NVIDIA to Counter Radeon HD 6970 ''Cayman'' with GeForce GTX 580

[ElSel10]

That card could be a full GF104 dual. 384 CUDA Cores x2, 64 TMUs x2, 256 bit x2 (marketing), 1GB GDDR5 x2.

Just a quick note... when using AFR rendering, you actually do get an effective memory bus size of double the single card. Just not memory size.

[SKYMTL]

So basically, at this point there is absolutely no way for Nvidia to catch up with AMD, and within a few generations of this lagging behind, they will likely go bankrupt?

I remember the same thing being said about ATI after they launched the 2900XT.

There's every possibility of them keeping up. The issue is that AMD is currently 1/2 of a product generation ahead but that doesn't mean NVIDIA can't compete. The GTX 460 series of cards and GF104 in particular has the has the ability of keep GeForce cards highly competitive at very enticing price points.

Remember, the real money in the gaming GPU market isn't made out of limited shipments of high end, expensive to manufacture cards but rather on large volume shipments of mid range and lower end cards.

NVIDIA also has the ability to leverage their near-monopoly in the professional GPU markets (the areas where Tesla and Quadro play) to take advantage of sky high profit margins. Now that volume shipments have begun of these parts and large OEMs like Dell, HP, Lenovo, etc. are starting to feature them in their workstation configs I think NVIDIA is in a pretty good position regardless of being 1/2 a generation behind in the standard desktop market.

[zalbard]

The issue is that AMD is currently 1/2 of a product generation ahead but that doesn't mean NVIDIA can't compete. The GTX 460 series of cards and GF104 in particular has the has the ability of keep GeForce cards highly competitive at very enticing price points.

Fermi arch is superior to 5xxx, too (as long as you can get chips made right so they don't overheat). So not necessarily.

[Dimitriman]

So basically, at this point there is absolutely no way for Nvidia to catch up with AMD, and within a few generations of this lagging behind, they will likely go bankrupt?

Please don't be upset with my comment. I said that nvidia will not be able to catch up with amd on anything built on the 40nm node. this round is a complete win by amd until next node is in production. for that time we can only speculate as of now about amd and nvidia next gen but i expect nvidia to be a lot more competitive on 28nm. still i dont expect amd to just sit around and do nothing either.

fact is - nvidia cannot make a functional 40nm fermi with 512cuda 128tmu and 512bit mem controller. and this means they are not likely to take the performance crown untill 28nm.

... and no nvidia wont be bankrupt cause of fermi

[SKYMTL]

Fermi arch is superior to 5xxx, too (as long as you can get chips made right so they don't overheat). So not necessarily.

A "product generation" does not refer to the underlying architecture. The architecture isn't an issue here, it is the fact that AMD is able to release a refreshed product lineup while potentially increasing their margins before NVIDIA can do the same.

[kaktus1907]

512-bit memory controller is way too high.

576SP/96TMU/384-bit would be nice

This

[Jowy Atreides]

Dejá vú?

I seem to remember this happening when the 5870 was released. Big promises and more of them and more of them etc.
Next thing you know, It's a different decade and 5 months into it when the card arrives.

[Bobsama]

So basically, at this point there is absolutely no way for Nvidia to catch up with AMD, and within a few generations of this lagging behind, they will likely go bankrupt?

Maybe not bankrupt, but nVidia is losing market share. Something needs to change at nVidia. They've been really off their game since HD5000 came last year, and took them quite a few months to release the GTX 400's. They're going to be squashed if HD6000 improves performance even 10-20%.

[Carfax]

They're going to be squashed if HD6000 improves performance even 10-20%.

Considering that Fermi is already 10-20% faster than the 5000 series in most games, I highly doubt Nvidia is going to be squashed..

Also, do we even know the slightest bit of info about Cayman, ie the amount of shaders it will have?

[Bobsama]

Last rumor was HD6900 would have 20% more shaders than HD5800 (~1920 for 6970, ~1730 for 6950). Add in more uArch improvements and 50% more VRAM and I'd think that 25-30% faster is feasible if nothing else.

[snoid_zero]

They're going to be squashed if HD6000 improves performance even 10-20%.

Probably true if nVidia is only about GeForce.
But they are not.

[zerazax]

Sounds like the continuation of BS from the green team. Let's see them deliver a full product at 40nm first, then we'll talk

[DilTech]

Please don't be upset with my comment. I said that nvidia will not be able to catch up with amd on anything built on the 40nm node. this round is a complete win by amd until next node is in production. for that time we can only speculate as of now about amd and nvidia next gen but i expect nvidia to be a lot more competitive on 28nm. still i dont expect amd to just sit around and do nothing either.

fact is - nvidia cannot make a functional 40nm fermi with 512cuda 128tmu and 512bit mem controller. and this means they are not likely to take the performance crown untill 28nm.

... and no nvidia wont be bankrupt cause of fermi

Not quite...the GTX 460 has been selling like hot cakes and rules the mid range with an iron fist. As such, you can't call it a complete win for AMD, but they definitely did surprise a lot of people.

Anyway, a dual GTX 460 will put up some numbers for sure. I'm not fond of dual-gpu cards regardless of brand(we all know that) but it is easily feasible to pull off and they scale like crazy. Especially if it's clocked right. Even more so if they go crazy and use the unlocked cores for this one(remember, the G104 has more to it then we see in the GTX 460).

One thing I will say though, a 512 shader card is EASILY within possibility if they base it off the GF 104 design instead of the GF100 design... The amount of space saved by cutting off the workstation features would allow for 512 shaders without issue and still be under the power draw of the GTX 480.

As for 28nm, didn't NVidia already say they have a new design sitting and waiting for when 28nm is ready? Not like I take either company's word as gold these days(EVERY company lies, we all know that), but sounds to me like this is just going to be a stop-gap tide over much like what AMD is doing.

Finally, about the renaming... NVidia would be stupid not to. With AMD toting their new cards as their next gen, people who don't know anything about graphics chips will think "well, that NVidia card is last years, while this AMD card is new"... Imagine the BS best buy workers would spew. If either company calls something a next gen part the other HAS to follow suit.

[Lanek]

Surely they will release something, ( i don't think Nvidia will stay without reaction to the actual situation with DX11 gpu's ) but it seems each time there's an annonces about AMD, we have a new " rumored " card from NV who surface somewhere .

[Lanek]

Surely they will release something, ( i don't think Nvidia will stay without reaction to the actual situation with DX11 gpu's ) but it seems each time there's an annonces about AMD, we have a new " rumored " card from NV who surface somewhere . Let's hope it's not only a rumor .

Not quite...the GTX 460 has been selling like hot cakes and rules the mid range with an iron fist. As such, you can't call it a complete win for AMD, but they definitely did surprise a lot of people.

Anyway, a dual GTX 460 will put up some numbers for sure. I'm not fond of dual-gpu cards regardless of brand(we all know that) but it is easily feasible to pull off and they scale like crazy. Especially if it's clocked right. Even more so if they go crazy and use the unlocked cores for this one(remember, the G104 has more to it then we see in the GTX 460).

One thing I will say though, a 512 shader card is EASILY within possibility if they base it off the GF 104 design instead of the GF100 design... The amount of space saved by cutting off the workstation features would allow for 512 shaders without issue and still be under the power draw of the GTX 480.

As for 28nm, didn't NVidia already say they have a new design sitting and waiting for when 28nm is ready? Not like I take either company's word as gold these days(EVERY company lies, we all know that), but sounds to me like this is just going to be a stop-gap tide over much like what AMD is doing.

Are you really sure GTX460 sold so much? We see many peoples on forums who have buy some lately ( i have buy some for friends and my brother ).... but last official numbers for AMD don't show this at all .... AMD have 90% of the DX11 market share it seems. ... 25millions of GPU ... I don't say they don't sell well, it's a good card, excellent price/perf ratio, but they come late, really too late, released in June2010, when AMD have a complete line up till Oct 2009.

Many peoples who wanted a middle range gpu's ( and price ) have wait after Fermi release in March .. They was no middle range gpu's at this time from Nvidia and 3 months more have been needed for see one surface, Many peoples have not wait and buy AMD gpu's during this time.... Look Steam and other hardware survey site: GTX460 represent just 5% of the DX11 cards used ( ofc they have a shorter life )

[Shadov]

I just hope we wont have an announcement of an announcement this time to stall competition... No wooden screws either!

[Monkeyface]

Lanek there is no way AMD has 90% market share, that would be ridiculous.

[Etihtsarom]

So will there be anything for Antilles?

[ColonelCain]

I call FUD, since TSMC doesn't have 28nm ready yet, and they are saying it will have a TDP close to the 480.

[iMacmatician]

So will there be anything for Antilles?

Exactly what I was wondering…

[Bobsama]

Lanek there is no way AMD has 90% market share, that would be ridiculous.

Perhaps 90% marketshare for DX11.
http://store.steampowered.com/hwsurvey/directx/

Quick estimate for high-end...

HD5800: 7.04%
HD5700: 6.82%
===13.86% DX11 cards

GTX460: 1.18%
GTX470: 1.14%
GTX480: 0.90%
===3.22% DX11 cards

So total 17.08% cards are DX11, giving AMD at least an 81.1% DX11 marketshare. That list doesn't include the very few lower-end nV cards or the lower-end AMD cards (for DX11 compatible), or any mobile graphics.

[RaZz!]

Lanek there is no way AMD has 90% market share, that would be ridiculous.

well, according to the steam hardware survey (http://store.steampowered.com/hwsurvey?platform=pc) 84% of DX11 GPUs are from ATI, only 16% from NVIDIA (desktop and mobile parts).

DX11 was what Lanek was talking about, so his 90% aren't that off.

[Tha Last Meal]

Perhaps 90% marketshare for DX11.
http://store.steampowered.com/hwsurvey/directx/

Quick estimate for high-end...

lol you are say that AMD Depend on steam When they talk about market share

This is really funny.... omg AMD now love steam
The number of active members of the Steam is Between 5-6 Million...no more

[QuadDamage]

Last rumor was HD6900 would have 20% more shaders than HD5800 (~1920 for 6970, ~1730 for 6950). Add in more uArch improvements and 50% more VRAM and I'd think that 25-30% faster is feasible if nothing else.

I'd say more like 10%. Look at the difference between the 5850 and 5870. 1440 vs 1600 SPU's and 100Mhz difference between the two and a measly ~5% performance edge over the 5850.

I'm thinking this Cayman might just match the GTX480.

[Bobsama]

lol you are say that AMD Depend on steam When they talk about market share

This is really funny.... omg AMD now love steam
The number of active members of the Steam is Between 5-6 Million...no more

I said no such thing. Steam has the most up-to-date gamer stastics that I've seen. Their sample size is enormous and easily large enough to represent all (upmarket) computer gamers.

If you haven't noticed, I'm not an AMD lackey and I'm not an nVidia or Intel lackey either. In the past few years, both companies have brought great products that I've advocated buying. When building for a client, I try to suggest what's best for their uses.