[SOLVED] Games only running on the lowest underclock possible (Zotac GTX 750Ti)

[Eternalll]

So, I bought a 750 Ti to attempt to fix (was 21$ shipping included..) basically, the problem was that the screen would ABSOLUTELY COVERED in artifacts (and then crash the PC), after playing some solution elimination rounds with this GPU, I found out it was being caused by a bad BIOS. So after the artifacts disappeared, I assumed I was done, but then it was still crashing, I assumed the clock was set too high (or low) so I checked GPU-Z, and there it was, 1375MHz, 25MHz above the max, so I opened MSI afterburner to turn down the clock. It seemed fine, and then the PC crashed again, so at this point I was like "maybe I need to lower the clock even more.." so I did, I dropped it to 609MHz Core Clock and 748MHz Memory Clock, then I tried to open Apex Legends.. And it actually worked, but as you can probably tell, this is still an issue, I can hardly play 720p with a constant FPS (36-55), I think the GPU may be severely over volted as it only gets to 4% usage before crashing (It's also constantly at 35⁰ C even though I just applied new thermal paste..)

Specs:
CPU - Xeon E3-1220 V3
GPU - Zotac 750 Ti 2GB DDR5
RAM - 2GB x4 (8 GB)
HHD - 500GB
PSU - 400w EVGA (more than 70 watts still free so this isn't it)

 

[kerberos_20]

2700mhz on gddr5 seems to be normal clock
some apps may report real clock (1350mhz), some reports it in its ddr effective form (2700mhz), and because gddr uses two data streams, some reports it as 5400mhz
which is valid in all cases
stick with gpu-z clock reporting, it should be showing real clock

 

[Eternalll]

2700mhz on gddr5 seems to be normal clock
some apps may report real clock (1350mhz), some reports it in its ddr effective form (2700mhz), and because gddr uses two data streams, some reports it as 5400mhz
which is valid in all cases
stick with gpu-z clock reporting, it should be showing real clock

Oh thanks, well, that's even more confusing then, crashing on all games at normal clock? I ran a test and it crashed at 4% usage 35 degrees, so you think it's getting overvolted?

 

[Eternalll]

Oh thanks, well, that's even more confusing then, crashing on all games at normal clock? I ran a test and it crashed at 4% usage 35 degrees, so you think it's getting overvolted?

also, both clocks are now at the regular clocks. like I said, games work with the clock as low as possible but crash at even 100MHz more than that..

 

[Eternalll]

Oh thanks, well, that's even more confusing then, crashing on all games at normal clock? I ran a test and it crashed at 4% usage 35 degrees, so you think it's getting overvolted?

Looking into undervolting but I can't find a software that allows it for a 750 Ti (MSI Afterburner only has overvolting which makes it crash faster (and with a black screen now) I brought it back down and tried only lowering the core clock but that also didn't do anything

 

[Eternalll]

I just realized, if it really was being overvolted, the new BIOS would have fixed that

 

[BFG-9000]

Did you get the BIOS from Zotac? You're not supposed to just flash randomly downloaded BIOSes because different board revisions may have different brands and models of memory chips. Instead you are supposed to open them just to view their values and transplant the ones you like into your card's original BIOS.

MaxwellBiosTweaker allows you to edit any of these parameters. You can disable boost and set any fixed clock you like, set the Power Table to under 100% or even manually adjust the speed vs voltage tables manually. Once the BIOS is modified and flashed, no overclocking software would be needed at all.

 

[kerberos_20]

if you remove cooler, then you can look at what memory ICs are being used, then you google those mem IC spec sheets, in them would be mem timings for those ICs, compare those with what your vbios has, if theres too big difference, then your vbios is not correct, manualy inserting ram timings should probably work...but maybe one or more mem ICs are busted? if one module is busted, im not sure if we can "hardware reserve" it in windows

gpu ddr (gddr) have ECC, which is error corrections, very low GPU usage due to ram errors before crashing would suggest either your vbios is not correct, or gpu ram is faulty

 

[Eternalll]

I just realized, if it really was being overvolted, the new BIOS would have fixed that

Did you get the BIOS from Zotac? You're not supposed to just flash randomly downloaded BIOSes because different board revisions may have different brands and models of memory chips. Instead you are supposed to open them just to view their values and transplant the ones you like into your card's original BIOS.

MaxwellBiosTweaker allows you to edit any of these parameters. You can disable boost and set any fixed clock you like, set the Power Table to under 100% or even manually adjust the speed vs voltage tables manually. Once the BIOS is modified and flashed, no overclocking software would be needed at all.

I got it from techpowerup's website, I got the exact model and the BIOS match all the way until the version number, I don't know if I should reset the BIOS back to the original though as that's what was causing the artifacting, but yeah I'll try this with my current BIOS

 

[Eternalll]

if you remove cooler, then you can look at what memory ICs are being used, then you google those mem IC spec sheets, in them would be mem timings for those ICs, compare those with what your vbios has, if theres too big difference, then your vbios is not correct, manualy inserting ram timings should probably work...but maybe one or more mem ICs are busted? if one module is busted, im not sure if we can "hardware reserve" it in windows

gpu ddr (gddr) have ECC, which is error corrections, very low GPU usage due to ram errors before crashing would suggest either your vbios is not correct, or gpu ram is faulty

Like I said, it matched pretty well only thing that was different was the version number, Zotac didn't actually have any of the BIOSes available to download so I got it from techpowerup.

 

[BFG-9000]

Zotac only list BIOSes for recent cards on their website but Support should be able to send you a link to it on their download server or email you one. However given that your BIOS from TPU actually boots and didn't brick the card may suggest it is indeed close enough.

I've rescued cards with dying memory before by underclocking that via the BIOS, but be aware that GDDR5 has some seriously effective error correction built-in, enough that overclocking it too far normally just slows it down. Too actually see artifacts may mean the memory is in pretty bad shape so don't hesitate to try underclocking it by a lot.

Of course it could also be a cold solder joint under one of the BGA chips that opens when it gets hot, in which case the last resort is to try baking to reflow the solder. It's only risking $21 and you've undoubtedly put more in labor than that into it, so it's just for fun and hey at least you might learn something that may prove useful in the future.

 

[Eternalll]

Zotac only list BIOSes for recent cards on their website but Support should be able to send you a link to it on their download server or email you one. However given that your BIOS from TPU actually boots and didn't brick the card may suggest it is indeed close enough.

I've rescued cards with dying memory before by underclocking that via the BIOS, but be aware that GDDR5 has some seriously effective error correction built-in, enough that overclocking it too far normally just slows it down. Too actually see artifacts may mean the memory is in pretty bad shape so don't hesitate to try underclocking it by a lot.

Of course it could also be a cold solder joint under one of the BGA chips that opens when it gets hot, in which case the last resort is to try baking to reflow the solder. It's only risking $21 and you've undoubtedly put more in labor than that into it, so it's just for fun and hey at least you might learn something that may prove useful in the future.

Thanks, this makes sense, I'm hoping it's solder-able, however Underclocking does fix some issues, so I'm assuming whoever had this card last was overclocking into the ground