My understanding of the motivation behind NV's Fermi and future product developments

[dev0lution]

It doesn't hurt AMD a bit as they stock more OEM systems.

That makes no sense. Even assuming they sell more to OEMs...

Today they would sell something like this in the mainstream PC segment:
1 CPU + 1 chipset + 1 mainstream discrete card
or for an intel-based system just 1 discrete card

Post-Fusion/Sandy Bridge they'll be selling:
1 CPU + chipset
and most likely nothing for an intel-based system because Sandy Bridge is "good enough"

Fusion just killed off 1 component of your platform sale and motivated your competitor to try harder which killed any discrete volume you had on theirs.

 

[Juncar]

How? If 99% of product A only results in a 5% Net Income vs. 1% of revenue for product B that results in 95% of your Net Income. Revenue only matter if the revenue stream had good gross margins and low associated expenses required to generate the cash flow associate with that revenue stream (i.e, brand costs, workforce costs, marketing and selling expenses, etc). Long term, revenue by itself is a meaningless metric, unless there is profitability. This is why companies discontinue unprofitable product lines even if they do generate solid revenue. GM and Chrysler quickly come to mind!

But that's not the point here - the integrated chipset market for NV is only about 10% of their profits vs. 30+% for professional graphics and 19% for discrete graphics, which includes laptops and desktops. Therefore, when the company designs their products within limited time and cost constraints, they will focus on their most profitable lines first and foremost, wouldn't you agree?

It would not really be a big problem if the unprofitable product lines were not such a large part of the company's overall revenue. If you were to discountinue a product line, the resources that were dedicated to that product line would be re-directed. But if the product line brought in 99% of the revenue, then you would end up with many idle employees and extra capacity. Without another product line to re-direct all these resources, the company would end up in a bad financial position.

As for GM and Chrysler, their unprofitable product lines had many redundant models, and we all saw what happened to either of these companies.

 

[sandorski]

10% of Nvidia's Revenues(Integrated) is pretty much gone. That in itself is a big issue.

19% of Nvidia's Revenues(Discrete Consumer Cards) is somewhat at risk of being gone. How much I don't really know, but if you look at the current crop of Consumer Fermi cards, that(assuming BD GPU is 5770 level) pretty much shows the Range that will be left for Nvidia to compete in, sorta. The 460/465 is really close to that BD GPU Performance level, so if they can exist one must ask, "At what Price?".

Then again, the question becomes, "What will be the Intel vs AMD Marketshare for these Products?" On the AMD side a 460 might make no sense, but they certainly might make sense on the Intel side. So I think Intel vs AMD will likely have the most affect on NVidias continued Discrete Marketshare.

31% Professional is certainly good for Nvidia, but then again they have dominated that Market so long, their opportunity for Growth seems nonexistent. In fact, it's just another Market where AMD is likely to start concentrating on, simply because it is there and AMD has good reason to continue pushing GPU Up Market both to feed future BD GPU Cores, but also for the $weet on the Professional level.

With nearly 30% of Nvidia's Revenues at Risk and the only potential for growth is in the Mobile space with completely different Established players, I really got to wonder how long Nvidia can maintain the lead in their Only current safe Market of Professional GPUs? I'm going to guess 5 years. If they don't make a breakthrough(actual Sales, not just Tech) in the Mobile space very soon, they are the Walking Dead.

Then again, AMD could choose to simply abandon the Professional Market. Time will tell.

 

[Dribble]

http://www.digitimes.com/news/a20100901PD215.html

imo that's not guaranteed to happen, it's just come out because they are asking manufacturers if they'd buy something like that.

Right now you have:
intel cpu+gpu, southbridge, nvidia optimus chip, nvidia gpu = 4 chips
I suspect their aim is to offer:
intel cpu+gpu, nvidia southbridge + optimus + gpu = 2 chips

So in theory lowering costs.

However it also defeats half the point of optimus which is to allow you to extend an intel integrated only solution with an nvidia gpu very easily as now you'd have to replace the whole southbridge too.

That said for a company like apple who always want an opencl powered gpu it might make more sense.

 

[golem]

19% of Nvidia's Revenues(Discrete Consumer Cards) is somewhat at risk of being gone. How much I don't really know, but if you look at the current crop of Consumer Fermi cards, that(assuming BD GPU is 5770 level) pretty much shows the Range that will be left for Nvidia to compete in, sorta. The 460/465 is really close to that BD GPU Performance level, so if they can exist one must ask, "At what Price?".

Then again, the question becomes, "What will be the Intel vs AMD Marketshare for these Products?" On the AMD side a 460 might make no sense, but they certainly might make sense on the Intel side. So I think Intel vs AMD will likely have the most affect on NVidias continued Discrete Marketshare.

31% Professional is certainly good for Nvidia, but then again they have dominated that Market so long, their opportunity for Growth seems nonexistent. In fact, it's just another Market where AMD is likely to start concentrating on, simply because it is there and AMD has good reason to continue pushing GPU Up Market both to feed future BD GPU Cores, but also for the $weet on the Professional level.

But if AMD makes their BD GPU too good, enough where it can threaten Nvidia's lower mid level discrete market, it also threatens their own discrete market. Intel's combined cpu/gpu puts a cap on how much AMD can charge for BD (even if BD is better), so if the profit from BD doesn't equal the profit from a non BD processor and a discrete video sale then AMD is in trouble too. The only way AMD can avoid losing revenue in this scenario is if they can win market share away from Intel, which is possible, but won't be easy.

If combined CPU/GPU become so good that the discrete cards market substantially shrinks, the only real winner is probably Intel since they don't have a discrete video card line. Intel is really the key here. If they don't charge too much of a premium, or no premium for their cpu/gpu chips then AMD is almost forced to not allow BD to be too good to preserve their video card business.

The professional market is all about developer relations and software/driver support. This is one area Nvidia shines in. Amd has good windows drivers but in almost any other software area and in developer relations they trail and sometimes by a lot to Nvidia. They can turn this around, but they have to really make an effort to do so.

 

[sandorski]

But if AMD makes their BD GPU too good, enough where it can threaten Nvidia's lower mid level discrete market, it also threatens their own discrete market. Intel's combined cpu/gpu puts a cap on how much AMD can charge for BD (even if BD is better), so if the profit from BD doesn't equal the profit from a non BD processor and a discrete video sale then AMD is in trouble too. The only way AMD can avoid losing revenue in this scenario is if they can win market share away from Intel, which is possible, but won't be easy.

If combined CPU/GPU become so good that the discrete cards market substantially shrinks, the only real winner is probably Intel since they don't have a discrete video card line. Intel is really the key here. If they don't charge too much of a premium, or no premium for their cpu/gpu chips then AMD is almost forced to not allow BD to be too good to preserve their video card business.

The professional market is all about developer relations and software/driver support. This is one area Nvidia shines in. Amd has good windows drivers but in almost any other software area and in developer relations they trail and sometimes by a lot to Nvidia. They can turn this around, but they have to really make an effort to do so.

AMD might not care much about Vidcard Sales loss. It really depends on how it's going to affect BD Pricing, Marketshare Gains, and also whether the Onchip GPU can be effectively Crossfired. They'll also still have the Intel system Vidcard Market to exploit.

As for the Pro Market and Developer Relations and such, that can change. Personally I don't care much about that Market as it has no bearing on my Needs. I could see NVidia become like Matrox, still existing, but mostly a non-Issue for most.

It's going to be an interesting next few years.

 

[Cookie Monster]

Fusion is going to perform at HD5770 level? Where are you guys getting this information from?

Fusion GPU core has rumored to consist of roughly 400SPs (you can sort of tell from the die shot). Its performance figures should be close or not to far behind a 5570. This is of course being optimistic since a discrete low end card with its own dedicated memory system is going to perform better than the integrated solution. AMD themselves already said that even with fusion family of products, they still will have discrete low end cards because they are for different markets.

Even though Sandybridge is far superior to its old GMA counterpart is still miles behind in performance to low end discrete cards from the other two vendors. The recent article of nVIDIA's mobile DX11 parts show how its lowest end part could be potentially three times as fast as SB. Unless theres a breakthrough in process technology and memory bandwidth or even processor architecture breakthrough(for example), discrete cards will always be here to stay.

 

[ModestGamer]

That makes no sense. Even assuming they sell more to OEMs...

Today they would sell something like this in the mainstream PC segment:
1 CPU + 1 chipset + 1 mainstream discrete card
or for an intel-based system just 1 discrete card

Post-Fusion/Sandy Bridge they'll be selling:
1 CPU + chipset
and most likely nothing for an intel-based system because Sandy Bridge is "good enough"

Fusion just killed off 1 component of your platform sale and motivated your competitor to try harder which killed any discrete volume you had on theirs.

You less chips at a higher margin lowering parts count and packaging cost. There is no downside for AMD whatsoever.

 

[ModestGamer]

[redacted]. Wave goodbye to your favorite company in every market but discrete GPU and Workstation.

You really are very ignorant about bussiness.

you comparing profit to revenue.


Enough with the personal attacks and insults. Not acceptable.

Moderator Idontcare

Dude I don't know how to spell it out for you any clearer.

This is what you said:

"integrated laptops - gone" = 4.5% of NV's profits

"integrated cell phones - gone" = Proof? Neither AMD nor Intel make anything remotely competitive in this market, dominated by ARM and Qualcomm. So you are saying that AMD and Intel have something better than Tegra 2? Really? What is it?

I guess you missed the memo where AMD said: "Advanced Micro Devices has no plans to develop direct competitors to system-on-chips (SoCs) based on designs developed by ARM holdings, such as those that power smartphones or the most popular tablet at present, the Apple iPad.
http://www.xbitlabs.com/news/cpu/di...M_System_on_Chip_Devices_Chief_Executive.html

What about Intel: They are just thinking about starting to compete in the mobile space with their recent acquisition of Infineon - http://www.dailytech.com/Intel+Acqu...ises+Not+to+Kill+ARM+Support/article19480.htm

"Integrated PCs - gone" = 5.3% of NV's profits

"Game consoles - gone" = I didn't know you have insider information on what GPUs will be in the next PS4, Xbox 720 and Wii 2. We all would love to know this info, so please share.

I am not going to continue to argue with you because you provide:

1. No facts/evidence to support your claims. (such as if 85% of users don't upgrade their computers, why have nV and AMD been in business of discrete graphics for the last 15 years? If this it to hurt NV, then this statement also implies that the majority of discrete graphics purchases are driven by upgrades and not OEM or new system builds?

Or, such as failure to backup your claims that integrated chipset business finances future product development at NV more than any other revenue stream as "integrated" is less than 10% of their cash flows...so how can that be?)

2. You just keep derailing the thread with "NV will die" attitude, without conveying what you are trying to say in a more coherent manner that won't offend posters in this thread.

 

[ModestGamer]

5670 levels is what alot of AMD source claim and looking at dies shots that is beliveable. which shuts out the gtx 220 and below market and AMD's own 5670 and below and HTPC cards for OEM desktops.

Fusion is going to perform at HD5770 level? Where are you guys getting this information from?

Fusion GPU core has rumored to consist of roughly 400SPs (you can sort of tell from the die shot). Its performance figures should be close or not to far behind a 5570. This is of course being optimistic since a discrete low end card with its own dedicated memory system is going to perform better than the integrated solution. AMD themselves already said that even with fusion family of products, they still will have discrete low end cards because they are for different markets.

Even though Sandybridge is far superior to its old GMA counterpart is still miles behind in performance to low end discrete cards from the other two vendors. The recent article of nVIDIA's mobile DX11 parts show how its lowest end part could be potentially three times as fast as SB. Unless theres a breakthrough in process technology and memory bandwidth or even processor architecture breakthrough(for example), discrete cards will always be here to stay.

 

[Genx87]

5670 levels is what alot of AMD source claim and looking at dies shots that is beliveable. which shuts out the gtx 220 and below market and AMD's own 5670 and below and HTPC cards for OEM desktops.

If it were released today, but what about in 12 months? We will have seen two new generations from AMD and Nvidia by then. A 460 could be a 100-125 dollar discrete card.

 

[Dribble]

Still think this discussion is somewhere slightly south of cloud 9.

Remember this is a CPU + gpu part. There is no indication the amd cpu is going to be any more competitive with whatever intel have then the current parts. Outside of gaming the cpu still matters a lot more then the gpu and if you were a serious gamer you'd get a discrete gpu anyway.

Combine that with the fact that amd have no hope of competing with intel on a marketing front, and for most users the intel gpu will work just fine you end up with fusion just being a budget part. Any powerful machine will always end up with an intel cpu. Any powerful gaming machine will also get discrete graphics.

Hence all fusion can ever hope to be is a budget part for cheap pc's and laptops. Imo all this talk of it competing in other areas is just fantasy.

 

[Wreckage]

5670 levels is what alot of AMD source claim and looking at dies shots that is beliveable. which shuts out the gtx 220 and below market and AMD's own 5670 and below and HTPC cards for OEM desktops.

So next year's Fusion may beat last year's GT220? I'm trying really hard not to yawn. Basically goes along with everything I have been saying. Fusion will replace the current integrated market, but will have little or no effect on the discrete market. That is if it even gets released.

If anything it could just mark the beginning of AMD's exit from the discrete market.

 

[ModestGamer]

q4 2010

If it were released today, but what about in 12 months? We will have seen two new generations from AMD and Nvidia by then. A 460 could be a 100-125 dollar discrete card.

 

[Wreckage]

If it were released today, but what about in 12 months? We will have seen two new generations from AMD and Nvidia by then. A 460 could be a 100-125 dollar discrete card.

Yeah if you read this thread it's looking more like 2Q2011, which could be nearly 12 months. These delays are going to sink them.

http://forums.anandtech.com/showthread.php?t=2102848

 

[evolucion8]

So next year's Fusion may beat last year's GT220? I'm trying really hard not to yawn. Basically goes along with everything I have been saying. Fusion will replace the current integrated market, but will have little or no effect on the discrete market. That is if it even gets released.

If anything it could just mark the beginning of AMD's exit from the discrete market.

Doubtful, AMD is so competitive in the discret market, why they would drop it?

 

[Wreckage]

Doubtful, AMD is so competitive in the discret market, why they would drop it?

They have not had the fastest chip in a generation since, what the x1900? 8xxx vs 2xxx, 9xxx vs 3xxx, 2xx vs 4xxx, 4xx vs 5xxx. All wins for NVIDIA. Either way, the high end discrete market is not a huge money maker. NVIDIA can keep making high end cards because they also make huge amounts of money from professional and server/super computer markets. AMD is basically not even remotely competitive in those markets.

 

[dev0lution]

You less chips at a higher margin lowering parts count and packaging cost. There is no downside for AMD whatsoever.

Margin will be the same as what they currently get for a CPU. Yes, cost would be lower but so would revenue because they aren't selling a GPU anymore. Net result is less revenue and profit.

Would it be better if I put it in a giant font size for you?

 

[Janooo]

They have not had the fastest chip in a generation since, what the x1900? 8xxx vs 2xxx, 9xxx vs 3xxx, 2xx vs 4xxx, 4xx vs 5xxx. All wins for NVIDIA. Either way, the high end discrete market is not a huge money maker. NVIDIA can keep making high end cards because they also make huge amounts of money from professional and server/super computer markets. AMD is basically not even remotely competitive in those markets.

4xx??
With very limited sales.
What would you suggest to compare to AMD 6xxx?

 

[SlowSpyder]

They have not had the fastest chip in a generation since, what the x1900? 8xxx vs 2xxx, 9xxx vs 3xxx, 2xx vs 4xxx, 4xx vs 5xxx. All wins for NVIDIA. Either way, the high end discrete market is not a huge money maker. NVIDIA can keep making high end cards because they also make huge amounts of money from professional and server/super computer markets. AMD is basically not even remotely competitive in those markets.

When is the last time you bought just a graphics chip? When is the last time you bought a graphics board? Who has the fastest single component on a graphics board doesn't determine who 'wins'. If it did I guess we could argue that AMD 'wins' because they have faster memory?

AMD's graphics division seems to be doing well, but their CPU division not so much. I see Fusion and SB pushing Nvidia out of the ultra lowend, but I certainly don't see them pushing out Nvidia... at least not in the near term.

 

[golem]

Margin will be the same as what they currently get for a CPU. Yes, cost would be lower but so would revenue because they aren't selling a GPU anymore. Net result is less revenue and profit.

Would it be better if I put it in a giant font size for you?

This.

The only way to increase volume and therefore revenue is to take it way from Intel.

 

[Dribble]

The only way to increase volume and therefore revenue is to take it way from Intel. Won't be easy because Intel CPU's are better and GPU will probably be good enough.

That was the plan, only for it to work amd had to beat intel to market with the combined cpu+gpu. They should have - they bought ati years ago so had all the bits but managed to mess it up sufficiently that intel not only got the combined cpu+gpu out first, but but managed a years head start. Now instead of taking the fight to intel all they can hope to do is stop the rot. Heads should have rolled for that failure, but I think coming second is almost so ingrained in the amd mentality they probably thought it was ok.

 

[RussianSensation]

19% of Nvidia's Revenues(Discrete Consumer Cards) is somewhat at risk of being gone.

I disagree. GPUs on SB will bring little difference to discrete graphics market. Right now the slowest 310M series from NV is 3x faster than Intel's HD graphics. 310M is about as fast as the 5470 from ATI: http://www.anandtech.com/show/3862/toshiba-t235d-amd-2010-ultrathin/5

SB graphics will be comparable to 5470. However, the slowest of the 400M GPUs should still pack quite a wallop when it comes to gaming. With three times as many cores as G310M and twice the memory bandwidth, the 415M will be at least double the performance of the 310M. That means the worst NV integrated graphics will be almost double the performance of SB graphics.

Also, I am not sure where you got that Bulldozer's GPU will be as fast as GTX460/465.... It will be at most 480 SPs. That means about 2x less performance than the 5770 which GTX460 handily smokes. Also by the time Fusion ships with 480sp core, GTX560 will replace the 460 and 6770 will replace the 5770. Discrete graphics won't stand still either.

One big reason most people didn't want discrete graphics was because it lowered the battery life. With optimus technology, that's a non-issue.

 

[golem]

That was the plan, only for it to work amd had to beat intel to market with the combined cpu+gpu. They should have - they bought ati years ago so had all the bits but managed to mess it up sufficiently that intel not only got the combined cpu+gpu out first, but but managed a years head start. Now instead of taking the fight to intel all they can hope to do is stop the rot. Heads should have rolled for that failure, but I think coming second is almost so ingrained in the amd mentality they probably thought it was ok.

Totally agree, since Intel already has their part out. AMD best hope isn't increased profits but just to not lose more to Intel. The losers in this are Nvidia and AMD. The winners are Intel and low/lower mid end customers. The computer makers as a whole probably stay even. They pay less for part, but because of competition will charge less too.

 

[Martimus]

Also, I am not sure where you got that Bulldozer's GPU will be as fast as GTX460/465.... It will be at most 480 SPs. That means about 2x less performance than the 5770 which GTX460 handily smokes. Also by the time Fusion ships with 480sp core, GTX560 will replace the 460 and 6770 will replace the 5770. Discrete graphics won't stand still either.

Why is everyone talking about the integrated GPU on a Bulldozer processor? There is no integrated GPU on the Bulldozer! Llano and Bobcat are the two AMD architectures with integrated graphics on die (the chips are called APU's).

There is a 20+ page thread about the Bulldozer architecture in the CPU forum if you are interested in finding out about that chip. There was a similar size thread about the Llano as well back when the architecture for that was revealed as well.

Here is a screenshot of the Llano die, which shows maybe 480 or 960 SP's (It shows 12 sets of shaders, but they may be 40 SP's or 80 SP's, or maybe some other amount each.)