NVIDIA 600-Series (Kepler) Overclocking Guide

Computer Topics
Post Reply
User avatar
Lawndart
Virtual Thunderbird
Posts: 9290
Joined: Tue Mar 29, 2005 8:36 am
Location: Mooresville, NC

NVIDIA 600-Series (Kepler) Overclocking Guide

Post by Lawndart » Tue Jun 19, 2012 12:44 pm

If you own a 600-series Nvidia (Kepler) card, this is one of the best overclocking guides I've ever used:
~~The GTX 670 Overclocking Master-Guide~~

Introduction:
With the mass-popularity of the new Nvidia GTX 670, and the constant influx of questions related to overclocking them, I've decided to write an all-inclusive master-guide to overclocking them that should help most people get off to a strong start with their new 670's. All of the Kepler-based GPUs (670, 680, and 690's) are a very unique breed of GPU. Gone are the days of manually increasing voltage to stabilize an otherwise unstable overclock. Now, the user must use a great deal of finesse, and a ton of trial and error, to maximize the potential of their overclock. We now have to worry about dynamic clocking, dynamic volt changes, temperatures, and power draw in-order to reach a maximum stable overclock.



Note: Even though it's written for a GTX 670, it applies to the 680, 690 as well as they share the same PCB and Kepler-based GPUs.


Here are my 3DMark11 scores: http://3dmark.com/3dm11/3689720 using a single EVGA GTX 670 FTW overclocked to 1238 MHz (Max Boost Freq) on my ol' trusty Bloomfield i7-950 that's also overclocked to 4 GHz.
(3DMark11 incorrectly lists the GTX 670's core clock at 705 MHz, but the benchmark was still run at full load with no step-down, thermal or power throttling. This is a reporting bug in 3DMark11 with Kepler cards).
User avatar
Teej
Virtual Thunderbird
Posts: 1533
Joined: Fri Jul 31, 2009 9:29 pm
Location: Milwaukee, WI

Post by Teej » Tue Jun 19, 2012 5:11 pm

To be clear...some 670s use the same board as the 680.

Not all.

The reference board for the 670 is much shorter. I don't think either uses the same board as a 690, since the 690 board has the 2nd chip mounted on it.

Image
User avatar
Lawndart
Virtual Thunderbird
Posts: 9290
Joined: Tue Mar 29, 2005 8:36 am
Location: Mooresville, NC

Post by Lawndart » Tue Jun 19, 2012 9:24 pm

You're right. Been reading so much about these cards and overclocking the last few days I completely overlooked that not all 670s have the same PCB.

A look at the reference GTX 670's PCB:
http://www.tomshardware.com/reviews/gef ... ,3200.html

The EVGA GTX 670 FTW (which I bought) uses the reference GTX 680's PCB:
http://www.fudzilla.com/home/item/27432 ... w-reviewed
User avatar
Ray
Virtual Thunderbird
Posts: 728
Joined: Mon Aug 17, 2009 1:21 am
Location: Las Vegas, NV

Post by Ray » Fri Jun 22, 2012 2:53 pm

That's an awesome guide, that forum has some great info. Badass card! 8)
User avatar
Teej
Virtual Thunderbird
Posts: 1533
Joined: Fri Jul 31, 2009 9:29 pm
Location: Milwaukee, WI

Post by Teej » Fri Jun 29, 2012 2:11 pm

Running water cooled, mine probably has a little left in it when I find time to play.

OOTB, the card (also the EVGA FTW) maxed out at 1175 in operation. I'm currently running PrecX set to +112 on gpu and +130 on memory.

During the 3dmax run, peak GPU temp was 35C. Peak power was 119 (and happened on the first segment of the test).

http://3dmark.com/3dm11/3769442
User avatar
Lawndart
Virtual Thunderbird
Posts: 9290
Joined: Tue Mar 29, 2005 8:36 am
Location: Mooresville, NC

Post by Lawndart » Fri Jun 29, 2012 3:20 pm

Interesting that you'd get 1175 MHz out of the same card I have. Stock with boost clock and Kepler boost added I get 1178 MHz (3 MHz more). :?

My current settings (totaling 1227 MHz Max Boost Freq):

Power target 121%
Clock offset +39 MHz
Memory offset +388 MHz


Aha... I just realized when you said 1175 you meant 1175 mV, not 1175 MHz. I have my voltage adjust set to 1175 mV, so it'll run steady at that setting if needed instead of trying to save power. I really see no reason why you wouldn't have the voltage set to 1175 mV, so it always taps full throttle whenever MBF is in use.

Power target only avoids the 13 MHz downclock increments if the peak exceeds the target in order to save power, so theoretically having it maxed out shouldn't hurt anything there either. I just set mine 5% higher than the highest peak power I noted in testing.

My peak temp during 3DMark11 was 64C (stock fan) and peak power 116%.

Your 3DMark score is very respectable for a Bloomfield CPU (even though it's mostly a GPU test)! ~500 more than me with pretty much the same mobo, CPU, GPU... just not the same RAM. That and of course you're crazy water cooling contraption you've got set up nowadays. :lol:

I'm amazed you got +112 out of your 670 FTW's clock offset! Wonder if that's a direct result of your peak temps staying in the 30s under load compared to mine being in the 60s? I kept crashing Heaven anytime I went above +52 clock offset (highest "stable" I've ran Heaven, 3DMark11 and BF3 has been +52/+582). Any time I'd set it to +53 clock offset or higher, regardless whether I offset memory at all, it would crash at least one of the benchmark runs.
User avatar
Teej
Virtual Thunderbird
Posts: 1533
Joined: Fri Jul 31, 2009 9:29 pm
Location: Milwaukee, WI

Post by Teej » Fri Jun 29, 2012 3:23 pm

No, I meant 1175 Mhz, not 1175 mv...although I do have the latter set as well.

The card I had the other day was trying to go 1280+ out of the box. Perhaps that's why it was crashing...but I never even tried to fiddle with the settings as it just wouldn't run "stock".

edit: I never tried to _raise_ the speeds. I tried radically underclocking it to make it work - it still didn't.
User avatar
Teej
Virtual Thunderbird
Posts: 1533
Joined: Fri Jul 31, 2009 9:29 pm
Location: Milwaukee, WI

Post by Teej » Fri Jun 29, 2012 4:16 pm

Cranking up the memory helped a little bit.

http://3dmark.com/3dm11/3770372

Seems to run fine at +550 (!) on the memory. I had the occasional flickering texture at +600.

Actual memory speed is 3649. Heh. Wow.
User avatar
Lawndart
Virtual Thunderbird
Posts: 9290
Joined: Tue Mar 29, 2005 8:36 am
Location: Mooresville, NC

Post by Lawndart » Fri Jun 29, 2012 4:43 pm

Teej wrote:No, I meant 1175 Mhz, not 1175 mv...although I do have the latter set as well.
Correction to what I said earlier... Stock speed with boost and Kepler boost added for the EVGA GTX 670 FTW is 1188 MHz (not 1178 as I said above). 1006 MHz core clock + 78 MHz boost clock + 104 MHz Kepler boost.

Now it's making more sense, and the reason you saw 1175 MHz is because your card was either thermal or power throttling the Kepler boost by subracting clock speed in increments of 13 MHz. Very unlikely that it's thermal throttling taking place in your chilled case, so at some point your peak power may have gone above the power target (even with default settings this could happen), hence why you'd see 1175 MHz instead of 1188 MHz on stock.
Teej wrote:Actual memory speed is 3649. Heh. Wow.
Depends how you read the numbers with DDR5. The actual memory clock is 1552 MHz on stock. The offset slider lets you adjust the memory speed starting from 3104 which is exactly twice the actual memory clock. To make it even more interesting they list the "effective memory clock" at 6208, four times the actual clock.

Your memory speed of 3649 MHz is actually 1824 MHz (not sure how your number came out an odd number when doubled, unless it's a speed reading in PrecX's performance log or GPU-Z... Or adding an odd digit in the offset would do it too I suppose).

Still wow, even without the multipliers! What you really should be saying though if you want to brag is: "My effective memory clock is 7296 MHz!" :mrgreen:

Highest run I've done in 3DMark11 and Heaven without any graphics artifacts is with my memory clock at 1845 MHz (equals 3691 MHz (with a +587 offset), or 7382 "effective memory speed")... :P

Just can't seem to get past a +52 clock offset without my card crashing though, which means the highest steady core clock without Kepler throttling I can achieve is 1240 MHz... :( I'm a little sad, but not really... anything over 1200+ MHz is ludicrous speed!
Post Reply