Overclocking my GPU is causing a lower result

[TheRealDeal]

You really want to keep temps below 80c at these voltage levels, the lower the better. Put on some headphones and forget about the noise, that's what i used do with my 290X reference cards.

Only had experience of my two pcs+ 290's which can handle a lot more volts on the stock profile. Best i had was 1220 core and 1500mhz at 200mv. I did run 100% fan for this run though. That card was better than my current one though which is not so stable. The temps are as good though. The Pcs+ cooler is among the best though.

 

[beany_bot]

PC completely crashed playing GTAV. Possibly the PSU. I've removed the overclock for now. Just not worth it.

 

[TheRealDeal]

PC completely crashed playing GTAV. Possibly the PSU. I've removed the overclock for now. Just not worth it.

It's not really tbh. I don't bench enough to be forcing Volts through my card. The 390x is still a fast card any how. Different games can make previously stable clocks flake out. My guess your crash was due to memory most likely. That's what crashes my card the most anyway.

If you fancy giving it ago again try gta v same core but with no memory OC.

There are games that don't see much benefit and others that can show a lot more.

I suppose this is what interests those that like to tinker the most as it takes more time to test lots of games to finally settle on an everyday overclock along with what the top end is on things like Firestrike.

 

[LtMatt]

PC completely crashed playing GTAV. Possibly the PSU. I've removed the overclock for now. Just not worth it.

Put the memory back to stock and focus on the core, Rome wasn't built in a day.

For finding a 24/7 stable overclock, the best way to do it is to pick a starting core clock, say 75Mhz over stock and test by gaming for 20 minutes. If it does not crash, put the core up another 25Mhz until it does. If it does crash, add voltage. If it crashes quickly, put voltage up three notches. If it crashes after 10 minutes, put it up two notches. Rinse and repeat till you find the maximum.

You'll need to be patient.

 

[beany_bot]

I appreciate the help but im going to leave it. Waiting for 4K and Vega or something that can run it. the 390x is doing me well at 1080p for now. No point chasing a few extra FPS. Also my PSU shouldn't really be doing this. Not with my CPU clock also.

 

[beany_bot]

Put the memory back to stock and focus on the core, Rome wasn't built in a day.

For finding a 24/7 stable overclock, the best way to do it is to pick a starting core clock, say 75Mhz over stock and test by gaming for 20 minutes. If it does not crash, put the core up another 25Mhz until it does. If it does crash, add voltage. If it crashes quickly, put voltage up three notches. If it crashes after 10 minutes, put it up two notches. Rinse and repeat till you find the maximum.

You'll need to be patient.

Thats exactly what I was doing all day....but in heaven. Had it totally nailed. But it appears GTAV is much harder on GPU's than heaven with everything maxed out....go figure. Honestly it's fine as it is.

 

[LtMatt]

Thats exactly what I was doing all day....but in heaven. Had it totally nailed. But it appears GTAV is much harder on GPU's than heaven with everything maxed out....go figure. Honestly it's fine as it is.

Heaven is useless, every 5 seconds you have a scene transition and the GPU gets a 1-2 second break.

GTA V is a good GPU bench test, as is Doom, Farcry 3, Sleeping Dogs.

 

[Gerard]

Be thankful its not an nvidia card you're overclocking, my experience of it seems to be the driver crashes and sticks at 595 mhz forcing you to reboot the system every-time an overclock test fails. Bloody irritating.

 

[r7slayer]

Overclocking a GPU properly can take more than a day! First day usually is finding that starting point getting your foot in the door so to speak. Then after that is just seeing how much you can open that door to that potential extra power.

Its a case of finding what you can get from stock volts, seeing what temps is like and cooling capabilities and if your bothered about noise how much noise is generated. Usually you just increase the power limits first (how much power the card can draw) and then increasing the core clock first upto when you see instability then you knock your clocks back down to what it was previously stable.

After that its a choice of if you want to proceed with adding voltage to help stabilise and push your overclock further within the cooling capability! You have to keep your card within a respectable temp. Once you have found a overclock your comfortable with then you can proceed with overclocking the memory which usually provides less gains than the core overclock. This is similar to overclocking the core. increasing untill you notice instability then dialing back to clocks to what was previously stable.


Each time you increase the core or memory however its worth doing a couple of runs of heaven for example to see if you see any instability issues. I use heaven as a starting point. Once i have found what i think is stable in heaven to where it appears might be my max overclock i then need to test with some proper gaming. This is why it can take days.


Your case appears to be the power limit being hit which is making the card throttle down to use/draw less power within the designated power limit.

 

[Ash0787]

As I understand it, sometimes when a microprocessor gets close to its limits
and there is signal degradation but not enough to cause a crash, it started to trigger some sort of parity checking system, it realizes it made a mistake and it has to redo it, thats why your framerate goes down, sometimes it doesn't detect it though and you just get visual artifacts. Not sure exactly how it works on an electronic level though.

 

[andybird123]

As I understand it, sometimes when a microprocessor gets close to its limits
and there is signal degradation but not enough to cause a crash, it started to trigger some sort of parity checking system, it realizes it made a mistake and it has to redo it, thats why your framerate goes down, sometimes it doesn't detect it though and you just get visual artifacts. Not sure exactly how it works on an electronic level though.

That tends to happen with memory overclocking - you keep increasing the clock rate and eventually the performance results will go down or itll just artefact.

If you read the thread though, you'll see he just hasnt increased the power limiter, so his card is throttlibg due to limiting, nothing to do with error checking

 

[r7slayer]

edit -0

nvm

 

[gupsterg]

Memory overclocking using works best in increments of 125Mhz, so I'd put it back to 1625Mhz and focus on the core instead, that's where most of your gains will come from.

These increments of 125MHz been shown by AMD internal testing to be better? cheers .

Why I ask is I've seen Hawaii factory OC'd AIB cards not adhering to this increment.

The 390X has a 384Bit bus so it has plenty of bandwidth already.

I thought it was 512Bit bus?

@beany_bot

The latest version of HWiNFO supports EDC counter for memory controller . This will allow you to see when OC'ing RAM if your encountering any errors, load card with a 3D app . Personally I found +50% PL wasn't needed on 4 differing Hawaii cards I owned / OC'd, generally a ~10% increase was enough (OCs upto 1150/1575 tested).

 

[LtMatt]

These increments of 125MHz been shown by AMD internal testing to be better? cheers .

Why I ask is I've seen Hawaii factory OC'd AIB cards not adhering to this increment.



I thought it was 512Bit bus?

@beany_bot

The latest version of HWiNFO supports EDC counter for memory controller . This will allow you to see when OC'ing RAM if your encountering any errors, load card with a 3D app . Personally I found +50% PL wasn't needed on 4 differing Hawaii cards I owned / OC'd, generally a ~10% increase was enough (OCs upto 1150/1575 tested).

Apologies for the typo, it is indeed 512bit. Ashamed i got that wrong when i used 290X QuadFire quite happily for many months.

As for the memory, something to do with the timings. I found from my personal testing that this worked best, but that could've just been the Hawaii GPUs i used.

 

[gupsterg]

No worries .

Thank you for the clarification , I now get what you posted .

Strap end 400MHz (40 9C 00) , Range = 150-400MHz
Strap end 800MHz (80 38 01) , Range = 401-800MHz
Strap end 900MHz (90 5F 01) , Range = 801-900MHz
Strap end 1000MHz (A0 86 01) , Range = 901-1000MHz
Strap end 1125MHz (74 B7 01) , Range = 1001-1125MHz
Strap end 1250MHz (48 E8 01) , Range = 1126-1250MHz
Strap end 1375MHz (1C 19 02) , Range = 1251-1375MHz
Strap end 1500MHz (F0 49 02) , Range = 1376-1500MHz
Strap end 1625MHz (C4 7A 02) , Range = 1501-1625MHz
Strap end 1750MHZ (98 AB 02) , Range = 1626-1750MHz

Above is an example of straps found in a Hawaii ROM, Grenada have 12 . Each strap has looser timings, so basically as RAM clock reaches near the end of a strap we are getting the best gain for the timings within that strap. ie 1250MHz will bench better than 1275MHz as 1275MHz starts using the looser timings of 1375MHz strap.

 

[Meaker]

Happens to my Fiji card's. Nano does not like anything over +12Mv. Adding more MV, while allowing me to stabilise a higher overall overclock, the overall results and FPS actually shows a net decrease in performance. I believe its fairly commonplace and a huge discussion on OCN forums.

Try upping your Mv in smaller increments and find the spot which if you go over it reduces the overall score.

Under water I have gone for stock fury x voltage on my nano to get 1120/565 24/7.