tubnotub1

Alright, now for an explanation of the results. For those of you that don't need one, you can just look at the spreadsheets and extrapolate the results on your own. This is for those that might not really be able to tell what all this means. First off, this experiment was to see how CPU bottlenecked the GTX (single and SLI) is at higher resolution and higher AA/AF settings. This test has real world value because most of us do not like playing our games at 0 AA, or for that matter at the stock resolution of the 3D MARK '06 Test, which is 1280*1024. Please understand that these results are only to be used as a reference for those with a monitor that is capable of 1920*1200 or above. It should also be noted that 3DMARK '06 would not allow me to run the tests at 16QAA, which would have lessened the CPU bottleneck even more. Now then, onto the observations:

For SLI, as seen by the information below, we can tell the dual card solution is indeed bottlenecked by a X6800 running at 3.4 GHZ. However, it should be noted that the bottleneck is not that large. By changing the CPU speed from 3.4 to 2.6 we see less then a 10% decrease in performance. Judging by the information I have collected I would say that the bottleneck would completely disappear around a CPU processor speed of 4.0+. Also, I would imagine if I had been able to run 16QAA we would have seen the bottleneck disappear around 3.6. For those of you with 30" monitors, you can rest assured that so long as you are running a processor @ 3.6+ you will be getting the most from two 8800 GTX's.

Attached Thumbnails

         

__________________

tubnotub1

A single 8800 GTX at 1920*1200 is no longer CPU bottlenecked around 3.2-3.4, a speed that is easily attained by most Conroes. Note that once again, by decreasing the CPU speed from 3.4 to 2.6 we see less then a 10% drop in performance. The performance drop is practically negligible until you drop from 3.0 to 2.8. For those of you with a single GTX, so long as you are running a 1680*1050, 16AA and 16AF, with a processor speed of 3.2-3.4, you can rest assured that you are probably getting the most out of your 8800 GTX.

Attached Thumbnails

         

__________________

The Hundred Gunner

Where!

__________________
"I just talk the way I see things... If that's offensive to you... I'm sorry; you're a loser." - Michael Savage

"But you would be amazed by how many people think that the only reason to have a computer is to play games, and that playing games is all that anyone with a computer does." - dangerousHobo

tubnotub1

They are coming... hold on =D

__________________

The Hundred Gunner

What does it mean? That since you can ~double the score w/ SLI, it's not bottlenecked?

__________________
"I just talk the way I see things... If that's offensive to you... I'm sorry; you're a loser." - Michael Savage

"But you would be amazed by how many people think that the only reason to have a computer is to play games, and that playing games is all that anyone with a computer does." - dangerousHobo

tubnotub1

No, what you need to look at is the change in the total 3DMARK '06 score, still setting everything up, give me 10 minutes and Ill have a full explanation.

__________________

ThaLoveDocta

dammit tub, you get everyone confused by posting poics with no explanation! lol
lets see whatcha got!

Oh and btw... don't you have a job or something what the hell are you doing ALL these benchmarks for with your time??

REP+

Edit: clean upi your desktop! there's install files all over the damn place!

The_Manual

The posted screenshots are consistent with what we currently know about information processing speed at high resolutions.

The greater the resolution, the greater the amount of data that the GPU will have to process, and store as Textures, and sub-routines.

Once the resolution increases above 1600x1200 the Central Processing Unit will have only a small impact on the frame rendering of the system as a whole. This will change with the addition of Direct3D 10.

The CPU receives little additional data to process at these resolutions. The frame rate is therefore mostly up to the GPU and GPM (Graphics Processing Memory stores).

If you were to run these benchmarks at 1024x768, then the Central Processing Unit will have a large impact on the overall frame rates.

However, in regards to what a bottleneck actually means.

A bottleneck is where a component, or multiple components, within a computer system are incapable of processing data at the speed of the higher end components within the computer system. They therefore prevent the system from operating at the speed which the faster components are capable of.

A bottleneck will not always be present. You can not just say "Do I have a Bottleneck?" without stating in which application you believe the bottleneck to occur.

For example:

Red Alert 2 (on my system): Bottleneck = False
3DMark 06 (on my system, 1024x768): Bottleneck = True

Just though I would add that to my post
Nice tables, and scores

__________________
I will be sufficiently rewarded if when telling it to others you will not claim the discovery as your own, but will say it was mine.

Fold for team 37726

Please Keep Spelling and Grammar Acceptable

tubnotub1

I completely agree with your post The_Manual, my little experiment above was meant to show people what they would need to get rid of the bottleneck, and to show, how in real world situations (I hardly call 1280*1024 a real world situation, 0AA) the 8800 GTX really isnt as bottlenecked as everyone makes it out to be. Assuming you have a 20.1 inch monitor it is probably safe to say that in most current day gaming applications you would see only a very small CPU bottleneck @ 3.4 GHZ assuming you were running 16AA 16AF. The same could probably be said about SLI with monitors above 24 inches, with the 16QAA/16AF.

__________________

Kramy

Pretty good comparison. More pertinent to the current games than all other 8800GTX benchmarks I've seen.

0xAA benchmarks are actually fairly important....just not right now. In theory an 8800GTX is 8 times faster with AA off, assuming a program/game can load up every single stream processor. For games like Alan Wake, which is supposed to run best on quad-cores and has massive CPU/GPU overhead, those extra 3Dmarks at 1024x768 will become quite relevent.

__________________
-Kramy

Athlon X2 ----- 3Dmark05 11569 ---- 3Dmark06 7138* ---- eVGA 7900GS ---- 738/1972mhz stable ---- [vmod]
Athlon XP ----- 3Dmark05 6845 ------ 3Dmark06 3519 ----- BFG 7800GS ------ 460/1450mhz stable
*1152x864


Check it out! - Videocard Power Consumption Graphs