000 profile by default, but it is actually 2T and 1.2V:
    [​IMG]

    [​IMG]

    No, I’m saying when it comes to laptops, it’s not the IMC or modules which are the limiter, it’s the mobo traces. 8600Ks and 8700Ks on desktops can handle up to 4000MHz or higher with good B-die. You can check your memory chips using Thaiphoon Burner.
     
  6. NomBidon

    NomBidon Newbie

    Reputations:
    0
    Messages:
    2
    Likes Received:
    0
    Trophy Points:
    5
    Hi fipz,


    I have increased the core clock by +230MHz from 700mV up to 787mV, then reduced the OC by 5MHz steps until 818mV (+200MHz) and then I reduced the OC to +175MHz from 825mV all the way until 1250mV. I am using a memory OC of 1100MHz, however, I have not seen a big effect on the benchmark results.

    I hope this helps. Let me know if you have any further questions

    [​IMG]

    By the way, this OC isn't stable to run the Shadow of the Tomb Raider in game benchmark. I had to dial it down a little. No issues for the other benchmarks.

    David
     
Loading...

Share This Page