R520 Infomania

That, and I don't see how ATI's multi-gpu method could be "a lot better than AFR or SFR".

 

Well ATI claims greater compatibility although from a technical standpoint I don't see how tiling is much superior to, or more compatible than SFR's horizontal split. It seems that load-balancing on a tiling implementation is free (assuming the most likely scenario of identical cards) and has lower driver overhead which could show small gains in cpu limited situations. I also have no idea if it's easier to join two parts of the screen (SLI-SFR) or sew a bunch of tiles together (AMR-tiling).

Also, according to the Inq AMR will support AFR as well. Not sure how ATI can differentiate their AFR from Nvidia's. It seems to be a straight forward process - only difference being the slave-board will be submitting its frame via the PCIe bus instead of the dedicated bus that Nvidia is using.

 

I heard/read/dreamt somewhere that ATI claims greater compatibility.

 

Aren't marketting departments wonderful?

 

We need to draw a distinction between "ATI/Nvidia claims" and "Inquirer claims".

 

Hehe. I knew I wasn't completely bonkers. A lot of people have been hoping/wishing/speculating that this is the case but I knew I saw something somewhere.

http://www.beyond3d.com/forum/viewtopic.php?t=22170&highlight=

http://images.tweaktown.com/imagebank/news_atitechday05.pdf

See slide 34. It has a quote from the Tech Report. I don't see why they would include that quote in marketing material if their alternative doesn't suffer from the same compatibility issues.

 

I would guess that the reason a few applications see no significant speedup on a SLI platform has more to do with how those titles were written and what features they use (particularly if they use any sort of runtime generated data objects) rather than it being NV's "fault."

ATI's multiboard solution will no doubt have its share of incompatibility. But we'll just have to wait and see...

 

Should that be called Crossfire from now on?

 

Hypothetical it is. I'd expect R520 to be still at 4 quads while increasing clockspeed to a quite high level.

Wouldn't that be quite a waste in R&D and resources? R5xx isn't according to my understanding just a high end accelerator, but an entire product line with mainstream and/or budget offerings appearing gradually as time goes by.

R480 fairs well against NV40, but what about the rumoured G70? Also NVIDIA intends from all indications to keep SM3.0 in the low end for at least 3 years.

Thanks to Xmas for clearing that one up. Didn't know that either.

 

An occlusion query can effectively disable AFR if the application wants the result back too early...
Post effects can probably hurt SFR...

 

Question would be if ATI would be able to overcome such hurdles and how.

 

Will the r520 support FP blending for HDR rendering? If it doesn't support FP blending what other methods could they implement for HDR.

 

The same ones that you can use now. Ping-pong'ing a few floating point render targets.

They'd be pretty slack if they didnt release a chip that could do floating point blending.

 

Considering Valve developed an additional level for HL2 with float HDR, I'd say it's pretty safe to assume that the support will be there.

 

I wasnt suggesting that r520 wouldnt have floating point blending, just that it would be err, sad if they didnt.

Who says that the new level (+code) is using floating point blending anyway?

 

Ask me the same question in some days, again.

NDAs sucks

 

Considering it should have been in the fudging game, I'm not so impressed with that support.

 

How many days?

 

Uh oh so are you guys suggesting it's just some sort of bloom effect?