Discrete GPU is dying? NVIDIA Disagrees

[Keysplayr]

+1



The PS4 APU in a PC will be enough for 9 out of 10 people. The GPU market, as we know it now, will just become uneconomical. Exactly how long before that happens? Not sure, but I'd imagine that by the time we get 14nm APU's not too many people will be interested in paying the money for a discrete card. Not enough to support the development, IMO.

And 99% of all statistics are made up on the spot.

 

[Imouto]

In a shrinking market Nvidia can feast on AMD marketshare for a fixed amount of time.

 

[raghu78]

In a shrinking market Nvidia can feast on AMD marketshare for a fixed amount of time.

yeah both Nvidia and AMD are going to see the dGPU market shrink. AMD will cannibalize it with their APUs and Intel will eat into both Nvidia and AMD dGPU sales below USD 100. Nvidia is riding a wave for the past 2 -3 years where AMD has lost massive marketshare in notebook because of Nvidia's superior Optimus solution. In 2010 AMD had 60% notebook dGPU market share and Nvidia had 40%. Now Nvidia has 70% notebook dGPU market share and AMD has 30%. AMD has also stopped going behind low volume design wins to cut costs as they are in bad financial shape due to a bleeding CPU division.

But guess what Intel is going after the dGPU market with Crystalwell GT3e (embedded DRAM) and next year with Broadwell. the company who will get hit badly are Nvidia as they have the lion's marketshare in notebook. traditionally desktop dGPU has been 60% nvidia and 40% ati. but with kepler Nvidia increased that too to 65%. You can expect AMD to fight back in desktop and get back to 40%. but notebook looks difficult for AMD. even otherwise Intel is going specifically after the highest volume segment of the notebook market. the entry to lower range of mid level graphics like GT650M. And Nvidia has the market share to lose.

lets revisit the situation when 14nm Broadwell is ramping. you will see the effect of Haswell on Nvidia and Broadwell will accelerate the trend.

 

[MrK6]

yeah both Nvidia and AMD are going to see the dGPU market shrink. AMD will cannibalize it with their APUs and Intel will eat into both Nvidia and AMD dGPU sales below USD 100. Nvidia is riding a wave for the past 2 -3 years where AMD has lost massive marketshare in notebook because of Nvidia's superior Optimus solution. In 2010 AMD had 60% notebook dGPU market share and Nvidia had 40%. Now Nvidia has 70% notebook dGPU market share and AMD has 30%. AMD has also stopped going behind low volume design wins to cut costs as they are in bad financial shape due to a bleeding CPU division.

But guess what Intel is going after the dGPU market with Crystalwell GT3e (embedded DRAM) and next year with Broadwell. the company who will get hit badly are Nvidia as they have the lion's marketshare in notebook. traditionally desktop dGPU has been 60% nvidia and 40% ati. but with kepler Nvidia increased that too to 65%. You can expect AMD to fight back in desktop and get back to 40%. but notebook looks difficult for AMD. even otherwise Intel is going specifically after the highest volume segment of the notebook market. the entry to lower range of mid level graphics like GT650M. And Nvidia has the market share to lose.

lets revisit the situation when 14nm Broadwell is ramping. you will see the effect of Haswell on Nvidia and Broadwell will accelerate the trend.

This and more. It will be interesting to see how much Intel flexes its muscles in the next year or two. It will come down to whether notebook manufacturers feel Intel's solution is superior or not to not offer an nvidia equivalent. I know I'm waiting for a nice power-sipping Ultrabook that can adequately game.

 

[f1sherman]

lets revisit the situation when 14nm Broadwell is ramping. you will see the effect of Haswell on Nvidia and Broadwell will accelerate the trend.

Fx1:

60.5% Margins - Nvidia helped by *&$#@*&( Maxwell

 

[raghu78]

Fx1:
60.5% Margins - Nvidia helped by *&$#@*&( Maxwell

I think people are underestimating the 800 lb gorilla of the industry - Intel. Intel muscled out Nvidia fom the chipset market by not licensing QPI to Nvidia and integrating the northbridge and memory controller and now the southbridge. Nvidia has made up for the loss of chipset revenue by growing GPU revenue. They have grown at the expense of ATI/AMD especially in the notebook market. Nvidia have everything to lose and Intel has everything to gain once Intel goes after entry level notebook dGPU market with Haswell Crystalwell GT3e. And if AMD put up a better fight in notebook dGPU with Volcanic Islands then it gets worse. 2014 should give us an idea. I have a feeling Nvidia has seen its best GPU revenue days and the future is going to be difficult.

 

[Grooveriding]

If Intel can manage to get an APU in their 14nm CPUs that comfortably runs games at 1080p with decent drivers, that will cause a big shift. Most people who buy discreet cards are spending about $250 or less to play at 1080p. They would be happy to just pay for the CPU that does it all and save the money.

It is reasonable to expect if Intel manages a chip like that they are going to lay the pain down on the competition. Enthusiasts will be left spending who knows how much for high end discreet cards at that point, well over a thousand at the rate it is going right now.

 

[f1sherman]

@raghu78

you are holding everything fixed (gaming progress, DirectX/software stack, display resolutions, Nvidia R&D (maxwell,denver,volta), new market opportunities)
while letting Intel's future chip evolve (to a point where they can run yesterday's games at 1080p)

GRID, GPU virtualization, Android, Shield, tablets, smartphones + $4B in a bank is a guarantee that NVIDIA can focus on any of these markets, and never look back at sub $100-150 GPU even if they lose all of it.


ohh... and Intel will be sued by their shareholders the day when they offer $200 worth GPU on $180 CPU die

 

[TheReaWarmonger]

Which is integrated into the CPU, which is I believe his point.

Which every APU has. It wont take long for CPU + GPU architecture to become unified as well.

I think in 15 years we'll be posting about "hey, remember when we all had discrete GPUs?".

With the going rate of silicon, we might be moved to a whole new fabrication process by then. I think Kaveri will be the first testament of where APU's are going, and how powerful they will become. I have high hopes that Kaveri will deliver with at least discrete HD 7750 performance graphics. I don't care if the flagship APU costs $169.99, if it can pack a punch like that along with faster steamroller cores it's well worth the money. I'd in all honesty would toss my old desktop setup, and jump on the FM2/FM3 bandwagon. Especially if HSAIL is taken by storm, with no latency between the CPU + GPU. Kaveri is going to be one huge compute monster, and I cannot wait to develop some software that utilizes it.

 

[raghu78]

@raghu78

you are holding everything fixed (gaming progress, DirectX/software stack, display resolutions, Nvidia R&D (maxwell,denver,volta), new market opportunities)
while letting Intel's future chip evolve (to a point where they can run yesterday's games at 1080p)

GRID, GPU virtualization, Android, Shield, tablets, smartphones + $4B in a bank is a guarantee that NVIDIA can focus on any of these markets, and never look back at sub $100-150 GPU even if they lose all of it.


ohh... and Intel will be sued by their shareholders the day when they offer $200 worth GPU on $180 CPU die

http://www.anandtech.com/show/6911/...usiness-haswell-gt3e-to-integrate-128mb-edram

Ok lets take it purely on tech. Intel is roughly a process node ahead of the foundries and more if you consider advancements like FINFET. Intel can easily go upto 200 - 230 sq mm die size on their high end quad core die with their flagship iGPU Iris Pro. So you are looking at 14nm FINFET Intel cpu with Iris Pro and EDRAM against a 20nm Maxwell.

Intel's current Haswell core is 14.2 sq mm .

http://semiaccurate.com/forums/showthread.php?t=6857

at 14nm it would be around 8 sq mm. four 14nm shrunk haswell cores would take just 32 sq mm. even with 1 MB L2 and 8MB L3, the high end iGPU will be more than 100 sq mm on a 14nm FINFET process.

and don't forget Intel is charging for every sq mm of silicon. the anandtech article says they are charging USD 50 for just the edram. obviously they will add the Iris Pro cost into the core i7 models and compete directly against Nvidia dGPU models on performance, beating them on power while significantly undercutting them. Can Nvidia's entry level TSMC 20nm 100 sq mm GPU die models compete against Intel's 14nm FINFET 100 sq mm Iris Pro iGPU. I don't think so.

as for Tegra they had a miserable quarter. down 22% YoY compared to same quarter last year and 50% down QoQ compared to last quarter. read Q1 FY2014 presentation

http://phx.corporate-ir.net/phoenix.zhtml?c=116466&p=irol-IRHome

With Qualcomm sweeping the next gen Google Nexus and Surface RT, Tegra 4 does not have a single major high volume/high profile design win.

http://www.digitimes.com/news/a20130201PD209.html
http://www.digitimes.com/news/a20130318PD225.html
http://www.fudzilla.com/home/item/31138-nvidia-scores-sole-tegra-4-design-win-courtesy-of-zte

so if anything the road ahead is more difficult than ever. T4i is expected to soften the blows but it remains to be seen how this unfolds.

also tech advancements like stacked DRAM will happen earlier with Intel which is a IDM and the world's best semiconductor company than with Nvidia which depends on TSMC. Volta is a 2016 product with stacked DRAM. what I am saying is Nvidia will face Intel's superior manufacturing force. Also Intel will continue to invest in GPU tech as will Nvidia. But when the battle is fought on the silicon playground its always an uphill battle when your competitor has a process node lead and is willing to spend similar die size or even more further down the road.

 

[Imouto]

Intel may be ahead in its process node but lacks in projection big time. The 800 lb gorilla needs a lot of effort to switch directions. Last time I checked it was 3 years behind in the mobile space and that doesn't speak any good about about their roadmaps and planning.

And Intel won't give up easily on their margins, ever. Intel is like a train on the wrong tracks that missed quite a few crossings for the right path.

 

[raghu78]

Intel may be ahead in its process node but lacks in projection big time. The 800 lb gorilla needs a lot of effort to switch directions. Last time I checked it was 3 years behind in the mobile space and that doesn't speak any good about about their roadmaps and planning.

And Intel won't give up easily on their margins, ever. Intel is like a train on the wrong tracks that missed quite a few crossings for the right path.

With silvermont core and baytrail SOC Intel is going to give the ARM crowd a run for their money. the ARM crowd is going to face what AMD faced. a company with the best manufacturing tech on earth and R&D resources to dwarf them all.

http://www.anandtech.com/show/6936/...tecture-revealed-getting-serious-about-mobile

"Intel’s 22nm process should give Silvermont a lot of runway to use. If it can quickly follow up with 14nm, Silvermont’s power advantage could end up being akin to Conroe’s performance advantage in the mid-2000s.

Even so, Silvermont is long overdue. It’s the first mobile architecture where Intel really prioritized smartphones and tablets, and on paper, it looks very good. Now it’s up to Intel to turn a great architecture into great design wins. From what I’m hearing, we may actually see that happen."

with 22nm baytrail in late 2013 and 14nm Atom SOC in late 2014 Intel is going to turn the screws on the competition with their relentless tick tock cadence.

 

[R0H1T]

With silvermont core and baytrail SOC Intel is going to give the ARM crowd a run for their money. the ARM crowd is going to face what AMD faced. a company with the best manufacturing tech on earth and R&D resources to dwarf them all.

Did you miss this part by any chance ~

And Intel won't give up easily on their margins, ever. Intel is like a train on the wrong tracks that missed quite a few crossings for the right path.

Also their next node shrink will be 1.5~2yrs, depending on market conditions, after their previous one meaning that ~

with 22nm baytrail in late 2013 and 14nm Atom SOC in late 2014 Intel is going to turn the screws on the competition with their relentless tick tock cadence.

It won't be as competitive as you might think it would be !

 

[raghu78]

Did you miss this part by any chance ~

who says Intel will give up margins. Intel is going to have Atom on the leading process node from 14nm. Intel's process is one node ahead of the foundries. so for the same amount of transistors they will need a much smaller die.

Also their next node shrink will be 1.5~2yrs, depending on market conditions, after their previous one meaning that ~It won't be as competitive as you might think it would be.

If Intel has problems with future node shrinks the foundries are in even worse shape. the ARM crowd depends on foundries and are much worse than Intel. Intel has mastered FINFET production while foundries won't have FINFET in volume production till 2015 and even then volume will be low.

http://www.extremetech.com/computin...rtex-a57-tape-out-chip-launching-no-time-soon

 

[R0H1T]

who says Intel will give up margins. Intel is going to have Atom on the leading process node from 14nm. Intel's process is one node ahead of the foundries. so for the same amount of transistors they will need a much smaller die.

And how do you think they'll recoup the cost of all that R&D, by pricing their chips in the 40$ range, however their competition(ARM) is at 20 bucks, or lower

 

[Sleepingforest]

Also their next node shrink will be 1.5~2yrs, depending on market conditions, after their previous one meaning that ~

Really? Anand seems to think differently; if Intel is "revving Atom yearly" then we'll see Airmont in 2014.

And how do you think they'll recoup the cost of all that R&D, by pricing their chips in the 40$ range, however their competition(ARM) is at 20 bucks, or lower

Probably by being in high margin $600 flagships and not in low cost phones. They cost more because they're more powerful (even per watt). Furthermore, ARM chips are going to get more expensive as things like the Exynos octocore processor (effectively 2 SoCs in one) becomes more common. Finally, ARM's R&D costs will be no less than Intel's in getting to 14nm.

 

[raghu78]

And how do you think they'll recoup the cost of all that R&D, by pricing their chips in the 40$ range, however their competition(ARM) is at 20 bucks, or lower

with 14nm FINFET (2nd gen) Atom SOC Intel will have a smaller die size than 20nm planar ARM SOC while beating the ARM competition badly on performance and perf/watt. Intel beats ARM A15 with their 5 year old Atom core. With a brand new microarchitecture starting with Silvermont, tick tock cadence and leading edge process Intel is going to take significant market share from the competition. Don't be surprised if Intel bags a Apple or Google smartphone design win in 2015.

 

[Imouto]

With silvermont core and baytrail SOC Intel is going to give the ARM crowd a run for their money. the ARM crowd is going to face what AMD faced. a company with the best manufacturing tech on earth and R&D resources to dwarf them all.

And yet again, no. Samsung, Apple, Google, Amazon and Qualcomm can safely show the middle finger to Intel any day of the week. They're not AMD, not this time.

This may be breaking news to you, but the mobile market has been growing without control without Intel. Hence Intel is disposable. Hence Intel needs the current players to break in.

Unless Intel brands its own devices I'm pretty sure they will have to lower their margins or remain on single digits mobile market share.

 

[R0H1T]

Really? Anand seems to think differently; if Intel is "revving Atom yearly" then we'll see Airmont in 2014.

And by that are you limiting the Tick-Tock to mobile chips only ? OR maybe you're feigning ignorance since they'll be using the 14nm node for Broadwell next year & mobile chips is a logical extension to that plot !

Probably by being in high margin $600 flagships and not in low cost phones. They cost more because they're more powerful (even per watt).

I've discussed this at great lengths in the CPU section that top phone makers(APPLE/SAMSUNG/LENOVO) have in house chips which will be preferred over these parts, even at the high(er) end of the smartphone market, so remind me again where will they get the critical volume from ?

Furthermore, ARM chips are going to get more expensive as things like the Exynos octocore processor (effectively 2 SoCs in one) becomes more common.

That makes no sense, chip costs are going down not up, its only the initial costs that're higher & Samsung already has their own fabs !

Finally, ARM's R&D costs will be no less than Intel's in getting to 14nm.

So TSMC pays ARM for their fab's R&D, yeah right

 

[raghu78]

And yet again, no. Samsung, Apple, Google, Amazon and Qualcomm can safely show the middle finger to Intel any day of the week. They're not AMD, not this time.

This may be breaking news to you, but the mobile market has been growing without control without Intel. Hence Intel is disposable. Hence Intel needs the current players to break in.

Unless Intel brands its own devices I'm pretty sure they will have to lower their margins or remain on single digits mobile market share.

who says Intel is not speaking with Apple for powering the iphone 7 in 2014 or iphone 8 in 2015. Apple switched from powerPC to Intel x86 because Intel provided them the performance within the power envelopes which only Intel could do. No other company could do and so Apple moved to Intel. Apple would not hesitate to switch to Intel if they know that the competition will be far behind in perf and perf/watt.

 

[R0H1T]

who says Intel is not speaking with Apple for powering the iphone 7 in 2014 or iphone 8 in 2015. Apple switched from powerPC to Intel x86 because Intel provided them the performance within the power envelopes which only Intel could do. No other company could do and so Apple moved to Intel. Apple would not hesitate to switch to Intel if they know that the competition will be far behind in perf and perf/watt.

Not sure if joking or serious :hmm:

You think Apple will sink their billions in R&D(that includes litigation money :biggrin on Swift for their man love on Intel :$

 

[Imouto]

who says Intel is not speaking with Apple for powering the iphone 7 in 2014 or iphone 8 in 2015. Apple switched from powerPC to Intel x86 because Intel provided them the performance within the power envelopes which only Intel could do. No other company could do and so Apple moved to Intel. Apple would not hesitate to switch to Intel if they know that the competition will be far behind in perf and perf/watt.

Do you realize this time the competition is Apple itself developing its own mobile SoC? Apple has the best mobile OS and can tweak its SoCs at will based on their needs while it would be the other way around with Intel.

Full control over hardware and software or getting a money grabber like Intel into your home?

 

[Sleepingforest]

And by that are you limiting the Tick-Tock to mobile chips only ? OR maybe you're feigning ignorance since they'll be using the 14nm node for Broadwell next year & mobile chips is a logical extension to that plot !

What are you trying to say here? Intel's Atom chips have a completely different roadmap than their Core counterparts (Bonnel-Saltwell-Silvermont-Airmont versus Sandy Bridge-Ivy Bridge-Haswell-Broadwell). They are planning to advance on a tick-tock yearly; next year is a die shrink to 14nm.

I've discussed this at great lengths in the CPU section that top phone makers(APPLE/SAMSUNG/LENOVO) have in house chips which will be preferred over these parts, even at the high(er) end of the smartphone market, so remind me again where will they get the critical volume from ? That makes no sense, chip costs are going down not up, its only the initial costs that're higher & Samsung already has their own fabs !So TSMC pays ARM for their fab's R&D, yeah right

Really? The cost of building and upgrading 14nm fabs is upwards of a billion dollars. Qualcomm, Samsung, Apple, and other SoC designers still have to design the SoCs, and that takes R&D too. You are the one that makes no sense; how can it magically cost Samsung and co. less to design and manufacture as a whole at 14nm than Intel?

I'm not saying that Intel will magically dominate. They have a hard fight ahead of them, which they may lose, especially since, as you point out, there are many established chip designs. But in tablets and some smartphones (HTC's phones, the Nexus lines, Asus tablets) they have a chance to gain some significant marketshare.

 

[raghu78]

Do you realize this time the competition is Apple itself developing its own mobile SoC? Apple has the best mobile OS and can tweak its SoCs at will based on their needs while it would be the other way around with Intel.

Full control over hardware and software or getting a money grabber like Intel into your home?

Option 1 - Apple ARM SOC on foundry process - 1 process node behind Intel. even more when considering advancements like FINFET. full control on hardware and software.

Option 2 - Intel SOC on Intel industry leading (n+1) process where n is foundry process. OS & apps optimized for x86 and Intel HD graphics. simplifies development as all devices from phones to tablets to laptops and desktops run on the same architecture. lower R&D cost too. The chance to whip the competition like Google and Samsung on perf, perf/watt , battery life and sleeker form factors.

time will tell what Apple will do. I say option 2. my guess is we will have an answer by 2015.

 

[aigomorla]

for every video card out there, there is 5 laptops without a discrete video card.

For every gaming system, there is 10 business machines or more with discrete video cards.

This statement is only telling half truth because both stages of the machines are having a very large climb in demand.

People learned by now, u dont need a machine which can do everything spam'd in every station.

Just machines which will do the job efficiently, and cost effectively, which typically doesnt have a dedicated videocard.