If that's true though insomniac why does the 7800 SLI do 84 while the single does 54 in BF2 1600x1200 4AA ??
AF perhaps, but not AA. AA is very heavy on memory. AA will likely never be free, unless the graphics card isn't the bottleneck. That's why there's 10 MB of eDRAM in the Xbox 360 (or PS3?)
Especially supersampling AA. The graphics card renders the scene at a higher resolution and blends the pixels to form a value for the smaller resolution. That's the main way of doing AA these days. We all know higher resolutions will never be free.
Perhaps I don't understand the impact of CPU limitation then...
My assumption would be that if a 7800GTX is CPU limited at 60fps, then an SLI shouldn't get a higher score, or marginally higher...
Is this incorrect?
Perhaps this would explain why at 1600x1200 w/ AA enabled n their benchmarks the 7800 does not get scores much higher than the 6800U?
Is AA perf. system limited to some substantial degree? (CPU/memory etc??)
My assumption would be that if a 7800GTX is CPU limited at 60fps, then an SLI shouldn't get a higher score, or marginally higher...
Is this incorrect?
Perhaps this would explain why at 1600x1200 w/ AA enabled n their benchmarks the 7800 does not get scores much higher than the 6800U?
Is AA perf. system limited to some substantial degree? (CPU/memory etc??)
Nope. It relies almost solely on the graphics card, unless bigger textures need to be loaded. It depends how the game works, but generally speaking AA is ALL graphics card work.
Keysplayr
Elite Member
- Jan 16, 2003
- 21,219
- 55
- 91
Actually, I wasn't even considering the 7800GTX with my remark. I am just looking at the 850XTPE.
Battlefield2 at 1600x1200 without AA/AF scores a 54.5
Battlefield2 at 1600x1200 with 4xAA/16xAF scores a 61.3
There is no issue here. Just pointing out an error.
Now the more than double performance of 7800GTX in SLI over a single GTX is another weirdism. UNLESS the DIII settings were to "ULTRA" and the 512MB total memory helps this. Just a simple guess and the first thing that comes to mind.
Keysplayr
Elite Member
- Jan 16, 2003
- 21,219
- 55
- 91
Why would you say that?
Each card shares the load and each card has 256MB of memory.
If what you state had any truth, All SLI setups would only utilize only the memory on one card or half of each card. No sense.
EDIT: But I am not saying there wasn't any error here in Xbits findings.
Look at this: Rollo's benches
Each card shares the load and each card has 256MB of memory.
If what you state had any truth, All SLI setups would only utilize only the memory on one card or half of each card. No sense.
EDIT: But I am not saying there wasn't any error here in Xbits findings.
Look at this: Rollo's benches
Don't know, just what I've heard. IIRC it was because one graphics card can't access the other's memory (the way SLI is currently set up). But as always I could be wrong. Hopefully someone around here knows for sure. I actually have no clue which way it is or if what I'm saying is even feasible, but I'm raising a possible doubt.
Keysplayr
Elite Member
- Jan 16, 2003
- 21,219
- 55
- 91
Each card has it's own job to do. Each has it's own GPU and that GPU works with it's own memory. I don't know what you've heard or from where, but I'm pretty sure all the memory gets utilized. The GPU on one card should not need to access the memory on the other.
I agree with the xtknight. I don't think it works just like that... I mean having 256x2=512. Right, they wouldn't have to access each other's memory.
But in AFR, for example, one card would have 256MB for one frame, then the other card has 256 for the next. This would be 512MB over 2 frames.... but it still isn't 512mb effective.
Just guessing, though.
But in AFR, for example, one card would have 256MB for one frame, then the other card has 256 for the next. This would be 512MB over 2 frames.... but it still isn't 512mb effective.
Just guessing, though.
Well think of it like this. Having two 500 MHz cards in SLI doesn't make it 1 GHz.
If the graphics system was asked to draw 2 lines, these are some possible levels:
One card could draw both (not even SLI but it would still be faster than the proceeding choices these days).
Each card draws a line.
Each card draws half of one line and half of the other.
Each card draws a point.
If one of the last 3 choices was true, the graphics cards would have to share their work (if only one graphics card sends the output, which is true in SLI's case). That work would be in their memory.
For a really simple operation like this, the first would be most efficient because drawing 2 lines is as easy as it gets. Truth is we have no clue how SLI works unless we work at NVIDIA (not me).
If the graphics system was asked to draw 2 lines, these are some possible levels:
One card could draw both (not even SLI but it would still be faster than the proceeding choices these days).
Each card draws a line.
Each card draws half of one line and half of the other.
Each card draws a point.
If one of the last 3 choices was true, the graphics cards would have to share their work (if only one graphics card sends the output, which is true in SLI's case). That work would be in their memory.
For a really simple operation like this, the first would be most efficient because drawing 2 lines is as easy as it gets. Truth is we have no clue how SLI works unless we work at NVIDIA (not me).
somethign is wrong with these benches...the xl somehow scores lower than a 6800gt in most of their games, and an x850xtpe BEATS a 7800gtx at 1600x1200 4xaa in FAR CRY??? even if it's only 2 frames....something is wrong.
Not necessarily. Theoretically an SLI setup could still process shaders, AA/AF, etc. faster and thus increase output.
Maybe CPU limited isn't the right word necessarily in this situation so much as CPU bottlenecked. Derek mentioned this (and the importance of hi-res) in his 7800GTX review.
It's redundancy. Each graphics card has to load the texture info for the entire scene into its memory. You can't have card #2 needing a texture from card #1 and using that high latency bridge (compared to onboard memory) to fetch it. That would bog performance like a motherscratcher with the was SLI currently does logic.
So no, you get 2xCore processing power, but you still have a 256MB card in practice, because all the texture information gets loaded, as a complete set, into each cards memory. Twice the memory, yes, but also twice the information. Net, nothing changes.
these are absolutely terrible benchmarks. i'm not a fanboy, but common! look at these
http://www.xbitlabs.com/articles/video/display/2005-17gpu_30.html
http://graphics.tomshardware.com/graphic/20050524/vga_charts-12.html
http://www.xbitlabs.com/articles/video/display/2005-17gpu_30.html
http://graphics.tomshardware.com/graphic/20050524/vga_charts-12.html
So when I enable 4aa from 2aa in BF2 and drop from 99 to about 70fps, do you think thats the 7800GTX just not fast enough, or is the CPU bottlenecking it as well / both...
Basically I'm a little disappointed in benchmarks showing at 4aa 16x12 res the GTX getting only 10-20fps higher than 6800u...if it were a CPU/system bottleneck I can understand... but otherwise it would seem the 7800 really gets very little boost over the 6800U at 4AA
AA is largely a function of the graphics card only. A 30FPS drop from 2 levels of AA sampling though sounds REALLY extreme though.
Not sure in your case. Would need more testing.
Could just be because BF2 at forced resolution is supposed to cause problems with anti aliasing...
But the thing is, in many of the benchmarks at 1600x1200, w/ 4AA, the 7800GTX only gets like 15fps more than a 6800 ultra, but often doubles it at above that resolution....
Any thoughts on why this may be? Frankly I think it sucks that the increase would be so small
Supersampling is one method of AA. If BF2 uses it like a lot of games, 4xAA means the scene is rendered 4x larger (in area) then blended then scaled back down. Each step of AA (2x,4x,8x,16x) is twice as intensive as the last unless I'm sorely mistaken.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter