What form will ATI's refresh take ?

[mactronix]

Ok so earlier in the year ATI said they would refresh during the second half of 2010.
Has anyone heard anything else regarding this ? I'm just wondering /curious as to how or what form of improvement to expect.
Could it be possible with the talk of the 6 series being 32nm or some are even saying 28nm, that we get another tester card like the 4770 ?
The idea of a 5770 on a 32/28nm process is one I like the sound of very much Especially if it could pull similar relative performance as the 4770 did.
Or should we just expect clock increases from a improved production process ?

Your thoughts please

Mactronix

 

[Mousemonkey]

[:grahamlv:3] I'm going to hedge my bets and say that it's going to be PCIe compatible and require something called electricity to function. [:mousemonkey]

 

[-Fran-]

[:grahamlv:3] I'm going to hedge my bets and say that it's going to be PCIe compatible and require something called electricity to function. [:mousemonkey]

Very cautious on your bet I see, Mr. MM. Well, I'll take my chances on my statement.

There was an article at XBitLabs saying TSMC will go directly to 28nm instead of staying on 32nm: link

So that leads me to believe ATI might do another 4770 experiment on 28nm, wich might as well be very nice.

I have my doubts about naming (5790? 5775? lol), but they will certainly make a card proving the tech, at least my hopes are very high

This might also mean that Fermi2 or an improved Fermi product might spawn right after this step to 28nm.

Cheers!

 

[welshmousepk]

i imagine something similar to the last gen 4890. (a 5890)

a slightly more powerful card, cooler and with more OC potential thanks to a die-shrink.

im pretty sure it will be at the high end though, since ATI will want to pull out all the stops to out perform fermi.

 

[Mousemonkey]

Very cautious on your bet I see, Mr. MM

Seeing as I can't even begin to pronounce Hecatoncheires I thought I'd give myself some room to manoeuvre. :lol:

 

[Griffolion]

This might also mean that Fermi2 or an improved Fermi product might spawn right after this step to 28nm.

28nm SHOULD have really been Fermi1, Nvidia were behind ATI from the off, they should have just conceded 2010 and got a 28nm Fermi wafer taped out by mid to late 2010 and have a release date christmas time. At least it would have given them time to actually get to grips with this new complicated chip design and optimise etc. Oh well.

 

[soundefx]

I can't remember which forum I read, but I thought that they were going away with the 5850, 5870 and 5970.

The only oneI can remember is that 5970 replacement was the 5990 or something like that.

In terms of the die size, I am not sure, but if they stay true to their track record, we will see ATi trying new things with new technology.

 

[welshmousepk]

the 5990 is not an ATI design, its just a 3rd party pre-overclocked 5890 with an extra 2gb VRAM.

i severely doubt that ATI would just do away with these cards, even if they replace them with better ones. not a great use of the R+d poured into them.

 

[mactronix]

Seems Yuka and myself have the same hopes, I must say that what welshmousepk posted does make a lot of business sense. I guess if they can shrink the die and get the clocks up they should end up with something capable of keeping the crown in the ATI camp. They can prove the process while making a limited run of top end 5 series cards to hold top performance untill the 6 series comes along.

Mactronix

 

[jonpaul37]

This might also mean that Fermi2 or an improved Fermi product might spawn right after this step to 28nm.

28nm SHOULD have really been Fermi1, Nvidia were behind ATI from the off, they should have just conceded 2010 and got a 28nm Fermi wafer taped out by mid to late 2010 and have a release date christmas time. At least it would have given them time to actually get to grips with this new complicated chip design and optimise etc. Oh well.

That could have worked but Nvidia had already given up on the 200-series cards and they were in short supply. So if they waited out 28nm, it could have been more disasterous.

 

[-Fran-]

That could have worked but Nvidia had already given up on the 200-series cards and they were in short supply. So if they waited out 28nm, it could have been more disasterous.

They prolly had a roadmap for the GT220/240 base chip (DX10.1), but the 5xxx series came out way early (for them) and had to force Fermi development with a gun at hand

Anyway, I just read this on TechReport: link.

Kind words are always great, but I wonder if they can turn nVidia's very tight tech way of developing (competition wise, not developers-TWIMTBP wise). On a simple analysis, I'd say nVidia gets more market when sharing tech and not overloading us with PR/Marketing crap.

Cheers!

 

[uncfan_2563]

Don't you think its a bit early to drop to 28nm, i mean, they just got done having problems with 40nm

 

[-Fran-]

Don't you think its a bit early to drop to 28nm, i mean, they just got done having problems with 40nm

That's a big unknown for me at least. It would depend on the lithography process itself.

Cheers!

 

[nfail]

nVidia are having serious execution problems, and have been for a while.

i think they have a problem with their philosophy and attitude more than anything else. they have to bite the bullet and accept that ATI got it right and they were wrong. they cant go on making huge chips that are 6-9 months later than ati's because that would be financially ruinous.

ATI wont change anything because their system is proven to be far better, so why should they change anything with a refresh? they will just keep doing what works.

 

[welshmousepk]

they cant just keep selling the same cards forever though nfail...

 

[nfail]

true, and they will change things a little. i just dont think we're gonna see huge changes while nVidia can't compete.

all i see from ATI for the next few years is more fps, better eyefinity etc. i think they will surprise the hell out of us within 2 years, showing the first 360 degree "holodeck".

 

[EXT64]

Now that I would like! But, would a console port look good on a holodeck? As that is likely all that will be left by then.

 

[uncfan_2563]

all i see from ATI for the next few years is more fps, better eyefinity etc. i think they will surprise the hell out of us within 2 years, showing the first 360 degree "holodeck".

Well duh we're getting more fps.. what the heck else would we get?

I don't agree with the opinion that NVIDIA isn't competitive. I think ATI definitely needs a refresh, like a 5890, to keep up in the high end single gpu cards IF the GTX 480 is really as fast as they say

 

[hopar]

I agree with the posters who think nothing major will occur for ATI anytime soon. All they need to be doing continue to improve drivers for the 5000 series, get even better yields and thus lower prices more. They might release limited quantities of the 28nm part in a limited edition card, but I dont see a full redesigned 6 series while there is no strong pressure from nvidia.

 

[EXT64]

Oh, the 6 series will be already far through development (done design wise essentially) at this point, so I imagine that will continue full speed ahead.

 

[shadow187]

If ATI's 6K series skips through the 32nm and heads to 28nm, that will be very good. Since ATI has always been the one to do new processes, ATI will have more experience, and can put better thins on the process. Ntm power consumption/yields.

Extremely cheap value/performance cards, very good p/p high-performance/enthusiast cards, and huge high-enthusiast/premium cards. Then the even bigger high-premium card made by its manufacturing partners.

Oh the joy.

 

[noob2222]

Well, last I read, GF is supposed to start testing the 28nm runs in june/july, seems to be just in time for this refresh. Wouldn't suprise me at all to see AMD start to dump off TSMC after the 40nm fiasco.

 

[EXT64]

"Testing" doesn't quite sound to me like the ability to start making extremely complex chips yet. Might not be until the following gen, but I guess we'll see.

 

[sabot00]

I'm guessing a 5790 or 5690 based on 32/28nm and a OC'ed 5870 into a 5890.

 

[nfail]

if ATI were going to release a 5890 they wouldnt allow their partners to overclock the 5870 to 1ghz already.