The NEXT LAST R600 Rumours & Speculation Thread

I wonder if Fuad took a look at this graph from NVIDIA:

http://enthusiast.hardocp.com/image.html?image=MTE3ODA2MTU5OGRmUGVmSGdkUU1fMV8xX2wuanBn

No doubt, the one huge drawback for the 2900 XT crossfire system is that power requirements will be absolutely huge, even though performance should be very good. An 8800 GTS SLI system on the other hand would have much more modest power requirements, but the Vista SLI drivers need major work.

On a side note, glad to see that this graph did not start at 147w, and then scale to 300w from there

 

If 240W is true, then the cooler design must be REALLY good in order to move all that heat and still ensure quiet operations.

It would also leave open the possibility that R600 is capable of much higher frequency and the power consumption is the major determining factor in default clocks.

 

It's simple, it can NOT be true due the fact it works with 2x6pin plugs

 

And then he writes...

So why the heck even include the numbers in the graph, if they know them not to be accurate.....

[edit]

My bad. I thought they were [H]'s slides.

 

Accepting the "one screen shot does not a test make" argument completely, and for what it is worth (not a lot probably)

at the same point with the same settings, except at 1680x1050 res, using

stock 8800 GTX with 158.18
athlon 4800+ x2 @2.55
2gb ddr400

43fps...

 

Case closed.

 

That part is funny too where the official slide show the ultra use less power than the gtx, reviews say different things Link Link Link Link

 

Because they are BS, me thinks. More FUD. [H]ard to believe.

 

They were... fudo just swallowed them, poo'd on his keyboard and then wrote that article about it....

 

I keep forgetting, what is PCI-E slot specked at, 75W? If so, then yeah, maximum power consuption could not exceed 225W.

 

75W slot and 75W per 6pin plug, that's correct (though I remember reading that the slot could actually provide 76W, but it might have been a typo and even if it wasn't, it's irrelevant)

 

There is a reson that they have clocked the card at 750, and the rumors said it will run just fine with a 2x6 pin, that alone says it would use under 225w if i am not way of.

 

Heh, yeah. So unless the figures are for an overclocked R600, or ATI has an on-chip power generator, it simply does not add up. Taken along with the fact that no website has been able to varify lower power consumtion by Ultra (did they mean "per clock" or something?) that chart is about as worthless as they come.

 

I get a 404 error when trying to load the G80 arc thread.

I would very much like from the guru's here if they can give me a breakdown of G80's texturing capabilties. Because I'm still in the dark...

So from what I can gather, G80 has 32 TAU's, 64 TFU's, and ?128? Texture samplers, and dedicated L1(how much L1?). Texture units clocked at 575mhz.

R600 has 32 TAU's, 16 TFU's, 80 texture samplers, dedicated L1(how much?), and shared L2(how much?). Texture units clocked at 750mhz.

Right? No?

 

The explanation is probably simpler. A histogram shows the R600 shot is clipped at both ends of the black/white range while the G80 is not. A quick play with lowered brightness ("-50" in ArcSoft PhotoStudio) and contrast ("-75") shows the G80 shot approaching the R600 in IQ and corresponding clipped histogram range (though G80 still shows more of a reddish hue). I'm guessing this is a simple case of a borked gamma setting.

But if a smaller HDR format (FP10, or that 9-9-9-5 format I think was in an R600 slide?) could produce the same results, then I guess we split the difference (MDR?).

Sound_Card, I think our front page still exists. Now, to give you a chance to razz me. From what I understand of what Rys & co. have gathered, G80's texture samplers consist of 32 TAUs and 64 INT8/clk (FP16 half-speed) TFUs, divided into eight groups corresponding to 8 shader processor cores, with 8kiB L1 cache per group (so 64kiB L1, all told). I'm not clear on the terminology, though, as B3D refers to G80's TAUs and TFUs collectively as texture samplers, while R600 appears to have distinct samplers.

So, your basic #s seem right, except that as far as I can figure G80 doesn't seem to have discrete samplers. Dunno how this figures into DX10 constants. Speaking out of my opposite end, and without figuring how L1/2 caches and memory bus widths affect efficiency, it seems like (surprise!) G80's geared for more texture manipulation and R600's equipped for more shader massaging per clock.

 

ahhh thank you so much!!! I see some light now..

Now I just need to know how much L1 and L2 R600 has. It would seem to me as well R600 has dedicated samplers and also has L2 cache.

R600's texturing capabilties should not be far off from G80 from what I can tell. Correct me if I'm wrong. R600 dedicated sampling units could be more effective, but that is unclear. Has the same amount of texture address units, but has 1/4 of the filtering units. Both GPU's have L1 cache but which one has more? R600 has L2(how much?), G80 does not. R600's texture units are clocked 175mhz higher than G80's.

 

I tend to refer to the entire exercise of fetch and filtering as sampling, rather than just the fetch part.

 

Interesting quote from Guru 3d's g80 ultra article

 

Just a question, being this not really clear in the info on G80 I saw in various reviews: in g80 architecture, it is possible to perform at the same time 64 basic Texture Filtering operations AND basic Texture fetch operations, or there are 64 total samplers that can perform texture fetch OR filtering operations up to the maximum of 64?

 

It all makes sense. Mystified by R600, Nvidia could not sit tight and fall behind in the much-coveted "WTF is this???" category. Their retaliatory strike will no doubt be a huge success: having released a card no asked for, wanted, or will buy they left everyone and AMD scratching their heads, bellies, butts and other scratchable parts. David Orton was seen wandering the hallways barefoot, drooling and muttering "8800 Ultra.. does it make sense to you? No? I thought so.... But it is... its is... what does this mean....? What...? It makes no sense, but here it is.... It's a trick... trick I tell you!!! Why don't you listens? Will anyone listen? Snakes... snakes.. I am covered in snakes! Where am I?"

Bravo Nvidia. Bravo.