Linus Torvalds: Discrete GPUs are going away

[3DVagabond]

I addressed that. They'll hit XBox One in 1-2 years, of course, but that's ignoring any other factors and just straight-up assuming that same hardware = same real-world performance = better sales for AMD APUs. It's ignoring any differences between PCs and consoles, the possibility of badly optimized PC games, the growth of 4K, tablet adoption, and console price cuts.

I'm not ignoring anything. If not for the issues you mention they could give us PS4 performance today.

 

[BrightCandle]

There are 600 million PCs in the world with discrete graphics cards. Admittedly I doubt most of them are recent but it should put it into perspective just how vast that market currently is. 81% of Steams users have either an AMD or Nvidia graphics card and that is 49 million gamers. That is a big enough market to maintain a discrete graphics card business for quite a while. Yes Intel GPUs are growing slowly on Steam but its a really slow grind and they have a long way to go from the 17% they hold today to dominating that market.

 

[ShintaiDK]

There are 600 million PCs in the world with discrete graphics cards. Admittedly I doubt most of them are recent but it should put it into perspective just how vast that market currently is. 81% of Steams users have either an AMD or Nvidia graphics card and that is 49 million gamers. That is a big enough market to maintain a discrete graphics card business for quite a while. Yes Intel GPUs are growing slowly on Steam but its a really slow grind and they have a long way to go from the 17% they hold today to dominating that market.

Only 65million PCs are shipped with discrete graphics a year(2013). And that amount is shrinking rather quickly. Its estimated the 2014 figure will be around 55 million.

 

[Insert_Nickname]

Well, it's amazing they sell any graphics cards then, isn't it? Actually I wasn't referring to those masses. The masses that want something better than a console.

Of course they do. Unfortunately quite a lot of them are unaware there are better IGP and discrete offerings available.

(do your bit and educate anyone who asks or seems to be in need of something better)

My point was more that every bodies doom-and-gloom scenarios for discrete GPUs have already happened. Despite which we still have discrete GPUs...

I addressed that. They'll hit XBox One in 1-2 years, of course, but that's ignoring any other factors and just straight-up assuming that same hardware = same real-world performance = better sales for AMD APUs. It's ignoring any differences between PCs and consoles, the possibility of badly optimized PC games, the growth of 4K, tablet adoption, and console price cuts.

Depending on how Carrizo turns out, hitting XBone-level could happen pretty quickly. If AMD can solve the memory bandwidth problems.

Only 65million PCs are shipped with discrete graphics a year(2013). And that amount is shrinking rather quickly. Its estimated the 2014 figure will be around 55 million.

That of course ignores the retail sales of GPU, which are likely substantial. Also a lot of discrete GPUs go into supercomputers, which doesn't figure there.

 

[Techhog]

I'm not ignoring anything. If not for the issues you mention they could give us PS4 performance today.

Not really. 8 Jaguar cores wouldn't cut it for anything but a console, gaming-wise. Even with Mantle, the bottleneck would be too big.

Depending on how Carrizo turns out, hitting XBone-level could happen pretty quickly. If AMD can solve the memory bandwidth problems.

In terms of theoretical performance, yes, that should happen next year. The question, however, is why would you buy a PC to beat XBOne but not PS4? On top of this, said desktop PC would cost about $450-500 vs a likely $350 console that you don't have to build. When people want "better than console" gaming, they typically mean 1080p max settings. That'll take more time.

 

[Qwertilot]

They could presumably top the CPU cores from the XB1/PS4 rather easily
(The bandwidth is much more of a problem!).

 

[Techhog]

They could presumably top the CPU cores from the XB1/PS4 rather easily
(The bandwidth is much more of a problem!).

The problem is having 4 big cores and what is essentially a 270X on the same die, since that's what's need to get settings similar to consoles. Even then, this is assuming that people want "better than console" settings but don't mind 30FPS.

 

[Insert_Nickname]

In terms of theoretical performance, yes, that should happen next year. The question, however, is why would you buy a PC to beat XBOne but not PS4? On top of this, said desktop PC would cost about $450-500 vs a likely $350 console that you don't have to build. When people want "better than console" gaming, they typically mean 1080p max settings. That'll take more time.

So in essence you'll be paying $100-150 more to have a fully functional PC+a console. Seems a fair deal, you're not going to get a lot of PC even for $150.

I agree that 1080p gaming is going to take more time at anything above minimum settings. Better then XBone/PS4?, I'd say 3-5 years at the current pace. By then we'll likely have affordable 4K TV/monitors. Try and run any 3D game at 4K on an IGP...

 

[Techhog]

So in essence you'll be paying $100-150 more to have a fully functional PC+a console. Seems a fair deal, you're not going to get a lot of PC even for $150.

I agree that 1080p gaming is going to take more time at anything above minimum settings. Better then XBone/PS4?, I'd say 3-5 years at the current pace. By then we'll likely have affordable 4K TV/monitors. Try and run any 3D game at 4K on an IGP...

This is assuming that you don't already have a functioning PC or tablet, so the potential market won't reach its full capacity right away. That's why it has to do more than match these consoles. There are more variables to consider, which is why certain people are trying to use sales projections to to basically say that AMD will no longer profit from GPUs and Nvidia will go bankrupt. It's certain that progress will slow down (hell, it already has, though mostly due to TSMC), but it's too early to make any projections of when it'll stop. That's also why certain people are making up things about staying in 28nm until 2017.

 

[AtenRa]

Your link shows 66 for some reason. Out of how many laptops?
http://www.newegg.com/All-Laptops-No...Category/ID-32

Not even to talk about the low volume of those actually selling with a dGPU. And you may note the majority of the laptops you link are MSI(35). MSI isnt exactly know to be the biggest laptop supplier.

Your link contains 1 Dell (Out of 458), 4 Acer(Out of 133) and 4 Lenovo(Out of 416).

Thanks for proving my point.

Number of Laptop models doesnt mean mobile segment is lost to the discrete GPUs. You can find every mobile dGPU available with large manufacturers like HP, Toshiba, Lenovo, ACER and Dell as much as with smaller ones like MSI and ASUS.
There are also businesses models with Quadro and MSI specializes more in Gaming Laptops lately.
Obviously there are more models and more laptops sold with iGPUs than dGPUs but that doesnt mean dGPU lost the mobile segment.

 

[ShintaiDK]

Number of Laptop models doesnt mean mobile segment is lost to the discrete GPUs. You can find every mobile dGPU available with large manufacturers like HP, Toshiba, Lenovo, ACER and Dell as much as with smaller ones like MSI and ASUS.
There are also businesses models with Quadro and MSI specializes more in Gaming Laptops lately.
Obviously there are more models and more laptops sold with iGPUs than dGPUs but that doesnt mean dGPU lost the mobile segment.

The numbers spoke for themselves. dGPU in the mobile segment is pretty much nonexistant.

I know you wanted to show something positive for the dGPU, but in reality you just showed how insignificant they are with a ever more declining position. We are talking about a couple of % that got dGPU.

 

[Insert_Nickname]

This is assuming that you don't already have a functioning PC or tablet, so the potential market won't reach its full capacity right away. That's why it has to do more than match these consoles. There are more variables to consider, which is why certain people are trying to use sales projections to to basically say that AMD will no longer profit from GPUs and Nvidia will go bankrupt. It's certain that progress will slow down (hell, it already has, though mostly due to TSMC), but it's too early to make any projections of when it'll stop. That's also why certain people are making up things about staying in 28nm until 2017.

I would seem we're pretty much in agreement.

The numbers spoke for themselves. dGPU in the mobile segment is pretty much nonexistant.

I think you need to get out more...

A quick walk trough a couple of random stores during my holiday revealed that ~50% of laptops displayed had some form of dGPU. I like to keep taps on what "Mr. Jensen" has access to. Granted, that may be something specifically Danish, or a freak accident, but it is food for thought...

 

[AtenRa]

The numbers spoke for themselves. dGPU in the mobile segment is pretty much nonexistant.

I know you wanted to show something positive for the dGPU, but in reality you just showed how insignificant they are with a ever more declining position. We are talking about a couple of % that got dGPU.

That was always the case in Mobile, even before 2011 and Sandybridge. iGPUs had the vast majority back then and now. But that doesnt mean that dGPU is non existent in the mobile.

What we see is the iGPUs have dethroned the entry level dGPUs like GT-620M etc. That is the reason of the declining dGPU total numbers, high end dGPUs are still going strong.

From NVIDIA Q1 2015

Notebook gaming has grown 51% CAGR for the past three years. Building on our gaming notebook momentum, we’ve also launched a new family of notebook gaming GPUs, including the first based on the Maxwell architecture. The performance, battery life, and thin form factors of these notebooks have built a great deal of excitement in the end market. Products are shipping now and will be available from every major OEM this quarter.

 

[Techhog]

The numbers spoke for themselves. dGPU in the mobile segment is pretty much nonexistant.

I know you wanted to show something positive for the dGPU, but in reality you just showed how insignificant they are with a ever more declining position. We are talking about a couple of % that got dGPU.

There's a problem of perspective here. You can't just look at the raw numbers; you have to look at how they've changed as well. If the number and sales of dGPUs in notebooks has been decreasing steadily, you have a point. However, it's also possible that they've just been low for years. I had a pretty hard time finding a decently-priced notebook with a dGPU 4 years ago. I think it may have decreased a bit since then, but that seems to mostly be due to the greater focus on thin, light, and low power. If that's the case and the performance isn't really a factor, then the market will stabilize at a certain point. That's exactly what we're seeing with the overall PC market. Everyone was sure that we had moved to a "post-PC world," but now it's looking more like tablets just took a bite out of the market while, at the same time, less people felt the need to upgrade. It won't be exactly the same with dGPUs obviously, but that's because of the fact that people usually only stop upgrading GPUs if they either don't use them anymore or nothing new is released. (Which I guess is why you're trying to stress the idea that there won't be anymore GPU node shrinks)

 

[ShintaiDK]

From NVIDIA Q1 2015

Notebook gaming. And using CAGR.

It says absolutely nothing about dGPU sales.

 

[ShintaiDK]

There's a problem of perspective here. You can't just look at the raw numbers; you have to look at how they've changed as well. If the number and sales of dGPUs in notebooks has been decreasing steadily, you have a point. However, it's also possible that they've just been low for years. I had a pretty hard time finding a decently-priced notebook with a dGPU 4 years ago. I think it may have decreased a bit since then, but that seems to mostly be due to the greater focus on thin, light, and low power. If that's the case and the performance isn't really a factor, then the market will stabilize at a certain point. That's exactly what we're seeing with the overall PC market. Everyone was sure that we had moved to a "post-PC world," but now it's looking more like tablets just took a bite out of the market while, at the same time, less people felt the need to upgrade. It won't be exactly the same with dGPUs obviously, but that's because of the fact that people usually only stop upgrading GPUs if they either don't use them anymore or nothing new is released. (Which I guess is why you're trying to stress the idea that there won't be anymore GPU node shrinks)

We know the dGPU have been in steady decline, even before the bubble. When dGPU sales peaked, it was 119 million a year. In 2013 it was 65 million. In 2010 the dGPU sales was around 75-80 million.

We also know that IGPs keeps increasing their graphics marketshare.

 

[n0x1ous]

We also know that IGPs keeps increasing their graphics marketshare.

Dont forget that every mainstream CPU comes with an IGP inflating its installed base. Certainly the majority of us on here with sandy/ivy/haswell mainstream parts are not using or have outright disabled the IGP, yet its counted anyway.

 

[ShintaiDK]

Dont forget that every mainstream CPU comes with an IGP inflating its installed base. Certainly the majority of us on here with sandy/ivy/haswell mainstream parts are not using or have outright disabled the IGP, yet its counted anyway.

The numbers are from every year. And since Sandy Bridge, the number still increases fast. So while they may be inflated till Sandy bridge. They still grow and show the difference increasing.

 

[Qwertilot]

Maybe AMD would have a problem with 4 big cores now, but the bandwidth stops it working anyway You can certainly see how they'll quite likely transition large chunks of their dGPU sales to APUs in 5(+) years.

Give them that new x86 thing of theirs and stacked memory. (3 years perhaps?). That probably gives them an efficient 30/40w CPU set up and no bandwidth problems for the iGPU.

At that point the APUs logically end up overlapping much of the low/mid end of their dGPU's. Probably displacing most of their sales over time.

That won't actually kill their high end dGPU's of course, because it'll just be moving revenue about. Might even let them finally get a bit back vs Intel/Nvidia.

 

[Techhog]

We know the dGPU have been in steady decline, even before the bubble. When dGPU sales peaked, it was 119 million a year. In 2013 it was 65 million. In 2010 the dGPU sales was around 75-80 million.

We also know that IGPs keeps increasing their graphics marketshare.

Of course there was a decline. It used to be that you couldn't even do basic things with iGPs. I just think your predictions of 28nm being here until at least 2017 and dGPU development halting permanently by 2019 (yet somehow sales won't end right after that... what?) are insane and baseless. There were people saying that mobile gaming would completely take over for many of the same reasons that you give for the death of dGPUs. Then, the new consoles came out and showed that, while the market did shrink, a lot of people were just waiting for something new. When 20nm hits next year, it's very likely that sales won't be quite as bad as predicted. I could be wrong, and perhaps you could even be right about 28nm still lasting all of last year. The thing is, we just don't know yet. What we do know for sure, however, is that people don't upgrade when there's nothing significant to upgrade to.

 

[blackened23]

We know the dGPU have been in steady decline, even before the bubble. When dGPU sales peaked, it was 119 million a year. In 2013 it was 65 million. In 2010 the dGPU sales was around 75-80 million.

We also know that IGPs keeps increasing their graphics marketshare.

You are correct, IGPs are playing a major role in mobile computing but in desktop computing, it is now as big of a contributor. If discrete goes away, you're suggesting that it is imminent and I do not agree. I think low end discrete will diminish for mobile platforms, but dGPUs will offer more performance for many years to come. It's going to happen (IGPs taking over), but not now. We're talking years from now.

I see IGPs eating more significantly into mainstream / low end discrete 5 years from now. This is not something happening overnight. Sure, discrete share has lowered, yet both AMD and NV are employing strategies to offset that. AMD is putting their GCN tech into their APUs. Whether that is successful in the long term, we'll see. Similarly, NV is employing their graphics architectures to their mobile lineup as well. On the same note, whether they're successful in the long term, we'll see. As well, high end discrete has high profit margins for both NV and AMD. So NV and AMD both get good margins from high and and professional products regardless of what IGPs are doing - and these professional cards employ the same architecture as dGPUs while being scaled up and offering software more fitting for the professional users.

Anyway, like I said, dGPUs disappearing is not happening tomorrow. We're talking years from now. dGPUs will continue to offer substantially increased performance for several years, so dGPUs will not disappear off the planet. People still want performance believe it or not. I do agree at some point years down the road, it will happen (mainly for mobile), but it is not imminent.

 

[ShintaiDK]

You are correct, IGPs are playing a major role in mobile computing but in desktop computing, it is now as big of a contributor. If discrete goes away, you're suggesting that it is imminent and I do not agree. I think low end discrete will diminish for mobile platforms, but dGPUs will offer more performance for many years to come. It's going to happen (IGPs taking over), but not now. We're talking years from now.

I see IGPs eating more significantly into mainstream / low end discrete 5 years from now. This is not something happening overnight. Sure, discrete share has lowered, yet both AMD and NV are employing strategies to offset that. AMD is putting their GCN tech into their APUs. Whether that is successful in the long term, we'll see. Similarly, NV is employing their graphics architectures to their mobile lineup as well. On the same note, whether they're successful in the long term, we'll see. As well, high end discrete has high profit margins for both NV and AMD. So NV and AMD both get good margins from high and and professional products regardless of what IGPs are doing - and these professional cards employ the same architecture as dGPUs while being scaled up and offering software more fitting for the professional users.

Anyway, like I said, dGPUs disappearing is not happening tomorrow. We're talking years from now. dGPUs will continue to offer substantially increased performance for several years, so dGPUs will not disappear off the planet. People still want performance believe it or not. I do agree at some point years down the road, it will happen (mainly for mobile), but it is not imminent.

I didnt say its imminent.

And there are 2 stages.
1 will be a stagnant dGPU development, but still selling.
2 will be a be when they stop all together.

And there may be the loss of one dGPU maker in the process.

 

[Techhog]

I didnt say its imminent.

And there are 2 stages.
1 will be a stagnant dGPU development, but still selling.
2 will be a be when they stop all together.

And there may be the loss of one dGPU maker in the process.

Since I'm sure you mean AMD, that makes no sense due to them offering double the price/performance on the IGP side. If they die, this process slows down.

 

[frozentundra123456]

Since I'm sure you mean AMD, that makes no sense due to them offering double the price/performance on the IGP side. If they die, this process slows down.

Maybe in the future, but right now, it is just the opposite. An 80.00 cpu like the Athlon 750k and a 100 dollar gpu like the 7770 or 7790 offers twice the performance of a 180.00 or so Kaveri. That is why except for very isolated uses like a SFF build an APU makes no sense to me in the desktop.

 

[Techhog]

Maybe in the future, but right now, it is just the opposite. An 80.00 cpu like the Athlon 750k and a 100 dollar gpu like the 7770 or 7790 offers twice the performance of a 180.00 or so Kaveri. That is why except for very isolated uses like a SFF build an APU makes no sense to me in the desktop.

I meant compared to Intel. Sorry.