Stop saying the NV30 will cost more than the R300!

[Uttar]

Hello everyone,

I've seen WAY too many people saying "Well, who'd want to pay for the NV30 since it's obviously gonna cost 100 bucks more than the R300 right now and..."

I think someone is forgetting some things here:
1. nVidia is still using a 128 bit bus.
2. 0.13 yields at TSMC will most likely be at around 25% when the NV30 goes in real production. AND, considering the same yields, it isn't more costly at all to do 120 million transistors than 107 million transistors - it's likely even less costly.

So, in my opinion, using a 128 bit bus will most likely compensate for low yields. And if nVidia gets lucky, and gets about 30% yields at TSMC when in real production, they could really begin to have a margin at $399.


Uttar

[volt]

Still stores will sell it for $450+ since it's going to be a hottie. Just like the Ti4600 at newegg haha

[DadGT]

Originally posted by Uttar

1. nVidia is still using a 128 bit bus.


Uttar

The NV30 is a still unannounced part. How do you know that it will be 128 bit bus? Are you basing it on the Daivd Kirk comment that 256bit is "overkill, at the moment." nVidia moves fast, that moment may have already passed.

I would not be surprised at all to see the NV30 with 256bit memory.

As for NV30 cost, 0.13 um process = more chips on the same size platter = lower cost per chip. Assuming of course the process has comparable yield.

[DIMA]

1. nVidia is still using a 128 bit bus.

Nvidia are using 256 bit bus, not 128.

[jbirney]

Nvidia are using 256 bit bus, not 128.

can you please point me to the nV Press Release that shows this? Oh thats right it does not exsit ;)

Anyways dont forget nV30 will be using high speed DDR2 which is still very costly at the moment. Also the yield point is really a har one. How good are the yeilds on the R9700? If they are decent then it may prove to be cheaper to make. The only advantage cost wise is volume of scale of the .13u process. The better the yeild the lower the cost. nV is going to lose money if they end up throwing 70% of the silicon away. Finally packaging can add cost. I have no idea how NV30 will be packaged. Too many unkowns for now...

[Uttar]

Originally posted by DadGT

The NV30 is a still unannounced part. How do you know that it will be 128 bit bus? Are you basing it on the Daivd Kirk comment that 256bit is "overkill, at the moment." nVidia moves fast, that moment may have already passed.

I would not be surprised at all to see the NV30 with 256bit memory.

My post was a reply to people using rumors to predict the NV30 will cost more than the R300.

And according to those same rumors, it'll use a 128 bit bus.

So, what i'm tyring to say is that according to the currently available information, it is not as likely as people believe that the NV30 will cost more than $400.


Uttar

[nutball]

I've said this before, but that's the beauty of cut-and-paste:

If NV30 spanks R300s butt, it will cost more than R300 (probably). If it doesn't, it won't (otherwise it won't sell, will it).

Simple market forces will dictate that.

[Lamentation]

If nv30 doesn't spank it, but is only the same, and cost about the same, I don't see why it wouldn't sell.

[sancheuz]

When Nv30 comes out, it will cost 400 at retail stores. Only stupid people go to retail stores(no offense). Ebay, 350... :-) smart people shop well..........OHh chocoalate doughnut, uhhhmmmmmm. Oh sorry. It dont matter how much nv30 cost, it will be flying off the shelfs i guarantee it.

[Nv40]

well i have no problems with the Price
Nvidia choose for the Nv30 ,if it goes over $400
;IF the performance of the card Smokes Radeon9700 like the Geforce4 to RADeon2 ..:)

The Thing is the Geforce4 already handle ALL avaible GAmes at high quality /max setting in 1600x1200 ,AA+AF 2x.. i play Ghost recon
/medal of honor/Jedignight2 at 1600x1200 with aniso from 2x to 8x(AA off).. and i only have
an atlonXp 1900+... :)

and even Geforce4 is fast enough to play UnrealT2003 at 1600x res -high settings
with a fast machine ,so until Doom3 i see no reason to pay $400 dollars for a video card
unless you are struggling with a low video card like Radeon2 or GeforceMx ...

You pay for what you get , and if you cant afford
$400 now ,go for the middle performace solution Nv28 or wait 6 months a get NV30 a lot more cheaper ..;)

[ZenOps]

Anything less than 50 percent yield is nothing to be proud of, it would be shameful for CPU makers Intel or AMD to release a chip with that low a yield.

If ATi is getting around 90 to 95 percent yield on a very mature .15, Nvidia would have to get at least 75 percent on the .13 to get the same number of chips per wafer (It is after all only 2 microns) At the end of the day, it comes down to how many wafers you end up throwing in the garbage.

With a .15 process which is mass manufacturable, could easily get 30 times more chips than .13 when demand kicks in. (Not sure of the real numbers, but 30 percent .15 manufacturing from TSMC would be reasonable, <1 percent for .13 looks positively anemic) Mass production = less expensive, always has.

One NV30 on the shelf to 30 Radeon 9700's... As I mentioned somewhere else, even though *a lot* of Radeon 9700pros were made in the first run, they have already sold out.

DDR2 is *very* expensive and rare, and from what I have heard the NV30 requires it.

There is no doubt in my mind the NV30, if released as specified now, will easily cost $100 more than the Radeon 9700pro.

[SavagePaladin]

Originally posted by ZenOps
Anything less than 50 percent yield is nothing to be proud of, it would be shameful for CPU makers Intel or AMD to release a chip with that low a yield.

If ATi is getting around 90 to 95 percent yield on a very mature .15, Nvidia would have to get at least 75 percent on the .13 to get the same number of chips per wafer (It is after all only 2 microns) At the end of the day, it comes down to how many wafers you end up throwing in the garbage.

With a .15 process which is mass manufacturable, could easily get 30 times more chips than .13 when demand kicks in. (Not sure of the real numbers, but 30 percent .15 manufacturing from TSMC would be reasonable, <1 percent for .13 looks positively anemic) Mass production = less expensive, always has.

One NV30 on the shelf to 30 Radeon 9700's... As I mentioned somewhere else, even though *a lot* of Radeon 9700pros were made in the first run, they have already sold out.

DDR2 is *very* expensive and rare, and from what I have heard the NV30 requires it.

There is no doubt in my mind the NV30, if released as specified now, will easily cost $100 more than the Radeon 9700pro.
You're forgetting that at the speeds required for a high end vidcard, even the Radeon 9700 is low yield, especially with so many transistors on it.
So can the 'if'...they aren't.

Also, hard as it is for us to believe, high end vidcards aren't very high demand. While this is undoubtedly due to the prices, you can't beat the competition and worry about prices too much.

[sbp]

Stop saying the NV30 will cost more than the R300! Ok it'll be more expensive. Better? http://sbp777.homestead.com/files/winkie.gif

[ZenOps]

Yeah, the 90-95 percent is a definite IF.

But considering Anand is overclocking these things to just under 400Mhz, I think it would be safe to assume that ATi is getting very high yields on the chips. If they were getting low yields, the chips that did make it though would not be overclocking very high.

[SavagePaladin]

Originally posted by ZenOps
Yeah, the 90-95 percent is a definite IF.

But considering Anand is overclocking these things to just under 400Mhz, I think it would be safe to assume that ATi is getting very high yields on the chips. If they were getting low yields, the chips that did make it though would not be overclocking very high.
I read that too, but given that I know very little about the process of making the stupid chips, I can't say if that means all that much wrt yield.

[PreservedSwine]

Considering that ATI will be releasing another .15 micron R300 based chip *before* the NV30 launch, We're aren't going to be seeing lower prices anytime soon. Unless you want to take that 2nd generation of products (Geforce4, R8500)

It's going to be all about $400+ video cards this X-mas, kiddies:eek:

[Uttar]

Originally posted by PreservedSwine
Considering that ATI will be releasing another .15 micron R300 based chip *before* the NV30 launch, We're aren't going to be seeing lower prices anytime soon. Unless you want to take that 2nd generation of products (Geforce4, R8500)

It's going to be all about $400+ video cards this X-mas, kiddies:eek:

Not really. The new ATI chip will be the Radeon 9500 - it'll be LESS costly than the Radeon 9700. No one knows how much it'll cost, but maximum $300 i'd say.


Uttar

[PreservedSwine]

Originally posted by Uttar


Not really. The new ATI chip will be the Radeon 9500 - it'll be LESS costly than the Radeon 9700. No one knows how much it'll cost, but maximum $300 i'd say.


Uttar

I should have been more specific:)

In addition to the R9500, rumours have it there is going to be another R300 based .15 micron chip released before the NV30- basically, a re-freshed R9700.Before its archrival Nvidia introduces its high-end NV30 at the end of this year, ATI will add one 0.15-micron-process R350 chip into its product portfolio, sources said, and later debut its first 0.13-micron, light version of the RV350.

from here http://www.digitimes.com/NewsShow/Article.asp?datePublish=2002/08/29&pages=04&seq=25

[tazdevl]

Originally posted by Uttar
Hello everyone,

I've seen WAY too many people saying "Well, who'd want to pay for the NV30 since it's obviously gonna cost 100 bucks more than the R300 right now and..."

I think someone is forgetting some things here:
1. nVidia is still using a 128 bit bus.
2. 0.13 yields at TSMC will most likely be at around 25% when the NV30 goes in real production. AND, considering the same yields, it isn't more costly at all to do 120 million transistors than 107 million transistors - it's likely even less costly.

So, in my opinion, using a 128 bit bus will most likely compensate for low yields. And if nVidia gets lucky, and gets about 30% yields at TSMC when in real production, they could really begin to have a margin at $399.


Uttar

A yield of 25% doesn't exactly impact nVIDIA. You have to remember that the foundry is bearing the cost of chip production and yield impacts their costs. nVIDIA sets a price for the production on a per chip basis. TSMC funded the cost of the shift to .13m, not nVIDIA.

If what you said was true and yields were that poor, the NV30 would probably be more expensive due to a few factors.

1) Economies of scale won't be achieved at a 25% yield... so chances are it would be more expensive on a per chip basis.

2) If nVIDIA were using that kind of pricing model, I'd assume that TSMC would factor in the costs of the .13 shift into the chips rather than treating it as a sunk cost.

3) Last time I checked, DDRII is a more expensive memory technology. If the NV30 is using it, that is going to drive the price a lot higher than the memory used on the 9700, which is a mature, efficiently manufactured product.

Digitimes is about a reliable as The Inquirer when it comes to speculation. I wouldn't give what they print too much weight unless they are quoting a specific, reputable source with a name that works at the company in question.

[jbirney]

A yield of 25% doesn't exactly impact nVIDIA. You have to remember that the foundry is bearing the cost of chip production and yield impacts their costs. nVIDIA sets a price for the production on a per chip basis. TSMC funded the cost of the shift to .13m, not nVIDIA.

I am pretty sure that is wrong unless nV has a special realationship with TMSC. TMSC sells them wafers at x amount of cost. If the yeilds are bad and nV proves its TSMC fault then they eat the loss for yeilds. However if not, nV eats the loss. I work for a company that in the top 5 manufactors of silicon ICs and have delt with fab issues before.

[jbirney]

Originally posted by Nv40
The Thing is the Geforce4 already handle ALL avaible GAmes at high quality /max setting in 1600x1200 ,AA+AF 2x.. i play Ghost recon
/medal of honor/Jedignight2 at 1600x1200 with aniso from 2x to 8x(AA off).. and i only have
an atlonXp 1900+... :)

You might want to double check some of those settings. Most reviews I have seen of the R9700 and Jk2 when they crank all of the setting up, the GF4 scores falls into the 20s (16x12x32 x4 fsaa and x8 aniso). SS:SE is another one.

Look at Anadtechs review and you we see that as good as the GF4 is, it just dose not have the fill rate or MB to keep some of the numbers above the 60 fps cap:

http://www.anandtech.com/video/showdoc.html?i=1683&p=18
http://www.anandtech.com/video/showdoc.html?i=1683&p=20
http://www.anandtech.com/video/showdoc.html?i=1683&p=21

So if you want all the eye candy and max IQ, the GF4 dose not haev enough power to do that.


Next,
yeild is roughly defined as the number of parts that pass the probe test done on the wafer (before they are cut and packaged). Usally here they will test each part to see if it meets some spec. Running at some stable clock speed is usally one of those tests. Usally a bigger yeilds imply that the process was not at any edge point and parts often exceed specs in some cases. A poor yeild means that the process may have hit some limit and a lot of parts dont pass the spec. However this is only a general rule of thumb, I have seen it go the other one once or twice.....