Extreme 4-Way Sli Tuning

Discussion in 'Videocards - NVIDIA GeForce' started by A M D BugBear, Mar 10, 2019.

  1. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    (UPDATED - 10/21/21):

    Cards used: 4 Gtx 970 Windforce(Gigabyte)

    Still Active for 2+ Years.

    4 TITAN Xp's INITIAL TRYOUT: PG. 38

    LONG STRESS TEST RESULTS/TEMPS(FOLDING@HOME), CHECK PG 24.


    UPDATE: LATEST PICS/VIDEO: PG 36.


    Migrated to new system, check pg 46.


    My 4-way sli project, I will continually update it, "ONLY" on guru3d.com and no where else.

    #1 Priority in any Quad Sli tuning setup: you absolutely "Must" keep those thermals under control, covering all aspects of the Graphics Card itself, absolutely "NO" Exceptions.
    ------------------------------------------------------------------------------------
    Latest Quad Sli Project: Warhammer 40k - Inquisitor/Martyr:

    Video Link(processing/please adjust your saturation accordingly as the saturation seems more higher when I encoding it through Hybrid, After handbrake, it seems saturation is lowered quite a bit, please just your saturation if need be, thanks.):



    Link to Handbrake version(Has frame pacing issues due to encoding to H.265 / .m4v format, I used the most extreme transcoding settings, in the end, there was only very minimal differences):

    Link(Copy and paste this after you type in youtube .com):

    /watch?v=ucog_yUuid0

    Final Conclusion as well as Performance Video's and Graph will be up later.

    Older 4-Way Sli project, 1st 4-Way Sli video, RE3 Remake(Recorded @ 30fps, old camera):


    ------------------------------------------------------------------------------------
    All these were tested at 4k(2400p) or stated otherwise:

    FOR MORE: CHECK OTHER PAGES.

    4 FUTURE PROJECTS/INITIAL TRY OUT: DOOM 3 + RESHADE(PLAYED UP UNTIL: ENPRO PLANT), BALDUR'S GATE 3(PLAYED 2hrs), WH40k: DOW II+RESHADE & WH40k SPACE MARINES+RESHADE =
    ALL 4: EXCELLENT SCALING.

    WARHAMMER 40K INQUISITOR - MARTYR + RESHADE:
    FINALIZING RESULTS(187+ Hrs in/GOOD/EXCELLENT SCALING). PG 42-48

    BDO + RESHADE:
    (37.2+ HRS. IN/EXCELLENT SCALING THUS FAR) PG. 42-48

    DS3+RESHADE:
    GOOD-EXCELLENT SCALING POTENTIAL, LOTS TO COVER (Graph/Videos Available) PG. 38-43. - FINAL -

    RE3 REMAKE:
    HOLY BANANA'S, A+ . Tons of stuff to cover. Check PG. 34-38(Graph/Video's Available) - FINAL -

    HELLPOINT:
    EXCELLENT SCALING!!(Depending on Resolution and Area of the game/Graphs Available), PG. 32-36 - Final -

    FFXIV: SHADOWBRINGERS BENCH: 2 Runs/4k(2160p)/14148 pts.) & 4400x2475/11333 pts.): PG 47

    FFXIV: HEAVENSWORD BENCH:
    2160p/MAX Settings: 14041 pts. PG 47

    FFXIV: STORMBLOOD BENCH:
    2 Runs/4k(2160p)/16380 pts.) & 4400x2475/10532 pts.): PG 47

    FFXIV: ENDWALKER BENCH (4k-2160p/MAX Settings):
    13905 Pts. PG 47.

    UNIGINE: SUPERPOSITION+TROPICS+SANCTUARY:
    PG. 35&46/47. - FINAL -

    3DMARK FIRESTRIKE BASIC+EXTREME+ULTRA:
    Basic: 31405/Extreme: 19131/Ultra: 10636 {PG. 47} - FINAL -

    UNIGINE HEAVEN 4.0:
    Good Scaling. 4-way Sli Score(2160p/Max Settings/8xAA): 1412 PG. 46-47 - FINAL -

    CATZILLA BENCH: (7/29/21):
    Excellent Scaling. Check Page 46. My Current 4-Way Sli Score: 4-way sli score(SFR): 16035

    MORTAL SHELL:
    EXCELLENT SCALING, DESPITE ISSUES, LOTS TO COVER - PG 31-34 - FINAL -

    ME 1(Revisted):
    (9/12/20: Mod+SGSSAA = EXCELLENT(Despite issues with stuttering with mod). PG 31-32

    ROTR - (3/29/19):


    The scaling ran eh to so so if only use smaa, now if you use SSAx2, the scaling is very good. I did however did another change in the sli hack bits, now with SMAA, the Scaling is an absolute beast but needs work around it, can't just turn on game then play, absolutely not. read later on this thread, describes vastly improved scaling I discovered, Highly recommended.. - UPDATED ON - (6/12/19) - (PG. 5 - FINAL -)/UPDATED ON 9/9/20: Re-tuned using AFR of SFR, HOLY $#|+ Scaling Output, CHECK PG 31. - FINAL - (Quick tested With 4 F'n Titan Xp's(Using Re-tuned Profile), 110% confirm working + 4x980ti's = Working)

    MECHWARRIOR V:
    Very Good/Excellent Scaling, PG 29-31. - FINAL -

    EDGE OF ETERNITY: Excellent in the Beginning. Lots to cover, page 28/29.(100% WORKING ON 4 980TI's).

    GHOST RECON: WILDLANDS(GOOD SCALING/SOME FLICKERING INVOLVED - SEE PG. 5 - FINAL - (7/23/19).

    DQXI: Decent - (UPDATED ON 6/21/19) - (PG. 5)

    GHOST OF A TALE:
    Decent - (UPDATE ON 2/10/20): Re-Tuned for SFR/EXCELLENT SCALING!(PG. 21) - FINAL -

    HEAVY RAIN: Tuning Sensitivity = WORST(game on hold).

    STAR WARS BATTLEFRONT(2015): Decent

    PHOENIX POINT: LOTS TO COVER - PG 21-22 For rundown. - FINAL -

    A PLAGUE TALE: INNOCENCE: Very good Scaling/Some Flickering - (pg. 4)

    THE TECHNOMANCER: (PG 21-23, EXCELLENT SCALING) - FINAL -

    BF 1:
    Eh to Decent(Depending on what you are doing in the game)

    COD - GHOST: Not really, have to use Nvidia inspector SGSSAA Injection to make better scaling, even then, eh.

    CRYSIS 2+MALDO MOD: Excellent - Pg. 4.

    CRYSIS 3: Decent/Very good

    ECHO: Very good/Excellent Scaling(Pg. 13-16)

    WITCHER 3:
    Scaling seems Great but tons of graphic flickering, and a whole bunch of other things, as of right now, only 2 way as of now.

    DARK SOULS 2(DX9 API): Game = 60 fps locked, 8xMsaa +8xSGSSAA to further stress the card down to its knees and make scaling much better, I would say not really, there is certain parts like if your viewing the waterfall in the beginning, it drops down to 15 fps per card, but soon as I use 4, I get 60 fps, other areas = 4-way not needed.

    YAKUZA KIWAMI 2 - A+(Except for Scaling drag issues(certain Spots) during 3-way and 4-way Sli/pg. 10-13.

    BATTLETECH(7/29/19 - Lots to cover - pg. 6-7)

    DYING LIGHT: Very good scaling, only problem is the blood splatter on the screen flickers alot, other then that, great scaling.

    REMNANT: BEYOND THE ASHES: Despite differences from 3 gpu's to 4 gpu's, 4 gpu performance is like 3 gpu's, check pg 8.

    KINGDOM COME: DELIVERANCE (5/16/19): PG. 4 - FINAL -

    FFXII - THE ZODIAC AGE(60FPS CAPPED REMOVED): EXCELLENT SCALING!!! and best of all Absolutely no flickering, no hitchin, hang, nothing, absolutely perfect, only one drawback if the fps cap is removed, the game moves incredibly fast, like in fast motion.

    GRID AUTOSPORT - Have to use SGSSAA injection otherwise the scaling is to weak, and the scaling after Sgssaa, I would say eh to so so..

    LOST PLANET 1 - (DX10 API): Capped at 120 fps, so I had to use SGSSAA to further push the cards down since without sgssaa, the card is using like 60% usage or so and caps at 120 fps, after testing with 8xMsaa+8xSgssaa = A+.

    DEUS EX - MANKIND DIVIDED: Overall pretty good scaling, certain things have to be shut off, Like TAA cause the edges of buildings,etc creates annoying haze like over it, best to turn it off, certain settings needs to be toned down a bit otherwise you will have problems, like motion blur and texture set to high.

    DISHONERED 2: Decent but certain things needs to be toned down due to 3.5-4gb vram on the card or certain settings that could cause the 4-way sli to react abnormal.

    GREEDFALL: DECENT/EXCELLENT(MUST USE SSAO, IF NOT, 3-WAY SLI IS BEST OPTION). PG 19-21, - FINAL -

    M.E. ANDROMEDA:
    If your in buildings,etc it acts like 3 way, but when your outside, decent scaling, inside areas NAH. (UPDATE ON 1/2/20/ON HOLD,PG 19).

    VANISHING OF ETHAN CARTER(UE4 VERSION): A+/PAGE 16-18 FOR RUNDOWN - FINAL -

    RE2 REMAKE(2019): Very good scaling but I can not get rid of the massive flickering and other major graphical problems like texture disappearing, and other things, NOT Recommended. (UPDATE 10/13-19): on hold for now)

    BF 3:
    Great Scaling! pg 3.

    SOTR: Very good overall scaling. 2 Profiles were custom made, one performance profile(some flickering involded) and the other a normal one(no flickering) - PG. 5 for rundown - FINAL - (7/1/19).

    DEAD SPACE 1 - (6/3/2019): EXCELLENT SCALING - (PG. 4).

    ESV: SKYRIM(SE): Some graphic problems here and there, overall scaling ok to decent.

    HITMAN: ABSOLUTION(ENHANCED) - UPDATE ON (6/5/2019): EXCELLENT, pg 4 for rundown.

    HELLBLADE - Acts like 3 way, although 2 way works perfectly but certain things needs to be changed in the configuration folder, TAA needs to be turned off and other things needs to be shut off, otherwise Tons of graphical problems, Overall experience with this game in 4 way = Pass, as it looks, PASS. - UPDATE(Revisted) ON HELL BLADE - (6/25/19) - PAGE 5).

    LORDS OF FALLEN = DAMN GOOD! - (UPDATE 9/25/19) - Pg. 9/10

    MGSV: THE PHANTOM PAIN = X

    FARCRY V = X


    DMC IV BENCH: Perfect Scaling!

    ALIEN ISOLATION: Seems I only get like up to 3 way performance, I have figured out how to use TAA under Sli and it works but despite the gpu usage all saying 96-98%, performance scaling is like 2-3 times over a single card.

    GOW 1: Very good Scaling, pg 3.

    F1 2014: SGSSAA Injection has to be used in order to scale well, overall = EH.

    RE 6 BENCH: some parts scales very well, some places no.

    PREY(2017) = X

    MIRROR'S EDGE CATALYST = X(May go back in the future)

    ACE COMBAT 7: Terrific scaling potential, major flickering(ON HOLD).

    COD: INFINITE WARFARE(Revisted): (UPDATE ON 3/8/20 - GAME IS NOW FINALLY RE-TUNED, GOOD/EXCELLENT SCALING, Pg. 22-23) - FINAL -

    NIER AUTOMATA:
    Very good scaling.

    METRO 2033 REDUX(DX10): I tried doing 4k without SSA = eh scaling, then I tried 4k with SSAX2 scaling is decent but takes to much of a hit in performance, so what I did is I tried @ 1600p resolution with ssaX2 or X4, scaling is alot better, overall good @ 1600p+SSA.

    NIOH : Excellent(3-Way Only/60 fps cap)/pg 13).

    RYSE: SON OF ROME: Behaves more like 3 way, I'll do more tuning on this game later, more then likely = X - (UPDATED ON 6-13-19 - pg. 5 - FINAL -

    CASTLEVANIA: LORDS OF SHADOWS: Very good scaling, use SGSSAA to make better use of the scaling.

    BF IV: EXCELLENT! PG 3.

    SHADOWS OVER MORDOR: So So/decent.

    QUANTUM BREAK = X

    DOOM: Best to use 2 way ONLY.

    THE DIVISION = X

    AC: ORIGINS: Scaling seems good but Massive flicker, I can not get rid of it, Not recommended at this time. (UPDATE: 9/3/19 - page 16(Game on hold).

    FARCRY PRIMAL:
    Up to 3-way only, pg 5.

    ME 2: (3/23/19):

    Mass Effect 2 - with Alot texture mod, modified .ini file. Very Very good scaling here.
    For some odd reason or reasons, Sparsegrid AA does NOT work above my native resolution, wether through custom resolution or through DSR. I tried different sli bits, aa bits = no go Either one card or sli mode, doesn't matter however, I did override AA with 2x2 Supersampling and it works great, confirm it works.

    ME 3: (3/19/19):

    Mass Effect 3: With Alot texture mods, and tons of other mods + modified .ini file.
    For some reason, It only scales from 50% to mid 70% on all 4 gpu's. I even tried to inject 2x2 SS or X4 Sgssaa+4xMsaa = no go. I tried doing some stuff to the .ini = Same Results. I tried many different variations within nvidia inspector SLI Dx9 Broadcast and Sli Dx9 Hack bits and changing many different sli profiles = same results. I literally gave up on this, kinda funny where I got ME1 and ME2 to scale very very well and this don't scale worth crap, if any, more like X 2.5 scaling over single gpu. I literally tried many variations, just gave up on it, worth trying though. I also tried to turn on through nvidia control panel, Compute Performance to ON, made it worse, infact, that particular settings can make many games lose performance so watch out, the only game that benefit this from what I have tried is The Division(in 2-way sli).

    Tried many different combinations = X

    I would stick with normal 2 way sli for this game. (PG. 6).

    YAKUZA KIWAMI 1 -
    DECENT, pg. 3 for rundown.

    GTA V: Very good to EXCELLENT/check pg 19. - FINAL -

    UPDATE ON FARCRY 4(4/23/19)
    :

    I will be needing more further testing on this, but early testing on this game, the scaling isn't all that good, scales like up to 60%+ on all gpu's, and initial testing shows and just like crysis3, If I use TXAA, the gpu scaling gets worse, I have to turn it off or use SMAA. Also, Farcry 4(Uplay), I have to set both of my cores to 2, so this means 4 cores, 8 threads, otherwise the game WILL NOT LAUNCH. Assassins Creed: Unity is the same thing but If I remembered Correctly, that game can go only high as 16 core, 32 threads, Otherwise the game WILL NOT LAUNCH, either 8 core or 16 core, I forgot but I will be testing that game later.

    The only time it scales like 98% on all gpu's, is at the title screen or while viewing the option window, other then that = X.

    Much more further testing is needed for farcry4.

    UPDATE ON FARCRY 4 - (4/24/19):

    Some scaling seems ok when your looking at ceiling, or in small rooms but when you start to move, at a certain point, it acts up, I changed some of my settings, the AA is causing some of the weak scaling, so I turned it off. I changed Both Sli hack and broadcast bits, changed sli dx11 profiles, did some workaround, = X.

    When you venture outside, it behaves like 2-way sli, in closed area, you might get up to like 3-way scaling. Also, you can inject Sgssaa in this game with enhance AA through nvidia inspector and use X2msaa ingame BUT it don't work when you use sli, single mode works perfect with Sgssaa.

    For 4-way sli = X

    UPDATE ON FARCRY 4 AGAIN ON (4/25/19):

    For the most part, I have gotten 4 way sli to work, However you MUST use SGSSAA+ingame MSAA to keep the scaling packed down, otherwise just straight 4k with no AA or just smaa, Scaling doesn't keep up well. How does it scale now? I would say all in all, Decent to Prettie good, But it depends what your doing and looking at in the game, its not constant good scaling, some areas are Very good scaling, actually almost perfect scaling, others areas and situation are Decent.

    One card, while playing in one area, I get 16-17 fps, with 4 card doing the same situation = like mid 50's to low 60's. In one area I was looking at the hill side, I got like perfect scaling.

    Remember This is @ 4k(2400p) With x2 SGSSAA + ingame x2MSAA, so there will be performance hit here.

    There were Couple instances where the game just went black screen, and I has to do hard system reset, I changed certain settings within nvidia inspector and change the voltage slider a bit within afterburner, seems to run alot better now, and yes I have gotten SGSSAA+MSAA to work under SLI, All in All, Give it a try.

    Yes I have Confirm that SGSSAA+MSAA is working under 4-way SLI.

    Overall, alright experience.

    KINGS OF LORN: LOTS TO COVER, PG 23-28. - FINAL -

    BLACK MIRROR IV = X

    AC: UNITY - (3/26/19):

    Game Tested 1600p resolution.

    Tried alot for this game. It only scales up to 3 way if possible, barely even using the 4th card if any. One thing I found out that If you inject through nvidia inspector SGSSAA+MSAA = Gpu usage fluctuate or behaves very abnormal, to fix this, use only enhance the aa within inspector and use SGSSAA only instead, then use ingame AA of your wish, try to stick with like within inspector for example, 4xSGSSAA then in the game, X4msaa, I confirm it does work, you can even enhance with ingame TXAA with SGSSAA within nvidia inspector, it works.

    Gpu usage is like up to 70's if lucky. I already got a baseline with one gpu, like 15 FPS(x4 SGSSAA+ingame TXAA), with 4 cards it gives me like 41-43 fps. I even just tried with ingame X4 msaa = no go.

    Added kicks, Compute perfomance to on = no change, did 4 way alternate frame rendering through nvidia control panel = no change.

    I even change the quality settings under advanced to see if the scaling is affected by any of these settings, nope, all good.

    Did Changes: Sli hack and Broadcast bits, Tried many different sli dx11 profile, along with my own custom that's not on the list, still = X

    At this point, I do NOT recommend at this time with 4-way sli, perhaps with 3 way but not 4 way.

    ALAN WAKE: Up to 3-way only, Pg 7.

    FARCRY 3: Blood Dragon DLC - UPDATE ON (4/9/19):

    Actually this Update is like a almost a week old of testing:

    Ok far as scaling goes, Certain parts of the game scales extremely well, Certain parts like in the beginning, when your getting near the Garrison Area, man, Scaling weakens BIG TIME.
    I tried changing all of my settings to the lowest then work myself up to see what is causing the scaling issues. At first, There is a setting for Water, this MUST set to low, other wise, it doesn't matter if your not near water area, The scaling weakens, so this MUST be set to low, no exceptions. In corridor areas, for the most part, scales prettie decent, The chopper part with the gatling gun, in the very beginning, don't scale worth crap, right when you get off the chopper, it scales rather well, it just doesn't scale well when you are getting near the garrison(base) area.

    Another area scales extremely well is right in the beginning of the a particular cave scene, namely the blood dragons, mission 2 area, that part scales extremely well, soon as you go outside, scaling starts to weaken a bit here and there, when you are getting near the base nearby, the scaling nearly cuts over half or so, there is something in the scene that's causing it to drop, I tried changing all my settings lowest to highest, by each individual settings, NOPE.

    So as it stands, Do I recommend it in a Long run? Nah, For testing purposes, yes. Do I recommend it at this point? Not really.

    PROJECT CARS 2: Good to Excellent(Weather/Track and amount of cars needs to be tuned correctly for Top Dog Scaling) pg 21. - FINAL -

    STAR CITIZEN: GAME ON HOLD. PG 22/28.

    SEKIRO: SHADOWS DIE TWICE - (4/9/19):

    For the most part, I have gotten the scaling pack down rather very very well, there is one problem, Flickering can be out of control on the main charc and some textures on the screen, I tried putting all settings to their absolute lowest and fine tuned each setting by themselves, nope, Even tried the old fashion method, Cntrl+ALT+Del, then go back into the game to see flickering goes away, NOPE, Actually at one point I got the flickering to go away completely, I was messing around with the camera with my mouse, moving around a bit, going into water, etc, changing angles, then all the sudden the flickering all went away, ever since then I can not re-create it.

    Is it playable in 4-way? Absolutely yes, but for most people, the flickering can be an issue, if you can deal with it other then that, it scales extremely well.

    Also, I tried Changing many Sli hack bits and broadcast bits, sli profiles, the flickering is still present and the only profile that can really push all 4 cards is the Battlefield V profile, I did try using the Dark souls III profile at first, There is no flickering at all BUT scaling on each card is like upper 20's to low 30's, Absolute Trash scaling, mise well be running one card, but BFV profile scaling wise is a Beast. UPDATE ON: 9-24-19, PG 8/9 - FINAL).

    UE4 Demo: Infiltrator - UPDATE ON (4/10/19):

    LOCKED to 2 gpu's.

    STRANGE BRIGADE - UPDATE ON (4/10/19):

    LOCKED TO 2 GPU'S.

    BRIGHT MEMORY:
    Some part's scaling is wicked, some still needs work on.(ON HOLD/PG. 8)
     
    Last edited: Oct 22, 2021 at 3:27 AM
  2. venturi

    venturi Master Guru

    Messages:
    344
    Likes Received:
    151
    GPU:
    2x 3090 RTX nvLink
    Only out of respect I ask:

    on occasion when scaling was so so, I resorted to SFR.
    Some engines have transient effects that go across frames / frame to frame. SFR seemed to solve those moments .

    I know you’ve become a master at coolbits, set force 4 gpu, set all the parameters, end the bits in 5 instead of 1 or 2.

    Lastly on heat-
    If the motherboard allows to configure it, make the primary card the lowest one in the stack. It will get the most amount of air. Then FLIP the sli bridge upside down. Reinstall driver and enable sli.

    Another choice is to disable CSM in the bios (all UEFI) and whatever card had the dp cable becomes primary.

    In bios set pci max read to 4096 and max load to 4096 instead of auto,

    The lowest card as primary seems to help heat. Also REMOVE the backplate; you’ll gain 2mm of airflow and the cards don’t have all that heat trapped in the backplate.

    Anyhow, you have probably tried all this stuff, just trying to help
     
  3. sverek

    sverek Ancient Guru

    Messages:
    6,073
    Likes Received:
    2,953
    GPU:
    NOVIDIA -0.5GB
    With such high resolution, wouldn't 4GB of GTX970 be an issue?
     
  4. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    Usually not an issue due to my 64gb of main ram and boy does it helps BIG TIME.

    If I only had 12-16gb of main ram, some games I would run into major trouble.

    I did quite a few games @ 8k(4800p) with my 1080ti's and never had a single problem, I have seen my main ram going to high as 40gb, Once the vram is out, it will use the system ram as buffer, But you must take this into account that I have dual quad channel memory so this will help out alot.
     

  5. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli

    Interesting about when you mention about the back plate, but that back plate is needed to secure the gpu properly.

    Also, some games, in the sli broadcast, I have to set it to:

    DYNAMIC_BROADCAST_RENDER_TO_TEXTURE - Use broadcast to render interframe depandent content

    Otherwise the scaling isn't all that well, same goes for like games like hellblade, the sli hack bits:

    FORCE_REPORT_NO_SLI_SUPPORT_FROM_NVAPI - If applition has buggy its own SLI support we need to do SLI magic in the driver and force app think it's run on single GPU

    This is a MUST when running 2 cards, Otherwise sli won't work worth jack, Speaking mainly about hellblade, certain games need it, NOT ALL, only certain games.

    There is a setting in the bios for these mobo's, above 4g decoding, this setting MUST be on otherwise you will get blackscreen during bios post at startup.

    I don't think I can set the bottom card as primary, I'll look into it again, I highly doubt it.

    Also I can not stress this enough:

    Lets say you have nvidia inspector profile backed up from 10xx cards and you want to try it on a previous generation like maxwell cards, don't even think about it, I thought this was the case, boy I was wrong, I tried DQ11 and hellblade profile backed up for my 1080ti's, then I tried it with my 2 gtx 970 in sli, it DO NOT WORK.

    I can only guarantee these profile that I fine tuned, to work ONLY for the gtx 970 family, I highly doubt it that it will work with 980ti's, cause I had some backed up, they don't work, so keep this in mind.

    Lastly, I'll try doing SFR with certain games, I know SFR in certain games for my 1080ti's works Excellent, I think it was Crysis @4k with Sgssaa, Works excellent but never tried with these cards.
     
    Last edited: Mar 13, 2019
  6. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    Be sure to check the first post, all the way to the bottom where it says updated, I'll be updating with games in that particular area.
     
    Last edited: Mar 14, 2019
  7. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli

    Hello again, Since gpu3 runs hotter then the others, I check out the backplate once again, and yes indeed, it can be used without it, gpu and the vrm heatsink screws down perfectly.

    As I originally expected, gpu3 had artic silver 5, the other had thermal grizzly, I knew one of them had AS5, just couldn't remember which one, so I decided to choose the one that run hotter, and I was right.

    I only took off the backplate on gpu 2,3, and 4, 1st one is not need as obviously there is nothing in the back of that card.

    Since I didn't make any recordings on the temps before I changed it, far's I remember, It dropped probably up to 5 celsius, gpu 3 was running like in the low 80's from what I remember, now its like running 74-75 C.

    Gpu1: Since the backplate doesn't have to be removed, I never changed the thermal paste, already had thermal grizzly kyronaut to begin with, so no need to change.

    If my room heater is turned off, gpu1: 72C, gpu2: 74C, gpu3: 77-78C(right after I changed the paste).

    After stress testing the system for a bit, turned off the system for quite sometime, re-observe the temps:

    This is while my room heater is on:

    Gpu1: 77-78C, GPU2: 75C, GPU3: 74C, GPU4: About 65C.

    If my heater is shut off: gpu1 and gpu2 is like 69C-71C'ish, gpu3 is like around 72(This is after the First stress test run for over 15 mins or so, then shut off the system, hrs later I turn it back on to re-observe if the temps improved).

    I am using the same cooling configuration as I did before I changed the black plate.

    I think from what I remember when the back plates were on, all three first gpu's were in the low 80's, I know for a fact that gpu3 was running over 80, If I remember correctly, like 84c+.

    Also, just to add very tad bit extra cooling, I added a 1.5 thermal pad on back of the gpu with a 1mm copper shim on all 4 gpu's.

    I did this on my 1080ti's, how much does this help, like 2C tops if any, not to much but anything more to help.

    All and All, I think there is differences here, Thank you for mentioning the back plate.

    Thermal Grizzly Kyronaut requires no break in time, I would disagree with that, I remember on the 1080ti, after couple days of applying the compound, the temps went down big time, and Other sites have mentioned about this too, I'll keep a keen eye out on the temps overtime and let you all know if any big changes, thanks.

    UPDATE 3/24/19:

    While under Extreme testing(All 4 Gpu's under Extreme Load) for about good 25-30 mins, same cooling configuration as before the backplate was used, same method were used.

    While my room heater is on:

    GPU1: 81c-82
    GPU2: 77c-78c
    GPU3: 76c
    GPU4: 71c-72c

    Also just heads up, GPU1: is the ONLY CARD that has a backplate on. I will be taking it off soon to re-observe temps. I will be re-applying Thermal Grizzly. Should see interesting results, in the meantime, these are the temps I gotten so far, hope this helps some.

    UPDATE ON TEMP RESULTS(3/23/19):

    Ok, I took out the 1st gpu back plate and re-apply thermal grizzly, in a room about 82 degrees fahrenheit of my room heater(same room temp as my previous results), Not enough break in time with the 1st gpu, but here are the results thus far and this was taken under very extreme heavy loading on all 4 gpu's:

    GPU1: 81c-82c
    GPU2: 73c-74c
    GPU3: 75c-76c
    GPU4: 72c

    UPDATE ON TEMP RESULTS(3/24/19):

    So something I wanted to do before but I was lazy, now I have done it. Since all 4 gpu's under load have different voltage load, What I did is 1st gpu to last gpu, I went from lowest voltage load to highest, Wanted to see results, thought it was interesting.

    Tested under Very Extreme heavy load on ALL 4 gpu's, for about good 10-15 mins. Fans on card blasted to 100% along with the same cooling configuration as I was doing in the very beginning of this 4-way test.

    Room ambient temp is about 82 degrees fahrenheit:

    GPU1 @ 1.193V: 80c-81c
    GPU2 @ 1.200V: 78c-79c
    GPU3 @ 1.212V: 70c-71c
    GPU4 @ 1.231V: 65c

    As you can see, GPU4 dropped about 7c and GPU4 dropped up to 5c.

    GPU2 increased its temps by up to 5c

    &

    GPU1 temps were roughly the same as before, 1c difference give or take.

    Whats so funny about this particular test, look at GPU4, it ran much cooler then the gpu I used before in the 4th slot and that was running at a lower voltage, very interesting results.

    All 4 gpu's were using factory windforce bios, @ +25(Voltage Slider). @ 1506mhz on core and over 7.7ghz on memory.

    UPDATE ON TEMP RESULTS AGAIN(3/24/19):

    I jocked down the voltage slider to +0 and its very stable @ 1506mhz(core) on all 4 gpu's.

    Here are the temps under very extreme heavy loads on all 4 gpu's:

    GPU1 @ 1.168V: 78c-79c
    GPU2 @ 1.175V: 76c-77c
    GPU3 @ 1.187V: 69c-70c
    GPU4 @ 1.206V: 63c-64c

    Very little difference from before but anything more to help bring it down a notch. Very stable @ + 0.
     
    Last edited: Mar 25, 2019
    fantaskarsef likes this.
  8. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    I'll be recording some games Very soon.

    And Just to let everyone know, I'll try to record @ 4k(2160p) through Geforce Experience, IF I lose more then 3-4 fps while recording, I will NOT record it and will go lower, I want to show the performance in 4 way as close as possible before the recording, Some games take Tremendous hit while recording at 2160p, some games NO, so therefore, if certain games doesn't take to much of a performance hit, I will record them @ 4k(2160p), if other games I record takes like more then lets say 5+ fps hit, I WILL NOT record it, I will then try 1440p, if its takes to much hit, I will do 1080p(bare Minimum).

    So I'll provide video links later, I am still trying to fine tune other games and other games I haven't tried yet, but I should start doing some recording like monday.

    If you all want me to record the ones that doesn't scale well at all, let me know please. I really only want to upload the ones that scales decent, but if anyone wants to see the lesser ones and the massive graphical problems during scaling such as RE2 remake(GOOD LORD/BAD(bad as in Major graphical issue while in 4-way), just let me know and you all can see while I play, thanks.
     
    SpajdrEX likes this.
  9. venturi

    venturi Master Guru

    Messages:
    344
    Likes Received:
    151
    GPU:
    2x 3090 RTX nvLink
    you know.... you've done some substantial effort here...good job!
     
    dark_surge likes this.
  10. cryohellinc

    cryohellinc Ancient Guru

    Messages:
    3,407
    Likes Received:
    2,760
    GPU:
    RX 5700 XT/ MAC M1
    My main concern is that you constantly have a lack of vRam, which results in your slower Ram taking action. This has an impact on overall performance, especially in such a setup. Would love you to test this out with something like 4x1080Ti!
     

  11. Kaarme

    Kaarme Ancient Guru

    Messages:
    2,626
    Likes Received:
    1,278
    GPU:
    Sapphire 390
    What a list of games. You must have burned through an SSD to test them all.
     
  12. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    Don't the 1080 themselves are locked to 2 cards??? I remember something needed a key from nvidia but this only enables for benchmark ONLY, it is totally cut off from games, please anyone correct me if I am wrong here.

    Look at this video please:




    Also get a load of this:

    https://forums.evga.com/3-way-SLI-GTX-1080-Ti-how-to-do-it-m2748218.aspx

    On that evga link, if you keep reading it all the way, the dude with the tri-sli 1080ti setup, managed to get it work properly, his firestrike ultra score is over 21k+, holy sweet upside down pineapple CAKE.

    I just took look at my 1080ti recorded firestrike video that I backed up, very close to 13k. So he managed to get it working.

    I can tell you now, with 4 of these, if scales very properly easily 60fps @ 8k, yum yum.

    Just 2 will give 8k up to 30 fps or slightly higher, depending on game and their settings of course.

    EDIT ON 4/20/21:

    Forget to mention this for quite some time now:

    Yes, there is a Way to get 3 or more Pascal based gpu's to work, I own 4 Titan Xp's(Best Pascal Based Gpu's), and it works.

    Its a particular parameter you have to change within the unknown ID settings.

    Give you all a hint, Its the most popular Benchmark and probably most widely used, the unknown ID settings pertains to that, if anyone is extremely into tuning, you would all know which one.

    If anyone is still needs more help on this, give me an PM and I will explain.

    It does work, 110%.
     
    Last edited: Apr 21, 2021
    fantaskarsef likes this.
  13. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    I have like 5 backup normal HDD drives, All my benchmarks from sli 980ti's, 1080ti's, 970's are all on the HDD.

    Also, All my games that I downloaded through Steam,Uplay,Origin,GOG,etc, ALL backed up on those hdd so no need to re-download them again, just transfer then play.
     
    Kaarme and fantaskarsef like this.
  14. fantaskarsef

    fantaskarsef Ancient Guru

    Messages:
    12,480
    Likes Received:
    4,788
    GPU:
    2080Ti @h2o
    This only makes me think: LUL
    Hence I switched my strategy to buy strong single GPUs... although I admire your testing @A M D BugBear so please keep it up! Love reading your experiences
     
  15. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    I have put so much work not just into these 4 970's, but the 1080ti's (4k to 8k res test), 2 980ti sli test, I have videos on the 980ti test, some only have 14 views on them, and its been like several years. Very low views, but its alright.

    I am thinking about re-doing some 980ti test in the future, cause I never really used afterburner, I like the osd much better and I have learned alot since then, especially messing with the 2 1080ti's.

    Heres some videos I did, Please video them.

    heres some Mass effect clip stuff, short clips, I think these were done on the 2 980ti's in SLI Mode, also, from 4k to 6k test.

    https://www.youtube.com/playlist?list=PLv7whYPHSlwTBJhqlAp7x65hR4FIJ-5lC

    Heres a Playlist for my 980ti VS 1080ti's:

    https://www.youtube.com/playlist?list=PLv7whYPHSlwQ6yVa28HyDXvEF92tuO0_Z
     
    Last edited: Mar 28, 2019
    fantaskarsef likes this.

  16. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    An old Video of mine. Take a look at these 3 Mass Effect videos, 2 980ti's

    Used single psu at that time, now using 2 for extra stability, Yes it helps out alot.

    Mass Effect 1 @ 4k(2400p) Along with + Override Nvidia Inspector: 2x2 SS):




    Take a look at this Clip, 5k but recorded at 4k(2160p). Mass Effect 2: Tali Loyality Mission part 3:

    Don't Believe the fps was uncapped, sorry about that. Top left, Fraps(Fps meter).

    Beautiful Video, have a look:



    Alot of texture mod was used in this video.

    Nvidia Inspector(Hardware Forced Sli+Anti-aliasing(1x2ss) Modified .Ini was also done.

    Mass Effect 3(4k/2400p):

    Update: doesn't appear alot was being used Properly in this video, sorry about that.

    Nvidia Inspector OVERRIDE AA(8xSQ)-[Combined: 2x2 SS + 2x MS]+2xSGSSAA For the ultimate Image quality that comes second none itself.

    Sorry for not using the Interface scaling mod, Didn't know at that time. So the hud is quite small

    1481mhz(GPU's)7961mhz(MEMORY)-(Consistent Throughout):





    All 3 videos, While playing, hardware used:

    Motherboard: Asus Z10pe-d8 ws
    GPU Used: 2x980ti Hybrids @ 1480mhz+(Core)/8150+{Memory)
    CPU: 2 Xeon e5 2678 v3 @ 2.9-3.4ghz(12 cores/24threads each cpu)
    PSU(Power Supply Unit): Seasonic 1200w gold(Single Rail)
    Main Ram: 64+GB Samsung ecc memory

    Also just to give everyone heads up, based on overall image quality from what I have observed:

    2560x1600 along with 2xmsaa + 2xSGSSAA is better then 4k(2400p/WITHOUT AA), This was while messing with Assassin Creed Unity.

    Not only that, You get vastly more Improvements in the FPS dept. if you do 1600p + x2 MSAA + x2 SGSSAA then just straight up 4k(2400p) with NO AA, but like I said before, this was just tested with that one game only, I have not tried it with others so I can not quote on that.

    I find it Quite interesting, very very interesting results.

    But this was just observed with one game, some AA bits react differently.

    For instance, If you use the bits in nvidia inspector, for Mass Effect 3, if you change it to 0x000000C1, the overall AA injection quality a little bit improved but Certain things, like some text just disappear, like in the option window, viewing certain things, like your journal, its missing, some adverse reaction from the AA bits take place in the game, like blank but the AA injection is stronger then the factory AA bits that normally comes with.
     
    Last edited: Jul 11, 2019
    fantaskarsef likes this.
  17. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    Another Great game, Lords of the Fallen @ 5k Max Except Texture:

    This is my recording:

    Bottom Left Fraps. I dont use fraps much anymore, I mainly use Afterburner Full OSD.

    1 1080ti:

     
    Last edited: Mar 28, 2019
  18. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    I highly agree with that and others will agree, get the best single card solution for your needs, and no need to tune it like me, lol, Takes TO MUCH TIME and work.

    But this is what I love to do. And never made a single penny from it from the very beginning, all this time and effort I did, not one penny.

    Although I was looking into something like that to help me along the way, but I ain't that kinda way, was looking into patreon or something along those lines, but I never do it for money, I did cause I love doing all this shi!, and give results to people whos interested. The only thing is that my some of my views are less then 15 in 2 years, lol, can't get any worse then that.

    I haven't record any games yet, My main focus now is to get 4 way running on all my games.

    Games I am interested to test:

    Sekiro: Shadow Die Twice(Just bought it, Downloading and trying it out later).
    Metro Exodus(Don't Own yet)
    Shadow of tomb Raider(Don't Own yet)
    Kingdom Deliverance(Don't Own yet)
    Anthem(Don't Own yet)
    Devil May Cry 5(Don't Own yet)

    And host of others, very much interested testing Sekiro, Will give heads up asap when I get a chance, Right now working on Far Cry 3's DLC: Blood Dragon.

    So far its working prettie good, just the scaling goes down when your in a certain area, then it goes back up again. Needs more tuning. Blood Dragon is good fps game, rather old but Very good.
     
    Last edited: Mar 28, 2019
    fantaskarsef likes this.
  19. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    Ok, I will be uploading some quick video record on temps, from 1 card to 4 card performance, game is going to be rise of tomb raider.

    Speaking of Rise of tomb raider, Even at 4 core/8 threads runs excellent, had to adjust my core setting for certain games, went on this game, had 4 core by accident, still ran great, supa smooth.

    This is just a quick video to show everyone real quick on temps and performance from 1 to 4.

    Will be uploaded in 1080p, takes to long at 4k.

    This is not really a game play video, this is merely just a quick test to show everyone on certain things.

    Actual video game play on games in general will follow later, thanks.
     
    fantaskarsef likes this.
  20. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    3,685
    Likes Received:
    477
    GPU:
    4 GTX 970-Quad Sli
    Ok something Extremely huge just came up, actually BREAKING NEWS.

    I recently did a 4k recording of the stuff I mention on last post.

    Listen to this, I tried recording 2160p,1440,and 1080p, the performance hit is nearly the same, My jaws went down.

    So I am recording Rise of tomb raider @ 2160p at bitrate of 90mbps, its about 7-10 fps hit in general...

    That wasn't the breaking news, the breaking news is this:

    I went back into nvidia inspector, I changed something within the sli hack bits and I said WTF, my scaling has vastly improved but went into the main screen, the whole screen became dark, but there is a work around it:

    I know I have used this Sli Hack bits before but I quickly Disregarded it because of the Overall brightness was Super Dark, I went back to the same hack bits again and did some testing.

    You have to load up a save file, then Change your nvidia hair to max to off, then back to the game, the screen will flicker alot, push control + atl + del, then back to the game, and walla, I had to reduce the brightness a bit cause it was kinda bright afterwards, bad thing about it though, Shadow Must be off during the benchmark session otherwise reduce scaling performance during certain scenes.

    Some tiny portions of flicker will still be present here and there but overall, I nearly got 99% of it out pretty much.

    Another thing, Absolutely no exceptions, Nvidia Hair MUST BE TURNED OFF, other wise your scaling goes out of wack, I am talking about my particular setup.

    I Observe 4 way then back to one, I almost died, basically like 75%-100% perfect scaling give or take.

    Before my sli hack bits was on:

    EARLY_PUSH_ALL_SUBRESOURCES - Defer early push of subresources with hope that we an early push the entire resource.

    This particual hack bits works prettie good but the average scaling is like 3 way'ish with a little touch of the 4th card, the only time you see excellent scaling is when you apply SSA to X2, then you see Really good scaling.

    But if your not using SSA and want to use SMAA, this is the one I used:

    SLI HACK BITS:

    FORCE_REPORT_NO_SLI_SUPPORT_FROM_NVAPI - If applition has buggy its own SLI support we need to do SLI magic in the driver and force app think it's run on single GPU.

    This is the one I dropped my jaws to the floor, although you have to work around the dark and flickering issue which I already know how to do it.

    I'll show it later, Scaling is an absolute beast. Also on top of that, I notice that even on SMAA, the overall AA is stronger on SLI HACK BITS:

    FORCE_REPORT_NO_SLI_SUPPORT_FROM_NVAPI - If applition has buggy its own SLI support we need to do SLI magic in the driver and force app think it's run on single GPU.

    Rather then using:

    EARLY_PUSH_ALL_SUBRESOURCES - Defer early push of subresources with hope that we an early push the entire resource.

    Speaking of which, The game Hellblade if Ran in 2 way sli mode, you must use That sli Hack bits: Force_Report_No_Sli_support_From_Nvapi, when you play the game, screen will flicker alot, just do cntrl+atl+del, then back on the game.

    And just to let everyone know, the temps are well under control, Actually prettie darn good, Full video on temps and 1-4 way quick video performance very soon, so check this particular thread often.

    I'll do my best to try to upload it tommorrow, Excellent thing I just found this out like just recently. Video later, thanks guys.
     
    Last edited: Mar 31, 2019
    fantaskarsef and SpajdrEX like this.

Share This Page