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

[raghu78]

http://www.techpowerup.com/204528/g...own-in-q2-nvidia-holds-market-share-lead.html

Key takeaways

1. Nvidia dominant with 62% market share , roughly flat y-o-y and down 2.9% q-o-q.

2. Total AIB shipments decreased this quarter to 11.5 million units from last quarter.

3. On a year-to-year basis, total AIB shipments during the quarter fell 17.6%, which is more than desktop PCs, which declined 1.7%.

4. The attach rate of AIBs to desktop PCs has declined from a high of 63% in Q1 2008 to 36% in 2014 2Q, down from 44% last quarter.

5. The change from quarter to quarter was significantly greater less than last year, down -17.5% this quarter compared to down -5.5% a year ago.

6. JPR has been tracking AIB shipments quarterly since 1987 - the volume of those boards peaked in 1999, reaching 114 million units, in 2013 65 million shipped.

The trend is clear. IGP or APUs are going to eat away the entry level dGPU market. HBM stacked DRAM solutions on IGP and APUs will be the final nail in the coffin for the <= $100 dGPU market (maybe even upto $120).

The effect of Intel Skylake and AMD's 2016 APUs with a new x86 micro-architecture and HBM (High bandwidth memory) on discrete GPU unit sales should be very interesting.

 

[ShintaiDK]

The trend was quite clear from Intels expanded graphics share in the first report. It could only come on a large difference in volume compared to dGPUs.

It seems the dGPU shipments this year will be between 50 and 55 million units.

 

[Silverforce11]

Most of the volume they lost is in low end, no disputing that. APUs (AMD & Intel) will eat away the low end, and cause a shift upward of prices.

The NEW low-end won't be a 100mm2 GPU, but a 200mm2 one which is the mid-range as we know it. So expect higher prices going forward for discrete cards.

It's all about the silicon, because next-gen APUs seem to go for massive iGPUs that take up most of the die. Obviously discrete will need to compete by being bigger & more expensive.

 

[VirtualLarry]

Most of the volume they lost is in low end, no disputing that. APUs (AMD & Intel) will eat away the low end, and cause a shift upward of prices.

The NEW low-end won't be a 100mm2 GPU, but a 200mm2 one which is the mid-range as we know it. So expect higher prices going forward for discrete cards.

It's all about the silicon, because next-gen APUs seem to go for massive iGPUs that take up most of the die. Obviously discrete will need to compete by being bigger & more expensive.

Succinct, and I agree with this analysis. IGPs are marching on, and dGPUs are slowly retreating.

I don't forsee dGPUs going away entirely. CUDA and GPGPU "compute cards" will continue to be made, for HPC applications. Only, they may start being priced at $10K and up, instead of whatever they are going for today. ($1K and up?)

Edit: Personally, I am in the process of possibly replacing my aging Q9300 @ 3.0 rigs, that use P35 chipset mobos without IGP, thus they require dGPUs to function, with AMD's A8-7600 65W APUs, rather than purchase new dGPUs for those rigs.

 

[96Firebird]

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

 

[Silverforce11]

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

All it did was for the time, it meant all produced cards went to mining instead of gamers (for USA anyway) due to gouging. Out of stock is out of stock. It's not like they can instantly make 5X more volume to cater to that craze.

Then the post mining boom, all the used cards are on ebay for cheap.

 

[ShintaiDK]

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

Out of stock is out of stock.

It was wastly overhyped into the clouds and beyond. It struck at the same time as AMD had made logistic mistakes for orders at TSMC and other component makers. And that was the primary reason for out of stock.

That some US retailers found the excuse to milk people that jumped on the hype is a whole other matter.

 

[Silverforce11]

AMD did gain 2.9% marketshare from NV this quarter. Definitely very competitive products with GCN 1.1 cards.

 

[ShintaiDK]

AMD did gain 2.9% marketshare from NV this quarter. Definitely very competitive products with GCN 1.1 cards.

You mean GCN 1.0?

 

[Silverforce11]

you mean gcn 1.0?

r290/x

Edit: and R270/X, launched in Q1. Thanks, 260/X.

 

[ShintaiDK]

r290/x

Edit: and R270/X, launched in Q1.

290 series is relatively very low volume.

The main driver for AMD is GCN 1.0 cards.

And the 270/270X is GCN 1.0. Its the 260/260X thats GCN 1.1. The only 2 cards besides 290 series thats GCN 1.1.

 

[sontin]

AMD's GPU business was down Y-Y and Q-Q. They are giving their GPUs away for free.

nVidia's business was up Y-Y while staying at the same market share.

 

[Noctifer616]

AMD's GPU business was down Y-Y and Q-Q. They are giving their GPUs away for free.

nVidia's business was up Y-Y while staying at the same market share.

GPU shipments for AMD are down Y-Y but up Q-Q. For Nvidia it's the same like last year but down compared to last quarter.

AMD GPU revenue was up Q-Q and Y-Y.

EDIT: GPU shipments were down, but market share was up.

 

[blackened23]

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

Yeah, that's what I was thinking. It is what it is really. AMD had new generation of products in Q4 which was bound to increase their sales somewhat, after prices from the mining idiocy died down. NV of course didn't have a new generation of products. I'd imagine the pendulum would swing back in NV's favor with the 800 series, especially if they're priced like the 600 series was (the 600 series were perennially sold out at launch and I think sold extremely well). Still, with miners buying cards in lots of 100 I figured that would have more effect on their share.

Personally, while I really like what NV has going on i'm happy that AMD gained dGPU share. It just helps competition all around. Maybe NV will see this and try to "bring it" with their 800 series. I don't know. Regardless of which side anyone prefers, everyone should want for fierce competition.

 

[ShintaiDK]

AMD's GPU business was down Y-Y and Q-Q. They are giving their GPUs away for free.

nVidia's business was up Y-Y while staying at the same market share.

GPU segment between AMD and nVidia in terms of marketshare is pretty much stagnant. Both losing badly to IGPs.

 

[frozentundra123456]

Discouraging trend that dgpu sales are continuing to decline. I dont think mid range and above cards are going away, so all it means is gaming cards are going to increase in price. I also think the fact that the dgpus on the market now are getting long in the tooth doesnt help sales. We have been on 28nm for what, nearly 3 years now? Perhaps 20nm, especially Kepler with the huge increase in performance per watt will give the market a shot in the arm.

 

[sontin]

nVidia's GPU business is up 2% from last year while staying at the same market share.
AMD is down from last year.

GPU segment between AMD and nVidia in terms of marketshare is pretty much stagnant. Both losing badly to IGPs.

How is nVidia "losing badly to IGPs" when their GPU business is up Y-Y and they did their best Q2 in history?

 

[ShintaiDK]

How is nVidia "losing badly to IGPs" when their GPU business is up Y-Y and they did their best Q2 in history?

Does that include Tesla?

But the volume is still shrinking. Nomatter how you try and twist and turn it. The amount of buyers is decreasing rapidly.

 

[RussianSensation]

How is nVidia "losing badly to IGPs" when their GPU business is up Y-Y and they did their best Q2 in history?

Simple, just like Apple sells record number of iOS devices but loses overall market share to Android - because the competitor sold even more relative to the total market. Also, NV talks about revenue gains, not shipment gains. When your ASP go up because you move mid-range from $250 to $350-450 and high end from $500 to $700, you could sell less units and have revenue growth since your ASP increase offsets your unit volume decrease.

As other have said, this is going to make it harder and harder for AMD/NV to keep historical pricing. I suspect bifurcation of a generation is one successful strategy of keeping margins and ASP high. This way Nv can sell mid-range GM204 at $400-500 and then launch high end $600-700 GM200, instead of launching everything top to bottom and crushing your margins. With lower nodes being more expensive and taking longer, it kinda makes sense to release GPUs 25-40% faster every 15-18 months rather than waiting 24-30 months and wait until you can achieve 60-80%. This strategy worked wonders for Kepler and GCN 1.0-> 1.1 so far.

NV's record profit margins are an indication that despite shrinking dGPU market, the higher ASP of $250+ products relative to their cost of sales (manufacturing cost) is healthier than ever, which is necessary to offset the lower demand for gaming graphics at this time.

 

[blackened23]

Does that include Tesla?

But the volume is still shrinking. Nomatter how you try and twist and turn it. The amount of buyers is decreasing rapidly.

NV and AMD do more than discrete GPUs, i'm pretty sure that NV aren't idiots and they see the writing on the wall with discrete GPUs. That said, there is still money in dGPUs as a lot of people still want them. And they will still want them several years from now....iGPU won't take over for several years.

So despite the dGPU market shrinking, it's not like the market is going to die immediately. dGPUs will still be here for years. You act like dGPUs are going to die out tomorrow. Not the case.

dGPUs will be less of a factor 5 years from now, but these companies aren't idiots. They know the dGPU market is shrinking. AMD's bet is on their mobile APUs, however their performance per watt is just not good compared to intel and their graphics performance hasn't been able to win many high end designs, which is what they need. Needless to say their CPU performance isn't pretty either, which is important for mobile devices. Hopefully they can have a breakthrough in architecture design with much better performance per watt, perhaps GCN 2.0. They need it if they want to win on their bet with APUs, as of right now they're selling on bargain bin prices for their mobile APUs and nothing else. NV's bet is on android, automotive, mobile dGPUs, automotive and scientific applications with quadro / HPC. And by the way they use the same architecture in all of these products: Kepler and Maxwell. Tegra has had growing pains but they're making a ton of money in all of their other segments. So we'll see I guess. I think AMD needs to improve their performance per watt and stop relying on Windows, because Windows 8.1 is a poor touch based mobile experience. AMD needs android compatibility, it's the bigger market for mobile. As far as NV goes, they need to do better with Tegra.

Anyway, we'll see what happens in a few years. But you can rest assured that both AMD and NV are aware of the situation and are placing their bets as they see fit to expand past dGPUs. Both are slightly flawed (AMD not having android and having terrible PPW in their mobile APUs, therefore not selling with worthwhile margins....and then nvidia with their poor T4 sales), but we'll see what happens.

 

[tviceman]

Most of the volume they lost is in low end, no disputing that. APUs (AMD & Intel) will eat away the low end, and cause a shift upward of prices.

The NEW low-end won't be a 100mm2 GPU, but a 200mm2 one which is the mid-range as we know it. So expect higher prices going forward for discrete cards..

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.

 

[Phynaz]

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

Not sure this adds anything to conversation besides flamebait.

 

[96Firebird]

Not sure this adds anything to conversation besides flamebait.

Care to elaborate?

 

[dangerman1337]

I wonder how much this is due to the fact early 2012 GPUs haven't been improved on in their targeted market. Performance Range GPUs have literally not improved much if you have either have a 6900, GTX 580/570, 7900 and 670/680 there isn't anything that has come out for the $500 or under that is a worthy upgrade (except maybe 6900s to 580/570 > 780 or 290x right now).

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...)

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

 

[monstercameron]

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