Originally Posted by Johnny C
Here's what I'm betting:
The GTX 480 will be about 10%-15% faster than a single 5870 1GB overall. However in extremely tessellation limited situations, AKA the Heaven Demo, it will actually be as fast or slightly faster than the 5970 dual card. However in your average gaming situation it will not be anywhere near the 5970 unless x-fire isn't functioning. Fermi will also have some issues with driver/game support similar to the 8800GTX at launch.
The GTX 470 will be about as fast as the 5870, again faster in some situations or in specific games yet slower in others. In some situations it will be bested by a 5850 at stock or slightly overclocked.
If/when AMD releases the 5890 2gb(I'm betting 2gb frame buffer) it will include a larger tessellation block and some manufacturing tweaks and will be at least 10% faster than the GTX480 in all but very limited situations. Also there will be more availability of the 5970 with some cards being in stock at most major retailers. AMD will shift their marketing to this card...and defend the performance crown until we see a dual Fermi card.
As far as availability goes: the GTX 470 will be available shortly after launch in low but existent quantities...however the GTX 480 will be scarce and they will be at a premium price until June.
Dual GPU Fermi Card: not gonna see it till yields improve and either a complete re-spin is done (B1 revision) or 28nm solves some of the transistor leakage and defect issues.
I've been contemplating starting a thread here titled:
"My GTX 480/470 is overheating" but I think that may be jumping the gun and would be considered trolling despite the humor associated with it.
After all is said and done everyone will realize that Fermi is not a Graphics chip at all, it designed for something different, it's made for GPGPU and has been pulled off the bench to do double duty against AMDTi's truly dedicated graphics chip.
I believe Fermi really is a giant step forward....but that we won't see the tangible benefits to the gaming community until it's next gen.....and/or a die shrink.
Honestly i don't know anymore,as there's been so many predictions made so far,covering the entire performance envelope,that someone will get it right and and "I told you so"...
Having said that though,and keeping Guru3D's own comments into consideration,which the author really has to know the actual performance by now,even though he can't reveal exact numbers,he simply stated it isn't going to be a knockdown blow relative to the HD5870 and he seems certain of it,so i'm i'm guessing that it'll be close to the same performamce difference that happened between the HD4870 and the GTX 280,or the HD4890 versus the GTX285 cards....In short,about 10~15% better for Nvidia in this case on average.
If that is the case though,will it be enough to keep it from getting roasted on reviews,given the delays and much higher power consumption though?,or that ATI can lower prices on the HD5870 cards,still make money on them given the smaller cypress dies and 256 bit memory bus,which needs less memory modules and uses a simpler PCB?,or even really require them to release a refresh card at all?
I'm sure that they can release a refesh of cypress,and for all we know,they could be stockpiling cypress chips that can clock higher than 850mhz,and do it reliably,to quickly recover on that 10~15% performance difference anyhow,by releasing an HD5890 card,with no other changes required to keep it a tight race in actual games,rather than synthetic benchmarks using polygon budgets we won't see in actual games for a couple more years,as consoles are still stuck with DX9 level GPU's will make sure of that,and pretty much all new game releases are multiplatform these days,with maybe a few quick DX10/DX11 tweaks added for the PC version at most....Hardly anything that will put the tesselator to the test in either Fermi's or Cypress chips as it is,even if Fermi's has more performance there.
Then there's the 28nm bulk process happening late this year,or early next,which an updated Fermi card with even more hardware resources is being developed right now for sure,and ATI is also doing the same with their next generation as well,regardless if it's still based off Cypress or using a new architecture built from scratch....That's the common point between both companies,the actual fabrication process for their GPU's,which comes from the same company for the last 10+ years(TSMC),and the design rules are same for both,hence why the competition in terms of performance has pretty much always been pretty tight overall.
I see this competition as being fairly over in realistic terms,as by the time Fermi is out in decent volumes for both the GTX470 and GTX480 cards,we should be hearing the first rumors on the next generation products built at 28nm anyhow,maybe even news about those next generation GPU's already gone thru the tapeout phase already,which usually means 6 months from that point to retail launch if nothing screws up too badly.