What do you expect for R650

[nicolasb]

Adding more ALUs is rather pointless at this time, although it's the easiest path to take on. It is evident, that the bottleneck is not there, as the numerous vertex/geometry benchies are showing considerable headway to G80.

While R650 could certainly do with some more texturing ability, it is beginning to sound as though the reason why the AA performance is so bad on R600 is because the AA is done in shaders; and it's also beginning to sound as though this was a deliberate design decision by ATI, not a response to a hardware bug (or at least not entirely). So additional shader-power might perhaps significantly boost AA performance as well as conventionally-measured "shader" performance, which would be cool.

But texturing probably ought to be a higher priority.

 

[pelly]

3dfx had some interesting ideas with the VSA-100 (Voodoo Scalable Architecture)...Have 1 VSA-100 chip for an entry-level card and add more chips for more performance...

If you had great margins on that chip.....you'd have obscene margins for a flagship-level card....

Perhaps this is a road ATI (er..AMD)...and NVIDIA are looking at....

 

[nicolasb]

R700 is very likely to be a multichip (or possibly multicore) solution, but I don't think R650 will be.

 

[Kaotik]

Fudo reckons R700 will be 55nm:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1028&Itemid=1

Consensus still seems to be 65nm for R650, though.

I almost hope R700 won't be 55nm, because, if it is, that probably means ATI will simultaneously be changing chip design and process, which is always a recipe for trouble. Unless, of course, something else between now and R700 actually is 55nm. That'd be nice.

I'm quite positive, without re-watching the Q&A session after the [something]690 chipset release, that they confirmed R650 to be 65nm

 

[Star_Hunter]

Not to mention if R650 was multi-chip it would likely end up late since it would be a total redesign and take too long since they would be converting a big chip into several small ones along with adding and changing stuff compared to R600. It much better suited to do that with R700 which looks like they went multi-chip from the start. Although you likely are right about lower end cards in the R7xx series just using all the same chips and the only difference being the number of them between high and low ends(ATI already does this by simply taking out quads).

 

[trinibwoy]

I'm quite positive, without re-watching the Q&A session after the [something]690 chipset release, that they confirmed R650 to be 65nm

Yeah I thought Orton pretty much mentioned 65nm and R650 in the same sentence. So I'm not sure what Rys is being so mysterious about. Unless Orton himself doesn't know wtf he's doing

 

[Tim Murray]

Yeah I thought Orton pretty much mentioned 65nm and R650 in the same sentence. So I'm not sure what Rys is being so mysterious about. Unless Orton himself doesn't know wtf he's doing

Didn't Orton say 96 shader units for R600?

 

[leoneazzurro]

Didn't Orton say 96 shader units for R600?

I reckon he said "up to 96" in future releases

 

[fellix]

96 -- yay!
They can finally get rid of the ROPs, emulating them completely.

 

[ANova]

At this point I would be happy if they just increased the TMUs and ROPs and either fixed the AA or added more shader units, the former being a better decision imo. Those changes alone would likely increase performance fairly well along with the lower power consumption you get on 65nm, possibly a slight clock bump and faster memory to go along with it.

 

[elroy]

Or maybe people ought to spend more time considering what the second 1/2 of his formulation might be suggesting.

Well, Rys statement does suggest two things:

1) AMD are unlikely to come out with another large chip (the first part)
2) If AMD come out with another large chip, it won't be 65nm

I just chose to enquire about the first, as the second had pretty much already been covered. Though, considering R600 is currently getting beaten like a red-headed stepchild, I wouldn't be surprised to see them release a refresh (i.e. another large chip), and soon.

Maybe from now on AMD should design the initial chip and the refresh, then produce just the refresh??

 

[seahawk]

No, it could mean 2xRV670 > R650

 

[Silent_Buddha]

Well, I suppose Jawed could always get his dream of a 4xRV630 chip or card. Although at 45 watts per RV630...that'd be around 180 watts so not much of a power savings.

Regards,
SB

 

[Rangers]

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1042&Itemid=1

Fudzilla?

Sounds great if true.

 

[nutball]

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1042&Itemid=1

Fudzilla?

Sounds great if true.

Ummm... if you were just going to randomly guess what might change between R600 and R650 based on zero information other than past history, what would you come up with? Let's have a try shall we...

- maybe a smaller process (with the presumption that this would lead to smaller, faster, cooler)
- some architectural tweaks, especially fix any bugs with R600

 

[Arnold Beckenbauer]

Didn't Orton say 96 shader units for R600?

When did he say this? Before or after that trouble with AMD?

 

[Rangers]

Ummm... if you were just going to randomly guess what might change between R600 and R650 based on zero information other than past history, what would you come up with? Let's have a try shall we...

- maybe a smaller process (with the presumption that this would lead to smaller, faster, cooler)
- some architectural tweaks, especially fix any bugs with R600

If I was to just randomly guess, I would probably just figure on a higher clock speed and no architectural tweaks. Since you asked.

Yeah yeah, I know Fudzilla. However, he's the only one I see writing anything at all about R650, real or not.

The important thing from my POV is that it seems ATI recognizes the texture deficit problem.

 

[fellix]

I think, they had recognize it much before this point, but the 80nm fab didn't allow it for implementation at the time... and, IMHO, there was no way they would trade off ALU power for fill-rate HW, giving their pathological GPGPU obsession.
It's a clear, yet another R520>R580 case here.

 

[Kaotik]

When did he say this? Before or after that trouble with AMD?

It was a long long time ago in the galaxy far far away, same time when he said that R600 would be the fastest DX9 chip they have ever built (in addition to DX10 support ofc)

 

[Shtal]

All together this will be what R600 was suppose to be, a Geforce 8800 GTX / Ultra killer, but the problem is that Nvidia will also have something new by the time ATI finishes R650.

All depends by how much improvement Nvidia could make based on 65nm tech,

A.) If their will be no delays from Nvidia
B.) Depends if it is just a clock bump from G80Ultra
C.) Also well possibly ATI could response with extreme high clock R650+ or R680 GPU
D.) Depends by how much R650 will be killer for G80Ultra and how it will stack against G81 65nm