TR Forums

Spec
*CPU i7 7700k
*GPU: Asus 1080ti Strix X 2
*Motherboard: Msi z270 Titanium
*Corsair Vengeance LPX 32 GB
*Seasonic 1050W Snow Silent PSU


I am pulling my hair out at this point. I will try to provide as much information as possible.
So I had bought a second Asus 1080ti strix about a week ago; while eagerly waiting for the matched Ek waterblock to arrive, I plugged it into my rig to test out the sli configuration and the much anticipated improved gaming performance. From here on, for the purpose of convenience, I will call the old GPU that was being water cooled Oldie, and the new GPU that had the stock cooler Newie.

All was good and smooth until I noticed a sharp FPS drop whiling playing DOTA 2 ( a very low demanding game btw) from 178 FPS to 38 FPS and immediately noticed severe screen tearing. I had dealt with my fair share of graphics card problems. My first instinct was that it was either the GPU wasn’t outputting enough frames or vice versa, and since the screen tearing was coupled with a significant drop in FPS; I had concluded that it had to be the prior scenario. I initially thought, it was probably a non-issue, so I removed Newie and kept on gaming with only Oldie; life was good and fun for like 3 hours and the issue happened again, my fps would drop without warning to 145mhz and restart PC would not fix it (please note that for almost 2 years, I had never had this issue).

Funny thing was when I ran furmark benchmark test, the clock speed would randomly fluctuate from normal to low at uncertain intervals.

I opened MSI Afterburner and realized that the oldie was running at only 145mhz (GPUZ shows 100% load at 300mhz!! like WHAT??), the memory clock was running fine at 5005mhz; however Newie was running at normal 1750~1830 mhz range. So here comes the frustrating part that had me rolling in the dirt for the past 4 days, I had exhausted every last bit of the scarce knowledge I had, I just couldn’t for the **** sake of me figured out how to restore the clock to normal functioning level.

My core problem is this, I have never had any problem with Oldie before Saturday. I am in a state of denial at this point, I kinda refused to believe that my faithful companion Olide would just die on me like that.

There are some accompanying issues starting to show up that may or may not be related:
1) My monitor plugged into Newie will not have signal, only plugging into Oldie can (from prior experience, I don’t think this is necessarily related because I think it can be solved by reinstalling the driver and manually disable Oldie from device manager).
2) My monitor would not display bios setting page ( can show once it gets into windows). I knew the system was successfully booted into Bios because the z270 motherboard has a distinct message on the onboard display, just no image was shown on monitor.


The method I had tried:
1) The first thing I ruled out is thermal throttling. s Oldie never run hotter than 55 degree under 100% load.I also opened up the waterblock to make sure that all the thermal pads were correctly placed and I put heat sensor on the vram plates to ascertain that temperatures were at the normal range.
2) Uninstall the driver with DDU in safemode, perform a clean installation on the newest driver available.
3) Uninstall the driver with DDU in safemode, perform a clean installation on an archived driver dating back to earlier last year.
4) I had tried setting Nvidia control panel to prefer maximum performance and disable gsync.
5) Switching among 3 new sets of PCIE cables, with or without extension.
6) Unplug the PSU cable from UPS and plug it directly into wall socket.
7) Uninstall Windows, formatted all the storage drives and reinstall the windows 10 ( I don’t have windows 7)


UPDATE 04/06/2018
Hi, sorry for the 2 week hiatus.. I have tried the methods previous posters mentioned, I am very grateful for them and I have used them in this 2 week period.

So here are the new methods I have tried (thank you for people helping me in this thread providing advice).

1) i have tried switching pcie position between Oldie and Newie. Result was surprising, Newie will run fine but Oldie still has the same problem of randomly dropping core clock speed despite being moved to primary slot. This indicates that PCie position might not be the problem.

2) I have tried using 4 independent cables to power 2 cards. This does noticeably enhance performance when the card works but won't do anything when Oldie's core clock gets stuck.. so it doesn't solve the problem. However, this method brings up an interesting problem; oldie's voltage is consistently about 5~6% (780mv idle vs 820mv idle) lower than Newie.

3) I have tried reinstall windows for so many times my brain hurts.. For each time I have thoroughly formatted my drive using diskpart before installation. This had me questioned whether the nvidia driver was not the culprit like I had previously suspected. However, windows forces me to install a bunch of crap after installation so I am not sure those subsequent forced updates interfere with the driver.

Remeber i said that I found a driver that kept Oldie working? well it worked for like a week and the problem starts showing up again.....


I have moved from denial to acceptence in the stages of grief; Before officially declare Oldie to be dead, I bought a 17 dollars windows 7 thumb drive yesterday and decided to go for one last test.

I have a question though, have you guys ever seen a card die by stuck in low clock speed? I have never seen a card DIE by outputting monitor signal fine, but just gets stuck in idle clock speed. I will update pictures during the weekend, I am sorry for the lack for screenshots because it is a lot of work removing/installing water loops constantly





final diagnostic
card is broken 99% sure to be hardware problem

Did I miss the part where you concluded Oldie broke its hip, you physically disconnected it, and everything ran fine using only Newie?
Have you ruled out a PSU issue? For instance, have you tried swapping the aux power from one GPU to the other to rule out that one set of 8-pin PCIe didn't die on you? Are you using the 4 single PCIe 8-pin cables that came with your PSU and not the single-to-double ones? Or if you are using the double 8-pin cables, only one 8-pin connector per cable should be plugged into the GPU.

I agree with DPete, I would try physically removing Oldie and just run Newie, see what happens.

If you had to move Oldie to install Newie, it's possible something got damaged in the process. (Which would suck so I hope that isn't the case, but it's not impossible...)

Nvidia control panel messes up the clock speeds sometimes and it can get stuck in idle. Open nvidia control panel>Manage 3D settings>Set power management mode to high performance and reboot. Also what version of windows are you running?

thank you so much for your help guys.. it's an unusual problem, I really do appreciate every bit of input from you guys so here is some update

So I disabled SLI in Nvidia setting and Newie can run at normal clock speed and I can play games with it; however, Oldie keeps getting stuck at 100~200mhz clock but normal mem clock (5000).

I enabled SLI and both cards are down to 100mhz, which can't do anything with really, all the games are just stuttering..

I was messing with drivers yesterday and found a driver that made I stress tested it with furmark for 12 hours and both cards were running at max with a drop for that entire duration. The problem comes back this morning after I update to the new nvidia driver...


Okay so my biggest question maybe just in denial at this point, how can a graphics card die by locking at low clock speed? I have seen plenty of cards die by not booting at all but I have never seen a card "died" but still have clock speed (just really low)..

cuethenoise wrote:

Nvidia control panel messes up the clock speeds sometimes and it can get stuck in idle. Open nvidia control panel>Manage 3D settings>Set power management mode to high performance and reboot. Also what version of windows are you running?

Running Windows 10

DPete27 wrote:

Did I miss the part where you concluded Oldie broke its hip, you physically disconnected it, and everything ran fine using only Newie?
Have you ruled out a PSU issue? For instance, have you tried swapping the aux power from one GPU to the other to rule out that one set of 8-pin PCIe didn't die on you? Are you using the 4 single PCIe 8-pin cables that came with your PSU and not the single-to-double ones? Or if you are using the double 8-pin cables, only one 8-pin connector per cable should be plugged into the GPU.

I didnt' physically disconnect it since it's very labour intensive to do with a water loop, but newie runs fine just by itself when i disable sli

DPete27 wrote:

Did I miss the part where you concluded Oldie broke its hip, you physically disconnected it, and everything ran fine using only Newie?
Have you ruled out a PSU issue? For instance, have you tried swapping the aux power from one GPU to the other to rule out that one set of 8-pin PCIe didn't die on you? Are you using the 4 single PCIe 8-pin cables that came with your PSU and not the single-to-double ones? Or if you are using the double 8-pin cables, only one 8-pin connector per cable should be plugged into the GPU.

interesting, i have never thought about the cable thing before.

I do use a cable with 2 8 pins. My package came with 4 pcie cables they are all have 2 8pin; I dont have 4 cables with single 8 pin..

I thought Nvidia no longer supported SLI .. and are you using the "correct" bridge ....
e.g. https://www.amazon.co.uk/adapter-flexible-PC24-Shop-Service/dp/B00576ESX2

drwho wrote:

I thought Nvidia no longer supported SLI .. and are you using the "correct" bridge ....
e.g. https://www.amazon.co.uk/adapter-flexible-PC24-Shop-Service/dp/B00576ESX2

Plenty of people have SLI running well, especially 1080 Ti cards, for high-end 4k/5k gaming as well as VR. It's simply the fastest economical solution available; I run a single 1080 Ti with a factory CLC myself.

xiashilang wrote:

I do use a cable with 2 8 pins. My package came with 4 pcie cables they are all have 2 8pin; I dont have 4 cables with single 8 pin..

Just an FYI, I found that info in this Jonnyguru review of your exact PSU. Apparently the cabling requirement is in the "manual" (pssh, who reads those). The stock GTX1080Ti carries a 250W TDP

Hard to believe it's a cable problem based on your description, especially since the setup had worked fine prior with the same cabling configuration.

I vote broken hip... either the power supply or the card. If it is a thermal failure in the GPU, likely it will go from room temperature to meltdown in seconds, faster than the typical temperature monitoring software will ever catch. The internal throttling code will catch a thermal issue, and react accordingly.

drwho wrote:

I thought Nvidia no longer supported SLI .. and are you using the "correct" bridge ....
e.g. https://www.amazon.co.uk/adapter-flexible-PC24-Shop-Service/dp/B00576ESX2

https://smile.amazon.com/gp/product/B071RTWWXZ/ref=oh_aui_search_detailpage?ie=UTF8&psc=1
that's my sli bridge

DPete27 wrote:

xiashilang wrote:

I do use a cable with 2 8 pins. My package came with 4 pcie cables they are all have 2 8pin; I dont have 4 cables with single 8 pin..

Just an FYI, I found that info in this Jonnyguru review of your exact PSU. Apparently the cabling requirement is in the "manual" (pssh, who reads those). The stock GTX1080Ti carries a 250W TDP

oh wow i will definitely try it, good catch

No doubt there are other more probabilistic avenues of failure than cabling. I was reading the JonnyGuru review and saw that bit of info, which I figured few people would take notice of while building (I would personally steer away from that to reduce cable clutter without thinking much of it), so I figured I'd throw it out there as a PSA. Apparently the provided split 8-pin cables aren't quite thick enough to handle >225W "safely" in the interest of preserving cable flexibility. In that same vein, I wondered if the terminals on the back of the PSU have similar current limitations, so pulling more than 225W from a particular 8-pin terminal may be detrimental. Not to mention that with water cooling (presumably overclocking), the OP may be pulling >100W more than the 250W stock TDP.

OP, have you tried the " Open nvidia control panel>Manage 3D settings>Set power management mode to high performance and reboot." suggestion?

roncat wrote:

Hard to believe it's a cable problem based on your description, especially since the setup had worked fine prior with the same cabling configuration.

I vote broken hip... either the power supply or the card. If it is a thermal failure in the GPU, likely it will go from room temperature to meltdown in seconds, faster than the typical temperature monitoring software will ever catch. The internal throttling code will catch a thermal issue, and react accordingly.

I am already mentally prepared for the card to be broken.... but when it is running at the normal clock I stress test for 10+ hours with furmark and no issues.... it's just once it gets stuck at low clock speed (unpredictable), you can't get out of it. I actually prefer it just stay low at 200mhz so that i can give up all hope; it just keeps teasing me lol :<

DPete27 wrote:

No doubt there are other more probabilistic avenues of failure than cabling. I was reading the JonnyGuru review and saw that bit of info, which I figured few people would take notice of while building (I would personally steer away from that to reduce cable clutter without thinking much of it), so I figured I'd throw it out there as a PSA. Apparently the provided split 8-pin cables aren't quite thick enough to handle >225W "safely" in the interest of preserving cable flexibility. In that same vein, I wondered if the terminals on the back of the PSU have similar current limitations, so pulling more than 225W from a particular 8-pin terminal may be detrimental.

OP, have you tried the " Open nvidia control panel>Manage 3D settings>Set power management mode to high performance and reboot." suggestion?

yes, not only did i change the nvidia global setting, I manually change all the programs to prefer max performance... ~~

xiashilang wrote:

stress test for 10+ hours with furmark and no issues.... it's just once it gets stuck at low clock speed (unpredictable), you can't get out of it.

Missed that before in the Oldie/Newie discussion.

xiashilang wrote:

found a driver that made I stress tested it with furmark for 12 hours and both cards were running at max with a drop for that entire duration. The problem comes back this morning after I update to the new nvidia driver...

That would be the golden spike right there. Don't you think? Driver problem. Go back to that older driver. I'm pretty sure DOTA 2 is about as optimized as it can get at this point

I capture one instance of the problem.

guys check the GPU 2's clock graph, GPU 1 is running fine

it runs low at 200mhz but it will jump back to normal 1400+.

DPete27 wrote:

xiashilang wrote:

stress test for 10+ hours with furmark and no issues.... it's just once it gets stuck at low clock speed (unpredictable), you can't get out of it.

Missed that before in the Oldie/Newie discussion.

xiashilang wrote:

found a driver that made I stress tested it with furmark for 12 hours and both cards were running at max with a drop for that entire duration. The problem comes back this morning after I update to the new nvidia driver...

That would be the golden spike right there. Don't you think? Driver problem. Go back to that older driver. I'm pretty sure DOTA 2 is about as optimized as it can get at this point

thank you, i will definitely change it back..

I think only 1 of 5 patches worked out meant that maybe there are more intricate problems than just the driver; I found it highly unlike for nvidia to publish unstable drivers four times in a row. I did some digging it seems like the recent Windows creator update may also be the culprit. I am currently trying to roll back windows 10.

I've been avoiding this, but please tell me this (SLI) is worth all the trouble since you're using a 4k display.

DPete27 wrote:

I've been avoiding this, but please tell me this (SLI) is worth all the trouble since you're using a 4k display.

Only thing that'll get you close to running sub-16.7ms frametimes (solid 60FPS) is a US$3000 Tesla V. 1080Ti SLI is as good as it gets, and it's still not really fast enough.

[and SLI works great if it's even remotely planned for by developers]

Airmantharp wrote:

DPete27 wrote:

I've been avoiding this, but please tell me this (SLI) is worth all the trouble since you're using a 4k display.

Only thing that'll get you close to running sub-16.7ms frametimes (solid 60FPS) is a US$3000 Tesla V. 1080Ti SLI is as good as it gets, and it's still not really fast enough]

My statement was more along the lines of hoping the OP has a monitor(s) that requires SLI GTX 1080Tis and not going through all this headache to primarily play DOTA 2 at 1920x1080

Not trying to de-rail this thread if the OP still needs help.

Hi, sorry for the 2 week hiatus.. I have tried the methods previous posters mentioned, I am very grateful for them and I have used them in this 2 week period.

So here are the new methods I have tried (thank you for people helping me in this thread providing advice).

1) i have tried switching pcie position between Oldie and Newie. Result was surprising, Newie will run fine but Oldie still has the same problem of randomly dropping core clock speed despite being moved to primary slot. This indicates that PCie position might not be the problem.

2) I have tried using 4 independent cables to power 2 cards. This does noticeably enhance performance when the card works but won't do anything when Oldie's core clock gets stuck.. so it doesn't solve the problem. However, this method brings up an interesting problem; oldie's voltage is consistently about 5~6% (780mv idle vs 820mv idle) lower than Newie.

3) I have tried reinstall windows for so many times my brain hurts.. For each time I have thoroughly formatted my drive using diskpart before installation. This had me questioned whether the nvidia driver was not the culprit like I had previously suspected. However, windows forces me to install a bunch of crap after installation so I am not sure those subsequent forced updates interfere with the driver.

Remeber i said that I found a driver that kept Oldie working? well it worked for like a week and the problem starts showing up again.....


I have moved from denial to acceptence in the stages of grief; Before officially declare Oldie to be dead, I bought a 17 dollars windows 7 thumb drive yesterday and decided to go for one last test.

I have a question though, have you guys ever seen a card die by stuck in low clock speed? I have never seen a card DIE by outputting monitor signal fine, but just gets stuck in idle clock speed. I will update pictures during the weekend, I am sorry for the lack for screenshots because it is a lot of work removing/installing water loops constantly

Sorry for catching up late (and I haven't read every single post) - but are the BIOS versions the same between the two cards?

Waco wrote:

Sorry for catching up late (and I haven't read every single post) - but are the BIOS versions the same between the two cards?

no, they are not.. but i dont know how to update gpu bios for 1080ti strix

msi afterburner shows both cards have the same bios while asus gpu tweak shows one card has as15. the o ther has as11

There are a variety of ways to backup and flash firmware/BIOS. It might be worth asking ASUS if they have a preferred/supported tool. You could backup the existing BIOS on the oldie and flash the BIOS from newbie onto it to see if things change. I used to do this a lot when matching cards for SLI to ensure clocks were as identical as possible.

Waco wrote:

There are a variety of ways to backup and flash firmware/BIOS. It might be worth asking ASUS if they have a preferred/supported tool. You could backup the existing BIOS on the oldie and flash the BIOS from newbie onto it to see if things change. I used to do this a lot when matching cards for SLI to ensure clocks were as identical as possible.

i looked up guides and used nvflash to flash both cards to the same bios and problem persists.. I think it is hardware problem at this point, I am gonna open the card up and check if there are rust spots.... My detective sense is telling me, I might have dropped some coolant on it.. However, I gotta stop the project for my own sake, this is taking way too much of my time; water cooling adds so much to the difficulty when something goes wrong.. I might just wait for volta