Why have AMD APUs failed on the market?

Page 27 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

cbn

Lifer
Mar 27, 2009
12,968
221
106
. . . really? Value shoppers who buy in volumes sufficient to make or break the bottom line of a company like AMD generally do not give a darn whether there is "more CPU" or "less iGPU" or anything of the sort. They just want it to work well enough to carry out some fairly mundane computing tasks without being a pain to use. That is, they probably want something better than an E1-2500, that's for darn sure.

The price matters a lot and that is why the large iGPU needs to go.

By getting rid of the large iGPU, the price of the big core APUs drops and become closer to the cat cores.

....And a AMD quad big core APU is going to be a lot more pleasant to use than any cat core. (This because of the difference in performance of the cpu)

AMD would not gain much by making a chip to cater to a vocal minority of budget DIY overclockers who are essentially asking for Kaveri with 128 SPs. Or what have you. The volume would be very low for chips like that. They would rather sell you an 860k or an FX

I have to imagine Athlon x4 860K is a really low volume item. How many dies happen to have a defect in display, media, and decode/encode, etc (the only parts that if suffered a defect would render the entire iGPU unuseable)?

Furthermore Athlon x4 860K is not useable for OEM desktop due the lack of iGPU.

And by Q3 2016, FX is getting too old and expensive to make also. Same goes for the AM3+ motherboards which are less integrated than modern platforms.

if you're really that averse to a "big" iGPU, for whatever reason.

My reason for disliking the big iGPU is purely because it is too expensive. If that iGPU was reduced in size all the big core APU prices could drop by a very meaningful amount.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Doesn't matter. Even if they had started work on that chip when anyone, you or otherwise, started suggesting it on these forums, they'd have it out by 2016, which is right before Xen. They'll be on Excavator quads shortly thereafter as well. It's too late. If AMD wanted 3M Steamroller in any form, it'd be out by now.

It doesn't technically need to be Steamroller. I just use that as an example because it is a construction core that is known to clock well and has sufficient IPC. If Excavator is a better desktop core, then AMD can use that in a hexcore APU with small iGPU in 2016.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Who knows what they would want for an overclocked 5350, or a 35W Carrizo-L (which tops out at 25W anyway, according to AMD).

Not sure, it just seems like a way to make the existing silicon more desirable in some cases.

P.S. Raising TDP to 30/35 watts for AM1 isn't so much about raising clocks, it is about allowing more chips to make it into a higher bin.
 
Last edited:

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
The problems are multiple actually. First is the failure to have a leading edge process node available, and the second is failure of APUs to offer a compelling advantage in either gaming or compute. This is because HSA never really caught on, so the weak cpu performance is not balanced out by gpu compute, and the second is because bandwidth limitations have prevented APUs from being competitive with a similarly priced budget cpu plus dgpu.

But why didn't HSA catch with developers? Was it because AMD tried everything it could to spread its adoption, things like deploying tools, software libraries, bleeding edge compilers, all that, or was because AMD usual approach, develop and they will come, again backfired?

No, they can't, not now. As I have stated, they do not have the resources to go that route. Their course is set through 2016, done deal. No going back.

My two cents on this discussion: If AMD was free to choose a foundry they would make a smaller iGPU version of their chips. It is already quite clear that the iGPU is not a main driver for APU pricing, so the big iGPU becomes an unnecessary cost burden for the product.

But with AMD is having to play shenanigans to fill the WSA commitment, so making a smaller chip today doesn't really translate in smaller manufacturing costs, because using less wafers just means a higher take-or-pay charge in the end of the year.
 
Aug 11, 2008
10,451
642
126
It doesnt really matter why HSA failed to catch on does it. If it had, they would have had much stronger performance. But it didnt except for a few apps, so they are still handicapped by the relatively weak cpu performance. I suspect there are a number of reasons, including thar the programming is simply not that easy, amd,s lack of resources. and the fact that their market shsre has always been small.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
My two cents on this discussion: If AMD was free to choose a foundry they would make a smaller iGPU version of their chips. It is already quite clear that the iGPU is not a main driver for APU pricing, so the big iGPU becomes an unnecessary cost burden for the product.

AMD actually does charge quite a bit for the iGPU (since the A10-7850K is an Athlon x4 860K plus a 512sp iGPU):

A10-7850K: ($159.99 ) http://www.newegg.com/Product/Produc...82E16819113359

Athlon x4 860K: ($79.99) http://www.newegg.com/Product/Produc...82E16819113379

However, I am doubtful how many takers there are for that kind of $79.99 iGPU premium when relatively powerful GDDR5 R7 250X video cards are often on sale for $59.99 AR.

I'll bet we see the same fate for these expensive A10 APUs as we have seen for the Trinity and Richland APUs (ie. they will stay on shelves for a long time). Meanwhile notice that Athlon x4 860K at Newegg is sold out.

But with AMD is having to play shenanigans to fill the WSA commitment, so making a smaller chip today doesn't really translate in smaller manufacturing costs, because using less wafers just means a higher take-or-pay charge in the end of the year.

AMD is rumored to be fabbing R9 390/390X (aka Fiji) at GF. Each one of those Fiji GPUs is ~ 550mm2. So this should help the reduce the impact of WSA.

P.S. AMD should not just reduce the size of the iGPU on the APUs, they should add one more module as well. (especially with DX12 making multi-threading more desirable). IMO a good hexcore CMT chip is what is needed for gaming in order to help AMD differentiate from Intel at the lower price points.
 

jpiniero

Lifer
Oct 1, 2010
16,865
7,305
136
But why didn't HSA catch with developers? Was it because AMD tried everything it could to spread its adoption, things like deploying tools, software libraries, bleeding edge compilers, all that, or was because AMD usual approach, develop and they will come, again backfired?

Well, the obvious answer is "Because Intel doesn't support it". But if you look at even AVX2, there is basically nothing using that either.
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,695
136
However, I am doubtful how many takers there are for that kind of $79.99 iGPU premium when relatively powerful GDDR5 R7 250X video cards are often on sale for $59.99 AR.

Honestly, I think that statement sums up this hole thread. At least on the desktop side of things.

If they charged an additional ~$20 for the iGPU it would be more reasonable and more in line with the performance of the iGPU... :hmm:

As for CPU performance, its kind of sad AMD still doesn't have an FM2(+) APU that decisively outperforms my 6800K on all counts.

my 2c...
 

Erenhardt

Diamond Member
Dec 1, 2012
3,251
105
101
Honestly, I think that statement sums up this hole thread. At least on the desktop side of things.

If they charged an additional ~$20 for the iGPU it would be more reasonable and more in line with the performance of the iGPU... :hmm:

As for CPU performance, its kind of sad AMD still doesn't have an FM2(+) APU that decisively outperforms my 6800K on all counts.

my 2c...

additional ~$20 for the iGPU
 

cbn

Lifer
Mar 27, 2009
12,968
221
106

That iGPU is smaller than the one on the A10-7850K (384 sp vs 512 sp), and the cpu cores are locked and clocked lower than A10-7850K or Athlon x4 860K. It is also comes with a smaller, lower TDP heatsink.

Also, another thing to consider (with any APU value assessment) is the higher price of fast RAM for iGPU gaming:

-- 2 x4GB DDR3 2133 is $10 more than 2 x 4GB DDR3 1600.

-- 2 x4GB DDR3 2400 is $13 more than 2 x 4GB DDR3 1600.
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
As for CPU performance, its kind of sad AMD still doesn't have an FM2(+) APU that decisively outperforms my 6800K on all counts.

Maybe AMD should design a construction core hexcore with small iGPU with both dual channel DDR3 and DDR4 memory controllers so it works in both FM2+ and FM3 sockets?

The only downside is that the two memory controllers would increase die size.
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,695
136
Maybe AMD should design a construction core hexcore with small iGPU with both dual channel DDR3 and DDR4 memory controllers so it works in both FM2+ and FM3 sockets?

I've said the same thing myself on multiple counts. A "steamroller FX6350" with a 128/192SP IGP would be a very nice product if priced right.

:)
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
AMD is rumored to be fabbing R9 390/390X (aka Fiji) at GF. Each one of those Fiji GPUs is ~ 550mm2. So this should help the reduce the impact of WSA.

P.S. AMD should not just reduce the size of the iGPU on the APUs, they should add one more module as well. (especially with DX12 making multi-threading more desirable). IMO a good hexcore CMT chip is what is needed for gaming in order to help AMD differentiate from Intel at the lower price points.

I wouldn't be so optimistic, especially with both their CPU and GPU business plunging the way they are. Around Q214 volumes the console chips could provide volumes enough for the WSA quota to be filled. With another > 15% drop in their CPU + GPU business combined, I don't think it will be the case anymore.

AMD sorely needs another high end embedded volume, and even then it will just kick the can down the road, because of the embedded product life cycle.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
I wouldn't be so optimistic, especially with both their CPU and GPU business plunging the way they are. Around Q214 volumes the console chips could provide volumes enough for the WSA quota to be filled. With another > 15% drop in their CPU + GPU business combined, I don't think it will be the case anymore.

How many wafers per quarter on 28nm does AMD need to in order to fulfill the WSA?
 

MiddleOfTheRoad

Golden Member
Aug 6, 2014
1,123
5
0
By the same token, do you call it a "success" to be barely breaking even on a good year when intel is making billions per quarter?

Barely breaking even sure beats the billions they were losing just a few short years ago. They lost 3 Billion in 2008 alone back when they were aggressively trying to compete in desktop CPU performance.

Considering Laptop sales are on the rise and Desktop sales continue to shrink -- I think their APU strategy is probably the right plan, but I seriously doubt they'll ever be the profit monster that is Intel. AMD has never been a particularly profitable company for their entire history.
 
Mar 10, 2006
11,715
2,012
126
Barely breaking even sure beats the billions they were losing just a few short years ago. They lost 3 Billion in 2008 alone back when they were aggressively trying to compete in desktop CPU performance.

Considering Laptop sales are on the rise and Desktop sales continue to shrink -- I think their APU strategy is probably the right plan, but I seriously doubt they'll ever be the profit monster that is Intel. AMD has never been a particularly profitable company for their entire history.

I would very much enjoy an explanation as to what drove those $3 billion in losses in 2008.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
I would very much enjoy an explanation as to what drove those $3 billion in losses in 2008.
Goodwill write down from ATI acquisition. In plain english, they paid too much and value they said was on the acquisition simply wasn't there.
 

Valantar

Golden Member
Aug 26, 2014
1,792
508
136
Regarding the performance improvements in the DX12 Star Swarm demo for AMD APUs vs. the Core i3, isn't all we're seeing really how (in GPU heavy scenarios) AMDs lackluster single core performance is now negated by being able to spread the load out to more cores? I.e. total CPU load for the AMD APUs would still be FAR higher than the i3, even with ~the same performance? I'd like to see a benchmark like this factor in power draw and CPU load at the same time.

In other words, I'm imagining it looks something like this (yes, these numbers are completely made up):
DX11:
AMD: 30FPS @ 100% load core 0, 0% core 1,2,3 (CPU limited)
Intel: 50FPS @ 100% load core 0, 0% core 1 (CPU limited)

DX12:
AMD: 60FPS @ ~80% load core 0,1,2,3 (GPU limited)
Intel: 60FPS @ ~50% load core 0,1 (GPU limited)

Or am I wrong?

If this is the case, any game with any significant CPU load in addition to graphics draw calls would go back to giving Intel a very significant advantage once again.


On the other hand, I'm hoping for HSA making a splash in gaming - wouldn't handing over physics processing, for example, to the iGPU make sense?


When it comes to AMDs current lineup, the cat cores are really just not good enough. They always end up in SKUs priced so that one (at least I) would expect ~Celeron levels of performance. Instead what you get barely touches Atom. That simply isn't good enough. Also, when you're talking about 25W parts not being viable in 18-20" AIO PCs ... what? My 5-year-old Lenovo X201 has a 35W CPU and handles that just fine. The cooling solution in that is about as simple (and small) as you get, perhaps with the exception of the fins being copper. How on earth can they not match that kind of cooling in a chassis that much larger?
 

Shivansps

Diamond Member
Sep 11, 2013
3,918
1,570
136
Star Swarm is crap, ive said this a lot of time, its really unrealistic for gaming, yes you can get more draw calls from your slow cpu, but the other wall is Physics, AI and game logic, and Star Swarm have none of that, basic Physics and basic AI, in a limited number too, as Star Swarm has wings, in with only 1 very rudementary AI is running on them, with a lot other "fighters" in "follow the leader" "mode".

You cant make any conclusion based of that, other than it allows more draw calls than DX11... now, when the cpu cores will be loaded with other real stuff you gona need in order to run a game? well...
 

DrMrLordX

Lifer
Apr 27, 2000
22,967
13,063
136
The price matters a lot and that is why the large iGPU needs to go.

"Big iGPUs" aren't selling to volume consumers at low price points, so whether or not AMD drops that lineup is irrelevant. AMD is selling them cat cores.

By getting rid of the large iGPU, the price of the big core APUs drops and become closer to the cat cores.

By getting rid of "large iGPUs", AMD has to tape out new designs, the cost of which must be recovered through the sale of resulting chips. They already have various Puma designs (Beema/Mullins) taped out, so they can keep pimping those as a replacement to the dual core Jaguar crap.

....And a AMD quad big core APU is going to be a lot more pleasant to use than any cat core. (This because of the difference in performance of the cpu)

Volume customer notices this too infrequently to care. OEM wants to sell that "more pleasant" experience at a price point over $400. Again, look at what they are charging right now for an E2-3800.



I have to imagine Athlon x4 860K is a really low volume item. How many dies happen to have a defect in display, media, and decode/encode, etc (the only parts that if suffered a defect would render the entire iGPU unuseable)?

Quite low, in fact. That ought to tell you how much our personal preferences matter in the overall scheme of things when it comes to volume chip sales.

Furthermore Athlon x4 860K is not useable for OEM desktop due the lack of iGPU.

Nobody ever suggested that it was useable in OEM desktops.

And by Q3 2016, FX is getting too old and expensive to make also. Same goes for the AM3+ motherboards which are less integrated than modern platforms.

FX is going to die before then. They just want DIY users to pick over what remains of it shortly after they stop stocking the channel.

My reason for disliking the big iGPU is purely because it is too expensive. If that iGPU was reduced in size all the big core APU prices could drop by a very meaningful amount.

It could, but AMD didn't do that, and it's too late for that. Technically they gave us a desktop CPU with small iGPU (AM1 Kabini), though it wasn't Steamroller. Adoption of the part has been slow to say the least, despite the fact that you can even OC it to 2.7-2.9 ghz with a lot of effort. People still don't want the things.

Regardless, if AMD had socketed or BGA Kaveri with 128 shaders (native), I think you can tell what OEMs would have done to the chip price-wise, when they want $500+ for an E2-3800. Pfft.

It doesn't technically need to be Steamroller. I just use that as an example because it is a construction core that is known to clock well and has sufficient IPC. If Excavator is a better desktop core, then AMD can use that in a hexcore APU with small iGPU in 2016.

They have an Excavator part coming in 2016 for desktop: Bristol Ridge. If they really want to dick around with 128, 256, or even 384 shader variants of Bristol Ridge, they've given no indication of that. Expanding the Excavator lineup shortly before the Zen launch just seems like a poor idea all around.

Not sure, it just seems like a way to make the existing silicon more desirable in some cases.

P.S. Raising TDP to 30/35 watts for AM1 isn't so much about raising clocks, it is about allowing more chips to make it into a higher bin.

It can be about raising clocks if you're going to keep the bin counts the same. But that is irrelevant. It is important to see what the market will and won't tolerate. I've done my best to show you what OEMs are interested in selling today. 25W Kabini quads in BGA form or AM1 socket - either one - would have been just dandy for AiOs. We know the die sizes are tiny, we know they run cool . . . OEMs just like to be dicks sometimes.

It's almost painful watching people go through gymnastics to try and figure out how AMD can tune their APUs to fit lower price points, when it's already obvious that the OEMs would just gobble up the savings and stiff consumers.

Or they could try to sell them to DIY buyers, but they'd probably wind up in the same position as AM1 Kabini: somewhat respected some of the time, but generally unloved.

My two cents on this discussion: If AMD was free to choose a foundry they would make a smaller iGPU version of their chips. It is already quite clear that the iGPU is not a main driver for APU pricing, so the big iGPU becomes an unnecessary cost burden for the product.

But with AMD is having to play shenanigans to fill the WSA commitment, so making a smaller chip today doesn't really translate in smaller manufacturing costs, because using less wafers just means a higher take-or-pay charge in the end of the year.

GF's 28nm planar is part of the problem, though I ask you this: right now, who besides Intel has a suitable high-performance node smaller than 32nm? Samsung? TSMC?

MAYBE AMD could have gone 3M/4M Steamroller at TSMC, assuming TSMC could have handled the volume. IBM just dumped their fabs so let's not go there. Samsung . . . if insiders are to be believed, steals all their process tech from TSMC. So hey there's your second source after TSMC!

Regardless, when you have AMD producing processors on a node that suits GCN cores/GPUs more than Steamroller cores/CPUs, what to you expect but CPUs light on OoO processing power and heavy on shader counts? I'll restate: a 3M Steamroller on GF 28nm planar would be a 125W TDP chip all on its own. There is a reason why they jumped ship on that idea some time ago. Several, actually, but whatever.

Honestly, I think that statement sums up this hole thread. At least on the desktop side of things.

If they charged an additional ~$20 for the iGPU it would be more reasonable and more in line with the performance of the iGPU... :hmm:

I don't know about anyone else, but my 7700k cost me ~$104, and I could have gotten it for less from TigerDirect had I been on the ball. The 7650k is stepping into that price slot "real soon now" with almost identical specs to the 7700k.

As for CPU performance, its kind of sad AMD still doesn't have an FM2(+) APU that decisively outperforms my 6800K on all counts.

my 2c...

I was under the impression that the 6800k, at least for "enthusiast" use, was a bit slower than Kaveri in most things. Hmm, tell you what. Do you OC your 6800K? I would like to know how it compares to my 7700k.
 
Last edited:

Shivansps

Diamond Member
Sep 11, 2013
3,918
1,570
136
AM1 arrived for 3 reasons only:

1) Because FM2 failed on small form factors, whiout AM1 everything outside mATX whould be Intel, with a small number of AIOs running mobile APUs.

2) OEMs where reluntant to make those soldered Kabini ITX boards, that also failed for several reasons.

3) Intel had won the low end pc market with G1610, they NEED to have the cheaper option avalible to sell a lot.
 
Last edited:
Mar 10, 2006
11,715
2,012
126
Goodwill write down from ATI acquisition. In plain english, they paid too much and value they said was on the acquisition simply wasn't there.

Exactly. So, that's why it's not really apples-to-apples to look at today's results with significantly slashed operating expenses (and thus future R&D pipeline quality) and no ATI write-down and compare it against the results in 2008 where the company's overall scope was much broader and it had taken a big non-cash charge.
 
Last edited:

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
Exactly. So, that's why it's not really apples-to-apples to look at today's results with significantly slashed operating expenses (and thus future R&D pipeline quality) and compare it against the results in 2008 with a big non-cash charge.

Yeah, accounting 101, really, but what can we expect from someone who recommends a Piledrivet server?
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
I'll restate: a 3M Steamroller on GF 28nm planar would be a 125W TDP chip all on its own. There is a reason why they jumped ship on that idea some time ago. Several, actually, but whatever.

If AMD can make a 95 watt hexcore on 32nm, they can surely make a 95 watt hexcore on 28nm.

Furthermore, AMD is already making quad core steamroller with 512sp iGPUs at 65 watts. Why would a hexcore Steamroller with small iGPU be tough at 95 watts? Answer: It isn't