NVIDIA GF100 & Friends speculation

[Alexko]

Seriously wtf, nV finally got their naming act together, nevermind the small issue of jumping to next hundreds quite a bit too easily perhaps, but nooo, they have to change that yet again, and GTX 560 will be GTX 560 Ti instead?

From what I understand, there will be a GTX 560 Ti with high clocks (823MHz for the core, twice that for the shaders) and a "standard" GTX 560, presumably with more reasonable clocks.

So yeah, after removing suffixes and turning them into prefixes, they're reintroducing suffixes again. In other words, it's probably worse now.

 

[trinibwoy]

I haven't seen anything indicating that 820Mhz isnt reasonable.

Also 2GB 560: http://www.chiphell.com/thread-155922-1-1.html

 

[Silent_Buddha]

It's going to be interesting to compare 1 Gig 6950 to 2 gig 6950 and 1 gig 560 to 2 gig 560. Far more enlightening with regards to how memory reliant each architecture is and the benefits of memory than what HardOCP tried to do with the 2 gig 69xx versus the 580/570.

I'll be especially interested to see the comparison in something like Metro 2033, where the 2 gig 69xx cards show a massive jump at very high settings compared to 1 gig 58xx cards.

Regards,
SB

 

[Xenus]

Nice a 2GB 560 might replace this old 8800Gt that or a 2.5GB 570 if they come.

 

[Squilliam]

I mean if Nvidia can make GTX460 so "tiny" wrt to modern gpu... GF114 Fermi must be a pretty economical chip..I think Ti competes with 6950 1GB...Nvidia again sending out OC versions...TWIMTP games...and i think Nvidia will have another hit...Cayman oh Cayman how fast it will be "sidelined" by the market forces...all IMHO.

The third largest chip in production at the moment is what you'd call tiny? Also one product which isn't out on the market competes with another similarly not released product?

Plus, if NVIDIA can sell GTX 460s for $118 (!!!) after MIR and with Just Cause 2 included, AMD should be fine with the 6950, even at $200.

It's not Nvidia clearing product, the board partners are responsible for the majority of the specials and discounts IIRC.

A salvage part they were previously selling for more money with the vanilla 6950. The margins would definitely have to take a hit for ATi to price it low enough to compete with the 560. If the 560 is all it's cracked up to be a 1GB Cayman Pro for $30-40 less than the 2GB version would obsolete the 2GB version. Essentially AMD are losing that revenue where before they were selling the same chip in a card with $15 more RAM for $30-40 more.

The other problem I can see is that 1GB 6950 could have lower clocks and lower memory clocks to make sure it doesn't eat into 2GB 6950 sales, but that would mean having to sell it significantly cheaper than 560 and that again will eat into their margins, selling a larger chip with worse performance would put AMD into a bad position similar to how they have been operating in their processor division. Nvidia were able to do that because they had fat margins on their pro products, they seem to get a better deal from TSMC, and they have a massive cash pile to fall back on.

I don't understand what you're all going on about personally. AMD makes a chip, they sell it to the board partners sometimes as a kit which is what you see with the referrence design and at other times they just sell the chip. A lower price doesn't always mean significantly lower margins for AMD given the fact that the board partners can also have some of their margins eaten into as well. If the margins are healthy for board partners then I would suspect that the board partners, at least with some of the early price cuts for competitive reasons take more of a hit to their margins than AMD would.

 

[gongo]

Nope.

I think you've missed some fundamentals of how this business operates.

Yeah, 6950 1GB is the full Cayman Pro chip with half the memory chips and .... single bios(!)
http://www.hisdigital.com/ca/product2-589.shtml
I wonder if it has a shorter PCB and reduced power delivery system....vapor chamber probably gone in HIS version....though the amount of heat i got @ 6970 speed..and similar reports in forums...i wonder how effective has AMD implemented vapor cooling for Caymans....

The third largest chip in production at the moment is what you'd call tiny? Also one product which isn't out on the market competes with another similarly not released product?

I meant the whole card itself...GTX 460 was really short and uses a simple HSF..

 

[Dave Baumann]

It's the same number of memory devices chips, just lower density devices. And the vapor chambers are excellent for the products they were designed for.

 

[Squilliam]

Poor Dave, I wonder sometimes whether being the one eyed man in the land of the blind is more burdensome than the metaphor made it out to be.

@Gongo: Even if the PCB is short it doesn't provide significant insight into the cost of actually making it.

 

[DegustatoR]

Thats a fully functional GPU compared to a salvage part. I don't anticipate a problem with gross margins under such a scenario.

Salvage part cost the same to produce as a fully functional GPU though -- in theory.

Plus, if NVIDIA can sell GTX 460s for $118 (!!!) after MIR and with Just Cause 2 included, AMD should be fine with the 6950, even at $200.

NV and AMD pricing schemes are very different. You can't make anything regarding AMD's prices from a price of some NV product. In other words: if NV has an ability to sell GF104-based product for such price that doesn't mean that AMD has the same ability even with a comparable in die-size GPU.

 

[AnarchX]

http://translate.google.de/translat...d-156557-1-1.html&sl=zh-CN&tl=en&hl=&ie=UTF-8

Gigabyte GTX 560 Ti SuperOC with 1000/2000/1154MHz (203W TDP)?

 

[psolord]

http://translate.google.de/translat...d-156557-1-1.html&sl=zh-CN&tl=en&hl=&ie=UTF-8

Gigabyte GTX 560 Ti SuperOC with 1000/2000/1154MHz (203W TDP)?

Actually they mention that the 1Ghz version will be at 233W. The 900Mhz one will be at 203W.

At these clocks, the 560 should be breathing down 6950's neck.

 

[AnarchX]

Actually they mention that the 1Ghz version will be at 233W. The 900Mhz one will be at 203W.

Did you translate itself?

According to Google the 203W TDP was reached through a better PCB. A public version (reference design) would be 233W at 1GHz OC.
Probably Gigabyte did also some selection of the GPUs, to reach this TDP @ 1GHz.

 

[psolord]

Did you translate itself?

According to Google the 203W TDP was reached through a better PCB. A public version (reference design) would be 233W at 1GHz OC.
Probably Gigabyte did also some selection of the GPUs, to reach this TDP @ 1GHz.

Ah yes, missed some letters on their coding. Impressive to say the least.

The 560 @ 1Ghz should be around (8/7) * (1000/675) = 68% faster than the stock clocked 460 1GB. In non bandwidth limited situations, it should be as fast as (or faster than) a 570! wth?

 

[Blazkowicz]

the power use is nearing that of a 570 too.
for fun I've computed 448 / 384, that's only a ratio of 1.17.

 

[Silent_Buddha]

Salvage part cost the same to produce as a fully functional GPU though -- in theory.


NV and AMD pricing schemes are very different. You can't make anything regarding AMD's prices from a price of some NV product. In other words: if NV has an ability to sell GF104-based product for such price that doesn't mean that AMD has the same ability even with a comparable in die-size GPU.

Salvage parts will cost less due to the fact that you can recover/use a GPU that would otherwise have been thrown away. In effect you can consider it one of two ways.

1. The cost of the GPU for a salvage part for AMD is 0 as it was going to be thrown away and thus cost of wafer was only factored into the non-salvage parts.
2. The cost of all GPUs is reduced as you can now refactor the cost per GPU from a wafer. More likely, IMO.

Problems arise, of course, if you have to use non-salvage chips for a salvage part.

Cost of the board and assembly is going to remain the same, unless you use different parts. For example, different VRMs or memory grades (IE - 6950 versus 6970). As well different coolers, different memory size, maybe a less complex PCB, smaller PCB, etc...

As for cost of GPU's for NV versus AMD. That's going to be almost purely based on yields as to how much each GPU costs per mm squared. Wafer costs will be roughly equivalent for both. Smaller GPUs will tend to not only make more efficient use of a wafer, but also yield better and thus will generally yield lower cost per mm squared.

Regards,
SB

 

[gongo]

http://www.xtremesystems.org/forums/showpost.php?p=4708609&postcount=154

More info on GTX 560 1GB....at 900mhz SOC speed, Gigabye rates it at close or equal to HD6950 2GB....the chinese website have prices of it at Y1900 while HD6950 2GB goes for Y2250..I am predicting stock GTX 560 SRP will be at $259, can AMD make Cayman Pro 1GB as affordable? Nvidia is meaning business with Fermi 2.0.....and do we need to wait till the 25th for GF114 to go on sale, just release the damn thing this week!

 

[Silent_Buddha]

Ah yes, missed some letters on their coding. Impressive to say the least.

The 560 @ 1Ghz should be around (8/7) * (1000/675) = 68% faster than the stock clocked 460 1GB. In non bandwidth limited situations, it should be as fast as (or faster than) a 570! wth?

Things like that make me wonder if Nvidia are looking to un-officially phase out the GTX 570 for a variety of reasons.

GTX 570 launching 1 month later than GTX 580 suggests that yields on GTX 580 are good enough that it can't support a healthy ecosystem of GTX 570's without using non-salvage parts. IE - it took an extra month to get enough chips that aren't fully enabled GF110's to support a launch.

If that's true, then having super overclocked GTX 560's at similar price points and performance to GTX 570 could bleed off some demand of GTX 570.

Which would serve two purposes. Reduce demand such that Nvidia won't have to resorts to using GF110's for GTX 570 that could have been used for GTX 580. And allow for a cheaper GPU to move into the price bracket held by the GTX 570, thus increasing margins and profit for board vendors and Nvidia.

All speculation, of course, since we can only speculate on performance of GTX 560 at this point. And if super overclocked GTX 560's can be similar in speed to GTX 570, Nvidia would have be careful not to allow any variants of that to go much faster, otherwise it could make GTX 580 incredibly unattractive.

Regards,
SB

 

[Squilliam]

Salvage part cost the same to produce as a fully functional GPU though -- in theory.

Tinned fruit cost the same to produce as store bought fruit, the same applies to fruit which is turned into juice. The cost to the end user is significantly different in each case. You could say that tinned fruit and juiced fruit are the salvage parts of the horticultural industry. The expectation is that the fruit bought by the end user at a fruit stand ought to cover the cost of operation. The tinned and juiced fruit are there purely as a means to increase profit/revenue on otherwise unsellable product. By no means would a grower consider tinned fruit to be unprofitable even if on average each piece of fruit doesn't cover the cost of production.

NV and AMD pricing schemes are very different. You can't make anything regarding AMD's prices from a price of some NV product. In other words: if NV has an ability to sell GF104-based product for such price that doesn't mean that AMD has the same ability even with a comparable in die-size GPU.

They both operate using the same fab, with sometimes the same AIB partners and probably the same kind of deals between themselves and their partners. Most likely if NV can sell a chip for X @ Y die size and with Z board components, if X, Y, Z are the same then AMD likely can do likewise.

 

[gongo]

Things like that make me wonder if Nvidia are looking to un-officially phase out the GTX 570 for a variety of reasons.

GTX 570 launching 1 month later than GTX 580 suggests that yields on GTX 580 are good enough that it can't support a healthy ecosystem of GTX 570's without using non-salvage parts. IE - it took an extra month to get enough chips that aren't fully enabled GF110's to support a launch.

If that's true, then having super overclocked GTX 560's at similar price points and performance to GTX 570 could bleed off some demand of GTX 570.

Which would serve two purposes. Reduce demand such that Nvidia won't have to resorts to using GF110's for GTX 570 that could have been used for GTX 580. And allow for a cheaper GPU to move into the price bracket held by the GTX 570, thus increasing margins and profit for board vendors and Nvidia.

All speculation, of course, since we can only speculate on performance of GTX 560 at this point. And if super overclocked GTX 560's can be similar in speed to GTX 570, Nvidia would have be careful not to allow any variants of that to go much faster, otherwise it could make GTX 580 incredibly unattractive.

Regards,
SB

According to Gigabyte in my above links...GTX 560 SOC is only able to compete with Cayman Pro and....at 900mhz, which is the upper acceptable TDP limits for Fermi @ 40nm (generally what i have seen in forums)....hmm, i would say 900mhz is where 40nm high end AMD/Nvidia gpu can clocks...wonder if we will see a 900mhz Cayman Pro SOC too!

 

[AnarchX]

According to Gigabyte in my above links...GTX 560 SOC is only able to compete with Cayman Pro and.

In your links its not the SOC, its only the OC version.

This is SOC: http://item.taobao.com/item.htm?id=9101414085

In Vantage Extreme it should be 3% faster than GTX 570.

The earlier benchmarks and screens of GTX 560 Ti showed a voltage of only 1,0V for basic 822MHz,