AMD: R8xx Speculation

[AnarchX]

59 FPS @ Vantage perlin noise indicates 800SPs@700MHz.

 

[leoneazzurro]

59 FPS @ Vantage perlin noise indicates 800SPs@700MHz.

The tested card is a 5750, not 5770 (it is written in the 3Dmark graph).

 

[Tim]

How so?

AMD has in the past had RV570 which was 3/4 of the R580. If Juniper is 14SIMDs then it is 2.8/4 of Cypress.

edit: Juniper having 14SIMDs is a bigger surprise for me than the 20SIMDs in Cypress.

The problem is that Juniper seems to have 70% of the shader and texture units of Cypress but only 55% of the die size.

 

[leoneazzurro]

The problem is that Juniper seems to have 70% of the shader and texture units of Cypress but only 55% of the die size.

If it would have also 3 RBE (24 ROPs) as rumored it will have about 70% of color fill power, too, so it will be basically 70% of the whole Cypress.
Unlikely? Who knows.

 

[Arty]

The problem is that Juniper seems to have 70% of the shader and texture units of Cypress but only 55% of the die size.

I'll give you that but I'm still not a believer in the possibility of Cypress having more than 20SIMDs and or large memory bus width.

Anyone care to guesstimate Juniper's die size from the above photos? Thanks The_Mask for the photos.

 

[fellix]

The chip looks quite stretched -- the same die as on this wafer?

 

[Arty]

The chip looks quite stretched -- the same die as on this wafer?

Seems like a match. CJ was right.

 

[leoneazzurro]

The chip looks quite stretched -- the same die as on this wafer?

Seems the same shape, dimension of the wafer chip was 181 mm^2 IIRC

 

[Kaotik]

The chip looks quite stretched -- the same die as on this wafer?

I did quick estimation using pixels and PCIE's 11.65mm long "fixed part" as reference, it's definately the same, in other words like reported back in the day, they showed off Juniper wafers.

edit:
Could the rumors off HD5890 coming quite soon mean that HD5870 doesn't actually have everything enabled, and HD5890 would have? It would also explain the lack of dieshots

 

[Alexko]

Seems like a match. CJ was right.

About what?

 

[-The_Mask-]

Thanks The_Mask for the photos.

Thank neliz.

 

[Arty]

About what?

When AMD showed off the wafers, everybody assumed that the dies were RV870 (Cypress) when KonKorT & co were in full force pushing the 1200SP rumors. CJ said that the wafers shown were Juniper and not Cypress.

Thank neliz.

Thank you neliz.

 

[neliz]

About what?

everything!

and no no, I'm not to blame for this leak, though I'll hit my wife a couple of times because she missed it. Can't trust a taiwanese...

 

[Tim]

I'll give you that but I'm still not a believer in the possibility of Cypress having more than 20SIMDs and or large memory bus width.

I am not convinced either way.

The die size(compared to the rumoured 180mm2 for Juniper with 1120SPs), the high pin-count, the fact that we have not seen any die-shots yet is pointing towards that there are something more to Cypress than we seen.

But there are a lot plausible explanations for the above.

Anyone care to guesstimate Juniper's die size from the above photos? Thanks The_Mask for the photos.

I would like that too, if Juniper is larger than 180mm2, I could stop posting (silly) theories about playing games.

 

[fellix]

Quick -- somebody to make a check if the PCB is sawed!

 

[3dilettante]

The die size(compared to the rumoured 180mm2 for Juniper with 1120SPs), the high pin-count, the fact that we have not seen any die-shots yet is pointing towards that there are something more to Cypress than we seen.

I've never seen a decent shot of R600 or RV670, either (RV740?). I don't think we'll be expecting much from them in the future.
The RV770 is the one I can think of.
CPUs aren't this secretive for released products, and I think there may be more examples of Nvidia chips than AMD GPUs having good shots.

 

[Psycho]

Resolution is a little low, while projection is good, but I read 37*48 pixels, meaning 12*15.5mm = 186 mm2, including packaging, but still pretty close to the old wafer based estimate.

 

[compres]

When AMD showed off the wafers, everybody assumed that the dies were RV870 (Cypress) when KonKorT & co were in full force pushing the 1200SP rumors. CJ said that the wafers shown were Juniper and not Cypress.


Thank you neliz.

 

[Tim]

I've never seen a decent shot of R600 or RV670, either (RV740?). I don't think we'll be expecting much from them in the future.
The RV770 is the one I can think of.
CPUs aren't this secretive for released products, and I think there may be more examples of Nvidia chips than AMD GPUs having good shots.

OK, I admit it is a rather weak argument. I could remember the RV770 shots, but it seems that it was quite unusual when they supplied those.

 

[Tchock]

EDIT: Damn, I'm late.

5750 castrated in terms of SIMDs? If 5770 is 850Mhz and still is in compact form, hello 9600GT redux.