View Full Version : Full 16 pipelines
I doubt it takes that much to implement low-k in their products skuzzy.. but I may be over-simplifying things a bit...
It is a process you have to design for Sazar. Great care must be taken in the length of interconnects and power to ground considerations must be observed as well. Dissipation of power is another design concern.
When you change the basic resistance constant of the material, you have to take that into design consideration.
EDIT: After reading this, I figured an easier way to say it; Changing the resistance of material alters the timing of all the circuits, so you have to design for that timing change. Sometimes it is as simple as a new layout, more often than not it is a redesign.
CapsLock
02-27-04, 04:21 PM
Originally posted by Skuzzy
It is a process you have to design for Sazar. Great care must be taken in the length of interconnects and power to ground considerations must be observed as well. Dissipation of power is another design concern.
When you change the basic resistance constant of the material, you have to take that into design consideration.
EDIT: After reading this, I figured an easier way to say it; Changing the resistance of material alters the timing of all the circuits, so you have to design for that timing change. Sometimes it is as simple as a new layout, more often than not it is a redesign.
C'mon Skuzzy. As much as I'd like to picture NV engineers with mallets and stone tablets :p All of your difficulties are fundamental design considerations no matter what kind of chip your making. Timing is always going to be difficult with clock shifts happening throughout the chip. Power dissipation is also always a major consideration as is the length of the interconnects. I'm not an expert by ANY means, but I doubt NV engineers, who moved to .13 first, aren't capable of learning new tricks.
Caps
I never said they were not capable. I question the press release information about it being too risky to move to low-k. I think there is more to it than that.
My other post was more in reference to Sazar's about moving to low-k. It's not as simple as taking an existing design and just compiling it. It has to be redesigned.
Speculation: I think NVidia tried to go low-k, but the design could not be manufacturered with acceptable yeilds. They did start with a complex design. ATI got low-k working, but they started with a much simpler design and that may give ATI a leg up in the next go round.
Let's put it this way; If ATI rolls the R420 out and it is low-k, why can't NVidia roll the NV4x out with low-k? It's all speculation at this point. What would your take on it be?
bkswaney
02-27-04, 05:09 PM
Well, My money is on 8x2. ;)
If so the R420 and NV40 should be very close in performance.
I just hope the NV40 brings better AA.
If not I'll go with the 420 just for the better AA.
Does anyone know for sure if Nvidia has shipped
any NV40's out to the review sites yet?
If so we should start hearing from some of the over seas
off the wall sites soon on specs. ;)
When is Cebit??
CapsLock
02-27-04, 06:28 PM
Originally posted by Skuzzy
I never said they were not capable. I question the press release information about it being too risky to move to low-k. I think there is more to it than that.
My other post was more in reference to Sazar's about moving to low-k. It's not as simple as taking an existing design and just compiling it. It has to be redesigned.
Speculation: I think NVidia tried to go low-k, but the design could not be manufacturered with acceptable yeilds. They did start with a complex design. ATI got low-k working, but they started with a much simpler design and that may give ATI a leg up in the next go round.
Let's put it this way; If ATI rolls the R420 out and it is low-k, why can't NVidia roll the NV4x out with low-k? It's all speculation at this point. What would your take on it be?
Now that is a good q. And it does blow a possiblly large hole in my reasoning. I'm afraid I just don't know enough to speculate why they might choose not to go that way. Perhaps there really is a yield problem with larger designs (edit:and Ati have decided for it because other counterbalancing reasons ie. clock speed or that new high speed trans process HB was talking about)? Is it more expensive a process? They weighed the engineering time involved and decided it would be more effectively spent elsewhere? TSMC's low-k production capacity? They think it sounds like a wimpy fad diet and thus is not easily marketed? :p
Got me.
Caps
Originally posted by bkswaney
Well, My money is on 8x2. ;)
If so the R420 and NV40 should be very close in performance.
I just hope the NV40 brings better AA.
If not I'll go with the 420 just for the better AA.
Does anyone know for sure if Nvidia has shipped
any NV40's out to the review sites yet?
If so we should start hearing from some of the over seas
off the wall sites soon on specs. ;)
When is Cebit??
Well NV40 must have new AA methods as the present AA which was around from GF4 is just not cutting it. We all know ATI's AA is better. Hopefully NV will go better or at least be the same as ATI's. That is also speculation as R420 might have a new AA unit too. But the use of high ass, i mena speed memory only make me keener to what is in store. I sure hope we get the higher forms of AA of less of a penalty due to the new higher memory bandwidth. :D
/me giggles about memory bandwidth and AA
ooops
:wtf:
On low K thing, NV will most probably use it in their refresh products using 110nm as they will be using TSMC's process as you guys know. Low K give you what exactly? Ability to run higher clocks with better thermal characteristics? At least thats what I thought. If NV can get away without using low K for NV40, hey no biggie. It all depends how it compares to a R420 ;)
Does anyone know how many FP op's NV40 can do per clock with pixel shaders? I really don't care how many pipes it's got, as long as it has a reasonable amount, I just hope it has better FP performance this time around.
Originally posted by Skuzzy
I never said they were not capable. I question the press release information about it being too risky to move to low-k. I think there is more to it than that.
My other post was more in reference to Sazar's about moving to low-k. It's not as simple as taking an existing design and just compiling it. It has to be redesigned.
Speculation: I think NVidia tried to go low-k, but the design could not be manufacturered with acceptable yeilds. They did start with a complex design. ATI got low-k working, but they started with a much simpler design and that may give ATI a leg up in the next go round.
Let's put it this way; If ATI rolls the R420 out and it is low-k, why can't NVidia roll the NV4x out with low-k? It's all speculation at this point. What would your take on it be?
just to clarify... I never said it was easy... I just said it was unlikely to be as difficult as you seemed to make it appear skuzzy :)
then again I may have just made an incorrect assumption on your part..
CapsLock
02-28-04, 07:27 AM
Originally posted by Sazar
just to clarify... I never said it was easy... I just said it was unlikely to be as difficult as you seemed to make it appear skuzzy :)
then again I may have just made an incorrect assumption on your part..
Since Skuzzy isn't putering right now (pfft, some people and thier so called "real lives"): His last Q to me was, so how come they didn't use it then? Which is a reasonable argument.
Caps
Voudoun
02-28-04, 08:10 AM
Originally posted by CapsLock
Since Skuzzy isn't putering right now (pfft, some people and thier so called "real lives"): His last Q to me was, so how come they didn't use it then? Which is a reasonable argument.
Caps
Might it be something as simple as saving face? It wasn't that long ago that nVidia said that low-k was dangerous.
Unless they've got some problem designing for it, it's hard to guess what else it might be. After all, aren't the benefits so high that it's ridiculous not to implement it if you can?
Voudoun
CapsLock
02-28-04, 09:32 AM
Originally posted by Voudoun
Might it be something as simple as saving face? It wasn't that long ago that nVidia said that low-k was dangerous.
Unless they've got some problem designing for it, it's hard to guess what else it might be. After all, aren't the benefits so high that it's ridiculous not to implement it if you can?
Voudoun
Well, as I was trying to say to Skuzzy, on the one hand the above seems evident, but on the other hand, how could NV engineers be that far behind Ati's? Designing the layout and principles of a chip is going to be very complex but employing a new process technique, especially when they were the first to have .13, how could they be that out of it? Some possible reasons why they might not want to do low-k I already posted. However I can't really say with any certainty if any of them are valid. Maybe NV engineers are lacking a skill level, and maybe they're not. :confused:
Caps
Originally posted by MUYA
On low K thing, NV will most probably use it in their refresh products using 110nm as they will be using TSMC's process as you guys know. Low K give you what exactly? Ability to run higher clocks with better thermal characteristics? At least thats what I thought. If NV can get away without using low K for NV40, hey no biggie. It all depends how it compares to a R420 ;)
low-k dielectrics allows for higher clocks... less cross-chatter of signals... increased interconnect performance et al.. and perhaps lower manufacturing costs (less cross-chatter == less layers for the pcb... PERHAPs.. this is hypothetical and possibly does not apply due to the nature of the dielectric material)
ati's 9600pro series showed that clearly with 2 things.. its uncomplicated pcb layout and the clocks it was easily able to obtain..
the more complex a card... the more complex the process as skuzzy has said and for a card like the nv4x lineup's big boy's you are likely to spend a fair amount of time dealing with it... I understand I think the point he was trying to make... sry skuz... mis-read your post :)
nv are not going to be able to just say they want to implement low-k and do it half way through... there are too many considerations...
if ati manages to release a complex part like the r4xx big boys using the low-k process they have a technological lead imo.. seems like it was a good idea to go with low-k on the 9600 series...
Originally posted by CapsLock
Well, as I was trying to say to Skuzzy, on the one hand the above seems evident, but on the other hand, how could NV engineers be that far behind Ati's? Designing the layout and principles of a chip is going to be very complex but employing a new process technique, especially when they were the first to have .13, how could they be that out of it? Some possible reasons why they might not want to do low-k I already posted. However I can't really say with any certainty if any of them are valid. Maybe NV engineers are lacking a skill level, and maybe they're not. :confused:
Caps
well to be fair with skuzz... ati have already shipped a full-working product using low-k... they started with the low-end...
if nv was to introduce low-k on a high end product it would be quite brave of them... low-k is not so easy as ati did find out (afaik there are tidbits bout delamination and additional layer/s to counter the fragility) but they managed to put forth a good product @ the end of the day...
CapsLock
02-28-04, 03:02 PM
Well as someone whose worked with plywood and counter top laminates, I can understand how de-lamination is not a good thing. But thats about as far as it goes. (:rolleyes: at self)
Caps
thatdude90210
02-28-04, 07:03 PM
There's a subject at Beyond3d that might shed some light why ATI has better luck with new processes, it's all about the IC layout: http://www.beyond3d.com/forum/viewtopic.php?t=10440&start=60 (scroll down to Basic's post).
I hope the 16 pipes rumor is true. It means that Nvidia will be really pushing hard to take back the crown.
Without low-k it will be very interesting to see what kind of clocks they can get. It will also be interesting to see if the R420 can match it assuming fewer pipes and higher clocks by using low-k.
It should be a very interesting and I suspect very close battle.
The good news is that no matter which side wins the eventual winner is the consumer! :D
Originally posted by Moose
I hope the 16 pipes rumor is true. It means that Nvidia will be really pushing hard to take back the crown.
Without low-k it will be very interesting to see what kind of clocks they can get. It will also be interesting to see if the R420 can match it assuming fewer pipes and higher clocks by using low-k.
It should be a very interesting and I suspect very close battle.
The good news is that no matter which side wins the eventual winner is the consumer! :D
soi is not a slouch of a process... and nvidia has had little problem ramping up clocks wrt ati...
however low-k maybe something they were interested in and had to rethink concerning ibm's implementation... afaik ibm did not go with black diamond (or whatever its called) and went with their own process and realised it wasn't gonna work so they re-did the whole thing...
@ the end of the day perhaps nvidia were relying on ibm's low-k process? who knows... I am sure we'll find out eventually..
Originally posted by Moose
The good news is that no matter which side wins the eventual winner is the consumer! :D
Hopefully yes we the consumers are. Suffice to say the top end cards will be uber expensive. Mainstream derivatives which are traditionally half the chip the high end ones are are gonna be intersting :D i mean if the nv4X mainstream has half the supposed pipelines etc of the nv40...that would rock. Something close to 8 pipelines etc and would give present day high end nv3X (or r3XX) costing more a run for their money. :D
GlowStick
03-01-04, 07:28 PM
Originally posted by Moose
I hope the 16 pipes rumor is true. It means that Nvidia will be really pushing hard to take back the crown.
Without low-k it will be very interesting to see what kind of clocks they can get. It will also be interesting to see if the R420 can match it assuming fewer pipes and higher clocks by using low-k.
It should be a very interesting and I suspect very close battle.
The good news is that no matter which side wins the eventual winner is the consumer! :D I belive low-k is not important at all, but you should also consider that IBM makes the G5 cpu's that run at 2GHZ and they say they can scale them to 3ghz, and also they have SOI!
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.