AMD RV770 refresh -> RV790

We have seen B3 crop up often in the discussion of GT200b, what is that supposed to imply?

Anyway, if the RV740 is indeed 256 bit, then I would expect it to have 10 * (55/40)^2 ~ 19 or 20 SIMD cores, just to avoid being pad limited. Though I was expecting them to call RV770's true successor as RV870. Wonder if it's a low/mid-low end chip.
 
I've got information from a insider about the RV740; the memory interface should be 256bit and the memory clock should be have 740 MHz (GDDR5) - 2960 MHz.
 
Is there even 740MHz GDDR5? And why do the current ES use 900MHz GDDR5? And how will they get the pins for the 256bit MC on a chip that small? Or is RV740 ~RV770 die-size?

So mayn questions :D
 
Salvadore said:
RV790 (so called from HWI) comes to Cebit 09 in 40nm as HD4900:
http://www.hardware-infos.com/news.php?news=2649&sprache=1

We've also got information, that the RV790 had tape out in 40nm for a while:
http://news.ati-forum.de/index.php/news/34-amdati-grafikkarten/237-rv790-zur-cebit-09-in-40nm
Click to expand...

Mr KonKort could you perhaps put in a percentage terms the likelihood that you think the RV790 will be a 40nm chip?

For me, at the moment, my personal percentages are:
55m GT 790 released somewhere between now and march - 80%
40nm still with a 256 bit bus - 5%
40nm and something else - 5%
cancelled or postponed - 10%

Am having my doubts about there being any extra units over the RV770, though have nothing conclusive one way or the other.
 
rpg.314 said:
We have seen B3 crop up often in the discussion of GT200b, what is that supposed to imply?
Click to expand...

hmm, seems no-one replied you here.

gt200 = 65nm gt200b is 55nm. the "b" moniker comes from the metal revision nvidia used for the 55nm part. gt200 ran as Ax, gt200b ran as Bx.

If you check the 9800GTX+ you'll see that it too has a "b" revision number to indicate it being 55nm all three those chips were developed simultaneously with the gt200b ultimately taking 3 spins (B1, B2, B3) to reach the consumer market (b2's were used in the workstation market.) If not nvidia would've launched gt200 at 55nm instead of the 65nm 260/280 that launched last year.
 
OK, I'm just being speculative here, but if RV740 is 16 RBEs/128-bit - i.e. 4 per 32-bit channel, then RV790, with the same ratio, would be a monster.

I dare say I'm coming to accept that there will be a refresh for RV770. Fingers-crossed it isn't just cost-cutting, R600->RV670 style...

Jawed
 
Jawed said:
I dare say I'm coming to accept that there will be a refresh for RV770. Fingers-crossed it isn't just cost-cutting, R600->RV670 style...

Jawed
Click to expand...

Cost cutting wouldn't make that much sence if they went for the performance-midrage segment with rv770, right? I mean, this would be.. cost-cuting-cutting. Ofcourse, I'm judging from a rather high, road-map-like POV. My point is that if they planed rv790 on the long term, it is likely they would have wanted it to be a better performer. If it's an on-the-fly (not literally) project, it can be anything.
 
entity279 said:
Cost cutting wouldn't make that much sence if they went for the performance-midrage segment with rv770, right? I mean, this would be.. cost-cuting-cutting. Ofcourse, I'm judging from a rather high, road-map-like POV. My point is that if they planed rv790 on the long term, it is likely they would have wanted it to be a better performer. If it's an on-the-fly (not literally) project, it can be anything.
Click to expand...
I'm thinking that if RV740 is for SKUs at <=$80, then RV790 as a pure cost-cutting GPU is for SKUs at >$80 <$150.

So HD4830 performance will be around $80 courtesy of RV740, HD4850 performance will be around $100 and HD4870 performance around $150, both courtesy of RV790.

Previously, I've suggested that by chopping off the CrossFireX Sideport, a significant chunk of die space could be saved, whilst staying at 55nm. In reality that could be 20-30mm2. Would that even be worth doing?

When considering 40nm as a cost-reducing process the bus width becomes a major constraint - 256-bit would require a die that's >200m2 roughly (assuming there's no CrossFireX Sideport - more if the sideport is present). That would inevitably lead in the direction of a "monster RV770" refresh - which is, arguably, no longer apt to be called a "cost-reduction" refresh.

If RV790 is a monster, then perhaps RV870 would be like the incremental performance update that R600 was when compared against R580. Sure there were a few areas where R600 was way faster, but in basic fixed-function units the performance growth was mostly pedestrian.

Jawed
 
Jawed said:
OK, I'm just being speculative here, but if RV740 is 16 RBEs/128-bit - i.e. 4 per 32-bit channel, then RV790, with the same ratio, would be a monster.

I dare say I'm coming to accept that there will be a refresh for RV770. Fingers-crossed it isn't just cost-cutting, R600->RV670 style...

Jawed
Click to expand...

This isn't exactly a bad thing if it also comes with the beefy power savings that the R600->Rv670 transition brought. Under load the 4870 compares well with the competition, but I'm still not at all happy with its idle numbers. Thank god I have Vista now and don't even feel the need for leaving my computer on for 24 hours as I did with XP (thank goodness for a working sleep mode). But I still spend the vast majority of my computer time at the desktop.

What would be interesting is if you could install two discrete cards, a low performance/low power one and a high performance/high power one and be able to switch between the two. Similar to how a lot of new laptops allow switching between GPU and integrated or how Nvidia allowed turning off power hungry video cards with certain of their IGPs.

Regards,
SB
 
Jawed said:
I'm thinking that if RV740 is for SKUs at <=$80, then RV790 as a pure cost-cutting GPU is for SKUs at >$80 <$150.

So HD4830 performance will be around $80 courtesy of RV740, HD4850 performance will be around $100 and HD4870 performance around $150, both courtesy of RV790.
Click to expand...

I'd be more inclined to think that Rv770 -> Rv790 will involve a performance bump. Unlike Rv600 -> Rv670, ATI isn't desperately in need of cost cutting. They already have the most cost efficient chip to build (in theory) with regards to the competition.

With a move to 40 nm, they'd have an opportunity to have 4870 perform across the board similar to or greater than GTX 280 (which may be needed if Nvidia also comes out with a more performant refresh of GTX 280) rather than only being faster in certain corner cases.

What I'm "hoping for" is that Rv790 will offer a modest performance boost (10-30%, granted 30% is a bit wildly optimistic) along with a bit of power savings. Too much to ask for? Well, I'll keep my fingers crossed.

Regards,
SB
 
Silent_Buddha said:
What I'm "hoping for" is that Rv790 will offer a modest performance boost (10-30%, granted 30% is a bit wildly optimistic) along with a bit of power savings. Too much to ask for? Well, I'll keep my fingers crossed.
Click to expand...
That, in my view, would be a 55nm refresh, say with CrossFireX Sideport deleted and higher core/memory clocks.

Anyway, right now, I'm leaning towards a 40nm monster, if there really is a refresh for RV770. 2x performance, across the board... If RV730 can get approaching 2x performance in its refresh, why not RV770?

Jawed
 
If at 55nm a ~250mm2 chip is pad limited, I'd expect it to be pad limited at 40 nm as well. At linear scaling that alone means a 1.7-2x increase in peak perf. My only doubt is that if the refresh is 2x faster, then what will they do with RV870, considering that it is their goal to increase perf by ~2x every gen and their die size is limited by their design choices.

OTOH I am fairly confident that this refresh chip too will show huge(1.7-2x) improvements if made at 40 nm. and not a small increase like 20% as has been rumoured in some places.
 
You must log in or register to reply here.
Back
Top