Originally Posted by jtom320
Well you two can go worry about Hardocp's bias I'll get back to the real world. 10%. That's was my only point. Just flipping thru the Tom's review as you made me curious and it's pretty much the same results not quite as wide a gulf but the review was in December when the drivers were really bad. Shoot that review has the 7970 beating the 590 in a couple games.
But they are probabally biased to.
Edit: When did I refer to you as bro? lol. I specifically said in this thread going from a 580 to a 7970 would not be a wise choice as there is not a huge gulf between the two. Only thing I said was that it was larger then 10% which it obviously is. What are you even arguing?
Well here is what TPU has for
Both of which are in relation to 1.5GB 580s. Now, that doesn't look like much, but the 1600 res is still impressive at 16%. That is not easy to push. Now all we said was that a 580 can make up that performance gap quite a bit with a 20~30% overclocks that they are capable. You also can remove the memory advantage by comparing to a 3GB 580.
That is all a lot of people are saying. Not sure what is so hard to believe about this. Now you go ahead and overclock the 7970 and you regain another gap but you are not going to make another 15~20% that everyone is claiming over a 580.
Originally Posted by xoleras
We've already established that 3gb does not provide a performance increase unless you are doing 3d surround resolutions.
For the HardOCP:
You do see that the 7970 is at 1,260 MHz over stock 925 MHz which a 335 MHz overclock or a 36% oc?
While the 580 has a 860 - 772 MHz which is a overclock 88 MHz right? Thats a 11% oc.
Problem with that review.
Secondly, your charts failed to capture one other important piece of info. You are hitting the GPU bottleneck before you reach the VRAM bottleneck. I have tested these settings extensively.
I have owned 3x 580s, 2x 6970s, 1x 6990, 2x 590s, and 2x 7970s (in my rig right now.) I have tested 5760x1080, 1920x1080, and 2560x1600 extensively.
3GB Vram is an issue but only after you have sufficient GPU power to push whatever it is you are pushing. There isn't a single DX9 game out there that can hit above 1.5GB naturally at 5760x1080, i.e. without modification with one and ONLY one exception is Witcher 2 with Uber Sampling. That eliminates half of those titles in your charts as a waste of time for benching 3GB. That leaves F1 2010, Metro2033 which didn't run without 3GB on a single GPU, and Crysis 2. So 33% of your titles didn't even run. The other two display a GPU bottleneck.
I can list 2 titles that can break 1.5GB 580 cards at that resolution off the top of my head. Dragon Age II and Civ 5. Witcher 2 is another strange story which implements very advanced features for a DX9 only title. Secondly, the 2560x1600 can easily enter territories (utilizing AA) well above 1.5GB and its only 4 Megapixels.
So I would say that your conclusions don't follow the data given, if you look real hard at it. I can see it easier as I have a lot of experience benchmarking the aforementioned video cards. Also having the 7970s in my possession now, I am giving you an unbiased opinion that they are faster than 580s, CFX is even faster than SLI 580s but they aren't destroying them, or at least what I was able to do with my own hardware (my OCed 580s.) I am just trying to share my information with those and hopefully help you see in a more balanced view the data you are providing.
One thing I have noticed is that the tessellation performance has greatly increased.Edited by RagingCain - 2/22/12 at 9:12pm