NVIDIA GF100 & Friends speculation

ares_ii_1.jpg


:oops:

Linky
 
Ah forgot about that. I think intel confused everyone because they switched fma encoding for AVX from 4 "operands" to 3 very late, but sandy bridge doesn't have it anyway (only the successor does, haswell).

Wait a Min, i thought FMA was scheduled for Ivy Bridge?
 
That's what I was thinking, too, but looking at the numbers it becomes apparent, that we're simply stagnating and he GTX260's price was actually nothing out of the ordinary when looking at it from a historic point of view:

Code:
Date   		Price	    Videocard             Performance improvement

April 2002:	~150€	GeForce 4 MX-440	     
April 2003:	~150€	9600 Non-Pro		    ~100%
April 2004:     ~150€	5700Ultra/9600XT	    ~50%
April 2005:     ~150€	6600GT/9800Pro	            ~100%
April 2006: 	~150€ 	7600GT/X1600XT              ~100%
April 2007: 	~150€ 	X1950Pro 512MiB             ~100%
April 2008: 	~150€ 	8800 GT 512MiB    	    ~25%
April 2009: 	~150€ 	HD4870/GTX 260 896MiB       ~60%
April 2010:     ~150€   HD 5770 1024MiB 	    ~0%

(Shamelessly ripped from here)


Looking at it from this perspective, we should have something faster than the GTX 260 1MiB for €150 by now and I don't really see the skewed value proposition.

Thanks for the charts, which basically conclude my feelings that graphics cards performance improvement has been slowing down. I hope we will Leap forward again sometime in the near future,
 
I'm not sure whether this has been discussed previously, but what are the chances that Nvidia release a GTX460 with all of the SM enabled clocked to 800MHz? That would surely be competitive with the 5870 and would force ATi to release 6000 series early and possibly rushed like Fermi was.

Also, is there any chance that the next full Fermi will be two GF104s stuck together for a 768SP part, or would that be too massive at 40nm, I suppose it's a better candidate for 32nm...
 
I'm not sure whether this has been discussed previously, but what are the chances that Nvidia release a GTX460 with all of the SM enabled clocked to 800MHz?
Why hasn't NVidia already released such a part at $350?

It's pretty clever to market the board as overclocking to >800MHz, don't you think? That way reviewers don't slam it for high power consumption, high noise levels. Take the halo from "800Mhz+" clock, but actually sell a pair of crippled parts - and while you're at it, call them the same name to further confuse customers.

That would surely be competitive with the 5870 and would force ATi to release 6000 series early and possibly rushed like Fermi was.
Why didn't NVidia launch GF104 in October 2009? There's no magic in there.

Also, is there any chance that the next full Fermi will be two GF104s stuck together for a 768SP part, or would that be too massive at 40nm, I suppose it's a better candidate for 32nm...
32nm node has been skipped. There's a decent chance GF104 was supposed to be implemented on 32nm.

28nm is the next node. Could the 28nm ramp be worse than 40nm?
 
I'm not sure whether this has been discussed previously, but what are the chances that Nvidia release a GTX460 with all of the SM enabled clocked to 800MHz? That would surely be competitive with the 5870 and would force ATi to release 6000 series early and possibly rushed like Fermi was.

Well, that would still be slower than the GTX 480 (most likely), and I'm not convinced that NVIDIA would be able to meet the demand associated with such a product, given that we know little about yields, other than the fact that the only GF104 product currently on the market is crippled.

Assuming that NVIDIA could do this, I think AMD would be fine with a price drop on the 5870, and possibly a 5890 at ~950/5000 MHz, if they felt the need to release it. I don't think they'd have to rush Southern Islands to market, especially since it's (apparently) due for October anyway.

Also, is there any chance that the next full Fermi will be two GF104s stuck together for a 768SP part, or would that be too massive at 40nm, I suppose it's a better candidate for 32nm...

I really don't see how that could fit on a <600mm² die at 40nm. As a dual-GPU solution, however, it's possible, but it would probably be a bit slower than the 5970 at 300W.
 
Wait a Min, i thought FMA was scheduled for Ivy Bridge?
Not sure but the only confirmation I found was Haswell will have FMA. Not much information seems to be available on Ivy Bridge, other than it's a shrink of Sandy Bridge...
 
Assuming that NVIDIA could do this, I think AMD would be fine with a price drop on the 5870, and possibly a 5890 at ~950/5000 MHz, if they felt the need to release it. I don't think they'd have to rush Southern Islands to market, especially since it's (apparently) due for October anyway.

Considering ATI remains supply constrained (fixed wafer allocation) and that's expected to continue through the rest of 2010, it's highly unlikely they'd reduce the price of 5870 even in the unlikely scenario that a "full and uncrippled" GF104 was released at high clocks.

That also means that it's unlikely AMD will rush SI to market. It's not like they are going to have a lot more wafers with which to make SI. And unless SI is the same size or smaller at 40 nm, they won't be able to sell more or even as many cards as Evergreen. SI is basically just going to be released during the Holidays to let consumers know that ATI isn't just sitting around twiddling thumbs while Nvidia continues to work on getting back on track.

Supply of those mythical "full" GF104s would have to be huge to lower demand enough for it to warrant AMD dropping price.

Regards,
SB
 
Considering ATI remains supply constrained (fixed wafer allocation) and that's expected to continue through the rest of 2010, it's highly unlikely they'd reduce the price of 5870 even in the unlikely scenario that a "full and uncrippled" GF104 was released at high clocks.

That also means that it's unlikely AMD will rush SI to market. It's not like they are going to have a lot more wafers with which to make SI. And unless SI is the same size or smaller at 40 nm, they won't be able to sell more or even as many cards as Evergreen. SI is basically just going to be released during the Holidays to let consumers know that ATI isn't just sitting around twiddling thumbs while Nvidia continues to work on getting back on track.

Supply of those mythical "full" GF104s would have to be huge to lower demand enough for it to warrant AMD dropping price.

Regards,
SB

Well, the GTX 460 must already be reducing demand for the HD 5830 and 5850 (hence the former's price drop), and it seems to be available in significant volume. NVIDIA is said to have ample allocation of 40nm wafers, so in theory, they could supply enough GF104s (full or not) to significantly reduce demand for the 5850 & 5870, which would prompt AMD to lower their prices.
 
Well, the GTX 460 must already be reducing demand for the HD 5830 and 5850 (hence the former's price drop), and it seems to be available in significant volume. NVIDIA is said to have ample allocation of 40nm wafers, so in theory, they could supply enough GF104s (full or not) to significantly reduce demand for the 5850 & 5870, which would prompt AMD to lower their prices.
I'm not sure if the recent prices of the 5830 are a response to the 460. I was already seeing a couple of them at $199 at newegg before the 460 launch. $239 always seemed a little high for it anyway. At various gaming forums, the advice was usually something like "just save up more for the 5850."
 
I'm not sure if the recent prices of the 5830 are a response to the 460. I was already seeing a couple of them at $199 at newegg before the 460 launch. $239 always seemed a little high for it anyway. At various gaming forums, the advice was usually something like "just save up more for the 5850."

I think that was just anticipation on AMD's part, in order to influence the conclusions of GTX 460 reviews.

I agree that the 5830 was overpriced, though. Actually, it still is in my opinion.
 
ATI still hasnt responded with price cuts. IMHO the 5830 should drop to $189 and the 5850 back to its launch MSRP of $259 to be competitive with the 768 and 1 GB versions of the GTX 460 respectively. But the rumours say AMD is still capacity constrained and have pushed some 40nm allocation to Cedar to fulfill notebook orders. So we may not see much of a price drop. Its been 10 months and Cpyress still contines to sell well above launch price. Its unprecedented in the graphics industry :???:

Back on track with speculation, how many SP's is GF106 supposed to have, 192 or 256?
 
Last edited by a moderator:
ATI still hasnt responded with price cuts. IMHO the 5830 should drop to $189 and the 5850 back to its launch MSRP of $259 to be competitive with the 768 and 1 GB versions of the GTX 460 respectively. But the rumours say AMD is still capacity constrained and have pushed some 40nm allocation to Cedar to fulfill notebook orders. So we may not see much of a price drop. Its been 10 months and Cpyress still contines to sell well above launch price. Its unprecedented in the graphics industry :???:

Back on track with speculation, how many SP's is GF106 supposed to have, 192 or 256?

You may want to check your sources(you can even read about it on this very page), you can get a HD 5830 for $199, down from $239.
Here's one at $184.99 with MiR http://www.newegg.com/Product/Product.aspx?Item=N82E16814161327&cm_re=hd_5830-_-14-161-327-_-Product
 
You may want to check your sources(you can even read about it on this very page), you can get a HD 5830 for $199, down from $239.
Here's one at $184.99 with MiR http://www.newegg.com/Product/Product.aspx?Item=N82E16814161327&cm_re=hd_5830-_-14-161-327-_-Product

The price had been dropping steadily for a while now, and had dropped to $199 before the GTX 460 launched. I suppose AMD cut the price in anticipation of the GTX 460 launch but there's been no major price cut otherwise. HD 5850 is still selling for $290-300
 
Back on track with speculation, how many SP's is GF106 supposed to have, 192 or 256?

we should expect 48SP per shader cluster now, thus I'd expect one GPC thing with four modules, giving a 192SP part, with a 144SP cut down part.
Probably using a 128bit bus and gddr5/ddr3.

I wonder about GF108 : 64bit or 128bit? a 64bit bus would make it lame.
I believe a 96SP, 128bit GF108 board with 512MB ddr3 would be pretty desirable and a significant step above using an IGP or an old card.

i.e. replace GT216 with GF108, GT215 with GF106 and keep that G98 on PCIe 1x crap for atom netbooks.
have a cut-down version of GF108 for cheaper/lower power needs?
 
Last edited by a moderator:
Back
Top