The ATI R600 Rumours & Speculation Centrum

The context switching is what I get the "bridge too far" vibes off of the most. Maybe it had significant transistor count/die area implications. Looking at G80 and NVIO and Inq predicting 720M for R600, and card lengths and power reqs. . . .it just seems pretty clear to me this gen until 65nm shows up we've officially arrived at "the ragged edge" of what the process techs can give us.

 

Isn't PCF (or a certain form of it) patented by nvidia?

 

Radeon x9700 or because nv using *800 number x9800, all 2 name sounds good when we checking back the gpu history

 

PCF was invented before NVidia even existed, so I doubt that.

 

I like the x2700pro idea my self.

 

There's a discussion here: http://www.hexus.net/content/item.php?item=884&page=2 re "Will it be part of DirectX in the future?"

 

Well considering that "the other IHV" can filter in a shader at very little cost with better results isn't hardware PCF blase now? (to borrow from BZB's dictionary )

 

I tried to mix up some of the latest rumours...

R600
80nm
720M transistors
750MHz (+-3,5%)
96 vec4 MADD (+scalar ADD?) unified ALUs
32TMUs

512bit mem-bus
2600MHz GDDR4 (+-3,5%)

260mm PCB
20.01.07

 

Well ATi is really pushing developers for code for soon to be released and developing games

 

Razor, what's the R600 info you have? I don't care if it's not "in stone", we understand that. Let's just see if your info comes true when it's released.

 

ATi doesn't seem to like the g80 much can't get into anymore details then that. But they want developers to open up code that developers and nV have worked on together, which nV has said ok to the developers to open up.

 

What are your sources? No company ever likes a competing product...

What do you mean by that? What code? ATi wants developers to Open Source their code...

 

ATi hasn't asked for developers to open up code ever, they always looked for ways around optimization for thier hardware. Or make a seperate path.

ATi wants dev's to open up code so they can optimize code for the r600 (code that was written by dev's and help by nV). I have quite a few sources that are saying this, its been going on for the past 2 months or so, and nV gave the ok open it up to them about a week and half back.

This could be due to the priority physics programming, I don't know, I will find out more soon.

 

What code? They can always get all the information from simply running the game/application... seems like a more efficient and a much easier solution.

Again what code and how is NVIDIA related to this?

Who are your sources if you don't mind me asking?

 

Yes, curious, what does NVIDIA have to do with ISV code...?

 

Well you should know this very well Dave , hmm game developer's programs come to mind doesn't it?

 

Sources: Developers.

You never get all the information from running the programs, you might be able to get shader information, but thats not all the information necessary to optimize paths.

 

Talk to your devrel teams then , something must be missing or they are worried about something. I don't have the specifics on what ATi wants to see, so the only guestimate I can make because of the recentness of "aquiring" certain knowledge, which was Oked by nV, which I confirmed from people inside nV, well not directly but dev's wanting to open up code was talked about, ATi wasn't mentioned, but how may times does something like this come up?

 

You get everything that goes through your driver.

 

I don't see where the confusion is...Razor...correct me if I'm wrong.

1) Developers (ISVs) have G80s to code to.
2) Being a new architecture, ISVs coding to the G80 (for compatibility and especially optimisations) work with nVidia on best practices / techniques. ISVs may have agreements with nVidia on code ownership for code that was worked on in collaboration.
3) The "G80" techniques may or may not be ideal for getting the best performance on R600 hardware, which ATI now (presumably) has in house up and running)
4) ATI requests that ISVs give access to ATI for this code...so ATI can evaluate how it runs on their R600...and ATI can turn around and recommend their own techniques if needed, or implement driver chagnes. (ISVs may then need to get permission from nVidia in the case of collaborated code.)

I don't know industry standard practices, but I would think that in most cases, even when there is collaboration with an IHV on coding techniques, the ISV likely retains full rights over the code...so in practice, nVidia shouldn't need to be involved at all...but I can see where the potential exists for nVidia involvement.