No announcement yet.

Guide to achieve a better rating for GPU on Performance Test with CPU as bottleneck

  • Filter
  • Time
  • Show
Clear All
new posts

  • Guide to achieve a better rating for GPU on Performance Test with CPU as bottleneck

    Hi there.
    I've been using passmark since it was version eight... and I'm here to help some people with a good video card and not so good cpu to achieve a better rating for GPU, if aplicable.
    First, I have a GTX 1060 6GB Zotac Amp! gpu (with another rig support pci-e 3.0) with a FX-8320E amd vishera cpu and a M5a78l-m plus/usb3 (very low end motherboard that support only pci-e 2.0).
    I did some "overclockings". But this is not the point of the topic. Even so, I'll put my clocks here, JUST FOR KNOWLEDGE, and not for using without prudence:

    Overclock (not the center of the topic):
    GPU OC via msi afterburner 5% plus voltage 150+ core mhz (I used to put 160+ but the Nbodygravity cpucompute score of GPU passmark test sometimes got a black screen, even with OCCT stable etc., and then I reduced 10mhz) and stable memory clock +450mhz (if I just put +400mhz I need only 2% plus voltage to be stable -). Concluding aprox. 2100mhz core and 4455mhz memory. BIOS: PCi-e 110 - originally 100 - (10% OC, because I've tried to get nearer to PCI-E 3.0' gpu passmark results of newer systems).
    CPU - fx8320e 3200mhz/3500mhz (turbo) 1.2V - I use 206 mhz for cpu clock, x20 to achieve 4120mhz total, vcore 1.403V, loadline calibration auto (i do not mess with that percentages!). everything else auto... obs I have a watercooler nepton 240m, and my cpu does not exceed 40C. I use also 8 mosfet heat sinks of copper (1cmx1cm) "glued" with a thermal adhesive, once the VRM of my low end mobo is the worst possible, and when TMPIN0 (socket temperature) achieves aprox. 52C thermal throttling occurs. My current CPU overclock was in sight of a ~1300 cinebench r20 score (aprox. 3min40s without thermal throttling).
    NB_CPU e HT_link frequencies- I used 10x (~2000mhz), in my case 2060mhz, and the results are lower than nb_cpu 12x (2400mhz) and ht_link 13x (2600mhz)). Stock voltages.
    Memory ddr3 8-9-9-24-33 g.skill ripjaws 1866mhz -> ~1920mhz (cpu base clock 206mhz), 1.58V to be stable.

    Now, nvidia control panel settings.
    Manage 3d settings: restore default. Now go to adjust settings with visualization and put performance. Now, turn back to manage 3d settings and: DSR 4x (for dx12 4k resolution if applicable), suavity 0%, lots of stuff to disabled (anti-aliasing, fxaa, multi sample AA, ambient occlusion, max framerate, anisothopic filtering). SEGMENTED OPTIMIZATION must be DISABLED: This is important,when the CPu is the bottleneck (my case, even with OC), makes gpu framerates of passmark somewhat of 1-5 fps higher. Energy management: prefer maximum performance. texture filtering: max performance.
    Configure surround, physx-> put CPU (in my case I've gained some 2 fps).

    Obs. For the directx 9 test achieve the best result, YOU MUST PUT VERTICAL SYNC: *use the configuration of 3d application* setting before starting PTest. If you put vertical sync off before starting ptest, directx 9 falls 10%. But, for optimal results (sometimes 1-2fps more), you, after ptest starts (with vertical sync controlled by application), put back vertical sync off. This stuff occurs from alooong time ago, with ptest 8, ptest 9, and now with ptest 10: "before starting ptest, vsync application controlled".

    Now, if you started the computer from a small amount of time ago, you must before enter ptest, open task manager, and wait 2 minutes for your cpu usage is being kept at 1% (when I perform logon, if I see task manager, my antivirus uses 10-20% cpu and turns then to zero%, the same with bank security softwares every logon, uses 10-20% cpu for a few seconds (maybe searching for updates), than 0%). With msi afterburner there is no way my cpu usage turns into zero percent for a long time, but 1%. Disable antivirus, parental protection... you may not disable firewall. Once occured that my baseline was not submitted because of my firewall, then now when I submit my baseline I for a few seconds disable my firewall).

    now, with 1920x1080 screen, open passmark performance test 10. If you change any nvidia control panel settings (except resolution) with performance test ON, it will not be applied (example: if you change in nvidia control panel from "controlled by application to 8x anti-aliasing" it will not change anything on your test if the ptest software is currently opened. You must leave/close ptest and enter again for all the nvidia settings fit for all ptest software tests.

    Edit-preferences: test duration: long (it is best for gpu tests, more fps than medium duration). Not necessarily better for cpu, hard disks, memory, 2d graph etc.
    If you click at 3d graphics mark for it to RUN everything automatic, you does not gain the best score. You should monitor your cpu temperature (in my case via msi afterburner, and wait 10 seconds from the end of one test to the beginning of the other 3d test)- more fps you will achieve, once you've made sure the cpu is not being used by another application nor is with high temperature.

    Results of all these things (note CPu speed of 4120mhz - gpu test 10082, if I raise cpu speed via overclock to 4400mhz the gpu tests achieve ~10370points). You of course must note that the best scores for my gtx 1060 6gb are 13000points, and the least are somewhat of 8000: it depends of your configuration of harware (bottlenecks) and software. A vishera amd cpu with a pci-e 2.0 motherboard will of course limit +-20% of the potential of this gpu (12000-13000 to 9500-10300).


    first result (1 click only at 3d graphics mark) with "application controlled settings", norton family and antivirus on, rivatuner disabled but installed etcx., cpu fx-8320e 4120mhz with nb_cpu and ht_link 2060mhz, physx GPU.
    MSi afterburner (GPU overclock etc.)

    Click image for larger version  Name:	1ptest 9854 controlado 100%aplicativo + performance.png Views:	0 Size:	183.2 KB ID:	49075
    SECOND test, physx to CPU, 9854->9890. (I cannot put more than 3 pictures here so I removed)

    THIRD TEST: now disabling vsync with ptest software active. And performing test ONE BY ONE(dont forget long duration tests), waiting 10 sec from one test to the other.
    9890->10028 points (I cannot put more than 3 pictures here so I removed)

    fourth and last test

    Raised nb_cpu and ht_link frequency from 2000mhz (2060 actually to;..)
    NB_CPU 2400mhz HT-link 2600mhz (in my case it is 2472 and 2678 respectively)...

    uninstalling rivatuner statistics server, and norton family. Stopping speedfan software and service (it is the TMPIN0 (41) temperature in yellow), disabling norton antivirus, and even as I use the software of norton utilities premium for CPU and POWER optimization (not using memory optimizer automatically), I use task manager to kill this "boost process" service (norton utilites premium), once it uses 0,5%cpu sometimes. Oh, before that, free memory of your pc.
    Obs, I have 16GB or ram ddr3 1920mhz and I locate 2048MB for RAMDISK (2GB pagefile), using dataram ramdisk 4.0.33 (freeware that enable using until 4GB of RAm for a ramdisk)... maybe it helps... but, actually, my best passmark memory results are with no ramdisk drive..

    Click image for larger version  Name:	10082 mesmas configurações com ht_link 2600mhz e nb_cpu 2400mhz.png Views:	0 Size:	185.3 KB ID:	49076

    My best gpu score...(gpu 10345) with this low end mobo (m5a78l-m plus/usb3) and mid range cpu (oc 4400mhz)... and this good gpu 1060 6gb zotac amp! with core of 2100mhz and memory of 4455mhz (and ssd kingston 128gb sata 3 in a pci-e to sata 3 adapter, since my mobo only has sata 2):
    I think this score is with nb_cpu and ht_link 2000 and not 2400-2600.

    It is not a high end cpu score, nor it has a good ssd to raise the score, but with the cpu FX-8320e it seems very good.

    Click image for larger version  Name:	my best gpu score.png Views:	0 Size:	132.4 KB ID:	49077

    Its just that...
    I must say that EVERYTHING stated here worked for my system... and for the computers Ihad before... these things I affirm might not work with other pc systems, even with cpu as a bottleneck and a good gpu, that was my case.

    Sorry if I posted something I shoudn't, I just use Passmark performance test for years and these tips may help some.

    I know, of course, the most reasonable setting for a normal 3d simulator is "application controlled settings", which implies , for example, DX10 AA 8x, Dx 12 AA 2x, etc. and not AA zero, as it is done with this settings.

    Thank you.
    Last edited by Roberto F Rossi; Nov-15-2020, 12:00 AM.

  • #2
    Edit. dataram ramdisk is that I use (not 4.0.33). "RAMDISK-4-4-0-RC-33.exe"
    edit. on Performance test 8, to use 100% of my gpu on DX9 I had to start the software with nvidia settings "balanced", and not performance nor quality... the result increased 25% (became normal). I do not remember that there was a vertical sync option to be controlled by application nor that this setting influenced something at that time...
    One more thing... I use nb_cpu and ht_link frequencies at 10x only (aprox 2000mhz), since my memory is less than that (1920mhz), and as I highten nb_cpu or ht_link to 2200 mhz, 2400mhz etc. or ht_link to 2600mhz;;; cpu thermal throttling occurs more frequently (and this really I dont want). I used those frequencies higher in the previous post to show that the 3d marks sometimes get a better score with that, but, in my case, If I raised those frequencies, thermal throttling would be more constant on games at high resolutions such as nfs heat (1080p). I would have then to reduce my overclock a little bit (example 4020mhz (206mhz x 19.5x) I just need 1.375V, there is no thermal throttling at all, only with 6 minutes of prime95 at maximum power, and then one throttle at 90 seconds (with prime95 at max)). etc.
    By the way, this software performance test is great.