NVidia's bumpy ride

[shangshang]

Until GPU computing becomes more pervasive in the masses such that it warrants the price of a huge monolithic GPGPU, NV strategy of satisfying both the gamers and the pure computer will fail. Most hard core enthusiat gamers, will only have a passive interest in GPU computing, so at some point, NV will have to split the monolith chip up somehow to make it more competitvely on the price side of things. NV's management right now have their eyes set on the mobile device market, and they would be more than willing to concede the gaming and PC markets to become one a dominant player in the mobile device space.

 

[SlowSpyder]

Originally posted by: BenSkywalker

but I thought it was odd that you seemed to think that Nvidia's total and complete silence in the midst of their competitor launching a next gen product was some how possibly a good sign, like they were playing possum.

If you look back through their history, the more worried they are, the louder they get.

Charlie has been bashed a lot, and probably rightfully so. But lately, even if he's not 100% correct, his stories seem to be headed in the right direction, so to speak.

Which story would that be? I didn't see Charlie talking about AMD abandoning the mid and high end market when they flushed the 38xx parts, nor have I seen him comment on the fact that they are currently flushing the 4xxx parts. I saw him lie through his teeth about the GT300 tapeout, even lied about it when he was showing proof that he was absolutely wrong. Charlie really hasn't been right about anything at all that I have seen, actually it seems that he has been as close to the opposite of right as you can get lately.

People are asking about why nV isn't die shrinking the GT200 line like that really needs to be said? How about we phrase it differently- why isn't ATi die shrinking the 4890? Exactly.

I don't remember really, did Nvidia make much noise before the G80? I don't remember a whole lot being leaked before the launch. But than again, that was a different situation, their nearest competiton was the x1950xtx which the G80 clearly outpaced in every way.

Just a few Charlie stories off the top of my head (and maybe there are others that I'm forgetting that you or others may know of that are indeed completely wrong that you can post) are the 'bump gate' story and the likely time frame for the launch of Fermi. I also remember his story stating that Nvidia could not do a daul GPU card using the GT200, while Nvidia obviously did create a dual GPU card, it wasn't until they went to 55nm. There is no questioning Charlies dislike for Nvidia, he seems to go out of his way to dig up negitive news about Nvidia.

Well, the 38x0 was never realy a 'high' end part. If AMD pulls the 48x0 cards from the market they already have the 5770 and 5750 launched. If they pull other lower end parts, at the very least they have replacements on a road map. I think it would be odd if the 48x0 cards were to start drying up before AMD had a replacement out. Nvidia has been pretty quiet about upcoming parts, we know Fermi is in the works, but that seems to be the only part we know of. Maybe Nnvidia will suprise us and launch a whole line of products, but it looks like by the time they may get Fermi out AMD will have their x2 part out and possibly the whole line up.

 

[zod96]

I'm sure they will have some of the new cards ready for Christmas. But I bet they will be impossible to get it and will be in very very short supply. I bet we wouldn't see good amounts of their new card until January...

 

[Stoneburner]

Originally posted by: BenSkywalker

but I thought it was odd that you seemed to think that Nvidia's total and complete silence in the midst of their competitor launching a next gen product was some how possibly a good sign, like they were playing possum.

If you look back through their history, the more worried they are, the louder they get.

Charlie has been bashed a lot, and probably rightfully so. But lately, even if he's not 100% correct, his stories seem to be headed in the right direction, so to speak.

Which story would that be? I didn't see Charlie talking about AMD abandoning the mid and high end market when they flushed the 38xx parts, nor have I seen him comment on the fact that they are currently flushing the 4xxx parts. I saw him lie through his teeth about the GT300 tapeout, even lied about it when he was showing proof that he was absolutely wrong. Charlie really hasn't been right about anything at all that I have seen, actually it seems that he has been as close to the opposite of right as you can get lately.

People are asking about why nV isn't die shrinking the GT200 line like that really needs to be said? How about we phrase it differently- why isn't ATi die shrinking the 4890? Exactly.

You talk funny... hahahaha

I think I have difficulty following your posts because you commas where you could use periods.

 

[jiffylube1024]

As others have said - EOL'ing their lineup of cards simply means they won't be producing more of them, and that they will use up their existing inventory over the next 4-6 months. This could very well create a leaner and meaner Nvidia come early 2010, but that would assume that Nvidia produces competitive cards that aren't overly expensive to manufacture on 40nm.

I'm wondering if Nvidia will design and manufacture a cheaper Fermi, a la ATI 47xx and 57xx series, because I don't see how they could make money just cutting down GT300 cards, with its predicted 467 mm² die size. Even given that Fermi is about 100 mm² smaller than GT200, it's still 130mm² bigger than the 58xx series, and ATI will have been producing 58xx series cards for 4-6 months by the time Fermi comes out.

 

[shangshang]

^ yep NV will going to need to design a cheaper and castrated Fermi if they wanna stay competitive price wise.

====

I don't think it'll take 4-6 months to clear inventory on the current GT200 cards either. I think max is 2 months! Any GT200 cards not selling after 2 months will be sold at a definite loss because in 2 months people will only gonna want to buy GT300-based and HD5xxx-based cards, not relatively expensive GT200-based cards

 

[Stoneburner]

Originally posted by: shangshang
^ yep NV will going to need to design a cheaper and castrated Fermi if they wanna stay competitive price wise.

====

I don't think it'll take 4-6 months to clear inventory on the current GT200 cards either. I think max is 2 months! Any GT200 cards not selling after 2 months will be sold at a definite loss because in 2 months people will only gonna want to buy GT300-based and HD5xxx-based cards, not relatively expensive GT200-based cards

Well, you forget BestBuy will sell the gts260 for $500 until 2010 and best buy customers will still buy them for their laptops.

 

[MarcVenice]

I simply wonder how bad this will hurt Nvidia in the long run. One bad qaurter and hell won't freeze over just yet, but what if AMD takes back 10% of market share from Nvidia, will Nvidia be able to take it back? What if derivatives from Fermi only come out beginning of Q2 2010?

I hope Nvidia doesn't forsake the gaming-market, because we need AMD and Nvidia to keep each other honest, like AMD keeps Intel honest...

 

[tviceman]

Originally posted by: Fox5

Originally posted by: GaiaHunter
Well I guess the only thing we can conclude from all these rumours is that NVIDIA doesn't think it is worthwhile, for whatever reason, to use the GT200 to erode the 5800 series price.

If you asked me a couple weeks ago, I would have expected nVidia to drop their GT200 price and then try to capitalize on brand recognition and features like physX and CUDA to keep a higher degree of pressure on ATI.

If nVidia can indeed do a hard launch by xmas, they probably haven't lost much, depending on how is their product price and its performance.

But from what I read, people like Anand and Kyle, don't put much money on that.

Well, the GT200 series might already be as low as nvidia can afford. They're already priced above the equivalent 48xx parts, going any lower may have just broken nvidia's back.
I don't see why nvidia didn't go for a GT200 die shrink though, but maybe any effort on that would have pushed GT300 back even further. Still, unless they rapidly get all the GT300 parts out, they'll have quite a gap in their product line up.

I read awhile back that they actually did try a die shrink to 40nm with the gt200, and the results were "disastrous." I can't remember from where, but the fact that gt200 was made originally for 65nm, it's not surprising they couldn't get it through two die shrinks without significant modifications to the chip's design.

 

[SmCaudata]

Intel is putting graphics on most of their mainstream CPUs. Larabee is touted as having strong graphics/gaming capability. AMD/ATI will likely move in the same direction soon. ATI has been pushing for more FPS and NVidia has been pushing for more GP. It is sort of poetic justice that the closed behavior Nvidia uses with all of their technologies could end up forcing them out of the market as the other providers have a more integrated solution potentially offering better performance. Eliminating the need for a PCIe BUS could potentially speed up graphics performace as long as they can keep the GPU-memory bus fast.

Nvidia is looking for their niche. There are a lot of GPGPU fans out there. For pure FPS per $$ spent nothing beats the ATI cards. That may likely continue through this generation.

 

[T2k]

Yes, I agree: short on being an x86 license holder NV has little options when facing down the barrel from both sides in the future, AMD/Radeon and Intel/LRB - they have to create their own market or they will shrink to a small discrete graphics card mfr, slowly spending more and more on R&D to keep up with the other two and eventually face extinction.

 

[BenSkywalker]

I simply wonder how bad this will hurt Nvidia in the long run. One bad qaurter and hell won't freeze over just yet, but what if AMD takes back 10% of market share from Nvidia, will Nvidia be able to take it back?

What it will come down to is very simple, how good Fermi turns out. ATi was 'late' with the 4xxx parts, did that mean they were horrible? ATi wasn't late with their 3xxx parts, did that mean they were superb? Obviously being out first has its advantages, but I can't recall it ever deciding a generation.

I hope Nvidia doesn't forsake the gaming-market

nVidia dropped over $1Billion on Fermi- anyone would have to be a bombastic simpleton to believe they are going to simply abandon the gaming segment because of a fab issue. If anything, they will abandon TSMC.

I read awhile back that they actually did try a die shrink to 40nm with the gt200, and the results were "disastrous."

Makes no sense to have your mid range a full DX revision behind. In the budget segment you can easily get away with it, but anythnig over $100 is pushing it, over $200 and you are out of your mind. nVidia didn't get to where they are out of stupidity.

slowly spending more and more on R&D to keep up with the other two and eventually face extinction.

This is hard to fathom given the financial history. While nVidia actually has money to continually increase R&D, AMD doesn't. nVidia's R&D budget has exceeded ATi's total revenue for years. Intel could continue to try and outspend them, but that would be rather futile when nV yanks their rights to make video chips at all.

 

[MarcVenice]

Originally posted by: BenSkywalker
I simply wonder how bad this will hurt Nvidia in the long run. One bad qaurter and hell won't freeze over just yet, but what if AMD takes back 10% of market share from Nvidia, will Nvidia be able to take it back?

What it will come down to is very simple, how good Fermi turns out. ATi was 'late' with the 4xxx parts, did that mean they were horrible? ATi wasn't late with their 3xxx parts, did that mean they were superb? Obviously being out first has its advantages, but I can't recall it ever deciding a generation.

Actually, I think when Nvidia launched G80, they completely mudstumped ATI for roughly a year-ish? Only when the 3800-serie came out could ATI somewhat compete.

I hope Nvidia doesn't forsake the gaming-market

nVidia dropped over $1Billion on Fermi- anyone would have to be a bombastic simpleton to believe they are going to simply abandon the gaming segment because of a fab issue. If anything, they will abandon TSMC.

It's not a fab issue? The fab issue has been solved, yields at 40nm are pretty much 'okay', yet not stellar. The issue has affected ATI as much as Nvidia, but Nvidia simply decided they had more time and ATI came out with a less complicated product, so they could launch it quite fast.

I read awhile back that they actually did try a die shrink to 40nm with the gt200, and the results were "disastrous."

Makes no sense to have your mid range a full DX revision behind. In the budget segment you can easily get away with it, but anythnig over $100 is pushing it, over $200 and you are out of your mind. nVidia didn't get to where they are out of stupidity.

This indeed makes no sense, I heard vague rumours about gt212 being pulled out of the fridge again, but I ignored them.

slowly spending more and more on R&D to keep up with the other two and eventually face extinction.

This is hard to fathom given the financial history. While nVidia actually has money to continually increase R&D, AMD doesn't. nVidia's R&D budget has exceeded ATi's total revenue for years. Intel could continue to try and outspend them, but that would be rather futile when nV yanks their rights to make video chips at all.

Uh, you think Nvidia can stop Intel from making video-chips? Which patents does Nvidia have, or which license does Intel have from Nvidia that could stop Intel from making video chips?

 

[BenSkywalker]

Actually, I think when Nvidia launched G80, they completely mudstumped ATI for roughly a year-ish? Only when the 3800-serie came out could ATI somewhat compete.

The big problem was that the 2xxx and 3xxx series sucked, not when they came out. That is a rather huge amount of time, if ATi had taken another 3 months and launched the 4xxx series parts when they launched the 3xxx parts don't you agree they would have wiped the floor with nVdia? The launch timeframe is of course an issue as time is money, but in terms of broader market acceptance being late and good>>>>>>>early and mediocre.

The fab issue has been solved, yields at 40nm are pretty much 'okay', yet not stellar. The issue has affected ATI as much as Nvidia, but Nvidia simply decided they had more time and ATI came out with a less complicated product, so they could launch it quite fast.

Yields are 'OK' on simple parts, I wouldn't exactly call that sorted out. Let me rephrase things a bit, if by some divine act Intel was fabbing for nVidia do you have any doubt they would be out right now? I honestly don't.

Uh, you think Nvidia can stop Intel from making video-chips?

Anything resembling what we consider a video chip today, yes. The punch cards was a bit overboard, but they would be limited to either a 2D only or a purely procedural 3D offering. nVidia owns all of their own and all of 3dfx's IP, there is no way you are making a 3D chip that will work with anything we use today without using some of their IP. If you want specifics, texture sampling hardware(there are a ton, that's just an awful big one to get around ). The same thing could be said about ATi, but they can not mess with their cross licensing as they need the x86 license obviously, the only thing Intel was giving nVidia was a license to make chipsets. Intel opened this can of worms on their own.

 

[MarcVenice]

But, does ATI have IP that Nvidia uses to make gpu's? I severely doubt Intel could be stopped from releasing LRB because it would infringe patents owned by Nvidia, but maybe I could be proven wrong.

And, yields are OK on more complex parts too, like the HD 5870. They were ok-ish on RV740. Thing is, TSMC simply wasn't equiped to pump out big volumes of 40nm wafers, they are in fact still getting up to speed. If by some divine act Intel was fabbing for Nvidia, Nvidia might have released right about now, but AMD would also have released 2 months earlier. Without a doubt did the 40nm-proces slow things down, but designing a chip for a new proces is always difficult, no matter who fabs it.

 

[BenSkywalker]

But, does ATI have IP that Nvidia uses to make gpu's?

Absolutely. They need each others cross license to make any parts.

I severely doubt Intel could be stopped from releasing LRB because it would infringe patents owned by Nvidia, but maybe I could be proven wrong.

You aren't appreciating the scale of the issue. At this point, they are going to court to pull their cross license agreement altogether- that is why they stopped even trying to make chipsets for the new Intel chips. It will go to court, but by letter of the law Intel won't be able to make any graphics chips when the case is over. They will still have the IP cross license from ATi, but it won't save even their integrated solutions from nV/3dfx's patents. I can't see Intel letting this happen, but I don't know what Jen Hsun is going to demand in order for them to rework the contract.

And, yields are OK on more complex parts too, like the HD 5870.

So where are the complex parts they are having decent yields with? I guess in relative terms the 5870 qualifies, but 330mm2 wouldn't have been considered large even a couple of years ago.

Without a doubt did the 40nm-proces slow things down, but designing a chip for a new proces is always difficult, no matter who fabs it.

Of course, but moving over to GF may be an attractive proposition once they gain a bit more distance from their former coworkers.

 

[spankure]

Originally posted by: BenSkywalker
Anything resembling what we consider a video chip today, yes. The punch cards was a bit overboard, but they would be limited to either a 2D only or a purely procedural 3D offering. nVidia owns all of their own and all of 3dfx's IP, there is no way you are making a 3D chip that will work with anything we use today without using some of their IP. If you want specifics, texture sampling hardware(there are a ton, that's just an awful big one to get around ). The same thing could be said about ATi, but they can not mess with their cross licensing as they need the x86 license obviously, the only thing Intel was giving nVidia was a license to make chipsets. Intel opened this can of worms on their own.

Don't Intel have a lot of patents about 3d that nVidia would need. When Intel bought the Ip's of Real3D i would think they got some IP's regarding the basic of 3d. They most likly have a bunch of IP's that nvidia need to be able to produce a working graphics card. You really think that the only thing Intel has to offer Nvidia is a DMI bus license ?

 

[MarcVenice]

Let's start a seperate thread and see what other people have to say, because this of course is mighty interesting.