The LAST R600 Rumours & Speculation Thread

Look at the picture posted, I smell a NV30

 

Well yes silly, but what sort of texturing? Extreme texturing? Ultra-texturing? Immoral-O-Texture? Or what?


Q: How much bandwidth?
A: huge lots. Many bandwidths.
Q: why? why so much? why nearly twice as much as their mortal enemies?
A: oh, because it'll be used to feed the TMUs
Q: that's silly. Why would they need twice as much as The Others? Are they silly?
A: No, they're not silly, they need the same much bandwidth per TMU, there's just twice as many of them
Q: That's goods. So they'res not teh silly. Why are there twice as many of them those there TMUs then?
A: Because they do twice as much texturing ... to do ... some things...
Q: What sort of things?
A: Naughty things

Q: Will it be naughty things they do with the TMUs? Is it naughty tiem?
A: Oh yes. It's naughty tiem. Definitely.

How much extra texturing is required, and when and where and how? compared to G80 that is... and why twice as much?

 

Because R600 is an AAAWWWWSOME immoral biatch!

 

Which would be a good thing..

To many current titles trust on shaders to look good, while they could really use some higher quality textures..!

 

I'm half serious Geo. A year out you nailed your flag to the mast re G80 and you were right(-ish), despite it being a total left-fielder for many architecture-wise. So, come on, give us a sig-worthy declaration. Why do they need twice as much? Eh? A(MD|TI) clearly has a left-fielder coming here, but what is it? How is this going to translate in to what's going on in my face that G80 can't deliver?

 

I sure as hell hope that hypothetical ~150GB/sec bandwidth has something better to show than 2x or 3x times the performance in 2560*1600 with say 8x multisampling, where the winner will after all lack in playability.

As for something being immoral here, it's most likely the neverending debate about inches

 

wouldnt this make it slower than g80? hard to believe it could be slower with all these ridiculous power numbers and board size rumors

btw anyone else find it weird that we can travel to different planets but we have problems developing gpus.

 

On G80 you'd actually put a 512bit bus to good use if you'd want extreme resolutions with 8x MSAA or even worse 16xQ. For the other two hybrid MS/CSAA modes you'd hardly need extravagant memory footprint or bandwidth.

I guess it's safe to assume that ATI also went for single cycle 4xMSAA, it then comes down how many cycles for AA would actually make sense and not for bandwidth alone. What's so horrendous about 8x sparse MSAA anyway? And it's not like marketing won't find ways to downplay coverage sampling as useless.

I'd exclude 16xMSAA. For one it sounds like a too large penalty for now and as a close second I recall a M$ presentation mentioning that "some" future D3D10 GPUs will go up to 16xAA.

 

this is ludicrous-speak isn't it? there's no mass market for gpgpu, unless that's going to come out of left-field too and ADMATI are going to announce gpgpu/physics support which hasn't leaked yet. gpgpu is a niche, it makes zero sense for AMD(ATI) to make such a major design decision (which I'm presuming will cost them major money) to satisfy a market which doesn't exist. What's the lifetime of R600? Six months? Three months, now that it's three months late? Where are the killer-app gpgpu apps coming from in the next three/six months? Planet Mars? gpgpu sounds like a made up stuff explanation. It's like coming up with Itanium2 as a HPC/desktop/mobile-one-size-fits-all processor, isn't it?

To me it's looking more and more like they've either screwed the pooch, they've got some IQ enhancement that no-one has thought of (which had better be damned good), or they're shooting to win the IQ crown at stupid resolutions that 1% of the population play at.

Well in that I think you are 100% correct. In interesting times, we live.

 

It's a niche, but a potentially very lucrative niche as a standalone market. I know we tend to think of these things as consumers, but there's more to it than that. For what's it worth (probably not a lot), here's a quote from Jen-Hsun Huang, Q3 2006 CC - I think it's fairly clearly they are considering GPGPU as a non-negligible revenue stream. As for Quadro Plex, think of what G80 does in terms of output (NVIO!) and what it means for scalability there...

It's fairly clear that with G80, GPGPU was a very real consideration, and so was imaging. Neither provides any real advantage for G80, but these architectural decisions will likely provide them with high-margin revenue streams in the hundreds of millions of dollars - if they manage to capture that market share, of course.

http://seekingalpha.com/article/20294

Uttar
P.S.: The point I'm willing to illustrate is that, for a variety of reasons, keeping GPGPU in mind when designing a DX10-level architecture is a Good Thing, and for a lot of reasons. I don't think designing a BOARD (->bandwidth) around it makes sense YET, though, unless aimed at the professional market.

 

Ahem..

Excuse me.. what other planets did WE travel to .. besides in our dreams..??

The harsh reality is that probably none of the current board members will be alive when someone actually sets foot on any other planet...

 

Even the Moon is in doubt

 

I suggest you stop posting crap.

50% faster? No way Jose.

 

should have been clearer, i just mean send rockets to other planets.

 

Picture! where?

 

Ok i have to admit that was a funny comment.



I remember the predictions made by movies/people for future technology of the human race past the year 2000. Arent we suppose to have flying cars by now?

 

Well, I think it's worth noting R580 was really held back by texturing. If r600 has 32 TMU's that's double. Plus add a little more for some more clock.

So I think something like 100%+ a little bit is as much faster as R600 could hope to be in general, than R580.

Which means it'll be solidly with G80 and all, but it's unlikely to be a performance stunner.