Anand says - now that NV30 is taped out - we can reveal...
- Thread starter g__day
- Start date
martrox said:Couple of interesting points:
1)IF the problem is not the .13 fab, and instead was the chip design, then maybe the rumors about a changeover from a 128 bus to a 256 bus due to R300 is right.
2)IF it's a 8x2 pipe architecture, maybe that rumor about 30 gig bandwidth is right. Otherwise, it will be a very unbalanced product.
Looks like it will be a few months before we really know the truth about NV30. Until at least the beta cards are independently seen & tested, I doubt you will hear anything except FUD on it.
One thing can be taken as fact from this, it that there are too many people that think the have more “knowledge†on this product than really do.
Bottom line: Believe only half of what you see, and none of what you read & hear! BTW, this applies to everything in life…… unless it’s something you wife tells you, then it’s the gospel!
FUD is always nVidia's PR. They have very nice hardware, but as a company they are detestable...
BTW Martox, what part of J-Ville you in?
asicnewbie
Newcomer
Joe DeFuria said:
TSMC's normal pricing schedule is based on fabbed wafer delivery. The contract is for a delivery of a fixed # wafers for a fixed price. If the yield is 100%, great for the customer! If the yield is 0%, too bad for the customer. (If the yield is 0%, the customer is going to investigate WHY the yield is 0%. If they can identify TSMC at fault, then TSMC will re-fab the number of defective wafers at no additional charge. The customer still 'loses time'.)
Other foundries (like IBM and LSI) have a standard payment schedule based on # "known good/packaged dies." In essence, the customer pays a fixed price for a fixed # of known good chips, effectively shielding the customer from the foundry's yield woes (if any.)
That said, who knows what kind of business arrangement NVidia and TSMC share. Perhaps TSMC is giving NVidia a pricing discount (due to supposedly low yield), or maybe not. If NVidia's design is simply 'too aggressive', then I doubt TSMC will give NVidia special consideration. TSMC claims they have had ">200 successful customer tape-outs at 0.13", but IMHO the statement's ambiguity sounds more like a marketing hook than any indicator of process maturity. On the other hand, all foundries (including TSMC) disclose its internal aggregate yield and process characteristics to its customers. (This data is collected and presented by the foundry.) So any customer planning on submitting a design will very well in advance, what sort of yield to expect.
According eetimes, TSMC's foundries are operating at less than 100% utilization. Anandtech's article implies TSMC's most advanced processes (0.13) suffer from a severe lack of customers. That being the case, there's no reason for TSMC to put NVidia's parts on the back-burner.
Furthermore, TSMC customers have the option to pay extra for a 'rush-order.' From what my coworkers told me, the 'rush-order' is very much like a lottery. If a customer pays for entrance into rush-order lottery, he has a chance to 'win' a 1/2/3 week schedule speed-up! Since the speed-up comes from recovering production-line slack (either due to another customer cancelling, a coincidental aligning of moons, etc.), the lottery's outcome is random. A customer could end up paying extra and get nothing in return, a possibility clearly stated in the contract.
OpenGL guy
Veteran
Any more unbalanced than the GeForce 2 GTS which was 4x2? Think about the GeForce DDR: 120 mhz engine clock and 150 mhz DDR on 128 bit bus for a 4x1 architecture. Next comes the GeForce 2 GTS: 200 mhz engine clock, 166 mhz DDR and a 4x2 architecture. So, you have ~66% increase in engine clock but only ~10% increase in memory clock for a new design with twice as many TMUs.martrox said:
AFAIK, GF2 has the same TMUs the GF1 has, they are just splitted.OpenGL guy said:
A GF1 TMU can do one trilerped fragment per clock, where a GF2 TMU can do 'only' one bilerped fragment per clock. I know, bandwith requirements are different in these 2 cases, but in fact I believe nvidia didn't add new functional units in their TMUs from GF1 to GF2 transition, they just revised their textures cache architecture.
ciao,
Marco
Prometheus
Newcomer
Why does everyone assume that nv30 will be 8x2?I haven't seen any official confirmation!
I'm no expert so I ask how much more features can you squess in 10 million trans(r300=110&nv30-120)or is it a matter of 0.15 vs 0.13 design?
I'm no expert so I ask how much more features can you squess in 10 million trans(r300=110&nv30-120)or is it a matter of 0.15 vs 0.13 design?
OpenGL guy
Veteran
Possibly, but I am pretty sure that it was marketed as being able to do 2 bilinear texels per pixel per cycle. However, the two units had to work together to do trilinear.nAo said:
Prometheus said:
Well, I'd just like to say that there would be little point to an 8x2 architecture unless nVidia used an 8-way crossbar and twice the texture cache of the GeForce4 Ti 4600. I don't think either is out of the question, but not entirely likely. I think an 8x1 architecture is the most probable.
On another note, hopefully the NV30 will have significantly increased pixel processing capabilities, given the massive number of supported pixel shader ops.
I am listening to the Nvidia CC and the CEO avoided directly saying the NV30 was taped out. LOL. incredible. They are really making a spin on this one. But on the flip side he says that they will have the NV30 available for Christmas and downplayed the value of the enthusiast market.
EDIT: Not only did he downplay the enthusiast market but he also refused to say that the NV30 would be shipping in any volume... not even for Jan or Feb.
IIRC the value of the high end market for Nvidia accounted for 10% of earnings once upon a time. This is a substaintial loss of revenue for nvidia. It seems now that they have been bumped out of the high end they are now downplaying that market. Further he goes on to say that the NV30 will be the "first" of the next generation graphics chips... hrm Radeon 9700 anyone?
EDIT: Not only did he downplay the enthusiast market but he also refused to say that the NV30 would be shipping in any volume... not even for Jan or Feb.
IIRC the value of the high end market for Nvidia accounted for 10% of earnings once upon a time. This is a substaintial loss of revenue for nvidia. It seems now that they have been bumped out of the high end they are now downplaying that market. Further he goes on to say that the NV30 will be the "first" of the next generation graphics chips... hrm Radeon 9700 anyone?
Richthofen
Newcomer
"
I am listening to the Nvidia CC and the CEO avoided directly saying the NV30 was taped out. LOL. incredible
"
Well you have to realize that CEOs no matter if they are from AMD, Intel or Nvidia do not like to comment on future products.
"
IIRC the value of the high end market for Nvidia accounted for 10% of earnings once upon a time
"
Well probably you should mention "past time".
With x-box, growing nforce/nforce2 sales and gorwing mobile sales the high end market gets more and more unimportant concerning earnings.
Of course it is important from a PR point of view but not concering earnings and profits.
"
Further he goes on to say that the NV30 will be the "first" of the next generation graphics chips... hrm Radeon 9700 anyone?
"
Well again you should consider why DX9 is not out yet
IF MS indeed updated DX9.0 specs to NV30 level Huang does not consider the R9700 a full DX9 part.
That's simply it.
I think you should get the big picture. Beside PR all that matters is money. And you earn money if you get OEM design wins. So you have to think about when the timeframe for OEM disign wins is. There are 2 periods over a year and Nvidia always hits that window.
The R9700/R9500 will not affect this at all.
The next period these products can effect is spring 2003 and guess what - Nvidia will meet that window again.
I am listening to the Nvidia CC and the CEO avoided directly saying the NV30 was taped out. LOL. incredible
"
Well you have to realize that CEOs no matter if they are from AMD, Intel or Nvidia do not like to comment on future products.
"
IIRC the value of the high end market for Nvidia accounted for 10% of earnings once upon a time
"
Well probably you should mention "past time".
With x-box, growing nforce/nforce2 sales and gorwing mobile sales the high end market gets more and more unimportant concerning earnings.
Of course it is important from a PR point of view but not concering earnings and profits.
"
Further he goes on to say that the NV30 will be the "first" of the next generation graphics chips... hrm Radeon 9700 anyone?
"
Well again you should consider why DX9 is not out yet
IF MS indeed updated DX9.0 specs to NV30 level Huang does not consider the R9700 a full DX9 part.
That's simply it.
I think you should get the big picture. Beside PR all that matters is money. And you earn money if you get OEM design wins. So you have to think about when the timeframe for OEM disign wins is. There are 2 periods over a year and Nvidia always hits that window.
The R9700/R9500 will not affect this at all.
The next period these products can effect is spring 2003 and guess what - Nvidia will meet that window again.
Richthofen said:
They don't? Then what was all the rabble about the NV30 being the best thing since sliced bread? But then a total avoidance on the question as to if the NV30 had been taped out. If it was indeed taped out then he would have made that clear.. gladly so. It is common for these sorts of questions to be answered at CC. What you have provided here for a case is poor at best. Further he was vocal about it before, now why all the hesitation? The only genuine valid conclusion is that certainly the NV30 is still not taped out.
Richthofen said:
I can assure you that these numbers of 10% of earnings from the high end market are not that old.. possibly six-nine months old. Further the Xbox is not really contributing much to nvidias bottom line as of yet. Nvidia received $400 000 000 US from Microsoft initially to develop the XGPU and sound chips. $200 000 000 at the beginning of their agreement and anther $200 000 000 when Microsoft actually made the resolution to go with nvidias solution in Jan 2001. In essence nvidia is paying this off at about $40 per Xbox.. that means alot of Xboxes must be sold before any real money from these sales actually contribute to nvidias bottom line. Further with the massive right off with regards to Microsoft changing the security codes nvidia is taking losses there as well. Nforce sales are poor as they expected AMD market share would have been better. Nforce 2 has yet to contribute to the bottom line... Nvidia has very little market share in Mobile arena, 5% or so.. it really doesn't contribute much to their bottom line.
Richthofen said:Well again you should consider why DX9 is not out yet
IF MS indeed updated DX9.0 specs to NV30 level Huang does not consider the R9700 a full DX9 part.
That's simply it.
Well I wouldn't expect that Microsoft alter DX 9 spec just for nvidias sake. Further I read somewhere that in fact Microsoft worked with ATI to have the DX9 spec fit the Radeon 9700 simply because I suspect that ATI was that much more ahead of nvidia with its focus on DX8.1. As for DX9 not being available until the NV30 is offered I think this is folly. http://www.theinquirer.net/?article=4998
Richthofen said:
You assume a lot of matters here without giving any solid evidence to back them up. I personally do think that the Radeon 9700/Radeon 9500/ Radeon 9000 not to mention new FireGL based on the Radeon 9700 core as well as potential Mobile Radeon 9000 chipsets will indeed have an impact on Nvidias bottom line..... As for spring 2003 unless Nvidias CEO says that the NV30 is in fact taped out without delay it appears that is just about the right time we will have the NV30 shipping on mass.(Good call there.)
Sabastian
Please excuse my vaguely fanboyish post. All in good fun, and I assume people with more serious interests have already bailed from this thread:
Or you could imagine that Huang is defining "generation" arbitrarily to mean what he wants it to mean.
What matters is marketshare. nVidia built marketshare by first acquiring mindshare among the PC and 3d buying public. It is now ceding some of this mindshare to ATI; the later the NV30 is and the more it's performance disappoints in comparison to expectations (I expect it will be slower than the 9700 in some current-game benchmarks at some settings) the more mindshare it cedes.
ATI hopes that by the spring the gaming thought-leaders (if that is not an oxymoron) are recommending ATI products to their friends. It is then ATI's job to have OEM products ready to convert that mindshare to marketshare.
Richthofen said:
Or you could imagine that Huang is defining "generation" arbitrarily to mean what he wants it to mean.
Richthofen said:
What matters is marketshare. nVidia built marketshare by first acquiring mindshare among the PC and 3d buying public. It is now ceding some of this mindshare to ATI; the later the NV30 is and the more it's performance disappoints in comparison to expectations (I expect it will be slower than the 9700 in some current-game benchmarks at some settings) the more mindshare it cedes.
ATI hopes that by the spring the gaming thought-leaders (if that is not an oxymoron) are recommending ATI products to their friends. It is then ATI's job to have OEM products ready to convert that mindshare to marketshare.
Joe DeFuria
Legend
Actually, unless nVidia launches the NV30 within the next 30 days, and has it shipping in October, they have missed the window.
The two periods a year are September/October and MArch/April.
OEMs need to know by September that the product is ready to ship, so that they can plan their holiday line-up. Even if NV30 ships in December, it's too late for that product to compete with the 9700 for "this period."
For "this period", rumor has it that nVidia will have the NV18 and NV28. We should know more next week.
Sabastian said:
In all fairness if they make 10% from the HE market, they make 90% from the other markets. So in terms of relative importance it is obviously much smaller than the rest. I mean the only other markets they have are Value, Mainstream, Performance, right? And I guess Motherboard chipsets now.
kid_crisis
Newcomer
Sabastian said:
Pure Nonsense. Did they comment on NV28 or NV18? No. Did analysts ask questions about them? Yes. So I guess NV18 and NV28 haven't taped out yet either, by your impeccable logic.
Doomtrooper
Veteran
It was interesting to hear about the enthusiast market, something we all knew but was interesting to hear. There isn't alot of money in the high end and even investors were more interested in TNT and MX sales
LittlePenny
Regular
My cat has just as good of logic as anyone when talking about unannounced products. Why can't you guys talk more about released/announced products, even if this means DX8.1, there is still much I would like to learn concerning those products.
