1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Graphics Overclocking in SLI

Discussion in 'Tech Support' started by ModSquid, 5 Mar 2020.

  1. ModSquid

    ModSquid Modder

    Joined:
    16 Apr 2011
    Posts:
    1,054
    Likes Received:
    120
    Here's a quick one - I'm assuming both GPUs in an SLI config need to be running at the same clock settings when overclocked, but in my case, this appears to be at much lower settings than either is capable of individually. Is this the case normally?

    I have a Galax Infin8 and an MSI Gaming (both 970s) and I can max each of them out in Heaven much further than I can when running as a pair. Currently at 1290 on the clock and 1843 on the memory, with a +37mV and +50mV applied respectively, but Heaven just about made it to the end then crashed whilst displaying the results. That's the most successful it's been when not at stock, which begs the other question - why doesn't it crash at stock when both are at default but different settings?

    I can get ~45fps min at 2560x1440 with all settings maxed out or ~55fps with the above overclock, which isn't the massive difference I was hoping for. Comparing to bit-tech's own tests, these results and settings were much higher on individual cards. CPU is at 4.6 stable.

    Thoughts, chaps?
     
  2. The_Crapman

    The_Crapman Don't phone it's just for fun. Staff

    Joined:
    5 Dec 2011
    Posts:
    5,893
    Likes Received:
    2,221
    Like Jeff said, it can be awkward getting 2 different cards to play nice. Historically you could overclock them both to their maximum, but they would both run at the speed of the slowest one. 1290 is pretty low though for a 970, or is that their rated clock speed, not actual boost clock when running? %There may be some settings that need tweaking in Nvidia control panel to get them to work properly, but you have done the usual stuff like setting power management mode to 'prefer maximum performance' and texture filtering quality to 'high performance'

    Trouble is Nvidia have pretty much washed their hands with SLI and left it up to developers to implement and support it, benchmarks are usually fairly well supported though. @true_gamer is an SLI pro, maybe he can offer some guidance.
     
  3. GeorgeStorm

    GeorgeStorm Aggressive PC Builder

    Joined:
    16 Dec 2008
    Posts:
    6,743
    Likes Received:
    420
    That won't be the boost clock (or at least I certainly hope it's not!)
    You can have cards run at the same or different speeds, should be an option in afterburner or a similar program to link them.

    I would expect you not to be able to achieve their individual max speeds when running them in SLI yes, how much slower they'd run not sure.
     
  4. ModSquid

    ModSquid Modder

    Joined:
    16 Apr 2011
    Posts:
    1,054
    Likes Received:
    120
    Thanks guys - some good insight there. I'll drop True Gamer a note and send him the link to this thread, maybe he can add some additional detail.

    I was also under the impression you could thrash both cards to their limit and they'd talk nicely to each other at a common speed, but I was having too many stability issues in Heaven and had a deadline looming, so I backed them both off to the same as each other until stable. Does make it tricky trying to find the maximums when it's not obvious which card to adjust (I did have the link option unchecked in Afterburner). Shame really, as the Galax is a beast for a 970. Those clocks are the base clocks as per GPU-Z, but both can go higher on their own.

    Is it right to assume the best way to approach this in the future is to install them both, then o/c one core first until stable, then the second to the same, then memory on one then the other?

    EDIT: can GPU issues also cause BSOD errors, specifically: System thread exception not handled, Page fault in non-paged area, Driver IRQL not less or equal and IRQL not less or equal?
     
    Last edited: 8 Mar 2020
  5. Guest-44432

    Guest-44432 Guest

    Morning!

    First, make sure no overclock is on the CPU etc, so everything runs stock.
    Always try to find the max overclock of a single GPU first, before trying to find it in SLI, as sometimes you may have to lower clocks a bit depending on your PSU in SLI.

    SLI your cards do not have to match the same speeds, they can run at different speeds with on issue, as long as it is the same GPU.
    Mixing Factory overclock Graphic cards with Vanillia cards in SLI is fine. They will boost to their defualt speeds.

    Usually a CPU Overclocking error. So your CPU overclock is not stable.

    Makes we wonder what PSU you are using, as you maybe overloading it too.
     
  6. ModSquid

    ModSquid Modder

    Joined:
    16 Apr 2011
    Posts:
    1,054
    Likes Received:
    120
    Hello hello! And also many thanks for popping along to assist - much appreciated indeed. I have some answers and additional questions below, in the hope of resolving this one (quite a bit of professional pride at stake, plus the machine is now based some distance away, so engaging with it physically is tricky):
    • Why would the CPU overclock affect things, out of interest? I thought that if that was stable, then it was just a case of cracking on with the GPUs. It was running SmallFFT on Prime95 comfortably - late 80s degree-wise flat-out at 4.6 on a 240mm AIO, for around 20-30 mins
    • PSU is a Corsair TX850
    • How do I know which card to back off when running two and seeing artefacts in Heaven? Should I do what I mentioned above and go back to increasing one clock then the other, then one memory followed by the other?
    • If I remote in to the machine via Team Viewer to run Heaven benchmarks etc., is this likely to have a detrimental effect on performance and clocks? I would assume not, as it's only the network side that's taking the strain of the remote connection, but that is a complete assumption...
    EDIT: CPU voltage set at 1.32v via UEFI.
     
    Last edited: 11 Mar 2020
  7. ModSquid

    ModSquid Modder

    Joined:
    16 Apr 2011
    Posts:
    1,054
    Likes Received:
    120
    Just wondering if anyone is still looking at this thread and whether they have any input to the above questions...?
     
  8. sandys

    sandys Multimodder

    Joined:
    26 Mar 2006
    Posts:
    3,969
    Likes Received:
    316
    CPU works harder with SLI due to driver overhead, so you could see stability issues arise, you will also be using more RAM most likely as they won't have a lot of VRAM, and stability testing CPU without GPUs is probably missing a load of heat from the system, heat leads to crashes if the high temps take your chip out of its comfort zone.

    artefacting in haven tends to be GPU memory in my experience, I tend to do as true gamer leave stuff stock, link cards and then focus on gpu memory and then core
     
  9. ModSquid

    ModSquid Modder

    Joined:
    16 Apr 2011
    Posts:
    1,054
    Likes Received:
    120
    Ah, righto - cheers sandys. In that case I'll back off the CPU and reset the GPUs to stock. By "link them" do you mean fix their settings to each other in Afterburner, or were you literally just talking about adding the SLI bridge? Only reason I ask that is because per true_gamer, you can happily pair up two differently-clocked cards.

    Is the answer then to o/c GPU RAM 1 then 2, looking for artefacts, then do core 1 then 2? I did bench the CPU without SLI active, but I've got three front intakes and three roof/rear exhausts (two of which are on the 240mm AIO for the CPU, admittedly), so I was rather hoping things were running at reasonable temps. I will double check that though.

    Will testing over Team Viewer be an issue ie. affect any results?
     

Share This Page