[KitGuru] Sales of desktop graphics cards hit 10-year low in Q2 2015

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Insomniator

Diamond Member
Oct 23, 2002
6,294
171
106
No reason to deal with the hassle of PC gaming when you can buy a console that lasts 10 years and never falls behind much in graphics anyway. Sure PC had Crysis pushing the envelope but now what? Can't even think of any major graphic powerhouse PC exclusive titles. The main PC exclusives these days can be run on almost anything - SC2, Dota, LoL etc. Its surprising to me that PC hardware hasn't fallen faster.

Its tough to convince people to spend 1000+ on gaming PC's only to deal with crashing, patching, stuttering, drivers, u play, origin, steam etc etc. to play the SAME game you can play on a Ps4 for 300 bucks @1080p/medium.
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
Next year 16nm FF High-end dGPUs will have 2x the transistor count than GM200 and Fury X, that is 2x 8B = 16B transistors. Performance will increase by almost 2x, especially in future DX-12 games.
Also, DX-12 games and 4K monitors will increase the GPU performance needs substantially.

With all that, 65W-100W TDP APUs from both Intel and AMD will not be able to catch up dGPUs not in a million years.

Which dGPU are you talking about ? Let me clarify. There are roughly 4 tiers of discrete GPUs for desktop and 3 for notebook.

100-150 sq mm - HD 7770, GTX 750 Ti,
200- 250 sq mm - GTX 660, HD 7870, GTX 960
300-450 sq mm - GTX 680, HD 7970, R9 290X, GTX 980.
500-600 sq mm - GTX 780 Ti, GTX 980 Ti, Fury X

Out of this the first tier's obituary is already written. HBM2 is what will enable the death of that tier of GPU. AMD Zen APUs with HBM2 in 2017 and Intel 2017 SoCs (remember they will be at 10nm) with HBM2 will essentially kill of the highest volume dGPU segment. In notebooks the effect will be even more disastrous as AMD/Nvidia are TDP limited and only sell 3 tiers. So to lose the highest volume notebook GPU segment is a huge blow. I am predicting 50% unit volume will go away and 30% revenue will go away from notebook discrete GPU market once HBM2 arrives in AMD APUs and Intel SoCs. The desktop market too will be affected but to a lesser extent than notebooks as they have the 4th tier. I foresee AMD coming out with Zen based FX APUs with 150-200w TDP and CLC kits too in the future.

Eventually we will see AMD and Nvidia just keep up the trend of higher $ per sq mm of GPU silicon which we have been seeing from the start of 28nm node. GPUs will be costlier and there is no running away from this reality.
 
Last edited:

Goatsecks

Senior member
May 7, 2012
210
7
76
I wonder how much of this decline has been because of the global financial crisis?

The general downwards trend appears to start around the same time. It would be interesting to see global trends in spending on luxury goods over the same timescale.
 

positivedoppler

Golden Member
Apr 30, 2012
1,100
171
106
No surprise here. Desktop monitors aren't getting much bigger. Only thing that will save dgpu is VR and both Nvidia and AMD has been pimping that for some time.
 

bononos

Diamond Member
Aug 21, 2011
3,877
137
106
I wonder how much of this decline has been because of the global financial crisis?

The general downwards trend appears to start around the same time. It would be interesting to see global trends in spending on luxury goods over the same timescale.

Or the relatively stagnant median wages of the average worker compared to the wealthy for the past decades. I don't think luxury brands are feeling the pinch at all.
 

AtenRa

Lifer
Feb 2, 2009
13,984
3,347
136
Which dGPU are you talking about ? Let me clarify. There are roughly 4 tiers of discrete GPUs for desktop and 3 for notebook.

100-150 sq mm - HD 7770, GTX 750 Ti,
200- 250 sq mm - GTX 660, HD 7870, GTX 960
300-450 sq mm - GTX 680, HD 7970, R9 290X, GTX 980.
500-600 sq mm - GTX 780 Ti, GTX 980 Ti, Fury X

Out of this the first tier's obituary is already written. HBM2 is what will enable the death of that tier of GPU. AMD Zen APUs with HBM2 in 2017 and Intel 2017 SoCs (remember they will be at 10nm) with HBM2 will essentially kill of the highest volume dGPU segment. In notebooks the effect will be even more disastrous as AMD/Nvidia are TDP limited and only sell 3 tiers. So to lose the highest volume notebook GPU segment is a huge blow. I am predicting 50% unit volume will go away and 30% revenue will go away from notebook discrete GPU market once HBM2 arrives in AMD APUs and Intel SoCs. The desktop market too will be affected but to a lesser extent than notebooks as they have the 4th tier. I foresee AMD coming out with Zen based FX APUs with 150-200w TDP and CLC kits too in the future.

Eventually we will see AMD and Nvidia just keep up the trend of higher $ per sq mm of GPU silicon which we have been seeing from the start of 28nm node. GPUs will be costlier and there is no running away from this reality.

2015
Broadwell GT3e iGPU @ 14nm = ~84mm2.
Broadwell GT3e iGPU @ 14nm = slower than 118mm2 GT740 @ 28nm

2016
Skylake GT4e iGPU 14nm = ~130mm2
Skylake GT4e iGPU @ 14nm = slower than 148mm2 GTX750Ti at 28nm
Skylake GT4e iGPU @ 14nm = slower than 80-90mm2 GT840 @ 16nm

How many Skylake GT4e do you believe Intel will sell and how many GT840 will NVIDIA sell ??
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
I am interested to see in what other people's thoughts are on why the desktop discrete GPU market has declined so dramatically in recent years? Please share your opinions.

I think one problem is that AMD reduced its presence in the low end of the market, leaving much of it to Nvidia.

Not sure why this is? Some have speculated that a reduction in Oland production was to help prop up demand for the desktop APUs. (re: having a high volume Oland would have contributed to erosion of Kaveri desktop sales)

If so, that is a shame because IMO there are so many worthy used SFF desktops (Sandy Bridge hardware,etc.) that could use a nice low profile card in the 25W to 40W range.

P.S. Right now, the best sub 40W card appears to be the GT 730 GDDR5 (GK208 with 64 bit GDDR5). The PNY version usually sells for $55 after rebate w/free ship, but at one time GK208 with 64 bit GDDR5 dropped as low as $40 AR (during the time it was called GT 640 GDDR5). I would like to see AMD offer something competitive in this area to give SFF gamers more options.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
P.S. Now one could try to make the argument that there are less gamers worldwide playing games overall but while the desktop discrete graphics card market declined from a stable 15-20 million sales per quarter to sub-10 million, the sales of PS4+XB1 are trending > 50% greater than PS3+Xbox 360:

The PlayStation 3 and Xbox 360 in their first 20 months sold a combined 24.23 million units, while the PlayStation 4 and Xbox One have sold a combined 37.34 million units.
Total Combined PlayStation 3 and Xbox 360 Sales: 24,229,386
Total Combined PlayStation 4 and Xbox One Sales: 37,342,738 (+54%)***

***Now I am not trying to start a PC vs. console thread but just using this as a point that there is growth in the gaming industry as far as unit sales are concerned which means there are still gamers/consumers interested in gaming as a whole but what are the possible explanations as to why the desktop discrete GPU market is getting wiped out this badly?

Regarding the console battle, eventually I would like to see what a Core i5-2500S (65W) could do with a good low profile 14nm card.

With a power budget of 40W for the 14nm dGPU, I'll bet it would start to get interesting for the PC.



P.S. I know you mentioned a lot of existing old cards out there being viable for gaming, but just remember these are usually full size ones. And the mass of old desktops floating around out there (off lease corporate units) appear to be SFF.
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
2015
Broadwell GT3e iGPU @ 14nm = ~84mm2.
Broadwell GT3e iGPU @ 14nm = slower than 118mm2 GT740 @ 28nm

2016
Skylake GT4e iGPU 14nm = ~130mm2
Skylake GT4e iGPU @ 14nm = slower than 148mm2 GTX750Ti at 28nm
Skylake GT4e iGPU @ 14nm = slower than 80-90mm2 GT840 @ 16nm

How many Skylake GT4e do you believe Intel will sell and how many GT840 will NVIDIA sell ??

First of all since AMD is not putting any pressure on Intel and was stuck at 32/28nm for roughly 4-5 yrs Intel has had to do the least in terms of their SoCs. That changes in 2017 with Zen based APU with GCN2 and HBM2.

btw lets wait and see when the first 16/14nm FINFET Nvidia / AMD GPU lands and when the first FINFET GPU <= 100 sq mm lands and more importantly the supply situation.

Also as Shintai said can you link to Skylake GT4e performance benchmarks ? Both AMD and Intel (with 10nm in 2017) and HBM2 will start to eat away at the low end GPU market. The 100 -150 sq mm discrete GPU is no longer viable as the AMD APUs with HBM2 will be on the same foundry node and make the dGPU redundant. The Intel 10nm SoCs in 2017 will definitely use HBM2 as the on die GPU (with a doubling of FLOPS over Skylake) is going to be handicapped if the bandwidth to feed it is not there. Intel can eliminate the eDRAM(which is built on 22nm) and keep costs under control even with introduction of HBM2.

btw why is dGPU sales falling and at a 10 year low ? :D
 

Red Hawk

Diamond Member
Jan 1, 2011
3,266
169
106
Boooo. Well I did my part by buying a new 290X, y'all need to pick up the slack! :colbert:
 

DustinBrowder

Member
Jul 22, 2015
114
1
0
Why are people still spreading the myth of the IGP as some sort of realistic GPU? The fact of the matter is low and mid range users won't be buying the latest generation Intel CPU's, they are not going to dish out every year or so $160+ on an Intel CPU that has 10%-15% improvement in GPU power over the previous one.

Even the latest IGP's are just garbage for any real gaming, if you really do want to game even an Nvidia GT 640 or AMD 250x is still better.

dGPU sales are dropping, as are dCPU sales, because we haven't had a real improvement in the past 4 years!

CPU's, GPU's have all been stagnant, this is why Intel I7 2600k can still hold its own with the I7 6700k. 4 generations later and the I7 6700 is barely 30% faster than a 2600k.

GPU's are the same, a GTX 780ti is barely 20% slower than a 980ti and this is with Nvidia's drop of optimizing for the 700 series and lower.

AMD's 7970GHz is 290x is 390x they all are about 10% difference of each other.

Why would anyone buy a new CPU or GPU if you already own a say Intel 2000+ series CPU or GTX 600 series or AMD 7000 series?

If you own a GTX 460 or 560 you might want to upgrade, but even so your options are expensive and bad.

Same for AMD. Unless there is a real new generation, that brings significant performance improvements we won't see any gains in the dGPU or dCPU.

We've reached a technological stagnation where in the past 4 years we've pretty much had the same cpu's and gpu's with terribly small improvements. Consumers are not idiots in the aggregate, individually there may be terribly uninformed and low IQ consumers, but in the aggregate consumers are smart and are not going to spend money for no gains.


 
Last edited:

n0x1ous

Platinum Member
Sep 9, 2010
2,572
248
106
GPU's are the same, a GTX 780ti is barely 20% slower than a 980ti and this is with Nvidia's drop of optimizing for the 700 series and lower.
We've reached a technological stagnation where in the past 4 years we've pretty much had the same cpu's and gpu's with terribly small improvements. Consumers are not idiots in the aggregate, individually there may be terribly uninformed and low IQ consumers, but in the aggregate consumers are smart and are not going to spend money for no gains.

980ti is closer to 40% and even higher with newer games and Nvidia drivers focused on Maxwell. With CPU's its a problem of no competition and/or a design issue. With GPU's is simply down to being stuck on 28nm. What the GPU IHV's have achieved being stuck on the same node since 2011 is remarkable. with 14/16nm next year the improvement will be what we would have expected in the past for a node jump - close to 100% or maybe even better since its a node and a half jump really.
 

poofyhairguy

Lifer
Nov 20, 2005
14,612
318
126
I am interested to see in what other people's thoughts are on why the desktop discrete GPU market has declined so dramatically in recent years? Please share your opinions.

All of PC sales are declining. People are reprioritizing their tech spending. Rather than have some monster $3k desktop that games, internets, and runs your life that budget gets cut into a Macbook Pro, a PS4, an iPad and an iPhone. Gaming shifts to mobile devices or consoles just so the primary PC in their life can have a better form factor. We all knew a priority on mobile was eating the lunch of dGPUs, that is why it has taken forever for new nodes- cell phones got first access.

To me the amazing story is not the decline of the dGPU or even the gaming PC, but how strong the PC gaming market has become despite this decline. We get more and better console ports than any time in the history of PC gaming when PC sales are declining. PC gaming quite frankly should be dead, but it is actually a pretty viable platform if you can wait a few months after a release to actually play a game.

What keeps PC gaming around if the costs of the hardware are going up? Unionization of buyer power via Steam? Fortune smiled when the consoles went x86? A locked in master race market who refuses to console game? I really wonder.

The big question for the future of course is "can VR save gaming PCs?" The irony in that is even though VR will run on gaming rigs and will be targeting gamers as early adopters, what would REALLY save the market is if non-gamers come in. Almost like with bitcoin mining if there could be a boom of normal people looking into VR for a Ready Player One-version of escapism then maybe, just maybe, there will still be dedicated GPUs to buy in a few years.
 

boozzer

Golden Member
Jan 12, 2012
1,549
18
81
Why are people still spreading the myth of the IGP as some sort of realistic GPU? The fact of the matter is low and mid range users won't be buying the latest generation Intel CPU's, they are not going to dish out every year or so $160+ on an Intel CPU that has 10%-15% improvement in GPU power over the previous one.

Even the latest IGP's are just garbage for any real gaming, if you really do want to game even an Nvidia GT 640 or AMD 250x is still better.

dGPU sales are dropping, as are dCPU sales, because we haven't had a real improvement in the past 4 years!

CPU's, GPU's have all been stagnant, this is why Intel I7 2600k can still hold its own with the I7 6700k. 4 generations later and the I7 6700 is barely 30% faster than a 2600k.

GPU's are the same, a GTX 780ti is barely 20% slower than a 980ti and this is with Nvidia's drop of optimizing for the 700 series and lower.

AMD's 7970GHz is 290x is 390x they all are about 10% difference of each other.

Why would anyone buy a new CPU or GPU if you already own a say Intel 2000+ series CPU or GTX 600 series or AMD 7000 series?

If you own a GTX 460 or 560 you might want to upgrade, but even so your options are expensive and bad.

Same for AMD. Unless there is a real new generation, that brings significant performance improvements we won't see any gains in the dGPU or dCPU.

We've reached a technological stagnation where in the past 4 years we've pretty much had the same cpu's and gpu's with terribly small improvements. Consumers are not idiots in the aggregate, individually there may be terribly uninformed and low IQ consumers, but in the aggregate consumers are smart and are not going to spend money for no gains.


I know you are trying to make a point, but downplaying the performance gaps between the gpus you mentioned just makes your entire post into fud.:thumbsdown:
 

tential

Diamond Member
May 13, 2008
7,355
642
121
980ti is closer to 40% and even higher with newer games and Nvidia drivers focused on Maxwell. With CPU's its a problem of no competition and/or a design issue. With GPU's is simply down to being stuck on 28nm. What the GPU IHV's have achieved being stuck on the same node since 2011 is remarkable. with 14/16nm next year the improvement will be what we would have expected in the past for a node jump - close to 100% or maybe even better since its a node and a half jump really.
If it's not a 100% jump (aka midrange of next year on par with fury x/980ti, then I won't be interested
 

DustinBrowder

Member
Jul 22, 2015
114
1
0
I know you are trying to make a point, but downplaying the performance gaps between the gpus you mentioned just makes your entire post into fud.:thumbsdown:

I mentions performance at the start, before Nvidia started gimping their 700 series to perform slower and pushing the 900 series to perform faster in order to get the idiots in the consumer aggregate to spend more money!

I'm talking on aggregate consumers are not idiots, idiots might still go and buy a new gpu or cpu every year, but on the aggregate consumers are not stupid and won't purchase new products just for the sake of it with no real value!

I mean few specific Nvidia games, few specific benchmarks in certain ways and the reference 780ti compared to a super clocked 980ti might even be 40% slower.

But the 780ti was faster than the 290x, with the 290x being just 20% slower than the 980ti that is the realistic difference, without the Nvidia tricks that the individual dumb consumers fall for!
 

4K_shmoorK

Senior member
Jul 1, 2015
464
43
91
I mentions performance at the start, before Nvidia started gimping their 700 series to perform slower and pushing the 900 series to perform faster in order to get the idiots in the consumer aggregate to spend more money!

I'm talking on aggregate consumers are not idiots, idiots might still go and buy a new gpu or cpu every year, but on the aggregate consumers are not stupid and won't purchase new products just for the sake of it with no real value!

I mean few specific Nvidia games, few specific benchmarks in certain ways and the reference 780ti compared to a super clocked 980ti might even be 40% slower.

But the 780ti was faster than the 290x, with the 290x being just 20% slower than the 980ti that is the realistic difference, without the Nvidia tricks that the individual dumb consumers fall for!

You sure about that? Or did the guy in the Best Buy laptop section tell you that?

http://www.anandtech.com/bench/product/1496?vs=1441
 

Rezist

Senior member
Jun 20, 2009
726
0
71
As someone in Canada the prices are getting really bad, but our loonie is tanking and the USD is high. That said however old my 7950 is now, 3+ years I believe, I actually can't get a card as fast as it for less then what I paid 3 years ago. So basically unless I'm spending 300+ it's pointless
 

5150Joker

Diamond Member
Feb 6, 2002
5,559
0
71
www.techinferno.com
The answer to the OP is very simple: PC AIB attach rate has been going down forever because more and more gamers are choosing to build their own systems rather than buy them from Dell, HP, Alienware, Acer, Asus etc. I'm pretty sure attach rate tracks OEM shipments rather than individual unit sales. That's why PC gaming as a whole is on the rise while attach rates continue to plummet because pre-made PC desktop sales are plummeting as tablets and smartphones take their place.

Eventually the market will stabilize but not enough to sustain two multibillion dollar companies and we're already beginning to see very clearly that the only company left standing making dGPUs for PC gamers will be NVIDIA. AMD is in financial ruins and will be forced to exit this market sooner than later, I can almost guarantee it. In fact, they probably would have exited by now if it wasn't for their console wins keeping them around. Get ready for a monopoly in the dGPU business, it's not anyone's fault (not even AMDs) because the market just can't sustain both of these competitors anymore.
 
Last edited:

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
The answer the OP is very simple: PC AIB attach rate has been going down forever because more and more gamers are choosing to build their own systems rather than buy them from Dell, HP, Alienware, Acer, Asus etc. I'm pretty sure attach rate tracks OEM shipments rather than individual unit sales. That's why PC gaming as a whole is on the rise while attach rates continue to plummet because pre-made PC desktop sales are plummeting as tablets and smartphones take their place.

AIBs are Asus, Gigabyte, MSI, EVGA, Sapphire, Colorful etc.
 

MajinCry

Platinum Member
Jul 28, 2015
2,495
571
136
Another fella's that part of the problem.

Sat on my 6670 DDR3 for a couple years, then moved onto a second hand 7850, which I've been using ever since.

For us lads that don't have much spendin' power and not a silvery-goldy-type-PSU, there's just not anything worth upgrading to.

Ya could point to the 960, but it's around 40% faster and costs around £130-£140. I paid £90 for me 7850, mind.

If us poorish lads got catered to, I dare say that the shipments could very well go up. The corporations' profits might not share the same fate, though.
 

Maximilian

Lifer
Feb 8, 2004
12,603
9
81
6) A lot of very casual PC gamers used to buy entry level GPUs to play any games at all. This was a larger portion of GPU sales than people give credit for. Now their Intel integrated graphics run everything they want anyway.

Yup, theses so much multiplayer free to play stuff thats actually good and dosent require a lot of graphical horsepower these days.

all the mobas, world of tanks/warships

Plus all the indie games that are around. Last graphically demanding game I bought was alien isolation in 2014.
 

5150Joker

Diamond Member
Feb 6, 2002
5,559
0
71
www.techinferno.com
AIBs are Asus, Gigabyte, MSI, EVGA, Sapphire, Colorful etc.

Just checked JPR and here's how they define it:

The AIB market now has just four chip (GPU) suppliers, who also build and sell AIBs. The primary suppliers of GPUs are AMD and Nvidia. There are 48 AIB suppliers, the AIB OEM customers of the GPU suppliers, which they call &#8220;partners.&#8221;


In addition to privately branded AIBs offered worldwide, about a dozen PC suppliers offer AIBs as part of a system, and/or as an option, and some that offer AIBs as separate aftermarket products.

http://jonpeddie.com/publications/add-in-board-report

Seems it can be both.