Basics of Overclocking review for gtx 470

[deltalives]

Hey all, I think I need a little help.

I have an EVGA GTX 470 and I tried overclocking it before. Made a few mistakes and learned from them. My overclock (pretty conservative 705/1410/1720 @0.987V) passed EVGA OC Scanner and the custom fan profile definitely helped keep temps under control. I was thinkin of going higher but ran into a problem with Furmark and Crysis

Furmark and Crysis kinda killed my oc . They both reset the core clock back to lowest speed (450mhz) after err....yeah I don't know what (though some ppl have said to not use Furmark for oc testing). I played Crysis and kept a sharp eye on the temps. Temps never went above 95c for either I believe. During gameplay or Furmark testing, the fans would just suddenly.....die. The clock rate dropped and fan speed pulled back and fps went into the single digits. I tried controlling heat and changing voltages but nothing seems to keep it stable in crysis.



So a basic question about overclocking. Why do overclocks fail?

I was under the impression that they fail either because of

A) Not enough power supplied to the card, so more voltage is needed

B) Too much heat, sometimes because of increased voltages


Any ideas? Stock runs fine but.....you know how it is when you start overclocking >_>

 

[deltalives]

Oh, jeez that might have been it. I do remember getting that several times when it was OC'ed but not since I've reverted to stock settings. Hmm, I still have the sneaking suspicion that maybe if I increased voltage more.....? Or it might be what you said, that there is a bug.

Do you know if the latest 256 drivers or newer version of afterburner have corrected that?

And seriously mate, where do you get the dough for quad sli gtx 470's and a 980x??? lol I'm jealous, I don't think I noticed those last time.

 

[wolfram23]

Start by ONLY raising the memory to find the highest stable mem speed. Memory sometimes won't OC very well, and voltages don't help it... well, it would, but generally you only get Core voltage adjustment so you're basically stuck with the mem. Like I say, OC that first. I'm not sure what stock is on the memory, but try lowering it by 10 so at 1710, and run FurMark (I think ppl don't recommend it because it really taxes the GPU, but that's kind of the point). If it doesn't crash, try 1715. If it's still ok, try 1720 again... if it doesn't crash then you know it was the core that's the issue. Anyway just keep raising the memory by increments. Just FYI it sometimes seems stable but really isn't, because it's GDDR5 it's harder to tell if it's stable or not, but if it's on the verge of instability but still running you should notice lower framerates and/or stuttering of the image.

 

[deltalives]

I guess I'll try that in addition to updating drivers. So drop core clocks back to stock and oc the memory first. Uhh, stress test the memory OC with...20 minutes of Furmark I guess? If it doesn't fail then, it "oughta" be ok and I can bump it up a notch?

What about when I find a stable memory overclock, how do the core overclock and mem overclock....uhh mesh? Is it like if you overclock both you have to find a happy medium? Or can you just slot in the highest stable core and memory OC's and they will both work at the same time?

 

[wolfram23]

You should be able to keep the mem whever it was stable, then start doing the core. I like to take slightly bigger jumps with core. If you get any artifacting or if the image has slight hangs or w/e, you need more voltage. I'd say if you're taking it easy like 20 mhz increases you only need 0.05V increases. When I did mine I jumped 50mhz at a time. Then bumped the voltage and tested, ran stable and temps were ok so I continued raising the core speed.

My 5850 was able to go from 725 to 875 core with no voltage increase, then I added 0.1V so it was at 1.18 and it was ok up to about 960, then after that it seemed like even up to 1.3V it was only stable up to 985 core, which I then backed the voltage off to 1.22 or something and it was stable at 985, but the temps were a little higher than I like. So, mostly I've been running at 1.15V 960 core but I think it might have issues (Cryostasis and SC2 random crashes) so I'm going back to the testing. I tried 1.18V and it crashed in Cryostasis again. That game just has issues lol. I've ran it at 1.15V in other games like Crysis with no issues.

EDIT: Psycho the reason I suggest memory first is due to a) it's limited by not having voltage adjustment and b) what I found out about it's affect on speed, which I made a thread about here: http://www.tomshardware.com/forum/259144-29-ocing-depends-core-memory#t1856092

 

[deltalives]

Ahhh yeah college grad and no dependents definitely helps lol. I'm workin at Staples right now and about to enter college so going for that icy vision cooler for the 470 or maybe the AC Accelero will cost some. Dunno if there's space in your case for coolers if you've got quad sli but they might be nice for temps, specially if you OC'ed yours too

Hmm, well I'm installing the drivers now, hopefully they're optimized better! I guess I'll start with the memory like you said Wolfram. Once I get that stable I'll try to OC the core clock to 725-750 ish with .987 or 1.00v? With a core increase of 150ish, you're sayin it'll probably need .40 more volts so 1.00 oughta cover it (my stock was .95 I believe). Thanks, now to try it out!

Edit: Haha about Cryostasis, I guess it's a good thing I don't play it. I know what you mean though, I tested out the oc on bioshock, Left 4 Dead 2, and TF2 with mostly maxed out settings and they were ok. Even Metro 2033 went sorta ok. Then Crysis just tore it a new asshole lol. 3 year old game and the 470's overclock took it on the chin


Oh, quick quick question. For Furmark, when I've hit artifact levels and am lowering memory clock and later core clock, what's a good test time for stability? I mean you said w/ GDDR5 there's no guarantees but would 30 mins or so be good?

 

[deltalives]

Mmm, not really. I don't think I have enough money for a 480. How come?

 

[deltalives]

Do you mean flashing the bios to unlock the cores??


Edit: I think you may be right Wolfram, it was the memory. Stock Memory is 1674. The OC at 1720 failed before when adjusting core clock and it's running pretty well now, 750/1500/1674.

Btw, what does overclocking memory clock accomplish?

 

[wolfram23]

memory speed helps when you use AA and in really heavy scenes, but core does most of the work as far as rendering goes. Memory also holds the textures.

 

[deltalives]

Ahh ok, that makes sense. I usually don't use much AA, maybe 2x or 4x but never over that.

Yeah, you were right about the memory, stock is 1674 and I could get it up to 1760ish stable but if I changed the core clock as well, it would fail. So I left the memory alone and got the core clock stable at 750/1500/1674 @1.00v. Maybe coulda gone higher but furmark had it peaking at 98c lol. Crysis only works it up to 88c so I think I'm good. Thanks!


@Psycho: What do you know about unlocking the cores in bios?


Edit: Hypothetical question: The Antec 300 case I have has a bottom mounted psu. The intake fan for it faces upward. The 470's intake fan is somewhat above the psu and is downward so the two intake fans are facing each other. Could this cause the two to be competing for limited amounts of cool air? I've got a 92mm fan on the side blowing air in towards the 470 but temps are still higher than other ppl's when furmarking (98-99c) *shrug*