View Full Version : R480 Taped out
Omega53
09-30-04, 08:16 PM
http://www.digitimes.com/news/a20040930A7056.html
In related news, the sources said that ATI has completed the tape-out of its R480 chip, which is due to launch in the fourth quarter. The R480 is being manufactured at TSMC, using a 0.11-micron process.
grimreefer
09-30-04, 08:46 PM
same time as nv48
With .11 micron process i think we gonna see some pretty high core clocks on the R480 .
Diamond_G
10-01-04, 03:55 AM
Isn't the the X700 made on the .11 process? If so I thought ppl were talking about how it doesn't have a very bright outlook for overclockability... Or maybe I am just wrong... who knows...
Omega53
10-01-04, 04:07 AM
Isn't the the X700 made on the .11 process? If so I thought ppl were talking about how it doesn't have a very bright outlook for overclockability... Or maybe I am just wrong... who knows...
You are correct. They are moving to .11 cause its a lot cheaper to produce versus .13 -low k and yeilds will be a lot higher. As far as overcloability .11 can hit about the same speeds as .13-low k meaning dont expect much of a core jump. Since the nv48 is also going to be using .11 that means we will see a jump over current speeds with core speeds reaching around the same as Ati and probably higher since nVidia is known to use massive heatsinks :D
AthlonXP1800
10-01-04, 04:22 AM
Isn't the the X700 made on the .11 process? If so I thought ppl were talking about how it doesn't have a very bright outlook for overclockability... Or maybe I am just wrong... who knows...
Indeed X700 on 0.11 micron have the same problem as 6800 Ultra on 0.13 micron, X700 can overclock by 20MHz on 400MHz core but it cant reach 75MHz need for XT version, ATI now redesign the core with A2 revision attempt to achieve 475MHz with good yield. However 6600 on 0.11 micron are excellent on overclockability, it capable to overclock by 300MHz on 300MHz core to 600MHz, NV43 is very efficient design.
mustrum
10-01-04, 04:55 AM
The current R420 chips are 0.13 low-k. The R480 will be 0.11 but without low-k.
Don't expect too high clock rates. ATI allways tests the new process on their mainstream cards I highly doubt that we'll see a 600mhz XT PE. Hell i'd be surprised if we see a lot XT PEs anytime soon! :retard:
Diamond_G
10-01-04, 05:20 AM
So is the R480 supposed to be a budget part or is it acutally supposed to be faster than the R420?
Johnmcl7
10-01-04, 06:32 AM
The current R420 chips are 0.13 low-k. The R480 will be 0.11 but without low-k.
Don't expect too high clock rates. ATI allways tests the new process on their mainstream cards I highly doubt that we'll see a 600mhz XT PE. Hell i'd be surprised if we see a lot XT PEs anytime soon! :retard:
They've been selling the .11 X300s for a while now, although with it being a budget PCI Express card, I've not heard anything of them or their overclocking, so presumably they've been going ok.
Diamond: I'm guessing since it's R480 rather than RV480, so it should be faster than the R420.
John
Diamond_G
10-01-04, 06:45 AM
They've been selling the .11 X300s for a while now, although with it being a budget PCI Express card, I've not heard anything of them or their overclocking, so presumably they've been going ok.
Diamond: I'm guessing since it's R480 rather than RV480, so it should be faster than the R420.
John
well that is going to be interesting. I guess we are in for a nice suprise...
mustrum
10-01-04, 07:33 AM
They've been selling the .11 X300s for a while now, although with it being a budget PCI Express card, I've not heard anything of them or their overclocking, so presumably they've been going ok.
Diamond: I'm guessing since it's R480 rather than RV480, so it should be faster than the R420.
John
That's true. A mainstream card should be called RV480 if ATI didn't change their naming scheme.
I doubt that core clocks will be a lot higher though. I think they'll up the memory speed a bit. It should be possible to squeeze more than the XT PE's 1120mhz out of 1.6ns chips.
My guess: 560/600 clocked card. 600/600 would eb awesome but i amnot sure if they would be able to put out enough 600mhz chips. I highly doubt that they can afford another pathetic paper launch.
Diamond_G
10-01-04, 08:01 AM
mustrum: so my question is if they seem to be having issues getting the X700's clocked past 500 (and it seems to have less transistors) then how would they get a part with more transistors to be clocked higher? Is there something that I am missing?
mustrum: so my question is if they seem to be having issues getting the X700's clocked past 500 (and it seems to have less transistors) then how would they get a part with more transistors to be clocked higher? Is there something that I am missing? they probably needed to do a respin and a newer revison to fix some internal issues. but the market being what it is ati probably didnt have time so they were happy with what they had for that market segment. it just sounds like pure economics.
mustrum
10-01-04, 10:49 AM
Where did you hear that they have trouble getting past 500mhz?
It's very hard to believe since almost all XT PEs can go past 550mhz and some even up to 600mhz.
My modded VIVO (that didn't pass the test to get a XT PE) can do 550mhz on the core and it's slightly undervolting even (1.38v instead of 1.40v).
Most VIVOs are not able to hit high enough memory speed even with the 1.6ns memory. I think that's where they need to optimize. (I doubt it's bad memory chips)
Omega53
10-02-04, 03:51 AM
Where did you hear that they have trouble getting past 500mhz?
It's very hard to believe since almost all XT PEs can go past 550mhz and some even up to 600mhz.
My modded VIVO (that didn't pass the test to get a XT PE) can do 550mhz on the core and it's slightly undervolting even (1.38v instead of 1.40v).
Most VIVOs are not able to hit high enough memory speed even with the 1.6ns memory. I think that's where they need to optimize. (I doubt it's bad memory chips)
Its because they are using .11 instead of .13 low-k. It seems .11 is having a hard time getting past 500mhz
Where did you hear that they have trouble getting past 500mhz?
It's very hard to believe since almost all XT PEs can go past 550mhz and some even up to 600mhz.
My modded VIVO (that didn't pass the test to get a XT PE) can do 550mhz on the core and it's slightly undervolting even (1.38v instead of 1.40v).
Most VIVOs are not able to hit high enough memory speed even with the 1.6ns memory. I think that's where they need to optimize. (I doubt it's bad memory chips)
.11 process has no low-k at this time the fab isn't mature enough to set up higher clocks, Its bad enough ATi has issues with clocks with low k which should theoretically increase thier clocks by 30% but now you drop microns and drop low K.
Thier .13 part without low-k take of a 30% boost, where are they right at 400, where was the 9800xt? 418....
The x700 is on .11 they have less pipes less transistors, only reaching a max of 525 with artifacts. I wouldn't be surprised if they they don't go past 450 for thier new card if they make it.
nIghtorius
10-05-04, 12:17 PM
.11 process has no low-k at this time the fab isn't mature enough to set up higher clocks, Its bad enough ATi has issues with clocks with low k which should theoretically increase thier clocks by 30% but now you drop microns and drop low K.
Thier .13 part without low-k take of a 30% boost, where are they right at 400, where was the 9800xt? 418....
The x700 is on .11 they have less pipes less transistors, only reaching a max of 525 with artifacts. I wouldn't be surprised if they they don't go past 450 for thier new card if they make it.
there are many more aspects to the "clockspeed" story.. Let me give a good example.. The Intel Pentium IV processor @ .13u clockspeeds ranging from 2.4Ghz up to 3.4Ghz.. when the 2.4Ghz were released there was no 3.4Ghz version. But by fixing stuff on the core design (revising) they could bump the speed higher and higher. the same story goes for AMD. their first .13u processors weren't as highly clocked as their latest .13u counterparts.
Low-K and stuff can boost performance without making big revisions. But the most gains can be found in the design. hell.. with your reasoning the Athlon64 wouldn't go past the 500Mhz barrier. But they do.. they reach as high as 2600Mhz (stock versions, not OC'd ones) that's 2100Mhz more.
there are many more aspects to the "clockspeed" story.. Let me give a good example.. The Intel Pentium IV processor @ .13u clockspeeds ranging from 2.4Ghz up to 3.4Ghz.. when the 2.4Ghz were released there was no 3.4Ghz version. But by fixing stuff on the core design (revising) they could bump the speed higher and higher. the same story goes for AMD. their first .13u processors weren't as highly clocked as their latest .13u counterparts.
Low-K and stuff can boost performance without making big revisions. But the most gains can be found in the design. hell.. with your reasoning the Athlon64 wouldn't go past the 500Mhz barrier. But they do.. they reach as high as 2600Mhz (stock versions, not OC'd ones) that's 2100Mhz more.
cpus have alot less transistos and there are no strict guidelines for API's.
Graphics chips are limited in the way they are built because of this, and to keep the level of performance up aswell. ATi by dropping low k lose thier 30% clock advantage as I explained earilier. R300 core is pretty much at its limits with the x800, if you drop low k you will see the frequencies haven't changed from the 9800 to the x800. Going to .11 ATi needs at least a 30% increase in core speed, which .11 on itself can't provide, as you can see with the x700's less transitors still at 525 max. Now if ATi has modified thier core structure, how much of a change will it be? Look at nV they have a better architecture but still they maxed out at 450 with .13, which is right around where the x800xt pe would be without low k.
ATi needs a totally revised core to get higher clocks, its not a simple tweak. If it was they would have done it for the .13 process instead of going to low K as its very expensive to use low k.
nIghtorius
10-05-04, 04:30 PM
ATi needs a totally revised core to get higher clocks, its not a simple tweak. If it was they would have done it for the .13 process instead of going to low K as its very expensive to use low k.
Now you are saying basicly what I am trying to say.. Core design has also a big influence on core-speed.
Now you are saying basicly what I am trying to say.. Core design has also a big influence on core-speed.
yes but the r480 won't have a different design it can't be modified too much either unless its rebuilt from the ground up or there will be no major changes.
Hey guys calm down.........sheesh. :ORDER:
I understand both of your positions, but I think we should clear some things up a bit. Talking to ATi designers about the difficulties that all chip companies go to when scaling down to smaller transistor sizes, a company doesn't necessarily mean that is has to change chip design, but more accurately has to use additional technologies to decrease thermal leakage & decrease thermal density to increase chip speed, such as using improved substrates, or using IBM's SOI.
So, in the example of the R480, ATi does not have to redesign the core, it just has to use different tech that the transistors lie on within the chip that ATi hasn't used before.
nVidia has demontrated that the 110nm process can yield 540Mhz chips in volume, ATi will do the same, as well:
Gainward 6600GT w/ 540 core standard (http://www.theinquirer.org/?article=18877)
Well thats the thing what fab processes does TSMC have with .11? Does IBM have a .11 with SOI? How about strained silicon, I'm pretty sure IBM has strained silicon not SOI, but cost of strained silicon is like x4.
Its not like they can just add in SOI or Low-k to the fab process at moments notice. All the machinary has to be made for it.
ATi's cards compaired to nV (both the 6600 and ATi x700) shows the potential of different architecture vs. possible clocks, Plus the 6600 has more transistors to begin with so they should be getting hotter but they aren't.
Btw we were just discussing :)
You are right, but that doesn't mean it is easy to go to 110nm. Whatever problems are in the A1 X700's, they will be fixed in the respin. Ati will then use it's experience when volume producing the R480. There will probably be a respin of the R480 after paper launch in 4Q '04, as well, resulting in a 1Q '05 volume (my gut feeling).
BTW, processes like 110nm, 150nm and 220nm are really just "tweaks" on the 130nm, 180nm and 250nm processes repectively - they are physically using the same tools. You'll note that the CPU vendors traditionally don't bother with these optical shrinks and stick with the base node - Intel certianly went 180nm to 130nm and the 90nm.
coldpower27
10-05-04, 08:02 PM
The GPU's from vendors nowadays like to stay just a little bit behind on the micron process now. Considering everyone is suffering somewhat from the associated with the 90nm process ATI and Nvidia both need time to get their products under control on this process.
Straigned Silicon according to Intel doesn't cost them much more to implement. SOI though AMD seems to be keeping tight lipped about. ATI is tight lipped about the costs of low-k as well. According to speculation low-k black diamond and SOI are expensive technology to implement but strained silicon according to Intel is not.
At this moment in time it just looks like that Nvidia has got their 110nm process under control just a bit earlier then ATI, but ATI will follow suit as soon as it can, it had to spend resources working with 0.13micron without low-k and with low-k, while Nvidia only had 0.13 micron without low-k and earlier to boot then ATI did.
Though we all know at equal memory, pixel rates and vertex rates the NV4x architecture seems more efficient. Fortunately for ATI, on the X700 XT they have a huge advanatge on vertex rate, and slight on memory bandwidth.
From what we can see even if clock rates are only the same on the 0.11 micron process as the 0.13micron with low-k, it makes sense to go to it as it will be significantly cheaper thanks to reduced die size plus lack of low-k. I am quite sure after ATI get it under control, they will have a nice product that isn't expensive to produce.
We are probably gonna get another intermediate process grade before we hit 65nm with the GPU vendors probably the 75nm process.
vBulletin® v3.7.1, Copyright ©2000-2009, Jelsoft Enterprises Ltd.