Specs for r300 and rv250

I find it more than hard to believe that we're going to leap from 32-bit to 128-bit color in one generation. Carmack can push all he wants, but he isn't going to get it if the companies can't profit. Maybe 128-bit using loopback of some kind, with extreme performance penalties? I don't see how this will be comparable to the current small performance delta between 16- and 32-bit, as companies are no longer optimizing for 16-bit.

I know nothing about chip design--is it possible for ATi to quadruple its bus width in a generation, and have the card working on its first tape out?
 
Same feelings here Pete.

32bit = 16.7 Million colours
64bit = 281.4 Billion colours

:rolleyes:
 
He was using the original and only logical definition of "billion", not the marketing version that then got promoted to american standard by the uneducated masses. Sadly more and more languages starts to get polluted by the idiotic definition "billion==10^9". :p

[edit]
Regardless of that, it's completely irrelevant that the number of colors is so large that it's difficult to grasp how big the number is. It can still be usefull.
 
Bah. I believe it's gonna work something around the basis of the RV250 besting the GF4 for the older DX8 games, providing a value alternative with speed for those OEM manufacturers like Compaq and Dell, although they'll probably still be biased towards Nvidia, just like Dell to Intel :)

The R300 is one that's still up in the air to me. A couple of people are still gonna have that whole "ATI has terrible drivers" fear, so they'll choose to go with the Nvidia Card. Personally, i'll look at the R300 and the P10, but the P10's lack of AGP 8x or FULL direct x9 support leave some questioning for it's abilities...
 
The part about Carmack using rev.1 of the chip is true which kind of points at the other stuff being true. The second revision arrived from the fab two weeks ago and solved a few problems regarding power consumption (the first card drew about the same amount as a GHz T-Bird across three rails). Performance estimates have been met though, so that's good news.

Regarding the color depth; that's obviously not a true 128-bit colour mode, but maybe the chip renders internally at 128-bit to preserve 10-bit per channel colour output for complex blends. The output stages could well involve 30-bit DACs so that will limit things anyway.

I doubt it's a symmetrical 128-bit mode though. 32-bits for alpha is totally unneccesary. In fact, not even the current 8-bit alpha channel (in 32-bit mode) is fully used by developers. That's why Matrox will probably get away with 10+10+10+2.

MuFu.
 
wow, that site link didn't last too long... anybody got it and could put it on a non-limited bandwidth? :)
 
Seems back up... I've seen some slides before but not others. For the RV250, the "Improvements over RV200" slide states "Double the number of rendering pipes" - "2 -> 4 parallel rendering pipes"... If this is legit, then it would seem that the RV250 is basically a die shrunk R200 running @ 300MHz.

As for the R300, it appears to be 350/400+ core/mem. Wait & see, I guess...
 
cellarboy said:
Hmm, does anyone else but me find the slides being dated June 18th 2002 awfully suspicious?? :-?
Click to expand...

Hmm, I thought so too... But I've seen a couple of the slides before. If they're PP slides, then you can enable the date & time to be updated automatically in the header/footer. If the caps were made today, then it pans out. The site only had a couple of hits when I was pointed there...
 
Yes, would anyone be kind enough to mirror those? It seems the site has exceeded its bandwidth limit for now.

BTW, 350MHz for 107mil. trans. on .15u sound a little high for anyone, considering Matrox is only hitting 220MHz with far less transistors? 400MHz 256-bit DDR sounds a little ridiculous, too--that's almost 50% over Parhelia's memory speed.

Still, I'd like to see the slides. :)
 
The other strange this is that the 150 mill polys a second seems to tie in with rumour (from 3DGPU) that the R300 is 15-20% faster than the GF4.

r300small.jpg


If that's true then it's not good. I want a reason to buy an R300.

Sure wish someone would mirror the slides.
 
Damn, those look legit (evrything matches what I have been able to find out from direct sources), BUT... the slides regarding R300 are only accurate if they have already secured the shrink to 0.13u!!!

I hope they haven't let TSMC decide R300's fate! I really didn't think 0.13u was ready to go for ATi yet. Hmm...

MuFu.
 
You must log in or register to reply here.

Similar threads

V
Strengths and weaknesses of PS2/GC/XB relative to contemporary PCs *spawn
Replies
3
Views
3K
Cyan
Cyan
D
AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion
299 300 301
Replies
6K
Views
1M
Bondrewd
Bondrewd
B
Why Barts is really VLIW4, not VLIW5 (and more on HD 5830/6790 being mainly 128-bit)
5 6 7
Replies
132
Views
37K
Alexko
A
G
Predict (for fun): Specs of Reality Synthesizer
Replies
4
Views
3K
kumayu
K
F
Sony's Next Generation Portable unveiling - PSP2 in disguise
42 43 44
Replies
871
Views
229K
liolio
L
Back
Top