SLI with different cards

[PKBO]

Win XP has never been supported by DifferentSLIauto.

With a huge stroke of luck however patching the 368.81 drivers might work with DifferentSLIauto1.5 (only if patch locations happened to be the same as in Win 7)

It has to work, i found Popraminos submition on HWBOT also with 285.52 even under 32bit https://community.hwbot.org/topic/175823-sli-of-gtx-285-problem/ Maybe some way, how to change path of files colation ??

 

[P!nkpanther]

Driver location can be found using Device Manager → Double click main display adapter → Tab: Driver → Driver Details and check the location of nvlddmkm.sys

Maybe you could use DifferentSliAuto 1.4 on 285.62 drivers?
Anyways, I can't help you with things related to legacy OSs.
You have to find out by yourself which driver versions can be patched with which DSLIA version. If patching is successful, you can proceed trying to make an install according to the install procedures posted in this forum.

 

[PKBO]

Anyone has working link for hyper SLI 1.0 or sli patch 1.5B please ? The one, provided by Anatolymik unforcenetly not working anymore.
I think that this should work under XP.

 

[Pretentious]

Hi, is it possible to patch also driver for win XP 64 ? how ? it doesnt have this name in system32..

What @P!nkpanther said plus maintaining offsets for Windows XP for just a selected few it's not really worth my time. I don't even use my own patches anymore, I only keep the current offsets up to date because there's a few people that still rely on it.

 

[venturi]

hi, yes, I'm stuck
way back I enabled 4 way sli with titan xp pascal (cool bits, inspector etc) wrote profiles and got quad sli working with scaling depending on game from 1 (no scaling) to about 3.3x.

now I am trying to get dx12 explicit mgpu to work but this requires sli to be enabled.

So I'd like to enable sli, even if the sli auto diff isn't exactly perfectly scaled, for the purpose of getting mgpu to work. If sli works as well, that would be great.

So I have 4 identical gpus. (One is a 32gb version, but they are identical chipsets).

so

can you please enumerate how to use the utility to get this to go, I've tried and had little luck. I'm not a newbie but can't seem to get there on this.

I also have a 4th titan v if i have to not use the 32gb CEO ed as primary (but would rather keep it).

My board is SLI compatible, no titan v does not have nvlink bridge, yes I have 32x/16x lanes per PCIE slot, yes I have over 100 lanes to the cpus, yes I have plenty of ram, yes I have plenty of processing power, yes, I have plenty of electrical power, yes I have adequate cooling, yes I have an OS and games that can use it, yes, once SLI is enabled I know how to enable 4 way.

For one moment I had it working on the 399.40 driver and the puff, never got it back on any driver after that. I'm trying to get it to work on 419.35. (I'd like to move on to a newer driver). I've figured out a few things with nvidia and the ray Tracing and I'm trying to experiment on a few things.


system spec:

Asus c621e sage
Dual Xeon 8180M
384GB of ram
4x Titan V including Titan V CEO ed 32GB
960pro nvme (OS)
10x SSD RAID (Apps and Backup 8x 2x )
1600W PSU (digital) noiseless
MS 2016 Data Center, Ubuntu
TT miniITX case



I'd rather learn to build the fire than have someone do it for me, I'm just a stopping point on 417.22, 419.17 and 419.35


Thank you in advance, really




https://www.techpowerup.com/forums/...st-powerful-god-box-in-a-miniitx-case.252663/

 

[topmysteries5]

hi, yes, I'm stuck
way back I enabled 4 way sli with titan xp pascal (cool bits, inspector etc) wrote profiles and got quad sli working with scaling depending on game from 1 (no scaling) to about 3.3x.

now I am trying to get dx12 explicit mgpu to work but this requires sli to be enabled.

So I'd like to enable sli, even if the sli auto diff isn't exactly perfectly scaled, for the purpose of getting mgpu to work. If sli works as well, that would be great.

So I have 4 identical gpus. (One is a 32gb version, but they are identical chipsets).

so

can you please enumerate how to use the utility to get this to go, I've tried and had little luck. I'm not a newbie but can't seem to get there on this.

I also have a 4th titan v if i have to not use the 32gb CEO ed as primary (but would rather keep it).

My board is SLI compatible, no titan v does not have nvlink bridge, yes I have 32x/16x lanes per PCIE slot, yes I have over 100 lanes to the cpus, yes I have plenty of ram, yes I have plenty of processing power, yes, I have plenty of electrical power, yes I have adequate cooling, yes I have an OS and games that can use it, yes, once SLI is enabled I know how to enable 4 way.

For one moment I had it working on the 399.40 driver and the puff, never got it back on any driver after that. I'm trying to get it to work on 419.35. (I'd like to move on to a newer driver). I've figured out a few things with nvidia and the ray Tracing and I'm trying to experiment on a few things.


system spec:

Asus c621e sage
Dual Xeon 8180M
384GB of ram
4x Titan V including Titan V CEO ed 32GB
960pro nvme (OS)
10x SSD RAID (Apps and Backup 8x 2x )
1600W PSU (digital) noiseless
MS 2016 Data Center, Ubuntu
TT miniITX case



I'd rather learn to build the fire than have someone do it for me, I'm just a stopping point on 417.22, 419.17 and 419.35


Thank you in advance, really




https://www.techpowerup.com/forums/...st-powerful-god-box-in-a-miniitx-case.252663/

Man, you really have a GOD OF PC.
Why you want SLI ? This thing don't require SLI. Single Titan V CEO is more than overkill for anygame.
BTW, if you want to SLI, then you need to remove Titan V CEO as it is 32GB VRAM, your remaining titans are 12GB each. Both titans have slight difference in ROP counts.
For SLI you need GPUs with same amount of memory ( exact same GPUs only).

 

[PKBO]

Driver location can be found using Device Manager → Double click main display adapter → Tab: Driver → Driver Details and check the location of nvlddmkm.sys

Maybe you could use DifferentSliAuto 1.4 on 285.62 drivers?
Anyways, I can't help you with things related to legacy OSs.
You have to find out by yourself which driver versions can be patched with which DSLIA version. If patching is successful, you can proceed trying to make an install according to the install procedures posted in this forum.

Win XP doesnt use nvlddmkm.sys but nv4_disp.dll (i think) (or nv4_mini.sys) thats why i was asking, if its possible to tweak patchers path / name of driver.. (but its probably totaly different)

 

[P!nkpanther]

Finding the correct offsets to patch will be the main problem, not just the file names.
In other words, if a patcher was made for a different filename, the offsets probably aren't usable either.

Some of the old sources and builds can be found here:
https://github.com/EmberVulpix/DifferentSLIAuto/releases

now I am trying to get dx12 explicit mgpu to work but this requires sli to be enabled.
So I'd like to enable sli, even if the sli auto diff isn't exactly perfectly scaled, for the purpose of getting mgpu to work. If sli works as well, that would be great.
So I have 4 identical gpus. (One is a 32gb version, but they are identical chipsets).

What you want is not SLI but Linked Node Mode (explicit LDA) for DX12 mGPU. But this depends on a hardware feature (NVLink) and that's precisely what's disabled in Titan Vs.
DifferentSLIauto only bypasses the checks for MoBo certification (and in earlier versions also the checks for different cards). So this will not help with your problem, which is enabling a hardware feature that has been blocked by nVidia (the NVLink fingers on Titan Vs are not even connected at all).

Anyways: Great comp you got there...!

 

[venturi]

Finding the correct offsets to patch will be the main problem, not just the file names.
In other words, if a patcher was made for a different filename, the offsets probably aren't usable either.

Some of the old sources and builds can be found here:
https://github.com/EmberVulpix/DifferentSLIAuto/releases


What you want is not SLI but Linked Node Mode (explicit LDA) for DX12 mGPU. But this depends on a hardware feature (NVLink) and that's precisely what's disabled in Titan Vs.
DifferentSLIauto only bypasses the checks for MoBo certification (and in earlier versions also the checks for different cards). So this will not help with your problem, which is enabling a hardware feature that has been blocked by nVidia (the NVLink fingers on Titan Vs are not even connected at all).

Anyways: Great comp you got there...!

thank you,
it is more complicated... the 2 nvlink in Titan V is hardware enabled. The driver has an "ignore" function (software disabled). Also, the card shroud is in the way of getting to the two nvlink per card. I used a temp cpu water block cooler on two cards to test and the nvlink is working. The driver.... is a different story. If you open up the card bios, the link asset is available. I'm considering a machine shop removing the 1.5mm lip of the front shroud to allow the dual links. I don't have the back plate on so that is not an issue.

I've tried the bios from the quadro equivalent (that was a lot of other issues) and the link is there as well. But the driver won't respect the quadro bios on the V card because the identifier is hardcoded / part of the card. So ---- I haven't given up yet.

yes, I've tested with the $600 a piece links... the whole debacle just feels like a lot of potential is being selectively prevented to not go near their higher end quadro tier and allow breathing room on the rtx series. ----just my opinion on the latter.

As a single card, each titan v is very competitive with the rtx titan. I had bought (then returned) 2 Titan rtx and my overall impression was that they are "less" of a card than the titan v. In 4k and 5k scenes with max eye candy (qualifying statement: I don't use anything higher than TAA, I don't use DoF, and I don't use motion blur, everything else is maxed) the titan v was 2-11% faster and more importantly more consistent on frames and just more "umph".--- maybe from the HBM2 or maybe because it a more populated chip. I realize everyone else says it can't do ray tracing, but in metro exodus it seems to run fine, albeit the inefficiency of ray tracing and the fact that at 4k with ray tracing on 'Ultra" the CEO ed V on metro exodus version 1.00 can only do 47-49fps (yes, the latest metro patch seems to have made improvements in ray tracing, but I have not installed that patch yet). lastly the rtx titans don't work as well on compute and this rig is not just about looking at Lara Croft's backside.

well, I'll quit lamenting.... just get frustrated with nvidia, their marketing strategy, etc etc.

 

[P!nkpanther]

Well, that does not sound good...
So I understand that it's actually only disabled in the driver as well as in the card's BIOS, but not in actual hardware (I read somewhere, that the NVLink isn't even connected on the TitanV, but that must have been false).
Then what you need is a 4xx driver which would fully work with the Quadro BIOS on Titan V, I guess.
Does not look easy...

 

[genesys]

I just tried to use DifferentSLI Patcher v1.4 to get my two GTX 1060 (6GB) to run in SLI, but when I run the patcher it says

[E] : Could not find patch #2. Please inform Ember @ techPowerUp! forums.

What's to do?

 

[Darekills4]

I just tried to use DifferentSLI Patcher v1.4 to get my two GTX 1060 (6GB) to run in SLI, but when I run the patcher it says

[E] : Could not find patch #2. Please inform Ember @ techPowerUp! forums.

What's to do?

Don't use the patcher. Read a few pages prior to this post, manually patch the drivers.

 

[P!nkpanther]

I just tried to use DifferentSLI Patcher v1.4 to get my two GTX 1060 (6GB) to run in SLI, but when I run the patcher it says

[E] : Could not find patch #2. Please inform Ember @ techPowerUp! forums.

What's to do?

Start here:
https://www.techpowerup.com/forums/threads/sli-with-different-cards.158907/post-3987274

Except using these offsets for 419.35:
https://www.techpowerup.com/forums/threads/sli-with-different-cards.158907/post-4007314

 

[Pretentious]

thank you,
it is more complicated... the 2 nvlink in Titan V is hardware enabled. The driver has an "ignore" function (software disabled). Also, the card shroud is in the way of getting to the two nvlink per card. I used a temp cpu water block cooler on two cards to test and the nvlink is working. The driver.... is a different story. If you open up the card bios, the link asset is available. I'm considering a machine shop removing the 1.5mm lip of the front shroud to allow the dual links. I don't have the back plate on so that is not an issue.

I've tried the bios from the quadro equivalent (that was a lot of other issues) and the link is there as well. But the driver won't respect the quadro bios on the V card because the identifier is hardcoded / part of the card. So ---- I haven't given up yet.

yes, I've tested with the $600 a piece links... the whole debacle just feels like a lot of potential is being selectively prevented to not go near their higher end quadro tier and allow breathing room on the rtx series. ----just my opinion on the latter.

As a single card, each titan v is very competitive with the rtx titan. I had bought (then returned) 2 Titan rtx and my overall impression was that they are "less" of a card than the titan v. In 4k and 5k scenes with max eye candy (qualifying statement: I don't use anything higher than TAA, I don't use DoF, and I don't use motion blur, everything else is maxed) the titan v was 2-11% faster and more importantly more consistent on frames and just more "umph".--- maybe from the HBM2 or maybe because it a more populated chip. I realize everyone else says it can't do ray tracing, but in metro exodus it seems to run fine, albeit the inefficiency of ray tracing and the fact that at 4k with ray tracing on 'Ultra" the CEO ed V on metro exodus version 1.00 can only do 47-49fps (yes, the latest metro patch seems to have made improvements in ray tracing, but I have not installed that patch yet). lastly the rtx titans don't work as well on compute and this rig is not just about looking at Lara Croft's backside.

well, I'll quit lamenting.... just get frustrated with nvidia, their marketing strategy, etc etc.

Does anyone know what happens if let's say you have two 1080TI's and one 1070? Would the NVIDIA control panel allow to enable SLI on the 1080TI's or would the option be grayed out?

 

[Leongel]

Sorry for interraption, but anyone knows how it works with redshift? What about it, I'm not sure that CUDA cores can be summed. I have 1080 for it, can I buy 1xxx or only 1070/ti?
Have anyone X399 motherboard like ASUS ROG Zenith Extreme, is it work?

 

[king of swag187]

Only GPU's with the same GPU core can be SLI'd

 

[topmysteries5]

Does anyone know what happens if let's say you have two 1080TI's and one 1070? Would the NVIDIA control panel allow to enable SLI on the 1080TI's or would the option be grayed out?

I did this with GTX 1050ti + GTX 1080 and GTX 1070 + GTX 1080, NVIDIA control panel allows to enable SLI (doesn't gray-out), but I get black screen, PC still on (no crash / freeze / BSOD).
If i use 2x GTX 1070 + GTX 1080, enabling SLI causes black screen. If GTX 1080 is disabled from device manager. Then GTX 1070 works in SLI without any problem.

 

[Coler]

I recently tried a SLI config with 2x1050 Ti on Win7 with old patch, and driver 388.13. Motherboard has a x16 and a x4 over chipset, both 2.0,
once in the past I did a similar thing on x16\x4 (1.1) and using SLI, performance was better, but in this case is worse.
Using x4 single card performance is almost the same as x16, so I tried to cover those pins to make both x4 , but it's even worse than before.
I already DDUed, and looking at some posts here it seems that something is wrong since it should have better performance than single card..
Also, running two game tests at same time, one on each gpu, performance is good on both. So power supply is ok.

Does anybody have an idea about this issue?

 

[topmysteries5]

I recently tried a SLI config with 2x1050 Ti on Win7 with old patch, and driver 388.13. Motherboard has a x16 and a x4 over chipset, both 2.0,
once in the past I did a similar thing on x16\x4 (1.1) and using SLI, performance was better, but in this case is worse.
Using x4 single card performance is almost the same as x16, so I tried to cover those pins to make both x4 , but it's even worse than before.
I already DDUed, and looking at some posts here it seems that something is wrong since it should have better performance than single card..
Also, running two game tests at same time, one on each gpu, performance is good on both. So power supply is ok.

Does anybody have an idea about this issue?

Try it with newer driver 419.35 on windows 10 only.
What motherboard and cpu you have ? Maybe you're getting CPU bottlenecked. Even i7 4930K bottlenecked gpus like GTX 980.
PCIe 2.0x16 is 8GB/s and PCIe 2.0x4 is 2GB/sec. This is why you're getting poor scaling.
Check Page-120 Post#3000 and Page-121 Post#3012, i get 100% scaling in 2 way and upto 80% in 3 way SLI with 1050ti.

 

[Coler]

Well cpu usage is higher using single card, since does more FPS, even running two tests at once on each gpu it runs better than a single one SLIed.
Maybe I still have to learn many things about SLI , I can imagine that syncing the two cards could cost some more but, does it need 4x or more times the bandwidth and cpu power than the single card to run well? or requirements are proportionated to the expected performance? Like for example, let's say 2x 1050 Ti SLI = 1070 , does it need so more cpu power and pci ex bandwidth the SLI config than a single 1070?

Thanks for explaining me.

PS: I'll try win10 with latest driver then!

 

[topmysteries5]

Well cpu usage is higher using single card, since does more FPS, even running two tests at once on each gpu it runs better than a single one SLIed.
Maybe I still have to learn many things about SLI , I can imagine that syncing the two cards could cost some more but, does it need 4x or more times the bandwidth and cpu power than the single card to run well? or requirements are proportionated to the expected performance? Like for example, let's say 2x 1050 Ti SLI = 1070 , does it need so more cpu power and pci ex bandwidth the SLI config than a single 1070?

Thanks for explaining me.

PS: I'll try win10 with latest driver then!

No, with multi gpu setups, games are rendered in "alternate frame rendering" method. one frame will be rendered by GPU1, another frame by GPU2, and your cpu must be fast enough to sync all these alternate frames. or you get stuttering / fps drops in games. You require higher pcie bandwidth bcoz u aren't using SLI bridge on your 1050ti as they don't have any sli connector
2x1050ti is faster than 1060, but not even close to gtx 1070. Try to get atlease PCIe 3.0x8 on both of your gpus. less than that results in poor scaling due to higher latency.

 

[Coler]

yep the 1070 was just an example, thanks for explaning.
Was expecting something like that, probably it's due the lack of SLI bridge

even if it still sound strange to me,
https://www.tomshardware.com/reviews/amd-crossfire-nvidia-sli-multi-gpu,2678-5.html .
http://www.hardocp.com/article/2010/08/25/gtx_480_sli_pcie_bandwidth_perf_x16x16_vs_x4x4

I know those are old cards, but doesn't explain the performance loss, there should be at least a little improvement even without the bridge, or I'm wrong?

 

[Heavenwood]

419,67
=====================
00000000002c3343: 75 С7
00000000002c3344: 05 43
00000000002c3345: 0F 24
00000000002c3346: BA 00
00000000002c3347: 6B 00
00000000002c3348: 24 00
00000000002c3349: 0E 00

Collected by analogy over

 

[Coler]

419,67
=====================
00000000002c3343: 75 С7
00000000002c3344: 05 43
00000000002c3345: 0F 24
00000000002c3346: BA 00
00000000002c3347: 6B 00
00000000002c3348: 24 00
00000000002c3349: 0E 00

Collected by analogy over

Thanks!

 

[DonnerPartyOf1]

Using Statelo's precompiled tool posted on page 123 and also when running the tools manually on LTSC 2019 I was still getting patchguard-related BSODs on a Q35 VM. I downgraded to LTSB 2016 and attempted it again and found that Windows would present an error screen about 15-20 seconds after running Sea.exe indicating that the Shark driver failed to install because it was unsigned. I ran Sea.exe again, waited another 60 seconds or so and received no additional error, then patched with dsefix. This worked. I wiped and did a clean install of LTSC 2019 and can reproduce this behavior: Sea.exe needs to be run twice before dsefix, even though Windows doesn't actually produce the error message on the latest version. Running it once results in a BSOD within a few hours.

Executing Sea twice before dsefix I've successfully run the VM for over 48 hours straight and for 2 24-hour periods without any issues and played several EAC and BE games. It seems like a stupid and simple thing but I'm posting here in the hopes that anyone still having issues with anticheat and PatchGuard might be assisted by the same fix.