View Full Version : R350 recently taped out
Bigus Dickus
11-22-02, 10:20 AM
From CMKRNL at Beyond3D, who has in the past been reliable information for tapeouts of both nVidia and ATi in overseas fabrication facilities, the following chips have recently taped out (last couple of weeks)
R350 on .15u
RV350 on .13u
NV31 on .13u
The NV31 and RV350 are hardly a surprise. The RV350 is probably similarly performing to the R300, and is aimed as a cheaper replacement to the R300 to accompany the R400. Smaller die, and possibly even a 128-bit DDR-II bus, depending on which route proves cheaper to acheive the target bandwidth (which I figure would be roughly the same as the current R300... by that time DDR-II would be considerably cheaper, so we'll have to see which route they take).
The real mystery here is the R350 on .15u. I had figured the R350 to be a myth, and that we would only perhaps see a clock bumped DDR-II equipped R300 doing battle with the NV30 in the first half of the year.
The R300 core is already seemingly pushing the limits on die size for that process, so you wouldn't think there is much room for large chunks of additional logic, like more vertex shaders or a second TMU per pipeline. They might try and match the advantage nV has in shader flexibility, but that is only matching PR and would give a marginal boost in performance at best. Unless they actually added more power to the pixel shaders, in the form of more operations per cycle. Still, the only way to improve current game performance is to increase raw VPU clock speed, memory speed, or effeciency. So long as ATi sticks with their sparsely sampled RGMS FSAA, it doesn't seem like huge leaps can be made in efficiency.
Perhaps it is only a R300 (with more shader flexibility, for PR sake) retooled for higher clockspeeds. Though this would have seemed unlikely when the R300 was first announced to be running at 325MHz, and people were surprised. But now that it has been seen that the current R300 core has a bit more room left... perhaps to 400MHz, then I suppose maybe 450MHz is not completely out of the question for the R350. A 450/450 part with a few more bells and whistles would probably compare favorably to a 500/500 NV30.
I'm still scratching my head on this one. Now that I know the R350 is not a myth, but is a real core on .15u, I'm having a hard time figuring out what it might be.
:confused:
jbirney
11-22-02, 12:05 PM
Well it will be an interesting ride thats for sure. I also am thinking CMKRNL either works at TMSC or has a hand in some dealings with them as he has been spot on in the last bits of info he has passed on.
Joe DeFuria
11-22-02, 12:54 PM
Yeah bigus, I have the same questions that you have. Allow me to explore some possibilities.
First of all, the "original" R-300 has been taped out for some time now. It has been at least 6 months since R-300 has taped out IIRC. Having said that, I don't predict any "major" architectural changes. Don't forget that the reason we were all surprised at the 325 Mhz clock, was because of the high 110 miilion transistor count. So no matter what R350 is...either faster clock or more features / transistors...it seems that ATI will have pulled another rabbit out of their 0.15u hat!
Now, I think it takes considerably more engineering to add new features, compared to tweaking the design for higher clock rate stability or bug fixing. So here's what I'm thinking:
1) No to significant new features. No second TMU. No advancement of pixel shader flexibility beyond R-300.
2) Yes to tweakage for clock rate, lower power consumption, and possibly the memory interface. (Similar to a GeForce3--3TI or perhaps even a GeForce3Ti-->GeForce4 Ti). We could not only be looking at a faster clock R-300, but one that is a bit more efficient clock-for-clock.
3) Outside possibility: fixing some things that "don't work right" on R-300. One of which might be supersample AA. Also possible that the pipeline moves from 96 to 128 bit color...as if that really mattered though except for marketing. But then...since when have these chip companies not been concerned about Marketing, right? ;)
So in a nutshell, I'm expecting R350 to be "the chip" paried with DDR-II memory. More or less an R-300 clocked at 400 Mhz plus, possibly more efficient clock for clock than R-300, and more remote possibility for pipeline tweakage for 128 bit color, supersample AA availability...
Joe DeFuria
11-22-02, 12:57 PM
450/450 part with a few more bells and whistles would probably compare favorably to a 500/500 NV30.
Almost forgot this one...
I would think a 450/450 R-300 would absolutely demolish a 500/500 NV30, not just compete favorably with it. ;)
suburbanguy
11-22-02, 04:13 PM
awesome. R350 taped out. i read this on Firingsquad.com the other day but it hadnt been confirmed then.
I think R350 will be a significantly enhanced core that's also clocked much higher and will have GDDR-3 or at least regular DDR-II as standard, and 256 MB of it.
* 8 pipelines / 2 texture mapping units per pipe
*more vertex shader pipelines
*many optimizations to pixel shaders
*additional shader functions to at least match NV30
*maybe 256-bit memory bus via 8x 32-bit controllers rather than 4x 64-bit controllers
*400~425 Mhz core
Bigus Dickus
11-22-02, 06:11 PM
Originally posted by suburbanguy
* 8 pipelines / 2 texture mapping units per pipe
*more vertex shader pipelines
*many optimizations to pixel shaders
*additional shader functions to at least match NV30
*maybe 256-bit memory bus via 8x 32-bit controllers rather than 4x 64-bit controllers
*400~425 Mhz core
I seriously doubt they will go to 2 TMU's per pipe, and they probaby can't squeeze more vertex shader units onto a .15u die (because the logical step is to 8 vertex units, or 6 at the very least).
I also doubt they will go with an 8 x 32 bit memory controller.
I do think they will probably increase the flexibility of the pixel shaders somewhat, but most emphasis will probably be on ramping clockspeed. The R300 is a very capable chip, already an equal architecturally to the NV30. If they could ramp core speeds to 450MHz, then it should have no problem competing.
Another thought is that perhaps they'll increase the internal bandwidth to 1024 bits, so that it can use DDR-II memory on a 256 bit bus clocked considerably higher than the core. A 400/550 card would be a monster.
ReDeeMeR
11-23-02, 10:18 AM
Um wouldnt it be logical to fully move to 0.13u? I mean wouldnt that be even more expensive to handle 0.15 and 0.13 production at the same time?
0.13u high end and low end R350 and R325? even with DDRII @ 256 bit Nv30 is going to be blown away.
StealthHawk
11-23-02, 08:59 PM
Originally posted by ReDeeMeR
Um wouldnt it be logical to fully move to 0.13u? I mean wouldnt that be even more expensive to handle 0.15 and 0.13 production at the same time?
no, how do you arrive at that conclusion?
there are a lot of other factors to consider besides manufacturing costs, and i don't even see why it would be more expensive to produce .15 and .13 parts at the same time. it shouldn't be. it takes time and money to develop and design a part for .13, so if there is still headroom in .15 it makes perfect sense to just release a slightly tweaked product at .15 instead of going for the more extensive process of moving down to .13
-=DVS=-
11-24-02, 12:21 AM
R350 = Dual chip R300 :eek: :D *DROLL*
that would be some monster :p
could do 8x AA easy :p
would have insane fillrate 325 core cloack x2 = 650 :rolleyes: (not that much over NV30 but still )
256MB of DDR or maybe 512MB :p
would cost 1 arm & 1 leg to buy :D
The Baron
11-24-02, 12:39 AM
From what I'm hearing right now, I get the feeling that R350 is being rushed as fast as humanly possible in order to compete with NV30 directly, not this 5 month later thing nVidia's doing with NV30... so I'm guessing it's just going to be a higher clocked R300. Nothing too special there :p
That's also the reason why R350 is going to be .15 instead of .13.. look at the delays caused by nVidia going .13. If R350 goes .13, it's going to have similar delays, during which time nVidia can and will play catchup.
Dual chip R300 is probably pretty much out due to heat and power consumption.
More than likely, it will have improved AA, maybe improved PS/VS performance/flexibility, but I'll be surprised if it has DDR2.
But, that being said, I'm going to guess that it launches about a month after NV30.
Bigus Dickus
11-24-02, 01:54 AM
Well, if they actually bump core speeds it almost has to use DDR-II to keep the part balanced.
Lol, a dual R300 chip. Your power supply would probably just put a message up on screen when trying to boot: "screw this, I give up."
:D
The Baron
11-24-02, 08:37 AM
Nah. No message.
The power supply just explodes as soon as you put in the card--no AC power required. :p
-=DVS=-
11-24-02, 02:27 PM
Originally posted by =SSC=The Baron
Nah. No message.
The power supply just explodes as soon as you put in the card--no AC power required. :p
Nah all you would need is Second Power Supply 1 for Dual Radeon another for rest of the PC :D :p hehe
The Baron
11-24-02, 02:52 PM
I'd much rather have a cold-fusion powered R9billion, thank you very much. :p
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.