AMD: R7xx Speculation

[Tchock]

This would also bode well for the "100Mhz+ clock, virtually double bandwidth" leading up to 30% performance. But in the case of 800 it should be more.


And the Vantage scores seem on the low side if so.

 

[trinibwoy]

4870 ~ X3400 (or higher see above)

so at least in Vantage the 4870 can approach a 9800GX2 (within 5%)

A GX2 scores ~X4500.

 

[CarstenS]

Yes I have to say my first impression is "that's big".

Seconded. A positively "big" that is at least from me. So, after all, maybe we're actually seeing 32 TMUs at work in that chip - a number I would not have deemed possible in ~250mm².

 

[trinibwoy]

Some one better versed with PS can probably change the angle (tilt) for RV770 and scale it better ..

Do you have one for RV670? RV770 certainly doesn't look small. But how can it really be that big on 55nm and not be faster than we're hearing?

I guess one way is 16TMU's and 800SP's

 

[Tim]

A GX2 scores ~X4500.

And 30% faster than the 4450 scores is x4400.

 

[Arun]

Side note:

625 (official clock for 4850) * 800 * 2

There's a magical number. I thought it would run for the highend, but looks like BOTH versions lived up to the PR.

Oh, you mean like this? http://forum.beyond3d.com/showpost.php?p=1166843&postcount=2058

 

[trinibwoy]

And 30% faster than the 4450 scores is x4400.

? 30% faster than what score?

 

[Arty]

Something is not adding up:

• Relatively large die size
• Performance estimates ~ 9800GTX
• 480 SP count

Could the drivers be intentionally borked so that performance is lower?

 

[trinibwoy]

Oh, you mean like this? http://forum.beyond3d.com/showpost.php?p=1166843&postcount=2058

No he means like this - http://forum.beyond3d.com/showpost.php?p=1163672&postcount=1781

300 Posts earlier

 

[Arty]

Do you have one for RV670? RV770 certainly doesn't look small. But how can it really be that big on 55nm and not be faster than we're hearing?

I guess one way is 16TMU's and 800SP's

You can find it here, but again different angle/tilt.

 

[Tchock]

Oh, you mean like this? http://forum.beyond3d.com/showpost.php?p=1166843&postcount=2058

OSNAP!

But I did suggest that such a scale was doable on 250mm^2 for a few times.

If my sloppy Photoshop skills are suggestive of anything (using the measurement tool on slightly off scales, might deviate +-10%) the die is about 30% larger.

I got a scale value of about 133%, of which 192*133% = 255.36 mm^2


Information from Chiphell previously indicated 800SPs, a 256mm^2 die size, and they had the PCBs in January. To trust, or not to?

 

[Jawed]

Not really that big after all:

In this picture "light grey" is RV670. I've superimposed RV670 on top of RV770 (which appears "black"). I created this picture by scaling a board shot of HD3850 so that the mounting holes for the heatsinks coincide.​

It looks like ~30% bigger, as rumoured.​

Jawed​

 

[CarstenS]

I got a scale value of about 133%, of which 192*133% = 255.36 mm^2

Damn, looks definitely larger than that. But then - eyes can be deceived quite easily.

 

[ShaidarHaran]

LOL, I can't believe we know more about GT200 than RV770 at this point... The two should basically be known quantities by now.

 

[Wirmish]

After some photoshopping...

 

[Tchock]

After some photoshopping...

My bet is that R580 had a different set of mounting holes.

770 and 670 both have the X underlay and with similar coolers which probably indicates the same set.

A GX2 scores ~X4500.

The review seemed like they used performance settings on the test, and just upped AA/AF.

Not comparable to the real "Extreme" which has better textures, shaders, shadows and etc.

 

[CarstenS]

My bet is that R580 had a different set of mounting holes.

That was my initial assumption also, since R580 is somewhere around 360mm².

But I just checked - at least R580 and RV670 are at ~535mm hole to hole (center).

 

[Mariner]

Could the drivers be intentionally borked so that performance is lower?

I'd tend to doubt it. Still, it's quite possible that the drivers could be unintentionally borked. :smile:

 

[trinibwoy]

The review seemed like they used performance settings on the test, and just upped AA/AF.

Not comparable to the real "Extreme" which has better textures, shaders, shadows and etc.

I think that's something I would have noticed. They used the extreme setting. Says so right above the first graph.

 

[roadie]

That was my initial assumption also, since R580 is somewhere around 360mm².

But I just checked - at least R580 and RV670 are at ~535mm hole to hole (center).

Correct, I have mounted a x1950xtx cooler on a x3870. The only mod needed was to saw away some of the plastic at the PCI slot cover end.