View Single Post
Old 01-25-12, 01:57 AM   #145
ninelven
Registered User
 
Join Date: Jan 2003
Posts: 132
Default Re: next gen kepler to support dx 11.1, also take a year to rollout all cards

Quote:
Originally Posted by shadow001
It is a known fact that AA will use up more memory bandwith period
I never said otherwise, so I am not sure why you are bringing this up.

Quote:
Originally Posted by shadow001
And makes the point of having more Rops kinda irrelevant now doesn't it
Yes and no. On the one hand, in bandwidth limited scenarios, the extra ROPs are not really useful. On the other hand, you are rarely bandwidth limited 100% of the time in actual games. Furthermore, the ROPs in Fermi are tied to memory channels. While Fermi could have used only 24 ROPs instead of 48, it is likely that the additional die space was negligible.

Quote:
Originally Posted by shadow001
Increasing the core speed had a big improvement though...
Then you were not completely bandwidth limited, so more ROPs would actual benefit you there.

Quote:
Originally Posted by shadow001
So where's the bottleneck?
Well, 9.75 Gigapixels/sec = 9750 Megapixels/sec. 2560x1600 is 4096000 pixels or 4.1 Megapixels, so 3 of those monitors is 12.3 Megapixels and 6 monitors is 24.6 Megapixels. Now, 9750 / 12.3 = 792 frames per seconds with 3 monitors and 9750 / 24.6 = 396 frames per second with 6 2560x1600 monitors. Thus, I would guess that whatever the bottleneck is, it probably isn't the ROPs.


Anyway, I've made my point here. If you are sincerely interested in the truth of the matter you can simply carry out the aforementioned test at your own leisure. Goodbye.
ninelven is offline   Reply With Quote