AMD: R7xx Speculation

Pretty weak clocks. I've written posts talking about AMD's seemingly increasing clock advantage, but that was assuming 850-1000 on 4870.

Hey, Fud also got the launch day right, on only his 8 trillionth different try!

I think, if the rumors of computex that they've got shedloads of 4850 just sitting around, and choose to delay to launch together, that once again AT is being really stupid. Now Nvidia will launch first which will be a psychological boost. And ATI missed out on 4 weeks of possibly hefty sales. You have to realize the average say, [H] forumer, the guy with the 2k SLI rig, really knows nothing about what these companies are doing, and they may not even know or be convinced ATI has anything coming down the pike (or vice versa). All they know it's whats out now and an uncertain future.

 

Launch was never delayed.

 

R5xx doesn't have double-rate Z with MSAA off (except for RV530/535) so of course the performance hit of turning on MSAA in R6xx is higher.

Building R600 with all that bandwidth and only 2x Z per clock was a crime. If only they'd not spent transistors on a 512-bit bus, just used a 256-bit bus and spent that die space on extra Z capability. If only...

Jawed

 

SPs or TMUs?

 

Interesting, but I'm going off certain computex rumors that a ton of 4850 were just sitting there (makes some sense being GDDR3) and ATi refused to launch them, preferring to delay to launch together with 4870, and ATI partners were angry about this.

 

Seconded and signed

 

Why "or"? Could be both, could it?

 

These clocks are, ahem, extremely surprising. I'm less surprised by HD4850's clock, but for both to have slower clocks than the GPUs they're replacing really is bucking a big big trend that ATI GPUs have had for a long time.

Jawed

 

Well its either one of those ..

 

I cant imagine whats different about the specs. 32 TMU and 480SP just fit so perfectly. 16/800 would lead to no performance increase.

I'm guessing 24 TMU's is a possibility, that's about all I can think of.

I cant even find what vr-zone had the specs as. I'm assuming the 32/480 common set.

It looks like vr-zone has been one of the places throwing around 800 sp's actually. Somehow I dont think wombat was reffering to that though.

 

Diidn't the Folding at Home guy pretty much confirm the SP count at 480?

 

The 800sp originated from chiphell, correct me if I'm wrong.

And then there's always Arun, trini throwing us off with their math.

 

Or did he maybe just repete the rumors he heard?

 

Well you guys can always start from scratch and try to guess how many clusters RV770 has, which would be a good starting point.

 

Sources have been claiming that a score boost will come with the final Catalyst. Why would AMD have so much trouble optimizing their drivers for the 4850 if it's only a single core resource bumped GPU?

 

Well, we could just boil this down to which is more "likely":

• 750 * 480 = 720GFLOPs
• 750 * 800 = 1.2TFLOPs

Jawed

 

Reading from rumor that R700 multiple GPU architecture will not be CrossFire on a card; this architecture is being built / designed from the ground up.

Let assume if this is correct!

Will Windows XP / Vista detect R700 as single GPU solution? - If so, you could probably have total 8-GPU's being detected by OS as 4 Quad-crossfire setup.

 

For that we have to disregard the different clock for shaders theory, isnt it?

 

How about

• 750 * 640 = 960 GFLOPS

 

If you want to say something say it.
Dont keep posting in the forum Spam like "may be, may be not" that don´t give nothing new to 108 page tread.

Understand this message please.