AnandTech NV30 article

[Nemesis]

Quite an intersting read, here (http://www.anandtech.com/video/showdoc.html?i=1678) it is...

NV30 probably in shops by December, faster than R300, and possibly expensive due to poor 0.13 micron yields at TSMC.

[jbirney]

Yeilds of 10 to 20 %? OMG thats crazy. I know its a low volume but still thats a lot of wasted silcon unless they can use some for some sub NV30 part.

Then another scary part was this:

The first NV30 silicon taped-out last week

:eek: A cip that complex could need another respin if they find bugs that are not fix able via Software. If that happens it will add time. nV can cut this time down by building wafers a head of time and holding them before the mask stage. Still its a few more weeks to the schedule.

Price is another scary thing. We know with poor yeilds it will drive the cost up. Top that with brand new DDR2 at 400+ mhz and you get expensive memory. Which will probably put this thing at or above the 400+ price range....


I am sorry the more I read about the NV30 the less impressed I am with it. Yes it will be faster than the R300 but thats practily about it. Any beyond DX9 stuff most likely will not get uses in the life time of the part (we have just started using TnL of good ol DX7 and have yet to really start on DX8 games). I am glad nV will have a monster chip on their hands but wow...scary stuff....

[SurfMonkey]

I get the distinct impression that gamers aren't supposed to be that excited by the NV3x feature set. It would appear that the people who should be getting excited are CGI specialists, film companies, and amateur flim creators.

It's surely is going to kick ass and be fast as well, but the CineFX architecture points more towards cgi content than game content. I remember seeing the Videotoaster running in an Amiga A3000 (whic I still have at home, toaster and all) and thinking that it was pretty amazing. And then along Babylon 5, ****e story but the cgi was amazing for the time. And cheap to produce.

When I look at the NV3x stuff I get the same kind of excited feeling I got when I first saw the VT, only it's going to be a whole load more powerful. And in relative terms dirt cheap. If gfx proffesionals take it up then the cut down versions that the main stream will have access to can only benefit from the higher cost of the mainstream part. More money at high level equates to a better margin at main stream level. And that'll be fine for me :)

[John Reynolds]

Originally posted by Nemesis
Quite an intersting read, here (http://www.anandtech.com/video/showdoc.html?i=1678) it is...

NV30 probably in shops by December, faster than R300, and possibly expensive due to poor 0.13 micron yields at TSMC.

Umm, no. Anand's predictions are all based on *0* bugs, no respins whatsoever. And what're the odds of that on a chip this complex being fabbed on a new process? Not bloody likely. The worrisome thing is that will Nvidia ship the first chips with bugs (they have before) in the metal since they're under more competitive pressure than they have been since '99? ATi is definitely in the driver's seat. . .can they stay there?

[nutball]

Well as an astrophysicist, and I can say I am very excited about NV30!

The potential it has to make a contribution to our work in both graphical (ie. scientific visualisation) and non-graphical (physical simulation) arenas is truly incredible for such a cheap piece of hardware.

Now I just wish they'd get the damned thing on the market, so I can start playing with it! :)

[Richthofen]

"
Any beyond DX9 stuff most likely will not get uses in the life time of the part
"

Who said NV30 is beyond DX9? :)
Just keep in mind DX9 has not been released and according to the Nvidia CEO NV30 will be released in the DX9 release timeframe.
Rumors say that MS updated DX9 to NV30 specs.
If that is true the R300 won't be a 100% DX9 part.
That would not be the first time MS does that.

We will have to wait and see.

[pelly]

I have heard that NV30 uses a Pixel Shader version which is beyond DX9......in addition to a few more features...

:D

It looks as though NV30 is the "Right" card......whereas the 9700 Pro will be the "Right Now" card. Anand has come right out and said the NV30 surely beats R300 on paper...I still think it will be tough for many to hold out for an extra month or two.

You can be sure that I'm saving my money for NV30 though... :cool:

[Fotis]

Let's be realistic here people.All that cinefx pr is only good on paper!I see this as another DX8 vs DX8.1 shaders.Ati screamed that their shaders were more flexible & had more instructions and nvidia said that they mean nothing as they will not be used in radeon 8500's life span.
The bottom line is this,is nv30 faster than radeon 9700 and if so how much?What about fssa & aniso speed?If nv30 is 10-20% faster without fsaa & aniso(according to specs) and radeon 9700 is faster using fsaa & aniso?If the rumor about 200million triangles for nv30 how does it stand against radeon 9700's 325million?
And last but not least what if nv30 ends up beeing 100-150$ more than radeon 9700,is it really worth the 10% additional speed?Price is essential!!
Many questions but no answers?

[Megatron]

Originally posted by pelly
It looks as though NV30 is the "Right" card......whereas the 9700 Pro will be the "Right Now" card. Anand has come right out and said the NV30 surely beats R300 on paper...


Remember how good 8500 looked on paper too( not that i dont think 8500 is a good card, but just making the point that paper tigers dont mean much)

[pelly]

Good point.....
The only way we'll know for sure is to get ahold of some actual benchmarks....

As for the price, I highly doubt it will be priced any more than $50 over R300.....I certainly don't call the shots at NVIDIA though so we'll just have to wait and see on that issue as well...

:D

[Megatron]

Yup..i wish the benches and reviews would start flowing. Perhaps in the next couple days they will with Quakecon coming and all.

Youre probably right on the pricing, Nvidia needs to keep it close to the 9700 in that area. I wonder how much of a hit they will have to take on the profit margins with the yields being reported to be as bad as they are.

[jbirney]

A little birdie told me the 19th is when you should see real reviews. Roumor has it that some sights are done with their reviews and waiting for that day.....

[pelly]

I came across a story at the Inquirer that talks about Fabs lowering prices by up to 10% to account for empty space ( not enough orders )

If this is true, it should make this hit a bit easier to take for NVIDIA.

:)

[imtim83]

The Nvidia NV30 will have zero performence hit with FSAA and AF on i believe. But remember i could be wrong.

[Fotis]

Originally posted by imtim83
The Nvidia NV30 will have zero performence hit with FSAA and AF on i believe. But remember i could be wrong.
Just the facts please imtim83!!

[Philibob]

Originally posted by imtim83
The Nvidia NV30 will have zero performence hit with FSAA and AF on i believe. But remember i could be wrong.

I'd love this to happen but I'm not expecting it.

[imtim83]

Fotis anandtech said that too btw. Its not just me.

[StealthHawk]

Originally posted by imtim83
Fotis anandtech said that too btw. Its not just me.

well, he certainly didn't say that in the article he just put up. it was just rampant speculation that was on the internet before. leftover technology that was acquired by 3dfx. we will see if it was true or not. if NV30 really has more bandwidth than R300 FSAA and AF performance should be faster on NV30 as well

[netviper13]

With the HUGE increase in bandwidth we're talking about for NV30, it is plausible for lossless AA to occur. With current games at least, because they won't even come close to using up all that bandwidth and fillrate.

[Kerbs]

Yep once again even at high resolutions the bottleneck may move from GPU/Bandwidth to CPU and the memory Sub-system at least in older games (quake 3 ect..)

[StealthHawk]

Originally posted by netviper13
With the HUGE increase in bandwidth we're talking about for NV30, it is plausible for lossless AA to occur. With current games at least, because they won't even come close to using up all that bandwidth and fillrate.

i'm hopeful that we'll have 8x FSAA or even something more advanced. i doubt that would be free :D with all that bandwidth(if true) it would be a waste to be constricted to 6x

[druga runda]

R300 is out now, beginning of September, and NV30 will be out at the beginning of December if they move so fast as they usually do.

That gives ATI 3 months advantage :eek:

Second even though NV30 will be more advanced part, the size won't be the most important saver, as they will have to compensate for low yeilds. The raw speed and bandwidth will be the order of the day to show for any advantages that the card might have over r300. And hopefully we will have DX9 3Dmark so that will show the advantage of the card.

But in next 2 years I can't see NV30 to be a much better performer than R300 if for any other reason - you will have to be struggling to find games that R300 doesn't run at max settings for playable framerates (call that 80 FPS or so) until Doom III comes out.

Even than R300 card will be quite good.

Sooo ... this makes the whole situation more interesting... Ati with a part that has good yields, massproduced, and Nvidia struggling to catch up, at least for the next few month... whoooo.

It seems that it will be as I expected, the product will be out in real quantites sometime next year, probably Feb (apart from reviewers/ lucky customers) And hopefullt the price will not be that astronomical (but maybe they will have to put the price up to justify low yields, and to subdue the demand, which will surely be strong comparing to the production volume as they will not be able to produce huge quantites of these processors at the beginning.) :hmmm:

[StealthHawk]

raw speed doesn't cut it anymore. R300 and NV30 will be horribly CPU bound for raw fillrate to ever rear its head i believe. it's all about eye candy now :D

[randsom]

didnt anand say the rumours that are floating around werent true, the real fact of the matter is nvidia is being way quieter about this gpu than any previous product they have designed, nobody that isnt under a nda has no idea of what the nv30 is all about, but there is a couple of things i think that are being overlooked

1.do you really think ati's getting great yields with the r300 on the .15um process, i doubt it.The r300 is pushed .15um process to the absolute breaking point.I dont want a gpu that runs that hot(my box is hot enough already) and needs external power to make fully functional.I mean take a look at the heatsink on it.Who knows about the drivers

2 I beleive the nv30 will have much better yields after the intial
alpha run. Nvidias a big player and tmsc or umc or whoever is going to be the founder for this gpu, wants them happy, and i think the .13 um process is alittle further along than people think.Nvidas got the edge in the software dept also/

All this speculation about the nv30 being late isnt warranted, do you guys realy think that nvidia didnt know the state of the .13um process. theyre probably behind theyre internal schedule, but not by that much . Ati probably surprised them by getting to the market first, but theres nothin nvidia can do about that, except design a better gpu.Nvidias the fastest from the foundry to the market bar none.

i think the r300 is gonna be a great card, i just think the nv30 is going to be better, in every dept

[StealthHawk]

seeing as how it's supposed to be a Fall part, and the fact that nvidia normally ships by October, it is late. by Anand's best estimates, a late November shipping is the earliest possible one.

the R9700 with 1-4 non-functional pipelines become R9500s, so that helps solve poor yields or R9700 by turning otherwise useless chips into sellable ones. if you look at the Anandtech article, he mentions that a slower, possibly castrated NV30 would be available months later than NV30, so nvidia might do that too. although with such a low yield, it would seem more prudent to launch the 'wasted' silicon along with NV30. i doubt R9700 is having as poor yields as NV30 is.