NV30 Do we have any info on it?

Intelman07

Senior member
Jul 18, 2002
969
0
0
I have seen alot about and i wanted to know if there are any facts? What do we know and what don't we know?
 

Rand

Lifer
Oct 11, 1999
11,071
1
81
Take your choice of rumours.... precious little has been confirmed besides that it will utilize DDRII.

The most poular theory at the moment....
400MHz+ core clockspeed.
500MHz DDRII
8 pixel pipelines w/2 TMU/pipe.
128bit memory controller.
3 vertex shaders units.
3D feature set slightly beyond DX9 requirements.
 

McManus

Junior Member
Apr 8, 2002
6
0
0
Sounds interesting... do y'all think they'll get 3-monitor support in? I think that's the thing I'm hoping for the most... i.e., something like TwinView, but instead of 2 RAMDACs, how 'bout 3? That would just really tickle my fancy, cuz right now I'm running two monitors off of my GF4 Ti4600, and I have a 3rd monitor coming off of a PCI GF2 MX... but I have stability issues with this setup :/

I also hope they keep up the awesome driver support (especially in Linux!)
 

WyteWatt

Banned
Jun 8, 2001
6,255
0
0
Rand yep the NV30 will be using DDRII now sense its offically from nvidia we know its not a rumor. But we still don't know if its 128 bit , 256 bit, or whatever. I guess we will see at the Comdex or offically by nvidia soon.
Rand did you like the last video?

 

Deeko

Lifer
Jun 16, 2000
30,215
11
81
Another long-standing rumor is that it contains 3dfx technology.
A more recent twist on this rumor is that that 3dfx technology is Gigapixel's deferred rendering techniques, which would have been utilized by the 3dfx Fear
 

chizow

Diamond Member
Jun 26, 2001
9,537
2
0
This is gonna be a crazy friggin card.....gonna cost another kidney :(

Chiz
 

WyteWatt

Banned
Jun 8, 2001
6,255
0
0
I am not holding my breath because i consider the article by the inquirer a rumor which it is. Its not officially by Nvidia themselfs. Except the DDR II thing and a few facts + quotes on the are you ready page are the only official things by nvidia that we know for sure. All then rest are rumors for now.



 

Deeko

Lifer
Jun 16, 2000
30,215
11
81
I don't feel like going to inquirer...are they claiming 256MB? Highly, highly doubtful. Do you know how expensive a car with 256MB of DDRII would be? Plus it just isn't needed. From the rumors I've seen, in current games, it may not be that much faster than the Radeon 9700. It is only 128bit, compared to the 256 in the 9700, and has fewer vertex units. However, it has a higher clock speed, and dual texture units. May be close.
 

Deeko

Lifer
Jun 16, 2000
30,215
11
81
Actually, it has been in the works for alot longer than 6 months, so it hasn't really been designed to beat it. Maybe held back and tweaked. It should be faster, and it better be, because a)it is coming out 6 months later and b) it will carry a price premium(probably like $400), and the 9700 will be well below $300 by then. It will definitely be faster in next gen games, but in current(Dx7) stuff, I don't think the gap will be as large.
 

Deeko

Lifer
Jun 16, 2000
30,215
11
81
That is going by the assumpion that it is an immediate mode renderer. If it is based on Gigapixel's deferred rendering, it will be much better.
 

Rand

Lifer
Oct 11, 1999
11,071
1
81
Originally posted by: Deeko
I don't feel like going to inquirer...are they claiming 256MB? Highly, highly doubtful.

Na, the Inquirer is claiming the NV30 has over double the memory bandwdith of the Radeon 9700Pro
rolleye.gif

Something that is even more dubious then 256MB in my opinion.

It is only 128bit, compared to the 256 in the 9700, and has fewer vertex units. However, it has a higher clock speed, and dual texture units.

Well we don't know for a fact any of that. It's only supposition based on nVidia's vague comments, wild rumours and leaked information and some data gained from nVidia's CG and CineFX showings.
Specifically I personally tend to doubt the widely held rumour that it will only have three vertex shaders units, though admittedly going by nVidia's claims of 2.5X the vertex shading performance of the GF4 Ti4600 this does seem the likely amount.

check out the inquirer article today

WOW 2x memory of the 9700

The Inquirer is twisting nVidia's claims.
ven nVidia themselves have NEVER claimed the NV30 has double the effective memory bandwidth as the R9700Pro.
They've claimed it has 46GB/s effective memory bandwidth... which is over double the R9700Pro's physical bandwdith but says absolutely nothing about what the R9700Pro's effective bandwdith is, which is what one would need to know to make a comparison.

If the Inquirer wants to count non-utilized bandwidth into it's effective bandwidth figure for the NV30 then they should do the same for the R9700.
Once you take into account occlusion culling, overdraw etc, the R9700Pro would suddenly put forth a good 55+ GB/s using the same "effective" bandwidth
scheme the Inquirer is using.
To be even more picky.... the Inquirer is not even talking about "effective bandwidth" at all, effective/real world bandwidth is at best equal to the maximum physical bandwidth. What they are referring to is closer to equivalent bandwidth.... equivalent in this case being relative to the bandwidth required to render a scene with X amount of over-draw at the same speed as it would take were there no over-draw at all.


If there is one thing that virtually everyone agrees upon it is that the NV30 has a 128bit memory controller...., if such is the case then it will need 620MHz DDR-II to match the R9700Pro's physical bandwidth.
Not even nVidia's wildest PR claims have ever mentioned beyond 500MHz DDR-II... and there is considerable doubt as to whether Samsung can even supply THAT in volume let alone 620MHz DRAM!

The Inquirer can twist their words any way they like but it's a virtual certainty that the NV30 will have less physical bandwidth then the R9700Pro if the widely held 128bit memory controller theory holds true.
The most popular theory claims 127bit memory controller with 500MHz DDR-II... yielding a peak physical bandwidth of 16.0GB/s, slightly under 4GB/s less then the R9700Pro is capable of. Slightly under 2GB/s less then even the reg. R9700 is capable of.
Now the NV30 may make much more efficient usage of it's available bandwidth then the R9700 does, which should make the "real world"/"effective" bandwidth reasonably close between the two.
 

Deeko

Lifer
Jun 16, 2000
30,215
11
81
Na, the Inquirer is claiming the NV30 has over double the memory bandwdith of the Radeon 9700Pro
Something that is even more dubious then 256MB in my opinion
Ah, yea that's even dumber.

Well we don't know for a fact any of that. It's only supposition based on nVidia's vague comments, wild rumours and leaked information and some data gained from nVidia's CG and CineFX showings.
Specifically I personally tend to doubt the widely held rumour that it will only have three vertex shaders units, though admittedly going by nVidia's claims of 2.5X the vertex shading performance of the GF4 Ti4600 this does seem the likely amount.
Yea, that's just assuming the rumors are true. We'll see.

What Rand said about memory bandwidth is correct. With half the bus width, the NV30's memory would need to be running at twice the speed of the 9700's in order to have even the same speed.
 

Deeko

Lifer
Jun 16, 2000
30,215
11
81
There is a thread about that article in GH. I've already posted my thoughts there, but I will here as well.

They claim NV30 has over twice the memory bandwidth of the R300. Total Bull.
NV30 - 500MHz DDRII, 128-bit controller
R300 - 310MHz DDRI, 256-bit controller
NV30 would need 620MHz DDRII to have ever the same bandwidth.

They claim that with FSAA on, it is 30% faster than R300. Even if that is the case, that is NOT that spectacular. With FSAA/ansio, R300 is 200-300% faster than the Ti4600. You want to tell me that that little of a performance gain is worth 2x the cost?

They use 3DMark!! That discredits them.

The NV30 will be faster than R300, and it better be, coming out 6 months later. But will it be enough? It has less memory bandwidth, and a weaker vertex unit. One thing that could save it, and totally eliminate the bandwidth issue is if it uses deferred rendering, acquired from 3dfx. I guess we'll see.
 

Lonyo

Lifer
Aug 10, 2002
21,939
6
81
Does the effective card bandwidth matter if the AGP slot doesn't have the bandwidth?
R300 will still be the better by cost wise when the NV 30 comes out, plus until Doom 3 the 9700 should hopefully cope, and then there's the revision of the R300 core, with smaller manufacturing process, possible DDR II and any other tweaks, which could happen around the same time as the NV 30 release.
 

Rand

Lifer
Oct 11, 1999
11,071
1
81
Originally posted by: Lonyo
Does the effective card bandwidth matter if the AGP slot doesn't have the bandwidth?

Yes it will. Bandwidth from memory to GPU is totally unrelated to the AGP bandwidth GPU to main system memory.


The NV30 will be faster than R300, and it better be, coming out 6 months later. But will it be enough? It has less memory bandwidth, and a weaker vertex unit. One thing that could save it, and totally eliminate the bandwidth issue is if it uses deferred rendering, acquired from 3dfx. I guess we'll see.


Personally, depsite the popular rumours I find it highly dubious the NV30 will have only 3 vertex shaders. In fact, anything outside of the typical order of 2 would be extremely unlikely. It'll either be 2 or 4 vertex shaders in my opinion... and I'd think the latter more likely.
Less memory bandwiodth... well I agree with that, as does virtually everyone else.

I'm betting against it going the route of an entirely deferred renderer. I'm expecting it'll still be an im,ediate mode renderer, but I do think quite a few bandwidth saving measures courtesy of GigaPiexel will find their way in.... enough so that I wouldnt be altogether suprised if the NV30 actually managed a very slight real world bandwidth advantage.
 

Deeko

Lifer
Jun 16, 2000
30,215
11
81
Personally, depsite the popular rumours I find it highly dubious the NV30 will have only 3 vertex shaders. In fact, anything outside of the typical order of 2 would be extremely unlikely. It'll either be 2 or 4 vertex shaders in my opinion... and I'd think the latter more likely.
Less memory bandwiodth... well I agree with that, as does virtually everyone else.

I'm betting against it going the route of an entirely deferred renderer. I'm expecting it'll still be an im,ediate mode renderer, but I do think quite a few bandwidth saving measures courtesy of GigaPiexel will find their way in.... enough so that I wouldnt be altogether suprised if the NV30 actually managed a very slight real world bandwidth advantage.
3 does seem a bit odd, I dunno, we'll see. As for deferred rendering, it's just a rumor, I remember back when it was a new idea, nvidia seemed extremely against the idea. So unless they've radically changed their stance, it is doubtful, but they do need SOMETHING to fix their lack of bandwidth.
 

JellyBaby

Diamond Member
Apr 21, 2000
9,159
1
81
Originally posted by: Intelman07
I have seen alot about and i wanted to know if there are any facts? What do we know and what don't we know?
It will own you. It will obliterate the competition. You suck if you don't get one.

These are all the facts you really need.