AA negates GPU OC?

[wolfram23]

Hi all, I was wondering if someone could explain what's happening here. I wanted to test how big of a jump my OC is over stock and over the CCC max OC. I also decided to compare results using different levels of anti aliasing.

Sapphire 5850
Stock 725/1000
CCC OC 775/1125
Max OC 985/1200

FurMark settings:
120000ms benchmark
1280x1024
windowed
no other settings enabled

Results (Score, Average FPS)
0xAA
stock: 10052, 84
CCC: 10791, 90
OC: 13221, 110

4xAA
stock: 5176, 43
CCC: 5698, 48
OC: 5970, 50

8xAA
stock: 2602, 22
CCC: 2885, 24
OC: 3060, 25

Clearly you can see that with higher levels of AA, the effect of the overclock becomes less and less. At 0xAA OC is 23.6% better than stock. At 4xAA it's down to 14%, and at 8xAA it's down to just 12%. However, the core speed is 26.4% higher and the memory 16.7%.

So clearly there has to be a bottleneck somewhere caused by the AA, right? But where? I know Furmark doesn't take advantage of multiple cores, but my turbo monitor shows it at 4.02ghz so it's probably not my CPU. My RAM is also pretty fast, I have it at 1750mhz 9-9-7-24 1T. My intuition is telling me that if I OC the GPU RAM more it would make a big difference, but unfortunately I can't as this is as high as I can put it and remain stable (possibly 1220, but I don't want more voltage than I have, at 1.237V)

Can someone try to explain this? Also, in theory, would using a 2nd 5850 in crossfire make a big difference at higher AA levels?

 

[wolfram23]

Seriously, nobody knows why?

 

[notty22]

Its the architecture of the card. Your asking it to do a different 'thing'. It is, the o/c or increase in core speed is showing less of a return. The power of the card with AA on does not scale the same as with AA off.

 

[fisshy]

Ahhhh, you Did open this thread...

I know it has something to do with the Ram, Its not the GPU clock That Will lower FPS on AA, Its either the Ram Size Or its frequency, My personal gut feel is that it wont make a difference if you have another 5850 in crossfire, the ram is still communicating at the same frequency.

I understand your Voltage Dillema, i am in a similar situation...

Good Luck

 

[wolfram23]

Well, I'm fine with 4xAA and all but maybe I'll do some quick tests with lowered core clock and see how the GPU RAM speed will affect AA performance.

 

[fisshy]

As far As i remember it should be pretty substantial (YMMV)

IT makes a Huge difference on my 4850...

I seriously need an upgrade

 

[fisshy]

^ Please don't Go frying your Card!

 

[wolfram23]

Haha I'll try my best not to! Don't worry, I'll monitor temps closely

My plan is this: I have the results at 725/1000 at 0x, 4x, 8xAA. I'll then try all three AA settings out with 775/1000, 825/1000, 875/1000, 725/1050, 725/1100, 725/1150. This should give fairly good results for determining which one affects AA the most. I don't know if it's necessary since I already have some results, but I could also try 775/1050, 825/1100, 875/1150 (this one might not work at stock volts) to see how the combination affects it compared to only gpu or gmem speed.

I think I'll have time to run these tests tomorrow.

 

[fisshy]

OK, good luck and keep us updated!

 

[wolfram23]

I posted it in a new thread http://www.tomshardware.com/forum/259144-29-antialiasing-core-doesn-matter-memory-baby

 

[wolfram23]

Best answer selected by Wolfram23.