R420, NV40 launch info at Anand's

[mreman4k]

(<---And it feels AWFULLY damned good to finally know something Dave don't and to be able to be annoyingly cryptic about it! Best birthday present ever

LOL

 

[PaulS]

I'm currently trying to establish when R500 will appear. I'd actually previously pinned it as a Longhorn / DX Next product, but that appears to have been very wide of the mark - seems that it will be here much earlier than I'd expected.

Yes, quite. (<---And it feels AWFULLY damned good to finally know something Dave don't and to be able to be annoyingly cryptic about it! Best birthday present ever. )

Don't know but if R20/R423 are launched in may june do you really expect R500 before spring 2005 or fall 2005 ?

Definately. 8)

Makes sense based on the original goals of R420, although it does raise some interesting questions about the (up to now) reasonably parallel nature of release dates of new architectures from both IHVs. Would throw it completely out of whack, because NV50 isn't due Q4.

I still don't expect R500 to surface this year though, despite what the internal roadmap says. I suspect it will be delayed to the next spring (and even that is pretty early, really - a mere 12 months after R420, which represents the standard new architecture speed boost, if not the technological boost).

Perhaps R500 will be considerably cut back from what you'd expect of an x5xx generation card? As in, I'd suspect it to be mostly new features, and not another 2x-3x speed boost (especially if it's before Spring 05).

Oh, and Happy Birthday apparently 8)

 

[digitalwanderer]

Makes sense based on the original goals of R420, although it does raise some interesting questions about the (up to now) reasonably parallel nature of release dates of new architectures from both IHVs. Would throw it completely out of whack, because NV50 isn't due Q4.

Yeah, but it would also nicely explain why ATi apparently isn't too worried about how it's going to be a close call for the next gen graphics king title rather nicely too...don't you think?

nVidia is coming out with a nice card this round, I know that; but the R420 is nice too. nVidia must have busted their freaking balls to get the nV40 out by the time they plan to and with the advances it's purporting to have, while ATi cut back their initial plans for it's competition by changing from the R500 to the R420...but they didn't stop developing the R500.

ATi will be ready with it a lot sooner than most think.

I still don't expect R500 to surface this year though, despite what the internal roadmap says. I suspect it will be delayed to the next spring (and even that is pretty early, really - a mere 12 months after R420, which represents the standard new architecture speed boost, if not the technological boost).

Now that I can't quite argue with yet because I don't think ATi has firmly decided it yet, I think a lot of the decision to release the R500 is going to be based on how the nV40/R420 match up goes and the yields/profits they make on the R420. (From some grapevines stuff I'm getting the impression that the R420 isn't going to be as profitable for ATi as the R3xx series was, and that the R500 might actually be more profitable...but again it's a bit too early to call)

It's like the R420 release. Do you think ATi was scrambling to meet nVidia's release date or holding back?

Oh, and Happy Birthday apparently 8)

Thanks, it's actually going rather swimmingly and I'm enjoying the bejesus out of it so far.

 

[ninelven]

My question is if there will be a refresh to the R420. It seems like there will be, but if ATI is launching R500 this winter then why bother?

 

[Evildeus]

ATi will be ready with it a lot sooner than most think.

Well, it all depends on the capacity to release this so call R500 in 2004 with 0.13 process (0.11 perhaps?)

Oh, and Happy Birthday apparently 8)

Happy birthday 8)

 

[digitalwanderer]

My question is if there will be a refresh to the R420. It seems like there will be, but if ATI is launching R500 this winter then why bother?

Is "sort of" a decent answer?

Ok this is all just me thinking out loud here and it ain't anything I can back up, I wanna make that perfectly clear first.

The R420 is going to launch with 3 of 4 quads enabled, and the "refresh part" is going to be all the R420's they've been binning that have 4 working quads along with better memory if it's available.

It just all kind of fits for me now.

Oh, and thanks Evildeus and I think the R500 will be .11

 

[Evildeus]

I thought the R400 was dumped because it would be too slow :?

 

[Joe DeFuria]

ATi will be ready with it a lot sooner than most think.

Well, it all depends on the capacity to release this so call R500 in 2004 with 0.13 process (0.11 perhaps?)

Yeah, I think actually the fundamental question has to be if R500 is targetted at 0.13 or 0.09 actually.

ATI has a history of getting their "feet wet" with new processes with less complex chips, so if history is repeated, you're right, ATI would go with 0.13 (possibly low-k) or 0.11 for the R500.

Though it's of course possible ATI might change its ways and try R500 on 0.09 first.

For 0.13 or 0.11, A Fall or x-mas launch of R500 could be physically possible. I don't think 0.09 would be possible until spring '05.

I personally have high doubts about a R500 launch this year...but with the way R420 is (apparently) in many ways a R300 refresh, and R500 has had some development for quite a time, history might not be a good guide for ATI over the next year.

One thing that I think both ATI and nVidia could and should consider, is that the high-end retail parts, from what I can tell, don't need to fall into some specific "fall" and "spring" cycle. While it's very important for ATI and nVidia to supply oems and builders with low-end and mid-range parts for these cycles...I can see both ATI and nVidia splintering off the high end parts and launching / shipping them at off-cycle times.

The people who buy these high-end cards don't tie themselves to any specific cycle.

 

[digitalwanderer]

I thought the R400 was dumped because it would be too slow :?

I thought the R400 was changed to the R500 because it was overly ambitious for their timeframe and they didn't think they'd need quite that much to match nVidia or that the new features wouldn't be used/needed yet and they made the R420 based around the R3xx design but highly modified/tweak/turbo-charged-with-bells-and-whistles. (The R420 is sort of like 4 R360s or RV360s on one die with 3 of them enabled is how I remember it being described to me)

 

[Evildeus]

I thought the R400 was dumped because it would be too slow :?

I thought the R400 was changed to the R500 because it was overly ambitious for their timeframe and they didn't think they'd need quite that much to match nVidia or that the new features wouldn't be used/needed yet and they made the R420 based around the R3xx design but highly modified/tweak/turbo-charged-with-bells-and-whistles. (The R420 is sort of like 4 R360s or RV360s on one die with 3 of them enabled is how I remember it being described to me)

Well perhaps, i just thought that they couldn't deliver or it would be much too slow if they could. Bur you are surely right
I would say that R420 = 4 RV360

 

[Evildeus]

ATi will be ready with it a lot sooner than most think.

Well, it all depends on the capacity to release this so call R500 in 2004 with 0.13 process (0.11 perhaps?)

Yeah, I think actually the fundamental question has to be if R500 is targetted at 0.13 or 0.09 actually.

ATI has a history of getting their "feet wet" with new processes with less complex chips, so if history is repeated, you're right, ATI would go with 0.13 (possibly low-k) or 0.11 for the R500.

Though it's of course possible ATI might change its ways and try R500 on 0.09 first.

For 0.13 or 0.11, A Fall or x-mas launch of R500 could be physically possible. I don't think 0.09 would be possible until spring '05.

I would say that 0.13 is out of question, if digitalwanderer's theory is right. Then so 0.11 or 0.09, but i'm not sure that it will be easy this year to do a 200m+ chip.

I personally have high doubts about a R500 launch this year...but with the way R420 is (apparently) in many ways a R300 refresh, and R500 has had some development for quite a time, history might not be a good guide for ATI over the next year.

I doubt also, R480 should be the next step, but as you say who knows

 

[Bouncing Zabaglione Bros.]

The R420 is going to launch with 3 of 4 quads enabled, and the "refresh part" is going to be all the R420's they've been binning that have 4 working quads along with better memory if it's available.

Are you sure that is not going to be the "non-Pro" and "Pro" versions (whatever designations they use this time around)?

 

[digitalwanderer]

The R420 is going to launch with 3 of 4 quads enabled, and the "refresh part" is going to be all the R420's they've been binning that have 4 working quads along with better memory if it's available.

Are you sure that is not going to be the "non-Pro" and "Pro" versions (whatever designations they use this time around)?

I think the Pro/non-Pro versions will just be variants on clockspeeds and they're saving the working four-quad cards for a quick refresh part a little later this summer.

 

[Pete]

Maybe more like X800 Pro now and X800 XT for the binned parts, with X900 reserved for "R480"?

Edit: Forgot to thank Joe for his explanatory post.

 

[OpenGL guy]

If we assume (like R300), that R420 does 2 AA samples per pipe per second

2 samples per pipe per second? I don't think AA performance would be very good in that case

 

[Joe DeFuria]

If we assume (like R300), that R420 does 2 AA samples per pipe per second

2 samples per pipe per second? I don't think AA performance would be very good in that case

Lol...yeah...make that 2 AA samples per pipe per clock.

 

[Mintmaster]

If we assume (like R300), that R420 does 2 AA samples per pipe per second

2 samples per pipe per second? I don't think AA performance would be very good in that case

Lol...yeah...make that 2 AA samples per pipe per clock.

I thought R3xx was capable of a full 6 AA samples per pipe per clock, and the performance drop with AA was entirely due to less than ideal compression and hence bandwidth or granularity limitation.

From everything I remember, the drop from 2xAA to 4xAA or even 2xAA to 6xAA is usually less than that from no AA to 2xAA. Most R350 packaging boasts 18 gigasamples/sec.

 

[Joe DeFuria]

From everything I remember, the drop from 2xAA to 4xAA or even 2xAA to 6xAA is usually less than that from no AA to 2xAA. Most R350 packaging boasts 18 gigasamples/sec.

Well, I'm going by the AA Sample fill rate as shown on the 3D Table here at B3D:

http://www.beyond3d.com/misc/chipco...derby=release_date&order=Order&cname=

In any case, if we assume the AA Samples per pipe per clock is the same with R350 as it is with R420 (which is just an assumption..it may very well change for all I know), then that doesn't change the fact that 600 MHz, 256 Bit DDR is needed to maintain the bandwidth / AA Sample fill rate ratio of a 12 pipe R420 running at 450 Mhz.

 

[AlphaWolf]

From everything I remember, the drop from 2xAA to 4xAA or even 2xAA to 6xAA is usually less than that from no AA to 2xAA. Most R350 packaging boasts 18 gigasamples/sec.

Well, I'm going by the AA Sample fill rate as shown on the 3D Table here at B3D:

http://www.beyond3d.com/misc/chipco...id=31&orderby=release_date&order=Order&cname=

In any case, if we assume the AA Samples per pipe per clock is the same with R350 as it is with R420 (which is just an assumption..it may very well change for all I know), then that doesn't change the fact that 600 MHz, 128 Bit DDR is needed to maintain the bandwidth / AA Sample fill rate ratio of a 12 pipe R420 running at 450 Mhz.

Well they could improve compression and increase effective bandwidth that way.

 

[Mintmaster]

You're absolutely right about the ratio, Joe. 600 Mhz memory would go well with a 12-pipe R420, at least for 32-bit rendering.

I was just curious about the number of samples per clock.