NV 55nm revision a preview to the gtx290?

ovaltineplease

Distinguished
May 9, 2008
1,198
0
19,280
speculation thread:


So, Nvidia released the 9800 GTX+ on a 55nm core. Largely speaking this card is in the territory of "just fair" - but more importantly it does imply to us that Nvidia is certainly looking towards die shrinks.


That said, I have a hard time imagining a gtx280 x2 (or anything of the same silly sort) - however what I can imagine is a gtx290 running on a 55nm core, with higher clocks, or maybe they'll call it the gtx280+ or something equally silly

The other possibility is a 9800GTX+ x2 - this, I could actually see happening given that 9800GTX+ SLI outperforms gtx280 - furthermore the 9800GTX+ isn't in the um.. well same power requirements/heat production as the gtx200s


I'm just basing these kind of predictions on how Nvidia lined up their last series of products; you can see something of a trend with how they have shifted pricing/performance and whatnot.

As it stands right now; if you made a parallel relation between the new product lineup and the old one you would have something like:

9600GT --- 9800GTX
8800GT --- 9800GTX+
9800GTX --- GTX260
9800GX2 --- GTX280

This is the current mainstream->high end product lineup from Nvidia in comparison to the old mainstream-> high end product lineup

As I already stated, I really cannot see the feasibility of a GTX280X2; but a 9800GTX+ x2 based on a 55nm and a larger framebuffer? Quite possibly, sure - I think thats a reasonable conclusion.

Where we stand now, AMD is most certainly the victor in the price/performance war; and with the 4870X2 coming, they will take the high performance crown as well. Would Nvidia continue on without trying to check AMD? Hard to say.
 
I would say that the GTX###GX2 wouldn't be happening either.

The question becomes though whether or not they use the 55nm revision to boost speed or boost yields. It's doubtful they could do both with such a limited time to optimize the design.

I have a feeling that they will do a minor if any boost in speed and instead focus on bringing the cost and price down quickly.

the biggest problem is that they are squeezed at both ends.
They likely could only boost performance about 10-15% at best which isn't huge, and they likely would only improve yields by maybe 60% from the current thrown around 60 (40%) dies to something like 100 dies if they yield above 50%. That would help bring down the cost, but it's not like it's cutting it in half, and it does nothing for the cost of the board which is also expensive (although intergrating the NVIO would help reduce that cost).

I think it may depend somewhat on how the HD4K-X2 does with regard to the performance 'crown', and if it's regainable by nV with a small 10-15% boost they may do that for a PR card, but I think they'll still focus their volume on selling a better yielding GTX-280 referesh. Kinda like the GF6800U/7800U was for show the GF6800GT/7800GTX-256 was for sale.

Price the OC'ed card at st00pid insane then you don't need to sell a bunch, but you still get the 'crown', and meanwhile sell a mint of those normal speed chips as GTX-270s or something and probably make alot more money than both GTX lines combined now.
 

ovaltineplease

Distinguished
May 9, 2008
1,198
0
19,280


Yea, true enough. I don't think its totally uncommon for people to buy into the mainstream sort of thing based on the psychological effect of "the crown"

I do like your point of view though, the yield angle seems quite sensible. I guess really you can only develope so many parts at once, being squeezed at both ends as you say.

I would think though that the 55nm revision biproduct of boosting yields would be making a sort of "better overclocking" card - some vendors are all about that sort of thing like BFG, XFX, Evga - the problem being though, its been shown that the gtx280 OC at least on the current model is not breaking any new ground in performance.

Of course, the gtx260 is a totally different story, it reportedly got some good gains from OC-ing in the HardOCP review.
 

jamesgoddard

Distinguished
Nov 12, 2005
1,105
0
19,290
You would have thought that Nvidia would have learnt from the mistakes that ATI made with the 2900's, ie a 512 bit bus is expensive and unmanageable - but no... What Nvidia need to do now is take a long hard look at the gtx280, shrink it to 55nm, get back with a 256 bit bus (and get working with that gddr5 that ATI have mastered) - then and only then will they be able to make a gx2 style card and get the performance crown back from the 4870x2 - anything else looks impossible...

One thing I would have to add, the thing that really sticks out is ATI's performance per transistor, the 48xx's have about 950mil, but the 2x0's have over 1400mil... That translates directly into production costs, then add that to the mega expensive PCB required for the 512bit bus - Nvidia don’t have a hope of matching the price / performance of ATI this time around... I expect production costs of the dual chip 4870x2 are even below a single gtx280 - Nvidia have been caught off guard big time..

Oh and add Nvidia's refusal to add any real development into the architecture since the 8800's - all they have done since is just make things bigger and bigger - not more efficient like ATI have managed to do - I won't be putting ay of my money in the direction of the green team..
 

ovaltineplease

Distinguished
May 9, 2008
1,198
0
19,280
Given that AMD had already mentioned a GDDR5 shortage contributing to the delays of the 4800 series - do you think a simultaneous GDDR5 launch was even possible in this time frame for both companies?

As Grape said, its more than likely NV intends to cut costs by refining the design. But, that will come later evidently.

Price/performance isn't 100% everything to everyone. I run an 8800GT SLI setup, do you think I am going to upgrade to 4850 CF ? No, because its a marginal improvement at best. Am I going to upgrade to 4870 CF? No, because its parallel to the gtx260 SLI, and I would have to buy an x48 board to appreciate (meaning, price/performance goes totally out the window)

AMD made a great launch, i'm truly glad because without competition the market would only sink more. The thing is though, that the AMD/Nvidia comparisons have already been made about 82 thousand times in other threads in the last couple days, and everyone has known what prices AMD was intending to launch the 4850, 4870, and 4870x2 at for months now. I mean that was words from AMD's mouths themselves.

I appreciate the thread bump, but it does just feel like we're going over well tread ground at this point.