AMD: R7xx Speculation

[Jawed]

480 ALU lanes confirmed

http://foldingforum.org/viewtopic.php?f=42&t=2454#p22137

Jawed

 

[w0mbat]

RV770 ES clock: 625MHz
Board no.: 102-B50102-00
Revision: A12

 

[Lukfi]

You have one of those?

 

[w0mbat]

Yeah, in my wet dreams

BTW: It uses 1GHz GDDR3 from Qimonda

HYB18H512321BF-10 512Mb, 16Mx32, 1.0GHz, 2.0V PG-TFBGA-136

 

[AnarchX]

Seems that ATi finally had learned, that clock-speed is not the only way to get performance.

So 88GT+25% ~ RV770@625MHz+1GHz GDDR3?
If Chiphells RV770@air 980MHz OC are true, it could be get interesting.

 

[Jawed]

RV670 dispose of 32 adressing units. Would it be impossible to share them to feed 32 bilinear texture filters?

I imagine it is possible to make all 32 capable of performing point-sampled or bilinear addressing calculations. But I have a feeling that the cache organisation wouldn't be able to use the extra throughput without doubling in width itself. Which it would need to do, anyway, for doubled filtering throughput.

So, it could be seen as a slightly cheaper way to get 32 filtered results per clock, instead of having 64 addressing units (32 point-sample, 32 bilinear).

It may come down to how tightly coupled the two types of addressing and caches are (for point-sampling and bilinear) - e.g. point-sampling may place the address units very close to the L1 point-sample cache.

Overall I'd say I was sceptical though.

Jawed

 

[Vincent]

Seems that ATi finally had learned, that clock-speed is not the only way to get performance.

So 88GT+25% ~ RV770@625MHz+1GHz GDDR3?
If Chiphells RV770@air 980MHz OC are true, it could be get interesting.

480 +320 ALUs

 

[AnarchX]

480 +320 ALUs

So you believe in the 320 SPs reserved for physics operations?

 

[Kaotik]

Yeah, in my wet dreams

BTW: It uses 1GHz GDDR3 from Qimonda

Are those how confirmed specs? I was under the impression they got A11 to production just like with RV670?

 

[w0mbat]

Are those how confirmed specs? I was under the impression they got A11 to production just like with RV670?

These infos are from AMD.

 

[Sound_Card]

What is RV770 ES?

Is that the top model?

 

[Lukfi]

Engineering sample.

 

[Sound_Card]

Engineering sample.

errr duh, . *hits head.

 

[Wesker]

RV770 ES clock: 625MHz
Board no.: 102-B50102-00
Revision: A12

R580's (XT) core clock speed, IIRC.

Is ATI looking for some more lucky charms? Wasn't the 777MHz core clock or the 666M transistor count enough in RV670?

But a chip which can give G92 level performance at 625MHz is very interesting. I wonder how far performance can go if ATI releases an XT or XTX SKU at over 800MHz.

 

[Jawed]

ES isn't necessarily at full clock though, is it?

Jawed

 

[w0mbat]

ES isn't necessarily at full clock though, is it?

Jawed

Right.

 

[Razor1]

ES isn't necessarily at full clock though, is it?

Jawed

no its not full clock. wombat beat me to it.

 

[LordEC911]

http://foldingforum.org/viewtopic.php?f=42&t=2454#p22137

Jawed

Hmmm... This caught my eye-
"Hartware.de says we can expect the 4870 to offer maximum floating point processing power of just over one teraFLOPS—roughly twice that of the existing 3870."

Which means at least a minimum of 1075mhz for the 480shaders.
So any more info on a separate shader domain? I really don't want one.

 

[entity279]

Hmmm... This caught my eye-
"Hartware.de says we can expect the 4870 to offer maximum floating point processing power of just over one teraFLOPS—roughly twice that of the existing 3870."

Which means at least a minimum of 1075mhz for the 480shaders.
So any more info on a separate shader domain? I really don't want one.

Or maybe the (presumed) extra shaders for the (rumored) ppu functionality are taken into account.

 

[w0mbat]

No, no infos on a shader domaine here.