Why have AMD APUs failed on the market?

[Idontcare]

Nada huh?

They seem to be looking for engineers.... lol

https://www.amd.apply2jobs.com/ProfExt/index.cfm?fuseaction=mExternal.returnToResults&CurrentPage=1

Join their workforce as a non-decision maker (engineer) and then wonder which business quarter in your near future will be the one in which you get laid off by management? Please, that's a fool's errand.

AMD's R&D budget is going down, pretty much every quarter. That is not an environment that communicates "career path potential" unless you are fresh out of college and desperately seeking a job, any job, so you can start paying off those college loans.

I noticed you didn't explain to your audience why you've elected to not apply. Surely you aren't expecting others to do that which you are unwilling, right? So how'd the interview go, don't leave us in suspense!

 

[AtenRa]

First rule of engagement : You cannot sell your products if you dont have any products in the market.

The APUs were and still are great for Laptops, yet there are only a hand full of AMD Kaveri APU laptops in Germany, USA but none here in Greece. You can only find one or two Dual Core overpriced Beema/Mullins laptops here. 99% of Laptops are Intel based here.

Second : Consumer perception
AMD products are known to be inferior (to the average consumer) against Intel/NVIDIA no matter what.

Not even a single Kaveri OEM laptop (Did anyone knows of the FX-7600p and how powerful it is at 30W TDP ???) review to be found anywhere in the net the entire 2014.
Internet review sites are not going to buy themselves the hardware to review you know, somebody should have to give them a review sample :whiste: ( yes im looking at you AMD not the OEMs)


Third : overpriced low end Laptops
Selling Dual core Beema/Mullins Laptops at Intel Core i3 prices is a no go from the start.
Personally, why even bothered to make a dual core 1.3GHz and let OEMs put them in Laptops in 2014-2015 anyway ??? Hell, Bobcat was 1.6GHz 2-3 years ago.
This only hurt the perception customers have for you.

Forth : The entire product stack was created for the retail DIY market in a time Mobile(Laptops, All in ones) was starting to bloom. When you loose the DIY perception meter due to Bulldozer, you taken the downhill from there on.
Not only that, but most of the average consumer was shifting to Laptops/All in Ones and the high-end Gamers to Intel/Nvidia. That made Desktop DIY AMD market share to take a direct hit.
Did they actually believed high-end gamers would actually give a dime about APUs ??? Yes APUs are fine for OEM PCs/Laptops/AiO but you have to make the distinction, you dont try to sell 100W TDP APUs to the DIY market.

If you cant do both, you only do one of them. It seams to me they are shifting more and more to Mobile and OEMs now.

The only product worth talking about and the one that maybe will make a new start for AMD is Carrizo.
Specifically designed and manufactured for Mobile/All in One as an SoC to reduce cost not only for the Chip each self but to the overall platform. This product should be very competitive in price/performance and perf/watt for Laptops and AiOs. AMD should focus on Mobile(high perf/watt and low cost) more than ever with products like that.

Just my opinion

 

[Ramses]

Check back, I did note I have not and why.
I've done this before, that was too big a hole to leave.
Good try though.

 

[ShintaiDK]

HSA has not arrived yet.

And will it ever? And for what usage?

Its the prime example of completely wasted resources.

 

[ShintaiDK]

Oh, and they screwed up again with their sales forecast, Kumar was saying AMD would gain share from Nvidia in H214, but instead what we saw is Nvidia record market share. They couldn't forecast that Maxwell would be that good.

Again the lack of performance/watt. We saw AMD release Tonga as the R9 285 only to fail completely flat due to yet again misunderstanding the market entirely.

 

[ShintaiDK]

Join their workforce as a non-decision maker (engineer) and then wonder which business quarter in your near future will be the one in which you get laid off by management? Please, that's a fool's errand.

AMD's R&D budget is going down, pretty much every quarter. That is not an environment that communicates "career path potential" unless you are fresh out of college and desperately seeking a job, any job, so you can start paying off those college loans.

I noticed you didn't explain to your audience why you've elected to not apply. Surely you aren't expecting others to do that which you are unwilling, right? So how'd the interview go, don't leave us in suspense!

Hillarious :awe:

Its always up to someone else to fix the flaws.

Plus its not exactly looking good on the CV, to be layed off again within a year or smimilar.

 

[Abwx]

Again the lack of performance/watt.

I would rather say that they lacked performance in selling overated specs, wich they never do...

Or eventualy debunking from the start Nvidia s shenaniganized specs, i mean they are advertising false specs in features as well as perf/watt, their 2x perf/watt is a total scam.

Other than that, there s no WE in Denmark.?.

 

[ShintaiDK]

I think we all know that you are in complete denial. And everything going wrong with AMD is someone else fault than AMD. Because in no way it could be due to AMD.

 

[cbn]

If we are in a race to the bottom then AMD APU is a bad idea because of the big die.

And I think needing fast dual channel to make better use of the extra die area only makes that worse.

Instead, if racing to the bottom (for the value end of performance desktop) I would like to see a more cpu heavy die (hexcore Steamroller with small iGPU works) that would be smaller overall than the current 245mm2 Steamroller quad core with 512 sp iGPU.

Ideally this more cpu heavy and smaller die would work well with a single DDR4 3200 stick.

Then combine with dGPU for those wanting the extra gaming performance.

 

[myocardia]

First rule of engagement : You cannot sell your products if you dont have any products in the market.

The APUs were and still are great for Laptops, yet there are only a hand full of AMD Kaveri APU laptops in Germany, USA but none here in Greece. You can only find one or two Dual Core overpriced Beema/Mullins laptops here. 99% of Laptops are Intel based here.

Exactly. Had AMD done 5 years ago with their APUs, what Intel did in the past year with
Baytrail tablets, even on a much smaller scale, they would be very relevant in the laptop market, and probably to a smaller extent in the tablet market, today.

They were already being forced to buy CPUs from Global Foundries. They could have offered to just completely give away a certain portion of their APUs, but only if they were going to be installed into laptops/AIOs that had dual-channel RAM that was fast enough to give the APU some performance, and a lower-than-1080P screen.

Had they done this with a mid-level APU, with a mid-level iGPU, every teenage boy on the planet would have demanded that his parents buy him THAT laptop, instead of the similarly-priced Celeron or i3-based laptop that they were considering.

Once you create a market, unless you screw it up, you usually continue to have customers in said market. As long as your customers are happy with the first product they buy from you/with your product inside of it, they generally will continue to buy your product in their subsequent purchases of like type, such as a laptop. That's assuming they aren't the posting on Anandtech CPU/GPU forums type, like you and I.

Second : Consumer perception
AMD products are known to be inferior (to the average consumer) against Intel/NVIDIA no matter what.

This is only because it's very obvious (to that same average consumer) that if AMD were a competitive product, either quality-wise or performance-wise, they would be MUCH better represented in the laptop/OEM market. Had they done what I suggested above, they would be well-represented in the laptop/AIO market today. Too bad that Hector Ruinz was never interested in making AMD a successful company.

 

[ShintaiDK]

And I think needing fast dual channel to make better use of the extra die area only makes that worse.

Instead, if racing to the bottom (for the value end of performance desktop) I would like to see a more cpu heavy die (hexcore Steamroller with small iGPU works) that would be smaller overall than the current 245mm2 Steamroller quad core with 512 sp iGPU.

Ideally this more cpu heavy and smaller die would work well with a single DDR4 3200 stick.

Then combine with dGPU for those wanting the extra gaming performance.

You want to replace one failure with another? The average consumers/OEMs dont want hexcores. And it would just erode the performance/watt further while keeping cost up.

If we look away from the WSA for a moment. Then Kaveri should have been a 150-180mm2 chip or so with 192-256SPs.

 

[cbn]

You want to replace one failure with another? The average consumers/OEMs dont want hexcores. And it would just erode the performance/watt further while keeping cost up.

Currently on top 20 CPU best sellers on Amazon FX-6300 is ranked 3rd, FX-8350 is ranked 5th, A10-7850K is ranked 13th and Athlon x4 860K is ranked 15th.

So hexcore is definitely in demand relative to other SKUs.

Furthermore, by getting rid of the large 512 sp iGPU and focusing on having smaller iGPU (64 or 128 stream processors) with reduced DDR PHY size (compared to Kaveri) die area would be pretty small.

 

[Fjodor2001]

You want to replace one failure with another? The average consumers/OEMs dont want hexcores. And it would just erode the performance/watt further while keeping cost up.

By that logic they don't need quad cores or beefy iGPUs either, because the average consumer does not need it. So Intel should scrap the 4C and GT2 and higher models.

In fact we might as well stop CPU and GPU desktop development right now, because the average consumer already has what it needs.

 

[ShintaiDK]

Currently on top 20 CPU best sellers on Amazon FX-6300 is ranked 3rd, FX-8350 is ranked 5th, A10-7850K is ranked 13th and Athlon x4 860K is ranked 15th.

So hexcore is definitely in demand relative to other SKUs.

Furthermore, by getting rid of the large 512 sp iGPU and focusing on having smaller iGPU (64 or 128 stream processors) with reduced DDR PHY size (compared to Kaveri) die area would be pretty small.

DIY market is insignificant to OEMs. And the fact that a product like the i7 4790K is the most selling should raise some alarm bells.

You didnt consider there may be a problem with that list at all?

 

[Abwx]

I think we all know that you are in complete denial. And everything going wrong with AMD is someone else fault than AMD. Because in no way it could be due to AMD.

I guess that you re talking of yourself at the third person, otherwise your senstence make no sense...

That said i m not saying that it s someone else fault, i did point that if the tables were turned the false specs would had been debunked much earlier given Nvidia tendency to viral market their competitor, you prefer things expressed this way..?.

By that logic they don't need quad cores or beefy iGPUs either, because the average consumer does not need it. So Intel should scrap the 4C and GT2 and higher models.

In fact we might as well stop CPU and GPU desktop development right now, because the average consumer already has what it needs.

You will soon understand that there s some "philosophy" that hang by there and whose motto is that what is good for Intel is forcibly good for consumers, this is what surfaced from the ongoing IGP relevancy debate..

 

[Fjodor2001]

I guess that you re talking of yourself at the third person, otherwise your senstence make no sense...

Maybe he was using the Royal we?

 

[Abwx]

Maybe he was using the Royal we?

Quite possible but i doubt that it will fill the void of the arguments, i mean companies like Intel, or Nvidia to a lesser extent despite the recent scam, is not what we could call a consumer friendly company, they could be but for this they first should be much less dominant than what they are.

It s not wildly speculating that the consumer would be better serviced should AMD increase their marketshare to more decent levels, in the current state of affairs Intel has the possibility to clear from the market products from the competition that are better than their own, rarely in this market the consumer interest has been as much jeopardised than it is currently.

 

[MisterLilBig]

And will it ever? And for what usage?

Its the prime example of completely wasted resources.

Fanboyism blinds the fact that similar technologies exist and are being pushed by the blue and green side. In your sense, everyone is wasting resources then.

But, to answer you. When? Carrizo supports HSA 1.0, so, when that gets launched. For what usage? Anything compute related. You know, like OpenCL 2.0 with shared memory that Intel has been pushing incredibly each generation also.

 

[ShintaiDK]

Fanboyism blinds the fact that similar technologies exist and are being pushed by the blue and green side. In your sense, everyone is wasting resources then.

But, to answer you. When? Carrizo supports HSA 1.0, so, when that gets launched. For what usage? Anything compute related. You know, like OpenCL 2.0 with shared memory that Intel has been pushing incredibly each generation also.

The story about wait and see, anytime soon is just getting boring.

 

[mrmt]

So hexcore is definitely in demand relative to other SKUs.

I don't really get the moar cores bandwagon. We all know that we have to trade core frequency and power consumption for more cores, it's a trade off that is really not small and simple, because depending on the workload we might end up with LESS overall performance than with a small number of beefier cores, as in 5820k vs 5960X.

Only because AMD chased a very bad, botched idea (cram a high number of of high speed, anemic cores on the die) moar cores work for AMD in terms of performance but cripples in terms of die size and power consumption. Moar cores isn't the real answer, a beefier core is.

 

[BSim500]

Isn't that because you can supplement GPU performance with a dGPU while you can't supplement CPU performance, as there are no substitutes for the CPU in a PC?

There is that plus recent slowdowns in year-on-year improvements have caused people to upgrade CPU & dGPU at different time intervals. Eg, of two people I know:-

Man A : He "wasted" his money with a "rip off" $180 i5-2500K + mid-range $200 dGPU compared to the "bargain glories" of 13fps slowdowns you can enjoy on an $90 A8-7600. Oh no - he's $290 out of pocket compared to buying a $90 APU, right? Thing is - that i5-2500K has lasted him 4 years and he plans to keep it another 1-2 years. End result - over 4-6 years that $180 chip has effectively cost him only $30-45 per year. He upgrades his $200 GFX card typically once every 2 years though ($200 each 2 years = $100 per year). Overall, CPU ($30-$45 pa) + dGPU ($100pa) upgrades cost him $130-$145 per year.

Man B : He's the only desktop owner I know that tries to run heavy games on an AMD APU, and so far he's blown over $400 upgrading an A10-5800K to an A10-6800K to an A10-7850K (and now hoping for desktop Carrizo after being frustrated with even 3 year old games not running at a consistent 1080p/30 and newer ones dipping down to 15-20fps). End result - he's spent over $400 in 3 years (and with a desktop Carrizo $550 over 4 years) = $133 per year. He's got some of that back by reselling his 2 old chips on Ebay, so overall it's nearer $90-100 per year.

Total actual premium = only $30-$55 per year (or about +6-10% of total rig cost). Total GPU performance difference that 6-10% premium buys = about +300%. Total enjoyment factor of having a consistent +50-60fps and not having to put up with frustrating 13-15fps slowdowns = 1,000,000%. (And that's with an R9 285 / GTX 960 class of card, whereas even a $70 R7 260 will beat an APU by +75% and if substituted above 1x CPU + 2x $70-100 dGPU's spread over 4 years would actually reduce Man A's gaming costs to $80-$95 pa or below that of Man B's "cheaper" annual APU churning). For desktop gamers, it's a "no brainer" to those capable of understanding the upgrade cycles of CPU & GPU's are now different and far more "staggered" vs 10 years ago, that people upgrading their dGPU will often reuse an existing CPU (and vice-versa), and that "buying a big CPU and making it last" often results in the biggest cost savings vs the annual "last year's APU vs newest games which cripple last year's APU's" rat race. And for non-gamers, excess performance of one brand's iGPU vs another in games is irrelevant.

The bottom line for many people is that APU vs dGPU is often not just about arguing over technical specs at all or comparing prices on any one single day - it's the "BS factor" that divides being able to barely run a game at subpar resolution & quality as some exercise in frugality vs actually being able to enjoy a hobby (even on a budget). Or buying "too cheap" that you end up needing to upgrade 2-3x more often wiping out most of the money "saved" during year 1.

 

[mrmt]

But, to answer you. When? Carrizo supports HSA 1.0, so, when that gets launched. For what usage? Anything compute related. You know, like OpenCL 2.0 with shared memory that Intel has been pushing incredibly each generation also.

If a market leader is already developing a similar technology on open standards, and AMD has access to that market leader patents, why bother with developing their own market standard? What are the chances of AMD making something mainstream? For a company as cash strapped as AMD is, it is better to just follow Intel lead down this rabbit hole and try to come with a more efficient implementation than to try to develop a new standard and then hope software companies will come.

HSA smells of another of AMD's fool's errands.

 

[cbn]

DIY market is insignificant to OEMs. And the fact that a product like the i7 4790K is the most selling should raise some alarm bells.

You didnt consider there may be a problem with that list at all?

Just because something would be good for DIY, doesn't mean it would be bad for OEM.

And if choosing between a very large iGPU tied to dual channel memory vs. an extra cpu module (that could be with single channel) I have to think the extra module would be far more beneficial. For one, an extra module adds far less die size than that very large iGPU and secondly with a hexcore as the starting point there should only be hex and quad cores in the mix. This, in contrast, to a quad core die where there will be some dual cores.

And while a person might say dual cores are appropriate for office machines, I'll bet they will lower the ASPs compared to what hexcore die would yield. Furthermore, I don't really see the point of dual core/single module anymore provided AMD was able to improve Carrizo-L (cat core with small iGPU) sufficiently enough.

 

[mrmt]

Man A : He "wasted" his money with a "rip off" $180 i5-2500K + mid-range $200 dGPU compared to the "bargain glories" of 13fps slowdowns you can enjoy on an $90 A8-7600. Oh no - he's $290 out of pocket compared to buying a $90 APU, right? Thing is - that i5-2500K has lasted him 4 years and he plans to keep it another 1-2 years. End result - over 4-6 years that $180 chip has effectively cost him only $30-45 per year. He upgrades his $200 GFX card typically once every 2 years though ($200 each 2 years = $100 per year). Overall, CPU ($30-$45 pa) + dGPU ($100pa) upgrades cost him $130-$145 per year.

Man B : He's the only desktop owner I know that tries to run heavy games on an AMD APU, and so far he's blown over $400 upgrading an A10-5800K to an A10-6800K to an A10-7850K (and now hoping for desktop Carrizo after being frustrated with even 3 year old games not running at a consistent 1080p/30 and newer ones dipping down to 15-20fps). End result - he's spent over $400 in 3 years (and with a desktop Carrizo $550 over 4 years) = $133 per year. He's got some of that back by reselling his 2 old chips on Ebay, so overall it's nearer $90-100 per year.

This is what every DYI player knows and that's why APU doesn't sell there, it's one of those cases where $10-$50 dollars make a lot of difference in the duration of the rig's life, spend a bit more of money now in order to save your customer a lot more in the future.

It takes really a degree of dishonesty to sell an APU to any serious gamer, I can't see how could any seller do that.

 

[ShintaiDK]

Just because something would be good for DIY, doesn't mean it would be bad for OEM.

And if choosing between a very large iGPU tied to dual channel memory vs. an extra cpu module (that could be with single channel) I have to think the extra module would be far more beneficial. For one, an extra module adds far less die size than that very large iGPU and secondly with a hexcore as the starting point there should only be hex and quad cores in the mix. This, in contrast, to a quad core die where there will be some dual cores.

And while a person might say dual cores are appropriate for office machines, I'll bet they will lower the ASPs compared to what hexcore die would yield. Furthermore, I don't really see the point of dual core/single module anymore provided AMD was able to improve Carrizo-L (cat core with small iGPU) sufficiently enough.

I think the market have already spoken and decided. And it wasnt in favour of hexcores.