JPR Graphics Add-in-board-market Q2 2014 sales

[Phynaz]

what do you guys think it'll take to rejuvenate the industry? it surely cant just be more performance and less power...can it?

Pretty much.

And software that can use the power.

 

[blackened23]

what do you guys think it'll take to rejuvenate the industry? it surely cant just be more performance and less power...can it?

Sorry but it's just a downward slope for desktpo GPUs. Don't get me wrong: dGPUs will be still be here for a few years but don't kid yourself into thinking that the glory days will ever return. Just aint happening. Mobile has had far too big of an impact. Nowadays every first computing device 99% of the time for students and up and coming teenagers is an ultrabook, macbook or tablet. 15 years ago, they were all desktop PCs.

Of the students I know, how many bought a desktop PC as their computing device? Maybe 1 out of 50, nearly all will be using some form of macbook, ultrabook or tablet.

Good news though. Mobile dGPus will always be faster than integrated. So if any of the above wants a gaming portable, there are options out there which AMD and NV provide.

Anyway....Computing has changed, buying habits have changed, and while dGPUs will be around for a few years yet. We aren't returning to the volume of dGPUs selling like they were 10 years ago. Not happening, sorry.

 

[tviceman]

I think we'll see Nvidia using TSMC 16nmFF in 2016 including their high end range if that paper which had that Q1 2016 for GP100 is true (most likely true for production or the Tesla/Quardo version of it and then a GeForce Version later in the year).

If Nvidia is skipping 20nm, then I think we'll see them migrate to 16nmFF sooner in a node's life cycle than what we've typically seen from Nvidia.

I hope there is a 512-bit memory buss with 3D memory GP100 :awe:.

Personally, I don't really care what the memory bus width of cards are. If one card has a 512-bit memory bus (R9 290x) and gets easily beat by a card with a 256-bit memory bus (hypothetical GTX 880), and you are going to buy one or the other, are you going to get the slower card with a 512-bit memory bus or the faster card with a 256-bit memory bus?

 

[tviceman]

Sorry but it's just a downward slope for desktpo GPUs. Don't get me wrong: dGPUs will be still be here for a few years but don't kid yourself into thinking that the glory days will ever return. Just aint happening. Mobile has had far too big of an impact. Nowadays every first computing device 99% of the time for students and up and coming teenagers is an ultrabook, macbook or tablet. 15 years ago, they were all desktop PCs.

Of the students I know, how many bought a desktop PC as their computing device? Maybe 1 out of 50, nearly all will be using some form of macbook, ultrabook or tablet.

Good news though. Mobile dGPus will always be faster than integrated. So if any of the above wants a gaming portable, there are options out there which AMD and NV provide.

Anyway....Computing has changed, buying habits have changed, and while dGPUs will be around for a few years yet. We aren't returning to the volume of dGPUs selling like they were 10 years ago. Not happening, sorry.

It will be a slow decline, but even in the face of lower volumes, Nvidia's total revenue is on pace this year to be the higher than it's ever been. I know that's because ASP's are significantly higher (when factoring in Tesla and Quadro), but all in all I think the dGPU market is still relatively healthy. We're stuck in a node process rut, we've had Geforce Titan-like performance (with Titan-like prices) since GTX690 came out back in May of 2012, so since then it's been relatively unexciting. Slap that on top of fairly undemanding games, the extremely slow transition to higher PPI displays, and the recent release of new consoles which temporarily draws money away from the PC market, and we have the exact situation we're in today.

There is Star Citizen, Witcher 3, and Grand Theft Auto 5 which will all push hardware significantly. Then you have the eventual next Fallout game (which may not be very demanding, but spurns hardware demand due to sheer popularity), and if Valve ever gets off their rears and ships a new Half Life game, it would overload the entire internet. Also Oculus is coming hard and fast (no pun intended), and pushing 2x 1080p screens is going to require more juice than what is typically needed now. When affordable 4k monitors are flooding the market ( < $300), when next generation games truly start taxing GPU's at 1080p (instead of just being terrible ports like Watch Dogs), and when true next generation cards offer 2x the performance of GTX 780 for $550, the upgrade cycles will click back into high gear.

If it was doom and gloom starting now, I think we'd have seen Nvidia already burning massively more money on hiring personnel and R&D for Tegra (to the point of breaking even or losing profits short term) and/or fielding offers from bigger fish to sell while their stock is still relatively high.

 

[blackened23]

I don't disagree. I guess I wasn't entirely clear: I don't think it's doom and gloom, I give the dGPU market a good 5-7 years. I guess what i'm saying is, that volumes will continually lower over the next 5 years, very very slowly. We will not return to the glory days of the early 2k years in which everyone and their brother had a dGPU, heck all PCs had dGPUs. But dGPUs will still definitely exist and will still have their niche. And then there's mobile dGPUs which aren't accounted for in the numbers in this thread: mobile dGPUs sell quite a bit as well and offer much more performance than iGPU.

If people want them, and they will, NV and AMD will sell them. Like you said it will force ASPs to rise and AMD and NV will find creative ways to offset increasing R+D costs.

For NV, their R+D applies to their entire product portfolio just like intel - from tegra to quadro. Similar for AMD but I don't think their hedging bets on Windows 8.1 is the best path going forward, at least not for low cost SOCs. Slight tangent but I still don't think windows 8.1 is a proper touch driven mobile experience, but Windows is the best when more power is needed and you have an ultrabook with a touchpad/kb or a desktop. Android/iOS is so much better as a touch driven mobile OS, when you dont' have a keyboard/touchpad/mouse. So hopefully AMD will get some android compatibility for their mobile APUs at some point. (and not bluestacks, because bluestacks is junk that barely works)

 

[Silverforce11]

148mm2 GM107 on now ancient 28nm process is what, 3.5-4x faster than the best IGP? 16nm will supposedly be in volume production in Q2 of next year, so it's possible Nvidia is skipping 20nm for GPU's. Speaking from memory (which means I could be a little off), 16nm will have 2.3x transistor density and 2x the power savings. Just shrinking GM107 and applying the power savings alone would put it at half the size and twice performance. And that is without architecture improvements (Pascal, Volta...)

So I disagree. GM107, on a long-in-the-tooth node process, crushes the best IGP solution available and there is no imminent release of anything that is going to catch it.

As for higher prices, unless Nvidia starts bleeding both volume and market share, premiums are the new norm. As RussianSensation has said, the current stagnation of not having any new high end, coupled with the glut of graphically undemanding or poorly coded games have attributed to this large decline.

Certainly, as we see integrated in the current gen without faster GDDR5 bandwidth its not competitive versus a 148mm2 part, but it is very competitive against <100mm2 parts which removed that market. It was only a few years ago where the low end was a tiny chip, now the low end is 148mm2. It doesn't take much to see in a few years time, the low end will be 200mm2.

Discrete low end needs to be much faster than integrated to justify itself in OEM builds, it cannot be slightly faster.

Once APUs get faster memory, they will be able to compete better on a direct mm2 vs mm2 (AMD's GPU can at least, obviously Intel keeps on improving their architecture too). Currently with poor bandwidth, its at a disadvantage.

Looking at NV, they sold less but made more money. It's obvious due to the prices being massively raised. Selling prior $200-300 mid-range for $500+ helps with revenue, no doubt.

So volume itself is pointless. It comes down to revenue and profit. As long as PC gamers are willing to pay extra for discrete performance, the revenue will be just fine (& discrete GPU isn't going the dinosaurs).

 

[monstercameron]

snip

So volume itself is pointless. It comes down to revenue and profit. As long as PC gamers are willing to pay extra for discrete performance, the revenue will be just fine (& discrete GPU isn't going the dinosaurs).

but how long until good enough status[with the current status quo]?

 

[blackened23]

Certainly, as we see integrated in the current gen without faster GDDR5 bandwidth its not competitive versus a 148mm2 part, but it is very competitive against <100mm2 parts which removed that market. It was only a few years ago where the low end was a tiny chip, now the low end is 148mm2. It doesn't take much to see in a few years time, the low end will be 200mm2.

Discrete low end needs to be much faster than integrated to justify itself in OEM builds, it cannot be slightly faster.

Once APUs get faster memory, they will be able to compete better on a direct mm2 vs mm2 (AMD's GPU can at least, obviously Intel keeps on improving their architecture too). Currently with poor bandwidth, its at a disadvantage.

Looking at NV, they sold less but made more money. It's obvious due to the prices being massively raised. Selling prior $200-300 mid-range for $500+ helps with revenue, no doubt.

NV sells visual computing across a variety of products, and AMD has expanded their product line and graphics technology as well. Profit wise? They're making money but they're not an Apple. NV made something like 128M in profit last quarter which in the grand scheme of things compared to intel or qualcomm is a drop in the bucket. Compare this to intel and qualcomm and apple who are making billions upon billions every quarter.

They're doing what they do to stay relevant and still be a company 5 years from now. Since they cannot make integrated x86 graphics, that's exactly what they're doing. For this company to exist, they must have cash for R+D and they must be making money. 128M is not a lot compared to other corporations such as Apple as I stated. AMD has the entire good guy marketing smokescreen going on, and try to paint themselves as being good and intel/NV as "bad" (major rolleyes) but they want your money just like any other corporation wants. Let's not kid ourselves: AMD cannot sustain losses quarter after quarter. So you can bet that AMD is trying to think of ways to make more money where they've failed in the past. They want your cold hard cash just like any corporation does, and their bets will be pretty good so long as their next architecture has the gains in PPW that they've claimed.

They're claiming 2x gains in PPW. Hopefully that will be the case, because despite misgivings with AMD in the past i'd like to see them strive. If they can make those gains in architecture - they can apply that to their entire product portfolio and it will help their mobile APUs. And it will help them sell a lot of chips for their mobile product stack, in addition to their dGPUs.

Anyway, these companies are expanding beyond their dGPU roots to be relevant several years from now. Unfrotunately since dGPU volumes are decreased, that means R+D has to be offset somehow. Currently that R+D is offset by expanding a product portfolio across a wide range of products (ie tegra, quadro for NV and APUs for AMD) while ASPs have to increase. Just think logically about how these companies are going to stay relevant without offsetting the fact that volumes of dGPUs are shrinking over time. If volumes keep lowering slowly something has to give. You want these companies to exist, right? Well, they're only going to exist with higher ASPs and wider product portfolios.

 

[Silverforce11]

but how long until good enough status[with the current status quo]?

Never. Certainly not in the next decades.

We cannot even push 1600p comfortably with a single top GPU.

4K? Not a chance. Not even 2 handle it well. In 6 years time, maybe 1 top GPU can drive 4k well, but that assumes games don't get progressively more complex.... which brings me to this point:

Gamedevs will always push the boundaries, that we can be certain of.

Then display technology will also improve and soon 4K will be the low-end.

Eventually, maybe in 100 years, we get digital holodecks that is fully interactive and we will game in simulations like on Star Trek. Even then, to fully max the realism, you will probably need Quad Fire/SLI (or whatever is available then)

It's a simple concept: as long as there's lots of individuals who are willing to pay a high cost for an uber gaming experience, discrete GPU will be relevant and able to bring in the profit.

 

[Silverforce11]

ps. The masses will be just fine with APUs.

 

[tviceman]

Certainly, as we see integrated in the current gen without faster GDDR5 bandwidth its not competitive versus a 148mm2 part, but it is very competitive against <100mm2 parts which removed that market. It was only a few years ago where the low end was a tiny chip, now the low end is 148mm2. It doesn't take much to see in a few years time, the low end will be 200mm2.

Discrete low end needs to be much faster than integrated to justify itself in OEM builds, it cannot be slightly faster.

Once APUs get faster memory, they will be able to compete better on a direct mm2 vs mm2 (AMD's GPU can at least, obviously Intel keeps on improving their architecture too). Currently with poor bandwidth, its at a disadvantage.

Looking at NV, they sold less but made more money. It's obvious due to the prices being massively raised. Selling prior $200-300 mid-range for $500+ helps with revenue, no doubt.

So volume itself is pointless. It comes down to revenue and profit. As long as PC gamers are willing to pay extra for discrete performance, the revenue will be just fine (& discrete GPU isn't going the dinosaurs).

Agreed with most of this, but it's a looong march. Had 20nm been on time, GM107 would be faster and significantly smalelr, alleviating some of the argument you put forth. As I said just a few posts above, 4K + VR + the never ending march of higher visual fidelity will keep the discrete market alive and strong for many more years. There is a glut in sales volume right now because there is a glut in graphical performance improvement.

Comparisons to the add-on sound card market are ill-conceived.

 

[Zodiark1593]

but how long until good enough status[with the current status quo]?

As I've said many a time. It will be good enough when we see real-time raytracing (spectral rendering at that) at 120+ fps, and w/e resolution is feasible at a given time.

Considering top-end dGPUs are very well below that mark, iGPUs have a very long way to go to be good enough for hobbyist gamers.

 

[tviceman]

I don't disagree. I guess I wasn't entirely clear: I don't think it's doom and gloom, I give the dGPU market a good 5-7 years.

It all depends on developers and where the money is for them. If Google Nexus TV catches on as an easy way to game and takes off, then it might accelerate a decline in hardware enthusiasts. It's an interesting situation long-term, for sure. PC gamers are likely an older demographic than console gamers, many brought up on buying boxes at Best Buy in the 90's and/or building their own since then. Since desktops are largely out of the picture at warehouse-style electronic stores, how is the younger demographic going to be exposed to PC gaming, building rigs, upgrading, etc.? Through notebooks? Having a dad (or mom) that games on a PC isn't going to be enough exposure to keep the market alive.

I think Valve is screwing up a huge opportunity with their steam machine initiative. They should have just built 2-3 spec'd machines put inside their great looking custom case with price tiers $400, $600, and $800, refresh the products with better hardware within the same price bracket every 18-24 months (kind of like phones and tablets), and either subsidize the cost of the hardware OR include steam wallet money with each purchase. Leave the ultra high end stuff, overclocking, and constant upgrading to the DIY and boutique crowds.

 

[3DVagabond]

I'm surprised the mining craze didn't help AMD more, as many people thought would happen.

We've discussed the mining craze to death already. I hope we don't go down that road and have to reread all those posts/opinions.

 

[blackened23]

I think Valve is screwing up a huge opportunity with their steam machine initiative. They should have just built 2-3 spec'd machines put inside their great looking custom case with price tiers $400, $600, and $800, refresh the products with better hardware within the same price bracket every 18-24 months (kind of like phones and tablets), and either subsidize the cost of the hardware OR include steam wallet money with each purchase. Leave the ultra high end stuff, overclocking, and constant upgrading to the DIY and boutique crowds.

I so much agree with you on the steam box thing. It had so much potential to help PC gaming but instead it has fallen flat on its face because Valve didn't put enough effort into it. I think if they want to do something they just need to go all in and put every ounce of effort into it. And when Valve does that, generally good things happen.

 

[PPB]

what do you guys think it'll take to rejuvenate the industry? it surely cant just be more performance and less power...can it?

Gsync/Freesync will give the industry a nice boost, then it will continue to fall off, even more pronouncedly that now because those 2 very techs will make the underlying GPU's lifespans even longer. What need to upgrade GPU if Gsync/Freesync allows you to have such a nice gaming experience at lower-than-60 fps?

 

[lefty2]

AMD’s shipments of desktop heterogeneous GPU/CPUs, i.e., APUs increased 16.7% from the previous quarter, and increased 10.3% in notebooks.

Could that be a sign that the new Beema and Kaveri APUs are selling well?

 

[Phynaz]

ps. The masses will be just fine with APUs.

It's those masses that fund GPU development. Once they are gone so are discreet GPUs.

 

[AtenRa]

It's those masses that fund GPU development. Once they are gone so are discreet GPUs.

The masses are fine with iGPUs for the last 10 years, I havent seen the dGPUs gone so far.

 

[AtenRa]

Could that be a sign that the new Beema and Kaveri APUs are selling well?

If well means better than Q1 2014 then yes.

 

[Zodiark1593]

It's those masses that fund GPU development. Once they are gone so are discreet GPUs.

The masses would probably be just peachy with an old GMA igp. Very few of the folks I've met would stress even that.

 

[Phynaz]

The masses are fine with iGPUs for the last 10 years, I havent seen the dGPUs gone so far.

10 years. Really.

And you haven't seen the fact that GPU sales have dropped by half? Read the first post.

 

[VirtualLarry]

A friend of mine has been using a 780G chipset IGP for the last few years, at least probably three years. Only recently has he mentioned slowdown.

 

[AtenRa]

10 years. Really.

And you haven't seen the fact that GPU sales have dropped by half? Read the first post.

GMA900 was released in 2014, it was fine for the masses at the time.

Where did you see GPU sales dropped by half ??

 

[Silverforce11]

It's those masses that fund GPU development. Once they are gone so are discreet GPUs.

This is a common myth.

The last time I saw a break down of discrete GPU revenue & profit from a financial report, most of it was from the mid-range and high-end to enthusiast segment.

The low-end have volume, but their margins are low. They got nothing close to the margins as a 780ti or Titan, or even a small chip gk104 being sold at $500+.

It's the core gamers that have always funded GPU development, ever since the 3dfx days, because only core gamers are willing to pay a high price a GPU, the rest, they are satisfied with cheap stuff or integrated.

Total discrete GPU sales are on a downward trend, due to the loss of low-end, but revenue is fine or up. This point was raised before.

Moving forward, core gamers will end up paying more.