AMD RV770 refresh -> RV790

But if we are looking on the "up to 950MHz" which AMD promised to the partners, the additional transistors might be used as mass transitors.
Click to expand...
Stretching the pipeline with more stages?
 
It was pretty idiotic of DT not to clean-up the chip properly if taking the heatsink off. Is measuring the "square" impression in the gunk on the heatsink productive?

For what it's worth my estimate is 284mm2, hohoho.

Jawed
 
fellix said:
For your measuring pleasure:

rv770790.png
Click to expand...

Shouldn't we line up the holes for the heatsink instead of the shim?
 
The 4890 picture is too much bended by the lens distortion, so I had to align & scale the GPU area to the substrate dimensions (e.g. the outer shim boundary), which is presumably similar to the RV770 ASIC.
 
From a chinese website:
AMD originally in February < 2000 wafers ordered.
This month 4000-5000 wafers ordered, in april hoping for 8000-9000 per month and in june 10000 per month.

Think above numbers are total order size, not sure what proportion would be RV790 and what would be RV740 and what would be more established products.

Taking a wild guess that around ~2000 wafers dedicated to RV790 launch. At ~200 valid dies per wafer implies a launch volume of around 400k.

[Edit - 1 minutes thought makes me think above is wildly optimistic, there appears to be no lower priced 4850 like variant, surely at most 1000 wafers dedicated to RV790?]

(Feel free to correct any inaccuracies in above numbers)
 
Last edited by a moderator:
Vincent said:
I am confused. How could DX10.1 hardware support D3D11 feature ?
Click to expand...

From that article appears that the talk will be heavily focused on DX11 tesselation while just touching on other features.

Also it appears that Rv770 has a different tesselation implementation than R600 or Rv670. It's possible that the version in Rv770 is much closer to what will be implemented in Dx11 that they were able to make Rv790 with Dx11 compliant tesselation.

Along with that, some previous rumors that the shaders were "improved" in some way for Rv790. There's the possibility this improvement may have been with regards to Dx11 hull shader and domain shader.

All that is pure speculation however...

Regards,
SB
 
Jawed said:
Would a different version of TSMC's 55nm cause a difference in die area?

Jawed
Click to expand...
Honestly I don't know. But then - do we have really comparable die shots at the moment? I mean, we're talking millimeters of even fractions thereof here. A slight skew in fotographs, heck even some thermal grease residue could lead to wrong conclusions.

Plus: Wouldn't it be a factor for AMD if all their RV770 and up (so basically EOL RV770 and only produce RV790 from now on) could come from the same wafer - being sold as 4890, 4870 or 4850 as binning allows? That sure would be easier, if they kept the die at least pin compatible to earliert products but wouldn't prevent them from having fixed some internals that previously limited clock targets.
 
Last edited by a moderator:
CarstenS said:
Honestly I don't know. But then - do we have really comparable die shots at the moment? I mean, we're talking millimeters of even fractions thereof here. A slight skew in fotographs, heck even some thermal grease residue could lead to wrong conclusions.
Click to expand...
The high resolution Daily Tech picture + the high resolution Tech Power Up pic are not too blurry - but yes, there's a difference in dimensions of ~4 pixels on the DT pic. Just can't believe the incompetence of DT.

Plus: Wouldn't it be a factor for AMD if all their RV770 and up (so basically EOL RV770 and only produce RV790 from now on) could come from the same wafer - being sold as 4890, 4870 or 4850 as binning allows? That sure would be easier, if they kept the die at least pin compatible to earliert products but wouldn't prevent them from having fixed some internals that previously limited clock targets.
Click to expand...
This was a basis of my 40nm argument: since RV740 provides HD4850 performance at ~$100+, AMD needs two SKUs at ~$150 and $250+, otherwise they've lost all the momentum they gained in the $200+ bracket. So it seems natural to expect two SKUs at least from RV790. It seems 850MHz and "faster" are what are planned.

Jawed
 
rjc said:
[Edit - 1 minutes thought makes me think above is wildly optimistic, there appears to be no lower priced 4850 like variant, surely at most 1000 wafers dedicated to RV790?]
Click to expand...

I think the inq could be right that a rv790 based 4870 (with quite a bit less power consumption and thus cheaper) will be the new salvage part. So it would be interesting to know if rv770 is EOL now (rv740 can take the 4850 spot).

CarstenS said:
But then - do we have really comparable die shots at the moment? I mean, we're talking millimeters of even fractions thereof here. A slight skew in fotographs, heck even some thermal grease residue could lead to wrong conclusions.
Click to expand...

I'm pretty sure about my 323 vs 305 mm2 numbers, the skew is taken care of and they should be 1-2% accurate for what it's measuring. Sure there is some grease, but you can still see a very straight and square form. I think it's safe to assume it's around 20mm2 larger.

neliz said:
well, their GPU-Z 0.32 is working while the rest of the world doesn't get 0.32 to recognize the 790. what do you think yourself?
Click to expand...

They somehow got the bandwith etc right now, but the "RV790" string is because they use like 2 months old drivers, it's called "4800 series" in the more recent.
 
Last edited by a moderator:
The GPU-Z according to Sampsa Kurri might be real, since Wizzard often builds the betas on the old versions without changing the version number before final new version based on his experience.
That, however, doesn't explain the "RV790" name when drivers call the card "HD4800 -series"
 
You must log in or register to reply here.
Back
Top