20 replies to this topic

[El_Capitan]

I thought this was peculiar. I think this is the first time I've had such a great overclocking card score so much lower than a not-so-great overclocking card.

Card #1
(Stock) 950MHz Core 1200MHz Memory @ 1.134V's
Unigine Heaven 2.1 = 39FPS


Card #2
(Stock) 950MHz Core 1200MHz Memory @ 1.134V's
Unigine Heaven 2.1 = 34.7FPS


*****

Card #1
(Stock Voltage OC) 1115MHz Core 1470MHz Memory @ 1.134V's
Unigine Heaven 2.1 = 45.7FPS


Card #2
(Stock Voltage OC) 1145MHz Core 1450MHz Memory @ 1.134V's
Unigine Heaven 2.1 = 40.9FPS


*****

Card #1
(Max Voltage OC) 1195MHz Core 1470MHz Memory @ 1.3V's
Unigine Heaven 2.1 = 47.8FPS


Card #2
(Max Voltage OC) 1290MHz Core 1450MHz Memory @ 1.3V's
Unigine Heaven 2.1 = 44.3FPS



I've never seen such a big difference even with different manufactured cards of the same type. The card at 1290MHz Core Clock at 1.3V's is scoring lower than the card at 1115MHz Core Clock at stock voltage. Anyone seen this before? Ignore the bus interface on GPU-Z, they're all running on the same system in the same PCIE slot.

Edited by El_Capitan, 09 October 2012 - 09:39 AM.

[Waco]

The stock scores being that different is surprising...perhaps it's a leakier and more error-prone chip (which causes the driver to recompute bad answers consistently)? Leaky chips do tend to clock higher though that kind of score difference is pretty drastic. At stock that's nearly 12%...

[El_Capitan]

The stock scores being that different is surprising...perhaps it's a leakier and more error-prone chip (which causes the driver to recompute bad answers consistently)? Leaky chips do tend to clock higher though that kind of score difference is pretty drastic. At stock that's nearly 12%...

I'm not sure, then it would be more prone to fluctuations in FPS and crashing, but that's not the case. Just has lower MIN and MAX FPS. It stays at a more constant in-between level than the higher scoring graphics card.

[Waco]

Perhaps this is a clue that Heaven isn't a very good benchmark.

Does anything else show this same score differentiation? Heaven on extreme is pretty math-heavy and it wouldn't surprise me if the driver hides some errors from the user if it catches them in time (which would manifest itself in lower scores when errors occur).

[El_Capitan]

Perhaps this is a clue that Heaven isn't a very good benchmark.

Does anything else show this same score differentiation? Heaven on extreme is pretty math-heavy and it wouldn't surprise me if the driver hides some errors from the user if it catches them in time (which would manifest itself in lower scores when errors occur).

Hah, probably true. I just keep using 2.1 since it's what I used to test and stress test past graphics cards, and works great for testing stability. I'll run a 3DMark11 extreme benchmark and compare it with the other cards once I'm through finding the overclocks on each of them.

[Waco]

Assuming the overclocks hold. Heaven hasn't been terribly reliable for me for final overclocks on more recent cards.

I find myself using it as a good starting point and backing down from it when finding errors in games.

[El_Capitan]

Assuming the overclocks hold. Heaven hasn't been terribly reliable for me for final overclocks on more recent cards.

I find myself using it as a good starting point and backing down from it when finding errors in games.

Same here. I usually end up dialing the final core clock down a bit when finding issues in games. I also use Performance over Extreme for 3DMark11 for stability. Some tests passed Extreme but not Performance. You'd think it would be the other way around...

[El_Capitan]

Ah, 3DMark11 Performance scores higher with the higher overclocked card.

http://www.3dmark.co...6/3dm11/4604584

It must be Unigine Heaven then. I need to explore this a bit more.

[El_Capitan]

Well, my 3rd card overclocks well and gives good Unigine Heaven results.

http://forums.overcl...dpost&p=2054666

[red1776]

You realize of course...that this is going to drive litwicki22 to the brink of insanity?

I'm kind of tracking with Waco, I wonder if the card is hitting the board TDP faster. Have you tried power tuning up and see if that helps?

[_TheAlexO]

You realize of course...that this is going to drive litwicki22 to the brink of insanity?

haha I was thinking the same thing.

[ccokeman]

Hah, probably true. I just keep using 2.1 since it's what I used to test and stress test past graphics cards, and works great for testing stability. I'll run a 3DMark11 extreme benchmark and compare it with the other cards once I'm through finding the overclocks on each of them.

I find that you can be Heaven 3.0 stable and still fail miserably in games. I use a mix of Heaven 3.0, DIrt 3, BF3, Metro 2033, and Batman to find out where my clocks end up. Its a slippery slope when you use only one benchmark to validate stability.