AMD: R8xx Speculation

[Silent_Buddha]

Would you be disappointed if the chip grew to.. say G92 size, while G300 stayed GT200 size?

I'll have a better idea once the card has been tested, prices set, and volume determined. Until then, there's just too much that is unknown.

I will say however if the 5870 (or whatever it ends up being called) ends up using a 2 chip MCM that does nothing better than AFR, I'm going to be HUGELY disappointed.

Regards,
SB

 

[Ailuros]

I think that it was back when the upcoming DX10.1 was confused with DX11 by some

I could come up with the same silly excuse for NVIDIA's codename macedonian salad; there was a time where some thought that just a tesselator would be enough for D3D11 and they even insisted in public that there won't be any changes. Unfortunately some way down the road some annoying shader shader stages like the hull and domain shader were added to the mix.

 

[MfA]

I will say however if the 5870 (or whatever it ends up being called) ends up using a 2 chip MCM that does nothing better than AFR, I'm going to be HUGELY disappointed.

I wouldn't ... for one I only buy cheap single chip cards anyway, for another I think it would be better for competition.

 

[no-X]

I will say however if the 5870 (or whatever it ends up being called) ends up using a 2 chip MCM that does nothing better than AFR, I'm going to be HUGELY disappointed.

Enything is possible, but it wouldn't correspond to current rumours. Huddy promised fast hardware. According to some rumours the >200mm2/1200SPs/256bit part was canceled (or postponed). According to rumours the 180mm2/800SPs/128bit(?) part is comparable to HD4870 in terms of performance.

That would mean, that the 180mm2 GPU is the fastest single chip part and the fastest (AFR based) board would be about as fast as HD4870X2. That isn't possible. Even HD2900XT was faster than X1950XTX

Either the >200mm2 wasn't canceled or the dual-chip board won't be based on AFR rendering.

 

[neliz]

can you please let me know where you heard the rumors about that because it would not make sense. if you take the RV740's size of 137mm2 and falling slightly below RV770 performance they would get 40mm2 extra on the die and less than 10% more performance. ... does not compute. I would put a 180mm part at least one step below high-end.

(and yes, in some cases a 1950xtx was faster than a 2900XT )

 

[Lukfi]

RV740 performance is comparable to a Radeon HD 4850, but a HD 4870 is 20 % faster.
Anyway, neither do I believe that a 180 mm2 chip is the best ATI has to offer.

 

[trinibwoy]

if you take the RV740's size of 137mm2 and falling slightly below RV770 performance they would get 40mm2 extra on the die and less than 10% more performance. ... does not compute.

Not making a statement about the validity of any rumours but does your calculation take into consideration the extra bits needed for DX11 compliance and better general compute support? I know most people expect minimal changes but I don't think we can assume perf/mm^2 will remain constant after those changes. In any case I thought the 180mm^2 part was rumoured to be 1200 ALUs?

 

[neliz]

RV740 performance is comparable to a Radeon HD 4850, but a HD 4870 is 20 % faster.
Anyway, neither do I believe that a 180 mm2 chip is the best ATI has to offer.

True, since it's missing 20% in the amount of shader processors. a 800sp part would still be below 180mm, maybe with dx11 added it would get close and then still be at most a mid-range part.

 

[w0mbat]

http://www.forum-3dcenter.org/vbulletin/showpost.php?p=7455220&postcount=1884

Cedar = 120mm², Juniper = 181mm², Redwood = 300mm², Cypress = 2*Redwood

 

[wishiknew]

What would hemlock be based on that?

 

[w0mbat]

4*Redwood? Or maybe the IGP part? Dunno.

 

[Jaaanosik]

What would hemlock be based on that?

Hemlock should be an entry, below Cedar I guess.

 

[neliz]

Hemlock should be an entry, below Cedar I guess.

nope.

 

[Jaaanosik]

never mind

 

[no-X]

http://forum.beyond3d.com/showpost.php?p=1316099&postcount=1256

 

[Jaaanosik]

http://forum.beyond3d.com/showpost.php?p=1316099&postcount=1256

Yeap, I forgot about that one

 

[Jaaanosik]

4*Redwood? Or maybe the IGP part? Dunno.

4*300? Is it possible on one PCB?

 

[digitalwanderer]

My prediction? The R8xx will make the sky break open and the heavens themselves weep.

Yup, you heard it here first.

 

[no-X]

Jaaanosik: Anyway, I don't believe that neliz would negate your statement by virtue of Charlies opinion

 

[Jaaanosik]

Yeap, I forgot about that one

[...] Hemlock.", is not a low end part, is the highest end part. Hemlock is the code for the dual Cypress board, likely to be called 5870X2. The low end parts are Cedar and Redwood, mid-range is Juniper, and the high end single chip is Cypress. Hemlock is above that, a dual Cypress/X2 board, not the low end.
[...]
I also confirmed that there is no disinformation campaign floating around, no leaked slides with the wrong info, just the correct information. More to the point, Sylvie was the only person to officially get the names so far.

There is some confusion here. Charlie called Redwood as a low end part. The new info says it's 300mm2. It's a mess