G71
- Thread starter trinibwoy
- Start date
caboosemoose
Regular
Ailuros said:
Bottom line is this - they didn't use the Ultra name. That was very, very likely so that they left space for an Ultra as and when neccesary. The current 7800 is single slot because it doesnt need to be dual slot, it's that simple in my view. If ATI had a faster card than the X850 when the 7800 launched then the 7800 would have been dual clot. As it was, NV could lunch a single slot 7800, show off about how power miserly it is, how cool it runs and how clever they are for doing a high-end single slot card. Hats off to them, they played 7800 vey well indeed. But I think its pretty obvious that if NV wanted, they could produce dual slot G70 with significantly higher core clock speed. Of course, the memory bandwidth problem remains. Indeed, the memory bandwidth problem will be very interesting when R520 launches.
Many people are making calculations regarding how high R520 would need to be clocked if 16 pipes to match or beat G70 in fill rate etc. That's really irrelevant, since G70 is basically bandwidth limited except when it's shader, or HDR etc limited. So given that ATI will have access to pretty much the same memory as NV, its awfully likely that R520 will have fairly similar performance to G70 in gmaes like Doom 3, Far Cry etc where G70 is bandwidth limited (without HDR on Far Cry of course - oh and I am assuming that the rumoured new memory interface/controler, whatever, on R520 only provides a marginal advantage, otherwise all bets are off).
Where things will get interesting is in games that are shader limited on G70, which Battlefield 2 appears to be for instance.
Finally, there's HDR issue, both in terms of pure performance and AA compatibility. With all that in mind, I think it's pretty easy to imagine an R520 with only 16 pipes and clocked at well under 700MHz that delivers very similar peformance to G70. I fancy it will beat G70 (in GTX trim) on balance, but that an Ultra G70 will give back the lead to NV pretty much immediately. Perhaps R580 vs whatever NV have going at that stage will be the closest fight of all.
Of course, I dont really know what I am talking about, but it was ever thus.
Makes sense.
If from each R520 quad comes out a similar result (as in artithmetic) as on a G70 quad, then the clockspeed advantage of the first will balance things out. In any other case the latter has more units, exept ROPs (which I hope ATI has optimized further this time).
Float HDR is important to me at least; I sure hope that R520 will be able to combine 64bpp HDR with MSAA. If it's FP10 only that's combinable with MSAA I'm not so sure I'll be that much interested.
I read over and over again why ATI hasn't released any preliminary performance numbers from it etc etc from others. It's not usual practice and no indication for anything in my book. What could be an indication though are developer reactions, since R520 prototypes are in major ISVs hands since March this year. They'll obviously not comment about it yet, but it would be the only source where something more reliable could leak out before the official announcement.
As for the hypothetical "lead" it's all about winning impressions isn't it? Some of those were already won by releasing months earlier with immediate availability. To make things clearer I don't know what NV really intends to do, but for the time being I don't see an absolute necessity; especially if they can have a 90nm Gxx variant on shelves before this year's holidays.
caboosemoose
Regular
Hmmm, I can't get my head around the idea that there's going to be a 90nm G70 this year (for sale). That feels too soon for me. I would expect the first 90nm G70 chip to be more of a mainstream performance part in the mould of NV42 (which was the first large NV4x chip based on 110nm) followed next spring by a high end 90nm chip. But I also feel that there will be surprises from both players.
caboosemoose said:
Reported by B3D on March 4th 2005:
Michael Hara/NVIDIA:
Spring turned out to be summer 2005....
Fall may turn out to be winter 2005....
http://www.beyond3d.com/forum/viewtopic.php?t=20937
Ailuros said:
I'm not much into blind faith, but I do believe in past as prologue. When thinking about FP10 MSAA, so far as how good the results will look in scenarios available today and the near future (say the next two years or so), do you take into consideration what ATI did with FP24 when the conventional wisdom was all about FP32? I look back at the famous debate with Sireric and Reverend, and I see an ATI that clearly spent a great deal of skull sweat working thru the theory in advance as to how much "eyeball impact" the reduced precision was going to have for the lifetime of the card at the top end.
So I apply that to the FP10 MSAA discussion and am comforted. Am I kidding myself?
geo said:
I don't think thwrowing anything with a "FP-preset" into the same pot and draw from there conclusions will work.
That conventional wisdom you're talking about is merely in one spot in the pixel shader for ATI and it aimed to replace the split FP16/32 precision of NV's approach.
Besides it'll come down to what developers think of a specific implementation; even if some consider it as a viable alternative, there are way too many games under development that include already 64bpp HDR.
caboosemoose said:
And they also haven't released a G70 quadro variant yet, and the card in this picture has a genlock connector, a feature which has absolutely zero use for home users
http://www.theinquirer.net/?article=24837
I guess we'll find out at siggraph.
caboosemoose
Regular
I'm not arguing tht card is 7800 Ultra, I am simply arguing that a dual slot 7800 Ultra is very likely to appear at around the same time as R520. And given that NV has invested in making this heatsink and fan solution, I wouldn't be surprised to see it turning up on the 7800 Ultra.
