[B3D Analysis] R600 has been unleashed upon an unsuspecting enthusiast community

Brilliant! Reading Rys' take on it now.

Oh, and will someone be a doll and lock that monster of a rumour thread?

 

If only they'd delayed it another month, I'd have gotten everything in there for once. If I missed something you wanted in there, my apologies. It's not as in-depth in places as the G80 arch piece, but it'll do to get us rolling.

Thanks to everyone that posted in the biiiiiiig R600 threads, there's a bit of each of them in each article, for those that took part in or followed the biggest consumers of DB space here since WaltC last talked about NV30.

 

Havent read it yet but thanks Rys

 

I'm pretty sure, but not quite so, that you meant the lifting of the NDA and not when the thing appeared in reviewers' hands!

Personally, I would've liked it if there were some quick gaming benchmarks thrown in this PartOne article. There doesn't need to be a lot of gaming benchmark analysis by B3D. Just give us some numbers basically, maybe at a single rez (whichever is the oft-used rez, via surveys, for each game you use, at "halfway-point" AA and AF settings since these are still the only really useful OOTB settings). But you know your target audience

There are a few uncertainties which I'll PM/email you but overall I enjoyed reading this, Rys.

Oh, and I still can't get over the weird feeling I get when I read "AMD" instead of "ATi".

 

I know it's come up before, so a hint: middle-click (or right click and select 'open in new tab') on the diagram and it'll open in a new tab, for easy reference.

 

Great as usual from beyond3d. I just hope AMD can come up with something better around winter time when I upgrade to vista and direct x10 card.

 

Great read, thanks Rys and all who contributed to the article! I was getting tired of STALKER and Oblivion benchmarks and this was a welcome surprise (didn't expect it until 9am).

So R600 is doing the MSAA resolve on the shaders even for the box filter? Well that would certainly help explain why performance tanks with AA applied. And thanks for the sampling chart - nice one stop shop for comparing different formats and point vs bilinear sampling.

 

Finally. Good job Rys (and everyone else involved in the process).

Just a quick question, has ATI mentioned any word on whether they plan on continued use/support of HDMI, or do they plan on switching over to DisplayPort later this year (early next).

 

Thanks -- it is a good start. I'll need to read (rather than skim) it tomorrow, but the more I learn about the samplers, the more puzzled I become. Why split the texturing units into four, and then map each to one part of each of the "clusters"? Isn't this going to take a routing and caching hit? As well as restricting your texturing unit count to your cluster ALU organization?

ALU capabilities are very interesting -- NV efficiency spin notwithstanding , the sheer number of them....

-Dave

 

Rys (and others),
A really brilliant article, completely in line with the quality and thoroughness I've come to expect at B3D.
Thanks! (Can't wait for the follow up articles!)

 

Does it need to be an either/or scenario...?

 

Nice detail, and commentary that's (in Beyond3D tradition) refreshingly absent of cheerleading while not refraining from pointing out shortcomings informatively. Please keep it up in the upcoming additional articles.

...

Perhaps I missed a link, but I think it would be a good policy to provide readily accessible links for downloading custom testing tools (and/or associated web pages if they are associated with a site that offers such), with source if available, whenever they are used in an article.

NOTE: A "maximum" on Page 7 should probably be "maximize".

 

Thanks Rys. I'd thank/congradulate personally but i never seem to catch you. Some interesting stuff in there. ((Alot met my expectations. And you didnt fail to cover the areas that interested me))

Chris

 

Yes!




In all seriousness, i posted the question after coming to the HDMI section, and was in a hurry to get back reading. Just wanted to know if ATI still plans on going with DisplayPort later this year.

 

Good read Rys.

I assume they haven't implemented EATM in the driver used?

 

I was wondering the same thing when I read about this. I don't know why they decided to split the texture units overall, but perhaps the reason for the inter-cluster splitting is because SIMD execution within a single cluster will create heavily unbalanced access to the texture unit across time. By sharing a texture unit across all less-synchronized clusters the accesses are more randomly spaced? Just a guess.

 

Think of it like this.

Visualize the SIMD’s as 4 columns, with a 5’th column for textures.

Each of the SIMD’s are comprised of 80 processors, that work on 16 elements – these elements are still arranged as quads; i.e. each SIMD is 4 quads, stacked high. The texture column is also arranged as 4 quads of textures and each of those quads serve a quad on a single SIMD at the same time. As all the processors in a single SIMD are going to be working on coherent data then this helps keep the texture coherency high.

 

From all I've read, the drivers are going to be the determining factor here. Thus, it means more waiting, until they mature. Although, given the potential and the price point, the 2600XT is probably a better deal than the 8800GTS. We'll see what happens.

Also, I would bet that the R600 will be better than the G80 for GPGPU.

Anyways, good article.

 

Can we please see some D3D10 benchmarks to prove once and for all were not nV 30...