View Full Version : The difference between NV31 and NV34?
So could someone please clarify me what's the difference between NV31 and NV34? NV31 is supposed to come in the first half of 2003 and NV34 a bit later (X-bit Labs (http://www.xbitlabs.com/news/story.html?id=1040641432) ). I hope NV31 is the more powerful version, cos I won't have the money for GFFX regular/ultra but I think I will get the money for GFFX 'mx400' ;) .
An old nVidia roadmap said the NV31 was mid-end while the NV34 was low-end. So, if you've got a budget between $199 and $299, it's reasonable to think the NV31 is for you. I expect to see 2 NV31 models ( $199 and $299 )
I'd expect the NV34 to be released fairly shortly ( 2 months maximum ) after the NV31, with two models too ( $99 and $149 )
Now, keep in mind that everything is speculative but the fact that NV34 is low-end and NV31 is mid-end.
Oh, and the reason he actually states the NV34 is expected to be after the NV31 is because this is a partly a financial conference. So, he says that because the NV31 is probably in Q1 and NV34 in Q2, thus giving financial contributions in different quarters and providing good financial results for both periods.
Uttar
Thanks Uttar :D .
So NV31 is going to be released shortly after the NV30? And how long is Q1?
It's going to be released in Q1 if Huang isn't too optimistic again. After all, before the Q3 conference call, he insisted the NV30 was to be in stores for the christmas season...
Financial Q1 for nVidia is from January 27 to April 28. Financial Q2 for nVidia is from April 28 to July 28.
Now, as to wether it'll be in stores in Q1, that's another matter. All that matters for the investors is that the financial contribution is in Q1. And that's when nVidia sends the chips to their partners. Not when their partners send it to stores.
Uttar
nutball
12-23-02, 12:02 PM
Right. Hopefully though given that at least one of NV31 and NV34 have taped out as of early December (IIRC), and being aimed at a cheaper segment of the market the cards should be easier to produce, the long painful birth of NV30 shouldn't be repeated.
Or am I being too optimistic? :)
Originally posted by nutball
Right. Hopefully though given that at least one of NV31 and NV34 have taped out as of early December (IIRC), and being aimed at a cheaper segment of the market the cards should be easier to produce, the long painful birth of NV30 shouldn't be repeated.
Or am I being too optimistic? :)
Well, yes, the yields are bigger when there are less transistors. However, for mainstream, the yields are required to be better to be able to sell it at a lower price point.
So, I really don't think this an advantage at all.
Actually, I think both NV31 and NV34 already taped out. It's just that, since NV34 will have even more volume than NV31, it's going to require more of them at launch. So there'll be more waiting time to get the chips & cards ready, since we need more of them. I'm really, really not sure of that, however. I'm not much into factory systems, so I don't know if it's really slower ( after all, each chip *is* smaller, too )
Uttar
I caught this from anandtech: "although NV30 (GeForce FX) was delayed - both NV31 and NV34 are on track for their scheduled release. NVIDIA's NV31 will be what's pitted against Radeon 9500 Pro but we'll have to wait and see about the performance of NVIDIA's as yet unannounced part." (AnandTech (http://www.anandtech.com/video/showdoc.html?i=1756&p=11) ) So based on "this" (my own guess), we could see NV31 in stores by the end of the Q1, or am I totally wrong?
And another thing in that statement caught my eye: "NV31 will be what's pitted against Radeon 9500 Pro" So what do they have against Radeon 9700 (in terms of performance and cost)? It's getting cheaper day by day so if they have NV31 only against R9500 Pro there would be a big hole in their market segment ( the 250 $ price range). Or will NV31 cost 250 $ and be significantly faster than R9500 pro?
nutball
12-23-02, 12:37 PM
No, from everything I've read, I think there's no real reason to think that either NV31 or NV34 will be (significantly) behind the schedule they would have been on had NV30 been available in September/October(-ish).
Clearly most of the problems with NV30 were been associated with TSMCs inability to deliver on 0.13u. There were of course these rumours about the number of pipelines having to be increased from 6 to 8, but for reasons of symmetry I don't believe them.
If the basic architecture of NV3x is sound (as in "it works"), and TSMC is now where they should be with yields on 0.13u, I don't see why there can't be a quick-fire series of launches in the high-, mid- and low-end over Q1.
This is one of the reasons I felt that the original slip in the launch of NV30 wasn't as hyper-critical to the future of NVIDIA as people made out. The real revenue earners remained on-track.
Originally posted by nutball
No, from everything I've read, I think there's no real reason to think that either NV31 or NV34 will be (significantly) behind the schedule they would have been on had NV30 been available in September/October(-ish).
Clearly most of the problems with NV30 were been associated with TSMCs inability to deliver on 0.13u. There were of course these rumours about the number of pipelines having to be increased from 6 to 8, but for reasons of symmetry I don't believe them.
If the basic architecture of NV3x is sound (as in "it works"), and TSMC is now where they should be with yields on 0.13u, I don't see why there can't be a quick-fire series of launches in the high-, mid- and low-end over Q1.
This is one of the reasons I felt that the original slip in the launch of NV30 wasn't as hyper-critical to the future of NVIDIA as people made out. The real revenue earners remained on-track.
I never said the NV31/NV34 would be delayed. I just said the yields problems are the same. So, if they're fixed for the NV30, they're fixed for the NV31/NV34. The only problem being that we got no 100% reliable source saying the yields at TSMC are fixed.
Maybe the 70% number is nothing more than a rumor and that nVidia realized they GOT to launch immediatly if they don't want to lose market share and reputation.
Uttar
nutball
12-23-02, 02:02 PM
Sorry Uttar, I wasn't disagreeing with your post when I said "No", but answering Netapes question "am I wrong". Guess I should have quoted! ;)
Originally posted by nutball
Sorry Uttar, I wasn't disagreeing with your post when I said "No", but answering Netapes question "am I wrong". Guess I should have quoted! ;)
Oh, sorry. Yeah, it's always better to quote, in case idiots like me look at your post :p
Uttar
What are the specs for NV31 (any rumors)? All I know is that it has DX 9 support :D
Originally posted by netape
What are the specs for NV31 (any rumors)? All I know is that it has DX 9 support :D
Well, I really don't think there are any such rumors.
The thing I wonder the most about, however, is the memory of the NV31 and NV34. DDR-II would sound way too costly... But then, if they used DDR, they'd have to redesign LMA around it. Probably just taking a modified GF4 LMA would be sufficent.
Uttar
Originally posted by Uttar
Well, I really don't think there are any such rumors.
The thing I wonder the most about, however, is the memory of the NV31 and NV34. DDR-II would sound way too costly... But then, if they used DDR, they'd have to redesign LMA around it. Probably just taking a modified GF4 LMA would be sufficent.
I thought it would be just like the NV30 but with lower clocks... Seems like I was hoping too much :D
Originally posted by netape
I thought it would be just like the NV30 but with lower clocks... Seems like I was hoping too much :D
Certainly not. If it only had lower clocks, it would still be a NV30.
For example, the GF3 Ti200 was still a NV20.
Uttar
Originally posted by Uttar
Certainly not. If it only had lower clocks, it would still be a NV30.
For example, the GF3 Ti200 was still a NV20.
But if they modify the LMA then it isn't a 'copy' of NV30 with lower clocks.
Originally posted by netape
But if they modify the LMA then it isn't a 'copy' of NV30 with lower clocks.
Yes, changing the LMA to be optimized for DDR-I would be sufficent to qualify it as a NV31.
However, that would still be way too costly to produce for the $199 price segment, even with significantly lower clocks.
Originally posted by Uttar
Yes, changing the LMA to be optimized for DDR-I would be sufficent to qualify it as a NV31.
However, that would still be way too costly to produce for the $199 price segment, even with significantly lower clocks.
So you'd think - but ATI managed to get the Radeon 9500 Pro down to that pricing, and all they had to do was a PCB change (did the number of layers decrease?). Here, nVidia can _really_ cut costs - I can't imagine GDDR-II is cheap stuff, and 128mb of it is quite a lot of money. If they were to simply cut it down to GDDR-I (or DDR-I, whatever), the cost savings would be enough to let them do it.
However, what they really need to do is come out with a GFFX MX at the $100-$150 price point. The GF2 MX was wildly popular because it was an amazing value at the time - they need to do that again to restore lost market share. If they were to take the GFFX, use DDR-I and eliminate half the pixel pipelines, I think they could cut costs to te necessary point. Just having DX9 compatibility will enable them to smack talk the Radeon 9000...
-Erwos
Originally posted by erwos
So you'd think - but ATI managed to get the Radeon 9500 Pro down to that pricing, and all they had to do was a PCB change (did the number of layers decrease?). Here, nVidia can _really_ cut costs - I can't imagine GDDR-II is cheap stuff, and 128mb of it is quite a lot of money. If they were to simply cut it down to GDDR-I (or DDR-I, whatever), the cost savings would be enough to let them do it.
However, what they really need to do is come out with a GFFX MX at the $100-$150 price point. The GF2 MX was wildly popular because it was an amazing value at the time - they need to do that again to restore lost market share. If they were to take the GFFX, use DDR-I and eliminate half the pixel pipelines, I think they could cut costs to te necessary point. Just having DX9 compatibility will enable them to smack talk the Radeon 9000...
-Erwos
ATI is *insane*
AFAIK, They're barely making any money on every card beside the Radeon 9500 non-pro and the Radeon 9000
The NV31 is not at the $100-$150 pricepoint. It's at the $199-$299 price point AFAIK.
The NV34, on the other hand, is at the $79-$149 price point AFAIK.
The NV34 is the true low-end part. The NV31 is still kinda mid-end.
Also, remember half the pixel pipelines does NOT mean half the transistors. Pixel pipelines, with DX9 GPUs, probably only take between 25% and 50% of the die.
That means a lot more got to be cut down, such as half the VS units and some Intellisample techs ( see my "NV3x: VSA Influences & Strategical Advantages" huuuge speculation if you're interested in what I think they might cut down and what advantages it gives them )
Uttar
Originally posted by netape
What are the specs for NV31 (any rumors)? All I know is that it has DX 9 support :D
I don't know of any rumours, but currently nearing qualification there are (at least) two NV31-based reference boards. They are succesors of the Ti4200 (NV25/28) and will most likely go head to head with the Radeon 9500 Pro and 9700.
The base model is clocked at 350MHz, the Ultra at 450MHz (Edit/Update - those are probably the targets, but I don't think they have fixed the clockspeeds yet due to a few problems). Both support DX9 (as you say) and the ASICs are fabbed on a 0.13u process. Both cards support dualhead output - a single or dual link panel and one or two CRTs (one is cloned, of course). As for other specs and the name I'm not sure - I'd guess they are based on the NV30 core with redundant/die space-inefficient functionality removed. The VS unit in particular may have been severely "retarded". GeForce FX5400/5400 Ultra anybody?!!
:confused:
MuFu.
Originally posted by MuFu
They are succesors of the Ti4200 (NV25/28) and will most likely go head to head with the Radeon 9500 Pro and 9700.
MuFu.
I hope they are competitive in price/performance ratio :D
I think we will see NV31 based boards for <200 dollars on pricewatch before the NV30 ones. Just remember how lots of people were left feeling ripped off after the GF4 MX was released. Wait for complete specs before you buy anything :)
Originally posted by Uttar
Also, remember half the pixel pipelines does NOT mean half the transistors. Pixel pipelines, with DX9 GPUs, probably only take between 25% and 50% of the die.
That means a lot more got to be cut down, such as half the VS units and some Intellisample techs ( see my "NV3x: VSA Influences & Strategical Advantages" huuuge speculation if you're interested in what I think they might cut down and what advantages it gives them )
Having recently taken a course in low-level design at college, I fully understand that it takes an ungodly amount of transistors to do anything efficiently.
However, I never ever meant that the halving of pixel pipelines was all they should do. Yes - chopping out some of the anti-aliasing stuff is another place they could improve. Removing some of the LMA stuff is another good place to go.
ATI doesn't need to make money off the Radeon 9700 as long as the Radeon 9500 and 9000 sell well. Do you really think nVidia's going to have huge profits off of GF FX? I doubt nVidia cares, the money's going to be from NV31 and NV34. I'd still enjoy hearing why you think ATI's making no money off the Radeon 9700, though.
-Erwos
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.