NV30 details - one monster of a VPU

[Dark_Archonis]

Well, I was going to wait until monday when Nvidia would officially release the NV30, but I just can't resist. It seems all the details (or most of them) of the NV30 are now known. First off, here are a few links about the details:

<A HREF="http://www.theinquirer.net/?article=6260" target="_new">http://www.theinquirer.net/?article=6260</A>

<A HREF="http://www.geek.com/news/geeknews/2002Nov/bch20021115017345.htm" target="_new">http://www.geek.com/news/geeknews/2002Nov/bch20021115017345.htm</A>

<A HREF="http://www.geek.com/news/geeknews/2002Nov/bch20021106017159.htm" target="_new">http://www.geek.com/news/geeknews/2002Nov/bch20021106017159.htm</A>

Of course, the most exciting news of the NV30 is that the DDRII memory (running at 1ghz) Nvidia is using combined with LMA III will give a memory bandwith of <b>48GB/s</b> according to these sources. That is simply amazing! If this is in fact true, than means the NV30 will have twice as much memory bandwith as the 9700 Pro, <b>even though</b> it's using a 128 bit memory interface, compared to the 9700 Pro's 256 bit interface. Imagine if NV30 was running on 256 bit interface, OMG...

On top of this, the links say that using max FSAA AND aniso filtering, the NV30 will STILL be 30% faster than the 9700 Pro. This is taking into account that NV30 will use 64 and/or 128 bit color precision, and will support 128 tap aniso filtering along with FSAA 8X, which apparently leave ATI's methods in the dust.

Again, I know many people may have not wanted this to be spoiled, but it was just too tempting. I couldn't believe my eyes when I see the "supposed" bandwith.

The only thing that is NOT certain is the name. It's rumoured to be Geforce FX...

- - -
<font color=green>All good things must come to an end … so they can be replaced by better things! </font color=green>

 

[johnnyx]

holy F**k!

If an orange was driving a racecar would it peel out? www.jxfiles.com

 

[AMD_Man]

First of all, you're getting too hyped about some extreme misunderstand on the Inquirer's part, of all sites. First of all, the NV30 doesn't have 48GB/s bandwidth. It doesn't even have 32GB/s bandwidth. Assuming it has 1GHz 128-bit DDRII, then it has 16GB/s of REAL BANDWIDTH. The claimed 48GB/s bandwidth is the effective bandwidth with the bandwidth saving techniques at work. If nVidia is really going to keep that 48GB/s claim by Monday, then that is simply marketing PR non-sense. For all we know, ATI has 100GB/s effective bandwidth by the same rating scheme. We just don't know. There is no way to compare this bandwidth rating scheme with ATI's conventional rating scheme. If nVidia wants to artifically inflate their specs, then it's simply to hype the product.

On top of this, the links say that using max FSAA AND aniso filtering, the NV30 will STILL be 30% faster than the 9700 Pro. This is taking into account that NV30 will use 64 and/or 128 bit color precision, and will support 128 tap aniso filtering along with FSAA 8X, which apparently leave ATI's methods in the dust

I still don't understand...

You realize ATI already using 128-bit floating point colour precision internally?

If at max settings, the NV30 is only 30% faster than the R9700Pro a whole 6 months after the R9700Pro was launched, then it will not hold the crown for very long as ATI already has working DDRII samples of the R9700.


My advice, don't listen to the rumours...wait for the previews on Monday and the reviews in late January or early February.


Intelligence is not merely the wealth of knowledge but the sum of perception, wisdom, and knowledge.

 

[eden]

I remember hearing the NV30 is four times more performing than the GF4, based on Hard News. I am not sure how that reflects.

The bad news is that apparently running a Ti4600 in 640*480 in Doom III got 70FPS in 1 Hardware Light, and below 10FPS for 8. That's not good at all!

However, I trust nVidia will have a kicker card here, the features that were revealed so far seem very advanced, Cg also supports 65K vertex instructions compared to the current 128!
I am sure their strategy was rethought when they knew about the DDR II samples from ATi. The card will especially kick if it has 2 TMUs on 8 Pipes. That will leave ATi's MTxt performance in the dust with the Ti4600. Here's hoping this competition creates low prices and high tech cards we can get to play Doom III!
--
*You can do anything you set your mind to man. -Eminem<P ID="edit"><FONT SIZE=-1><EM>Edited by Eden on 11/15/02 11:37 PM.</EM></FONT></P>

 

[Dark_Archonis]

Yes, I realize that the 9700 Pro has 128 bit FP color already. I never said it didn't. And anyways, how are you so sure that the NV30 doesn't have this type of bandwith? The memory speed is not the ONLY factor that determines the bandwtih. You'd be surprised as to how much bandwith Nvidia can squeeze out of it's cards.

Anyways, I mean that with setting all the details and quality settings at max, the NV30 is supposedly 30% faster than the 9700 Pro. Realize that this includes 128 tap aniso, while the 9700 Pro only supports 64 tap aniso. As The Inquirer stated, the NV30 leaves ATI's methods of FSAA and aniso in the dust.

I have no doubt that Nvidia had a purpose when they decided to stick with the 128 bit memory system. In case competition got a bit too tough, they could moved it to 256 and double the bandwith. The VPU of the NV30 is a monster which has significantly more performance power than the 9700 Pro. It will take way more than an R300 w/ DDRII to take down the NV30.

- - -
<font color=green>All good things must come to an end … so they can be replaced by better things! </font color=green>
<P ID="edit"><FONT SIZE=-1><EM>Edited by Dark_Archonis on 11/16/02 00:04 AM.</EM></FONT></P>

 

[icy_oblivion]

These fanboy threads always give me a good laugh. It is just like comparing AMD and Intel's actual clock speeds. They are built on different technologies but people want to compare them directly.
No on ever though ATI would overtake the Ti4600, but guess what they did. And of course Nvidia will over take the 9700, it is the way of the computer buisness.
As others have said, I don't believe anything till I have seen actual preformance in a real user/gamer situation. Synthetic benchmarks are great for numbers, but what do they actually mean.
And who cares how much bandwidth the card has, can AGP 8x even transfer that much bandwidth.

Yah, I got the fastest card on the market, but it is bottle-necked by my AGP port. And there are no games avaliable, yet, that can make use of that much bandwidth even if all of it could be transfered by the AGP bus.

I really hate fan boy threads.

<font color=green>My other personality is schitzofrenic.</font color=green>

 

[TKH]

I totally agree with you. There is no point to compare between NV30 and R300. NV30 is definitely faster than 9700 Pro, if the perforamnce is no different from NV25 why should nVidia luanch it? Marketing strategy has lead many people to believe the 'performance myth' while the actual situation is not always the true. For example, ATI makes heavy claim on the DirectX 9 support, but does it come out yet? When DirectX 9 becomes mainstream, would anyone still holding the 9700 Pro? There must be a more advance graphic board out there and everyone will eventually turn to it. The same applies to AGP8X (Remember NV28 which is NV25 with AGP8X support and does it makes any difference?).

Plug & Play? How about Plug & Pray!

 

[Galvin]

To get more bandwith out of slow memory you dual channel it like Granite bay or Nvidia's own chipset.
Or even go Quad channel. basicaly 4x the bandwith.

So the only way they can get 35GB bandwith is to be doing somthing like dual channel or what not.

Anyway I thought this is how nvidia got like 6+GB bandwith out of the TI4400 cause they were doing something like crossbar getting more real bandwith out of slower ram?

 

[Dark_Archonis]

Sorry if I sound like a fanboy, but I'm just trying to show to AMD_Man that ATI is not the greatest. For AMD_Man, ATI is like a god, and he should wake up and realize that ATI is not that great. I admit I prefer Nvidia, and may sometimes even act like an Nvidia fanboy, but again, most of the things that I say do turn out to be true. If you state the truth, does that make you a fanboy? I have been labelled an Intel fanboy many times, and yet most of the things I post turn out to be true, yet there are still several ignorant people who think I make up everything I say simply because I'm a fanboy. You may not believe the NV30's "claimed" bandwith (The Inquirer's claim) but I believe that probably is the real bandwith. Most of the predictions I made about the NV30 came true. Heck, I even underestimated the NV30 in some cases. Even I did not expect the NV30 to be THIS powerful.

<i>Written by icy_oblivion</i>
As others have said, I don't believe anything till I have seen actual preformance in a real user/gamer situation. Synthetic benchmarks are great for numbers, but what do they actually mean.
And who cares how much bandwidth the card has, can AGP 8x even transfer that much bandwidth.

Just to let you know, I never said anything about synthetic benchmarks. And about the so-called "bottleneck" of AGP 8X, you are wrong. AGP 8X is not a bottleneck for the NV30, not even by a long shot. Based on your statement, you really don't know alot about the architecture of cards like the 4600, 9700Pro, or NV30. If you did, you probably wouldn't be making this statement. If AGP 8X was such a bottleneck, Nvidia would've been using AGP 16X for the NV30. Again, this all has to do with the architecture, and just to give you a hint, cards like the NV30 are designed to use the AGP bus <b>as little as humanly possible</b>. The memory bandwith of the NV30 is used to transfer data back and forth between the VPU and the DDRII memory on the card.

So before you call me a fanboy, get your facts straight, and try not to be ingnorant. Being ignorant is even worse than being a fanboy.

- - -
<font color=green>All good things must come to an end … so they can be replaced by better things! </font color=green>

 

[AMD_Man]

I wasn't being ignornant...the Inquirer says "Details emerge of 48GB effective memory bandwidth". The key word here is effective. This isn't 48GB/s of real bandwidth.

EDIT: The Inquirer is comparing nVidia's effective bandwidth to ATI's real bandwidth. That's just plain stupid, if you ask me.

Intelligence is not merely the wealth of knowledge but the sum of perception, wisdom, and knowledge.<P ID="edit"><FONT SIZE=-1><EM>Edited by AMD_Man on 11/16/02 10:52 AM.</EM></FONT></P>

 

[icy_oblivion]

Okay, I got a bit carried away with my statement above and did try to make an issue out of non-issues. I will readily admit I don't know a lot about the card architectures, but I have been reading some and hopefully won't make that kind of mistake in a post again.

And i agree the NV30 will be an excellent card.

<font color=green>My other personality is schitzofrenic.</font color=green>

 

[Unit01]

The inquierers statements should always.
Note: ALWAYS be taken with 3 buckets of salt.

And you should know how desperate nvidia fanboys are of these statements that the nvidia pr department spit out of them all the time. Ati must die
So the papers are their weapons

And it's often those who are the most ignorants. Those clinging to one company. Do you people win anything on being an Nvidiot or FanATIc ???
Be fair, don't bullshit. No company is without faults

 

[Lonemagi]

I agree, go with what works best for you. Look at performance, and price before you go and look at the label.

If AGP 8X was such a bottleneck, Nvidia would've been using AGP 16X for the NV30.

I do believe that AGP 3.0 is the last AGP, it was announced by Intel that it will be replaced by PCIexpress. I dont know if you are aware of this and just stating that to make a point, just thought I would mention it.

You dont have an <b>ANY</b> key? Your keyboard must not be Windows XP certified...

 

[Defective]

All the hints and such at the NV30s power mean nothing right now.

What counts is when its on the bench and we get to see what it can do. I am hoping for high marks because my next card purchase will be from this wave of cards. Whether it is the ATI that is best at the time or the Nvidia, I dont really care. I will want the most bang for my $300. Luckily I will get to wait till both ATI and Nvidia put out their best performing versions of these cards.

I think the Nvidia news right now is more hype than substance. The more substance there is to it the better, but we shall see. I am pretty sure ATI is holding the DDR2 version of the 9700 in the wings to be brought out as an answer to the NV30. I do like watching this fight though. It keeps the technology advancing at a faster and faster pace.

---------------

Stupidity should be painful.

 

[Ghostdog]

I wouldn´t go to far into believing what the Inquirer says.

It´s true that the tech-sites already have NV30-samples, but nVidia would have to be a band of monkeys to be stupid enough to give a sample to the Inquirer, ofcourse they would leak it, regardless of NDAs.

Also, have you considerd that the NV30 could be a defered renderer, requiring much less bandwidth.

<font color=red>I´m starting to feel like a real computer consultant.</font color=red>

 

[groth2757]

C'mon guys....as long as that hot fairy comes with every GeforceFX card I'm getting one....lol.

<font color=blue>Can anyone help me overclock my girlfriend's love toy?<font color=blue>

 

[phsstpok]

First of all, the NV30 doesn't have 48GB/s bandwidth. It doesn't even have 32GB/s bandwidth. Assuming it has 1GHz 128-bit DDRII, then it has 16GB/s of REAL BANDWIDTH. The claimed 48GB/s bandwidth is the effective bandwidth with the bandwidth saving techniques at work.

You are forgetting that DDR2 is not just a faster form of DDR. It is a new technology that produces twice the per pin bandwidth for a given clock speed. Where DDR, 128-bit, clocked at 500 Mhz (1000 Mhz effective) would produce 16 GB/sec of theoretical bandwidth, DDR2 will produce 32 GB/sec.

For the quoted 48 GB/sec number the extra 16 GB/sec is due to the LMA III architecture presumably.

See the white paper on DDR2 below. It explains how DDR2 produces twice the per pin bandwidth of DDR.

<A HREF="http://www-3.ibm.com/chips/micronews/vol7_no2/clinton.html" target="_new">http://www-3.ibm.com/chips/micronews/vol7_no2/clinton.html</A>


<b>99% is great, unless you are talking about system stability</b>

 

[johnoh]

To me the biggest news will be the price.

John A

 

[johnoh]

To me the biggest news will be the price.

John A

 

[Makaveli]

This memory does not transfer 4 times per clock but simply improves on the way data gets in and out of the chip, allowing for much higher clock rates. This should sound familiar as it is very similar to what ATI did with GDDR3.

As Stated in Anandtech's NV30 article!
<A HREF="http://www.anandtech.com/video/showdoc.html?i=1749&p=2" target="_new">http://www.anandtech.com/video/showdoc.html?i=1749&p=2</A>

 

[lhgpoobaa]

1. The molex power attachment is good. I don't know why ATI didnt use a more common molex connector, not a FDD connector.

2. Kudo's for getting 0.13u and DDR2 up and running.

3. It will be interesting to see the cooling solution. I really think the first PCI slot should be perminantly removed for better/more AGP cooling.

4. I dont like the name though. Geforce FX... Thats going to just confuse even more people into buying cut down MX versions or the sucessor to it.

5. Im also not pleased about the size of the card... compared to the nice and compactR9700 the FX is yet another long card. SO people will continue to have problems with it and ramslots/cables.

<font color=red><b><i>"Humans are just overclocked Monkeys!"</i></b></font color=red>
<b>Serving THGC for over 2 years.</b>

 

[AMD_Man]

1. The molex power attachment is good. I don't know why ATI didnt use a more common molex connector, not a FDD connector.

The R9700 does come with an FDD to molex adapter.

2. Kudo's for getting 0.13u and DDR2 up and running.

Indeed! The NV30 will scale up to a whopping 500MHz core! w00t!

3. It will be interesting to see the cooling solution. I really think the first PCI slot should be perminantly removed for better/more AGP cooling.

I just hope it doesn't heat up the PCI cards beside it.

4. I dont like the name though. Geforce FX... Thats going to just confuse even more people into buying cut down MX versions or the sucessor to it.

Lol, I can imagine people saying they're looking for a GeForce FX MX. I'm sorry that's a horrible name. Maybe they'll call it a GeForce FMx^2? LOL! :tongue:

5. Im also not pleased about the size

Yup, it doesn't look like it'll fit in a Shuttle box. Hmm...that might tick some enthusiasts off.



Intelligence is not merely the wealth of knowledge but the sum of perception, wisdom, and knowledge.

 

[flamethrower205]

Dun care what people say, I think it's a monster. Waiting to see more hehe. Made a bet with a guy it'd beat a R9700 @ school. Most of the guys thought he would be right. Cha ching!

Kentucky: 15 million people, 5 last names....

 

[lhgpoobaa]

Oh. I forgot an important point!

6. Due to timing/delays, the FX will completely miss the Christmas selling season, while the R9700 is here in good numbers.

<font color=red><b><i>"Humans are just overclocked Monkeys!"</i></b></font color=red>
<b>Serving THGC for over 2 years.</b>

 

[AMD_Man]

Dun care what people say, I think it's a monster. Waiting to see more hehe. Made a bet with a guy it'd beat a R9700 @ school. Most of the guys thought he would be right. Cha ching!

What a stupid bet...I wouldn't bet for the R9700Pro myself considering the NV30 comes out like 5 months later so the NV30 is definitely ahead tech-wise.

Intelligence is not merely the wealth of knowledge but the sum of perception, wisdom, and knowledge.