The LAST R600 Rumours & Speculation Thread

Yeah, it'll prolly do as much as WHQL did for ensuring bug-free drivers...

 

So, in reality, the feature is offering x6 AA quality over-all -- when it works in a title.

 

Allow me to express some thoughts on the rumoured RV630 specs.

It says Kohinoor GDDR4 consumes 127 W, I think this TDP is made up. The retail R600 board consumes 240 W when R600 is built on 80 nm RV630 is supposed to be 65 nm. An R600 built on 65 nm is supposed to consume 180 W, cut the chip by half to make a mainstream card and u will get 90 W. Am i right?

 

Errr.. No. It offers 16x edge sampling. And 4x Transparency sampling. And will offer in some cases "Reduced" sampling on very thin objects. 6xAA isnt even a remotely close comparison to 16xAA. Even if your trying too find some common ground to make a base comparison. There are times 16xAA can look better than 8xQ. And times it can look worse. Dependent exactly on whats being rendered.

 

If things were just that simple, yes, but sadly, they're not.

 

These are my theories on the delay.

1) Given to how close R680 is to completion(guessing), AMD decided to hold the R600 back for a joint launch along with their other respective family(RV610/RV630). A complete and total launch mind you if you call it that. AMD will be able to meet all price points of the market. R680 for $600, R600 for $500, RV630 for $200+, RV610 for $100+, and downgraded R600 variants meeting the $300, and $400 marks.

2) AMD actully has good working samples of 65nm R600 cores.

3) AMD is almost done with their 4x 16x PCIe slot chipsets, and for marketing reasons, decide to launch both at the same time to show the power of 4 R600's on editors day.

 

Maybe the R600 already the R680

 

I would imagine that the only way this could even have some weight of truth is if ATI was able to shrink the die to 65nm. Something i highly doubt they could accomplish right now with the R600 core.

 

And that would be a bad bad mistake on AMD's side I would say.

US

 

One might want to consider that there are no guarantees that RV610/RV620 will ship on schedule. There is no real logic for delaying just to ship a family. IF everything came together at once they could launch a family. ATI is racing to catch NVIDIA and its 80nm mainstream parts, which apparently have garnered the bulk of the Santa Rosa design wins. (NVIDIA captured these, btw, because samples were available a while back - this was indicated on the November earnings call.) When you know your competitor is ahead, perhaps there is a tendency to set aggressive production dates, and perhaps even rush things. Might this be one reason we are seeing the R600 delay now?

 

It is with very thin objects and stencils. I highly doubt x16 CSAA performance offers comparable image quality as X8Q over-all. Do believe it may in some cases.

 

Have you seen it? Theres millions of screenshots out there demonstrating it. The most notable problem with 16xAA is on very thin objects that cant be seen unless you blow them up. Stencil shadows are usually bordered by other objects as well. Even FEAR shows minor differences with 8xQ/16x/16xQ on stencil objects. 16xAA and 8xQ on non alpha test edges and looks pretty much the same. The point however is that 16x CSAA has absolutely zero comparable qualities to 6x multisample AA and is very poor comparison.

Chris

 

http://www.3declipse.com/images/stories/G80/AA/GT1/8xqplane.png

http://www.3declipse.com/images/stories/G80/AA/GT1/16xaaplane.png

The x16 AA clearly isn't offering better smoothing than the X8Q..........and the reason why this shot matters is because it is close to a near horizontal with nice color contrasts to gadge image quality in a static shot. This is easier to notice with a moving screen.

 

Both of you are way too much OT. Pauly get back to "base" if you want to discuss it. Chris can find his way to SC if he wants to

 

Maybe they aren't just looking at a family launch but an entire platform launch. CPU, GPU, chipset all at once. Just a matter of how close the chipset is and whether or not Barcelona is close. I can't imagine a quadfire setup not being CPU limited. Nearly 600GB/s of bandwidth should be enough for a little AA/AF at higher resolutions. I'm not sure you can even push the resolution high enough to make that thing break a sweat.

It's a little hard to stomp all over the competition if you're CPU limited so I'm sure they're pushing to get something out the door that can compete performance wise. Even if the CPU is bordering on a paper launch.

 

Be careful, when I suggested that possibility a few pages back everyone thought I was teh crazy.

The chipset was rumored to be either 2x16 OR 4x8 btw, now 2x16 + 1x8 (16x physical) + 1x2 (iirc), only two slots with 16x, or three really capable of using high-end graphics. I think the main goal of the chipset pertaining to GFX is to launch, support, and sell lower-end cards (RV610/RV630) alongside a crossfire setup for physics. This is not to say we won't see quadfire in a two-card setup though. Surely 2x 'x2900xtx2's' (or something similar) + another card for physics is possible, and would still more than likely be CPU limited...Let's hope this does exist and Barcelona does help that possibility.

I agree with everything else you said, though.

 

Barcelona's a server chip. The reduced clock speeds, smaller per-core cache, and registered DIMMs would hurt single-threaded game performance. In a Crossfire setup on any popular game, it would be even more CPU limited than a standard higher-clocked FX or X2.

AMD would be better off just using an FX chip at 2.8 GHz.
What might be possible is the announcement of a new CPU speed grade, maybe. I don't know when the next FX grade is supposed to be announced.

 

And I already explained that Opterons work in Quad-FX mobos.

Jawed