View Single Post
Old 07-10-08, 03:24 PM   #34
Xion X2
Registered User
Xion X2's Avatar
Join Date: May 2006
Location: U.S.
Posts: 6,701
Default Re: The right price for the GTX200 family...

Originally Posted by ChrisRay View Post
This "Ceiling" is not an advertisement. And its very real.
I'm not interested in playing the semantics game. I never said that it was an advertisement; I said that it was advertised. There's a difference.

I don't believe I ever said it wasn't "real," either.

Now that we have that out of the way...

The 512 problem has been there "Long" before the HD4000 series ever existed. And could be seen on 9800GTX 8800GT cards in comparison to 8800GTX cards in specific cases.
And "long" before we ever had cards with over 115GB/s memory bandwidth at their disposal.

As time has moved on these cases have increased. Memory amount can be a hard bottleneck. Specially when used. Once you exceed that framebuffer count. Your system memory starts acting as framebuffer. How ATI/Nvidia may allocate that system memory to framebuffer could very well be different. But the point is you cant get past the Zstorage of AA data. And its expensive.

Just some framebuffer numbers to chew on.

4xAA/16xCSAA @ 2560x1600 250 Megs

4xAA/16xCSAA @ 1920x1200 140 Megs

8xAA/16xQCSAA @ 2560x1600 370 Megs of Memory

8xAA/16xQ CSAA @ 1920x1200 280 megs of memory.

Thats just memory dedicated to Zstorage alone.. That does not include HDR which also increases storage data. It also doesnt include textures. Uncompressed ((or compressed Normal maps)) either. Its a pretty inescapable fact that once you run out of onboard memory. You start touching system memory. I'd rather not do that it any event. Using system memory as a means for storing framebuffer is one of the largest causes of micro stuttering that exists.

I just got finished testing Crysis on 9800GTX + SLI cards. And they cant even produce 15 FPS @ 1680x1050 with 16xCSAA and Very High settings due to running out of memory.
I'm aware of how it works, but thanks for explaining. In your example of the 9800GTX, I would say that you're overlooking memory bandwidth rates yet again. There's not a big difference in that between 8800 and 9800 series, but on the flip side of the coin you have the 4870 which has over a 45 GB/s advantage in that by comparison.

i7-2700k @ 5.0 GHz
Nvidia GeForce 570 2.5GB Tri-SLI
Asus P67 WS Revolution (Tri-SLI)
OCZ Vertex SSD x 4 (Raid 5)
G.Skill 8GB DDR3 @ 1600MHz
PC Power & Cooling 950W PSU
Xion X2 is offline   Reply With Quote