Any plans on ATI's part for a 5830 or 5790?

[GodisanAtheist]

Basically is there anything in the cards (I made a pun!) for ATI to fill the ginormous performance gap between the 57xx series and the 58xx series? Just like the 5850 takes a 10% SP hit from the 5870, perhaps a 5830 with another 10% off for a 1280 SP part or a 5790 with a 10% added on for a 960 SP part.

With the 4890 being EOL'ed soon and the 4xxx's series habit of introducing a variant for every market segement, half segment and quarter segment it makes sense that ATI would release one of these mid-high cards.

 

[Arkaign]

It's possible once the supply issues are fixed. As it is, ATI is selling every single 5xxx series they can produce as far as I can tell.

 

[Borealis7]

if the price gaps justify it, they will.
check again in 6 months after AMD has had some time to deal with the release of Fermi cards.

right now its more of a "Ferma" then a "Fermi" (unsolved mystery, haha)

and yes, i know Ferma's Theorem was solved in 1993.

 

[betasub]

Fermat?

 

[xenolith]

These are the rumors from a not-too-reliable source...

5790 (Juniper XTX) at Q2 2010. It basically looks like a 5770 with double the memory bandwidth (256bit vs 128bit). I was hoping they would release it earlier since there have been no shortages of 40nm Juniper, but they may move it to a 28nm process, so it may take at least that long.

5830 (Cypress LE) at Q1 2010. The only thing I know about it is it may have increased memory bandwidth than the 58xx (384bit vs 256bit) and move to a 28nm process. If true then I doubt it will launch far ahead of Juniper XTX.

 

[GodisanAtheist]

These are the rumors from a not-too-reliable source...

5790 (Juniper XTX) at Q2 2010. It basically looks like a 5770 with double the memory bandwidth (256bit vs 128bit). I was hoping they would release it earlier since there have been no shortages of 40nm Juniper, but they may move it to a 28nm process, so it may take at least that long.

5830 (Cypress LE) at Q1 2010. The only thing I know about it is it may have increased memory bandwidth than the 58xx (384bit vs 256bit) and move to a 28nm process. If true then I doubt it will launch far ahead of Juniper XTX.

-The 5790 makes sense, as there is no reason a 5770 should be trading blows with a 4870 unless the memory bandwidth is holding it back.

The 5830 sounds a little out there though. A larger bus width on a lower end card? Makes zero sense. That sounds more like the fabled 5890 than anything else.

 

[xenolith]

The 5830 sounds a little out there though. A larger bus width on a lower end card? Makes zero sense. That sounds more like the fabled 5890 than anything else.

Yeah, that made no sense to me either when I saw it.

 

[Arkaign]

From what I understand, the whole reason the 4830 existed in the first place was to fill in a gap until the 4770 (cheaper to produce) came into being. Given that the 5770 already exists, I rather doubt ATI will sell a wide-bus 5830 card, as it would be about as expensive as a 5850 or 5870 to produce, but command a lower ASP.

Perhaps a 5790 will eventually come out though, still with the more limited bus but with better core clock / memory speed? That might happen after a die shrink, but that hasn't historically happened very often either.

 

[GodisanAtheist]

Like all things surrounding the 5xxx series, this will probably be TSMC's fault too. Until ATI has harvested enough "not quite good enough" cores off the 58xx series this segment will be filled by the 4890... and since yield issues are plaguing the 58xx series its going to either be a while or show up as a tester card for GF's 28nm process sometime in the distant future.

 

[bunnyfubbles]

These are the rumors from a not-too-reliable source...

5790 (Juniper XTX) at Q2 2010. It basically looks like a 5770 with double the memory bandwidth (256bit vs 128bit). I was hoping they would release it earlier since there have been no shortages of 40nm Juniper, but they may move it to a 28nm process, so it may take at least that long.

5830 (Cypress LE) at Q1 2010. The only thing I know about it is it may have increased memory bandwidth than the 58xx (384bit vs 256bit) and move to a 28nm process. If true then I doubt it will launch far ahead of Juniper XTX.

Neither of these scenarios seem likely, although the 5790 with a 256bit bus would be ideal, I don't think the GPU can even be paired with a 256 bit bus.

Also, the 5830 would never have a 384bit bus...that would give it more bandwidth than the 5870 or 5850...

That being said, I think the 5830 sounds most logical as it would be just a further cut down 5800 (1280 SPs, etc), a 5790 would be hard to put out without faster GDDR5 to further distinguish it from the 5770.

 

[Daedalus685]

Like all things surrounding the 5xxx series, this will probably be TSMC's fault too. Until ATI has harvested enough "not quite good enough" cores off the 58xx series this segment will be filled by the 4890... and since yield issues are plaguing the 58xx series its going to either be a while or show up as a tester card for GF's 28nm process sometime in the distant future.

I was wondering about that. When we here reports of "poor yields" are we talking chips that don't function at all vs ones that do to some degree, or are we talking about chips that make the cut at the desired specs? (a la 5850 or 5870)?

At any rate, there will always be a cut down part eventually. They need to sell what they can. I'm almost certain we will see a 58(<50) at some point, and have a feeling they will refresh the 5770 into a 5790 when they refresh the 5870.

 

[GodisanAtheist]

I was wondering about that. When we here reports of "poor yields" are we talking chips that don't function at all vs ones that do to some degree, or are we talking about chips that make the cut at the desired specs? (a la 5850 or 5870)?

At any rate, there will always be a cut down part eventually. They need to sell what they can. I'm almost certain we will see a 58(<50) at some point, and have a feeling they will refresh the 5770 into a 5790 when they refresh the 5870.

- I'm fairly certain poor yields in this case is referring to unsalvagable cores, not merely gimped ones, otherwise I'm almost positive we would have seen the 5830 already given how bad the supply issues are.

At this point I'm certain ATI could shit in a box and sell it for $300 bucks so long as it was part of the 58xx series, and it would be to their benefit to get whatever they can out before Christmas. The fact that cut down parts haven't already made it to channel (gimped 5870/5850 cores thrown onto a 5850 PCB and out the door) implies the bad yields are simply not functional cores.

 

[Daedalus685]

- I'm fairly certain poor yields in this case is referring to unsalvagable cores, not merely gimped ones, otherwise I'm almost positive we would have seen the 5830 already given how bad the supply issues are.

At this point I'm certain ATI could shit in a box and sell it for $300 bucks so long as it was part of the 58xx series, and it would be to their benefit to get whatever they can out before Christmas. The fact that cut down parts haven't already made it to channel (gimped 5870/5850 cores thrown onto a 5850 PCB and out the door) implies the bad yields are simply not functional cores.

That is what I figured, but even if they are getting some usable 5830's out of there, how long would it take to get a 5830 card out the door if they had only done enough work on it conducive to releasing it in the spring? It might take a while if they already locked it in for a release after the 5600s.