...what do ytou know about Radeon 9500?
So, MuFu, Now that R300 is not longer a mystery...
- Thread starter Dave Baumann
- Start date
LOL.
Wel... "9" meens DX9 or sumfink, err.. no, maybee not. "5" meens dat itz abowt middle-speed 4 itz genurayshun or sumfink. I dunno.
Hehe... seriously, I haven't heard of such a card, although I'm guessing it'll be a stripped down 9700, same as everybody else. ATi might not even produce one themselves. "9500" could be the name given to AIB boards that don't have an external power connector and are subsequentally limited to speeds under 300MHz. I can imagine a 3rd party board called "Radeon 9500" clocked at say 275/275MHz with 3.6ns BGA (routing for TSOP would prove tricky given the buswidth, and they will most likely stick to the 942 reference board design, anyway). They'd save hugely on the memory and a little bit on the board. They can also cut back on the DACs which are "of the highest quality" on the R300 and subsequentally quite pricey (watch out Matrox!).
Couple that with some low yield percentile R300 ASICs and you have a product that costs much less than 9700 to manufacture, with minimal development costs and could perform about 20% less; although perhaps not by that much in CPU-limited situations (which are plentiful, as I'm sure you all noticed!). I'm guessing just a BIOS flash is required to switch back to AGP power. These are probably the boards we'll see offered in DC combos. Notice Supergrace are already offering "optional" VIVO with their 9000s, no doubt through a DC.
What I really want to know is where 958; a RV250 board that uses BGA memory, figures in all this. I think it's only just been qualified (2nd revision) and wasn't mentioned at the launch. Perhaps Radeon 9200? Not sure.... it seems like a slightly confused product to me and I wouldn't be suprised if we don't see it at all (it will depend a lot on how quickly they purge the BBA 8500/7500 channels, of course). :-\
MuFu.
Wel... "9" meens DX9 or sumfink, err.. no, maybee not. "5" meens dat itz abowt middle-speed 4 itz genurayshun or sumfink. I dunno.
Hehe... seriously, I haven't heard of such a card, although I'm guessing it'll be a stripped down 9700, same as everybody else. ATi might not even produce one themselves. "9500" could be the name given to AIB boards that don't have an external power connector and are subsequentally limited to speeds under 300MHz. I can imagine a 3rd party board called "Radeon 9500" clocked at say 275/275MHz with 3.6ns BGA (routing for TSOP would prove tricky given the buswidth, and they will most likely stick to the 942 reference board design, anyway). They'd save hugely on the memory and a little bit on the board. They can also cut back on the DACs which are "of the highest quality" on the R300 and subsequentally quite pricey (watch out Matrox!).
Couple that with some low yield percentile R300 ASICs and you have a product that costs much less than 9700 to manufacture, with minimal development costs and could perform about 20% less; although perhaps not by that much in CPU-limited situations (which are plentiful, as I'm sure you all noticed!). I'm guessing just a BIOS flash is required to switch back to AGP power. These are probably the boards we'll see offered in DC combos. Notice Supergrace are already offering "optional" VIVO with their 9000s, no doubt through a DC.
What I really want to know is where 958; a RV250 board that uses BGA memory, figures in all this. I think it's only just been qualified (2nd revision) and wasn't mentioned at the launch. Perhaps Radeon 9200? Not sure.... it seems like a slightly confused product to me and I wouldn't be suprised if we don't see it at all (it will depend a lot on how quickly they purge the BBA 8500/7500 channels, of course). :-\
MuFu.
the 9500 will not be just a lowerclocked 9700. What it is ATI wouldn't say and really even ati hasn't settled on the final specs of it. Or so they say
anand guess it to be a 4 pipeline version of the 9700 which doesn't sound too far off.
As for BGA 9000, AIB vendors who use 128MB of ram on their 9000 designs I believe are using BGA memory. I should've looked more carefully at the boards the 9 manufacturers were using.
anand guess it to be a 4 pipeline version of the 9700 which doesn't sound too far off.
As for BGA 9000, AIB vendors who use 128MB of ram on their 9000 designs I believe are using BGA memory. I should've looked more carefully at the boards the 9 manufacturers were using.
Joe DeFuria
Legend
That would be an interesting part. If it's 4 pipeline, then that would pretty much negate the need for a 256 bit bus....so a 4 pipeline 128 bit DX9 card. That part would be interesting competition for the Ti 4xxx series.
Disadvantages: lower performing in DX8 apps, when not using filtering or AA
Advantages: as good or better performing when AA and aniso is enabled, DX9 support.
Of course, the 9500 might not be competing with the ti-4x000, but some other new nVidia part...
Joe DeFuria
Legend
Yes, 4 pipe, single TMU.
If a 8 pipe, single TMU can offer over double the performance of a 4 pipe, dual TMU chip, then there's little reason to believe that a 4 pipe, single TMU can offer better performance than a 4 pipe, dual TMU board.
Again, I'm talking about AA and / or anisotropic situations here. I would expect that in non AA / Non Aniso situations, the GeForce4 would out-perform a R-300 "cut in half" type part. But then, everyone who owns a GeForce4 or this class of chip does so so that they can turn on Aniso and/or AA.
You can't just look at the number of pipes / TMUs...but the architecture as a whole. It's seems pretty obvious that the R-300 architecture was optimized for AA'd / Aniso Filtered pixels....that's where performance really shines.
According to ATI, Aniso on R-300 takes zero performance hit when used in conjunstion with bilinear. And "minimal" impact when pairing aniso with trilinear. Of course, I'd like to see tests bear the "zero" claim out (as well as define what ATI means by "minimal"), but the relative perforance tests between GeForce4 and R-300 with aniso / AA speak pretty mcuh for themselves.
If a 8 pipe, single TMU can offer over double the performance of a 4 pipe, dual TMU chip, then there's little reason to believe that a 4 pipe, single TMU can offer better performance than a 4 pipe, dual TMU board.
Again, I'm talking about AA and / or anisotropic situations here. I would expect that in non AA / Non Aniso situations, the GeForce4 would out-perform a R-300 "cut in half" type part. But then, everyone who owns a GeForce4 or this class of chip does so so that they can turn on Aniso and/or AA.
You can't just look at the number of pipes / TMUs...but the architecture as a whole. It's seems pretty obvious that the R-300 architecture was optimized for AA'd / Aniso Filtered pixels....that's where performance really shines.
According to ATI, Aniso on R-300 takes zero performance hit when used in conjunstion with bilinear. And "minimal" impact when pairing aniso with trilinear. Of course, I'd like to see tests bear the "zero" claim out (as well as define what ATI means by "minimal"), but the relative perforance tests between GeForce4 and R-300 with aniso / AA speak pretty mcuh for themselves.
I knew I had seen something of it!
If you watch the online broadcast of the launch they also show slides next to the video:
Seeing its it going to be DX9 and a performance part (with a number closer to the 9700 than the 9000!) then you would think the configuration must be more powerful than a 4x1 pipeline. Could they have a 4x2 or even a 6x1!
If you watch the online broadcast of the launch they also show slides next to the video:
Seeing its it going to be DX9 and a performance part (with a number closer to the 9700 than the 9000!) then you would think the configuration must be more powerful than a 4x1 pipeline. Could they have a 4x2 or even a 6x1!
Joe DeFuria
Legend
Thanks for those shots!
Though I disagree that the 9500 must be more powerful than a 4x1 configuration. As long as it retains the same relative performance characteristics of the 9700, a 4 pipe 9500 would still be faster than a GeForce4 Ti 4600 with aniso and / or AA enabled, plus offer DX9 support.
That's a damn good card if it MSRP's at $250 Max.
Though I disagree that the 9500 must be more powerful than a 4x1 configuration. As long as it retains the same relative performance characteristics of the 9700, a 4 pipe 9500 would still be faster than a GeForce4 Ti 4600 with aniso and / or AA enabled, plus offer DX9 support.
That's a damn good card if it MSRP's at $250 Max.
Indeed, at half the performance it would be unlikely to hurt their high end (unlike situation NVIDIA has at the moment with ti4200). But should still be more than powerfull and feature rich enough to let them dominate the ti4200, especially if it can run Doom3 faster even without AA (Not a sure thing, but not wholly unlikely.)
I wonder if NVIDIA will be able to introduce low end versions of their new architecture closer to the high end than usual ... with them missing the cycle they better.
(Not a sure thing, but not wholly unlikely.)I wonder if NVIDIA will be able to introduce low end versions of their new architecture closer to the high end than usual ... with them missing the cycle they better.
Doomtrooper
Veteran
The online videos are very good and the speed is great, it felt like I was right there with RealVideo
Joe DeFuria
Legend
It certainly would make sense to intorduce the cut down R-300 on the 0.13 process to cut more costs. 4Q seems to be about the time that 0.13 will be ready too.
Though I would doubt that the 8500 would come equipped with really high-priced memory. (To address your question, I don't know DDRII pricing.) More likely, IMO, something like 300-325 Mhz DDR on a $200-$250 part.
MrB said:the 9500 will not be just a lowerclocked 9700. What it is ATI wouldn't say and really even ati hasn't settled on the final specs of it. Or so they say
anand guess it to be a 4 pipeline version of the 9700 which doesn't sound too far off.
As for BGA 9000, AIB vendors who use 128MB of ram on their 9000 designs I believe are using BGA memory. I should've looked more carefully at the boards the 9 manufacturers were using.
Ah, ok. Thanks MrB... those are probably 958 boards, then. They aren't any different from the 964 boards AFAIK, apart from the memory packaging type. Interesting 9500 slide which I hadn't seen. So a 4-pipe 9700 sounds the most likely possibility after all.
No more ATi stuff from me, anymore. Going to move on to nVidia next.
MuFu.
P.S. No way were Inq reading my posts; they posted that "315MHz" information and I was desparately trying to find out if that was true or not! Then when I got news back that "some samples are running at 325MHz" I just sat back and enjoyed the show like everybody else. Hehe.
MuFu said:MrB said:the 9500 will not be just a lowerclocked 9700. What it is ATI wouldn't say and really even ati hasn't settled on the final specs of it. Or so they say
anand guess it to be a 4 pipeline version of the 9700 which doesn't sound too far off.
As for BGA 9000, AIB vendors who use 128MB of ram on their 9000 designs I believe are using BGA memory. I should've looked more carefully at the boards the 9 manufacturers were using.
Ah, ok. Thanks MrB... those are probably 958 boards, then. They aren't any different from the 964 boards AFAIK, apart from the memory packaging type. Interesting 9500 slide which I hadn't seen. So a 4-pipe 9700 sounds the most likely possibility after all.
No more ATi stuff from me, anymore. Going to move on to nVidia next.
MuFu.
P.S. No way were Inq reading my posts; they posted that "315MHz" information and I was desparately trying to find out if that was true or not! Then when I got news back that "some samples are running at 325MHz" I just sat back and enjoyed the show like everybody else. Hehe.
Ahh, a foul weather fan, eh?
Something like that, lol.
Plus I have a Ti4400 at the moment, so no upgrade for me in the near future (unless I stumble across a large wad of cash). NV30 should be very good. It'll never join the "Voodoo1/Voodoo2 SLI/R300" group though; definitely three product launches that will always stand out in terms of "-factor" in my mind.
MuFu.
Plus I have a Ti4400 at the moment, so no upgrade for me in the near future (unless I stumble across a large wad of cash). NV30 should be very good. It'll never join the "Voodoo1/Voodoo2 SLI/R300" group though; definitely three product launches that will always stand out in terms of "
-factor" in my mind.MuFu.
Mephisto said:
Whilst attempting to decrypt ATi's naming scheme's may not actually be possible, judging by the naming of RV250's (all 9000's with Pro's or whatever) I'm not sure that would be the case. Given that its a different number and scheduled to come out later it wouldn't surprise me if this is a different chip using a cut down R300 architecture.
DaveBaumann said:
Basicly true, the naming scheme would be changed. But the R200 was in production for about a year and now gets replaced with the RV250. From an economical point of view, it would not make sense to design, implement, mask and debug a chip if you introduce a similar but different chip which is much more attractive just two or three months later.
A very old ATI roadmap (which was accurate for R300 and RV250, so we can assume it will be for RV300 too) mentions an RV300 (4 pipe/2 VS) and a RL300 (4 Pipe/1 VS). I don't see how such a product would fit into the just introduced RV250/R300 line. An RV300 would reach about RV250 performance.
