R350 finaly! It's faster than FX, but is so flexible? Aparently yes!

Some minutes ago, the first reviews of the new Radeon 9800, with uses the new R350 core are launched.

Aparently, it’s so flexible as GeForceFX, or even more and have the advantage to be faster with filters, it’s much more noiseless and consumes less power, this all for the same pice

What do you think about her?

Hi, this a OpenGL forum!. Anyway, the change for the queen’s graphics board is near…the RV350 is better than NV30 and NV34, but i think it in Direct3D…and OpenGL too?.

Yes, but my main doubt is about the suport to OGL 2.0…

Besides somethings else, like de Stencil Buffer aceleration is very important to Shadow Volumes rendering in Doom3 which is OpenGl…

Besides such important thing just happen sometimes, and i don’t think this will realy disconcert someone

Can you ladies point me towards the reviews? I haven’t come across any yet this morning. A few links would be nice. Will help to add some context to this discussion!

www.tomshardware.com www.hardocp.com
among others.

Even better, go to www.rage3d.com there are links to many other reviews. Unlucky in the momment I don’t find any review really good on techinnical details

But appearently, Radeon 9800 is amazing, full suport to OGl 2.0, multiple render targets, stencil buffer with aceleration and more

Unlucky again, i canno’t find any good schedule to compare the nunber of registers, temporary registers and other things of this kind

Please if someone find, put the link, and off coursem say what think about the new flagship of ATI

[This message has been edited by Kosh Naranek (edited 03-06-2003).]

From what I’ve read ( www.driverheaven.net/reviews/r350/index.htm ), the 9800 has an F-buffer which allows for fragment shaders of unlimited length. Very impressive.

They say that is this which give to the Radeon 9800 full OGL2,0 support, infortunate I was not
see this in details, however this is really drop-shot

Do you anybody know what “Shadow volume rendering acceleration” http://www.driverheaven.net/reviews/r350/index.htm means on their features list?

There’s some good information at Beyond3D: http://www.beyond3d.com/reviews/ati/r350/

– Tom

Do you anybody know what “Shadow volume rendering acceleration” means?

I guess it means support for some extension that allows shadow volumes to be rendered once (instead of once for front faces and one for back faces), hopefully EXT_stencil_two_side.

The card looks good. I’ve been a NV nut with my GF3, but ATI are looking better and better, especially wrt the NV30 vs new radeons.

the shadowvolume optimisation is part of hyperZ3, you can read about it on their page. it means that the superfast hyrarchichal z-buffer thingy they have is now optimized as well for the stencil operations normally done for shadow volumes, means compressing/hyrarching them as well… so much of the stencil-buffer-fillrate can get forgotten… and of course stencil-two-side is in as well

i am very impressed by the info so far. now i would like to see some specs, espencially for the pixelshaders and vertexshaders…

yeah, gibber, currently there is no point in getting a new nv card anymore, is there? high end is now radeon again, high cheaper end is radeon as well (9700 stuff), and cheapest high end is 9500, still a great card… the lower stuff is still dx8.1 only… but that’ll change sooner or later, too…

The only worry holding me off getting a Radeon is the Linux support. How good are their Linux drivers these days? NVIDIA dropped the ball with their last Linux set IMO.

Oh, and I’d really like 32-bpc FP too…

[This message has been edited by nutball (edited 03-06-2003).]

Sounds like a pretty good board but it’s performance seems close to the R300.

I noticed that the transister count hadn’t changed between R300 and 350

And what pisses me off is that I don’t see a list of supported extensions. Tons of benchmarks but a simple list can’t be uploaded?

Originally posted by V-man:
I noticed that the transister count hadn’t changed between R300 and 350

… The review on AnandTech says that there are no new features except for the “Hyper-Z III” improvement and some color compression optimizations. Huh?!?

The features mentioned in the other reviews sound very impressive, though!

And what pisses me off is that I don’t see a list of supported extensions. Tons of benchmarks but a simple list can’t be uploaded?

Amen! Most video card §reviews seem to consist of a rehash of the marketing blahblah, followed by approximately 25 pages worth of benchmarks and FSAA/AF comparison screenshots, neither of which I ever even so much as look at.

What I want to know before buying anything is what the extension string looks like, whether all new functionality is exposed, AND if all relevant extension specifications are publically available by the time I get the card in my hands.

– Tom

How can they support “full” OpenGL 2.0. As far as i know 2.0 isn´t completely ready, yet. And if it were, than there would certainly be a final spec, or so.
I mean, if i by a GF fx, or a 9800, than i am not able to program and use 2.0 programs, am i? I still need the headers, and they are not out yet, are they?

Jan.

Have you tried e-mailing them? They’re probably not aware the extension list is important.

[This message has been edited by Adrian (edited 03-06-2003).]

This article (in French) says Radeon 9800 Pro has 117M transistors (107M for 9700 Pro). http://www.hardware.fr/articles/456/page1.html
I note that THG’s article put a ? next to 9800 Pro transistor count.

The 9800 implements an f-buffer which means it can support shaders of any length. You can problaby even support full dynamic branching in fragment shaders with clever use of discard and the f-buffer. It won’t be fast but it’ll work. The other new stuff seems to be mostly efficiency improvements, better HyperZ and a tuned a memory controller. Of course, it also runs at a faster clock rate with faster memory.

All I can read about R350’s F-buffer is that it “allows infinite instruction count”. But instruction count is not the only limited resource of FP, there are others:

  • texcoord count
  • texture unit count
  • dependant read complexity
  • temporary register count
  • constant register count

If you exceed any of these, you have to do multipass too. If I understand correctly the Stanford paper (didnt read thoroughly), the “true” F-buffer would remove all above limits, because it combines results of muliple passes, and each pass is given full set of resources for use.

So, the fundamental question is: which limits of FP are actually relaxed in R350?

If this is only instruction count, as reviews seem to suggest, then this is nothing to be excited about. In practice difference between Nv30’s 1024 vs. infinite will not matter, especially in situation when neither HW supports true (non-unrolled) loops in FP (otherwise ps3.0 support in R350 would be hyped). Then R350 is just as much GL2-ready as nv30.