Why have AMD APUs failed on the market?

[cbn]

You often won't be paying "only" $90 though as many pre-built Kaveri's have a silly price premium in many areas (because they aren't selling very well and don't benefit from economies of scale).

Example - EBuyer UK:-

£190 Cheapest Pentium G3240, 4GB RAM, 500GB HDD, no ODD, Win 8.1
http://www.ebuyer.com/643320-zoostorm-desktop-pc-7260-8001

£350 Cheapest A8 7600, 8GB RAM, 1TB HDD, DVD Writer, Win 8.1
http://www.ebuyer.com/662321-zoostorm-desktop-pc-7260-2024

Component balancing:-
£26 4GB RAM
£12 DVD writer
£ 8 difference between 1TB vs 500GB HDD
£46 Sub-Total

£350 - £190 - £46 = £114 Kaveri premium (enough for a "free" 2GB VRAM 260X with 2.5x the A8's performance with change to spare and native 1080p capability at decent +40-60fps on Medium in many games). And that's ignoring that for many games you've posted (eg, BI), 4GB RAM + separate 1-2GB VRAM is entirely playable, but you still need 8GB with an iGPU no matter what... Others have posted similar examples from other countries. Simply shouting "what can you buy for $90" over & over doesn't change anything for most people, since:-

1. Most non-tech savvy people buy computers not CPU's and the nearest consistent dGPU-less price equivalent to cheap Pentium "pre-builts" in many countries are stuff like the A4 / A6's 4xxx/6xxx, not the A8 / A10's 7xxx.

2. Most tech-savvy people interested in building their own gaming rig that they want to last for the next 3-4 years will have figured out a dGPU is better long-term value (perf-per-$ / perf-per-watt) even if it isn't the lowest absolute price, ie, +$70 in Year 1 may also = -$70 in Year 3 due to not having to upgrade so often whereas APU iGPU's are outdated almost every 12-18 months. (Hint : Look at the number of people on Youtube with older A10-5800K CPU's who've ended up adding a dGPU anyway for 2014 games...) "Buy cheap, buy twice..."

3. $90 isn't everyone's "hard" budget limit. (Why this obsession with "exactly $90" over & over? Before the A8 was released, and prior to AMD's price-cuts you were happily promoting $180 A10-7850K's, so your own budget seems to be remarkably "flexible" rather than being any serious "budget"...)

4. It's 2015 and many people simply aren't interested in gaming PC's with sub-console performance for heavier games even to save $50-$70 (and for lighter flash games, it's irrelevant). If an A8-7600 is "good enough", then why didn't AMD stick one in the PS4? I think we both know the answer to that...



^ This. The whole hype over HSA simply hasn't been matched in reality. Why? If you want software coders to rewrite everything for your 15% market share, then you have to put serious money into it (on the software support side). So APU's (on a market level) aren't actually that "cheap" to persuade everyone to buy at a premium vs a Pentium for the bulk of non-gaming usage and then demand everyone rewrite their whole software suites which doesn't need doing for the +85% of the other brand, and is irrelevant to most people who don't have giant +5,000 row complex spreadsheets or whose "photo editing" is mostly simple crop / resize / red-eye reduction / color adjustment on photo's taken on their mobile phone rather than perform complex blurs on multi-layered RAW images taken on their semi-pro "Four-Thirds" dSLR. In fact, AMD had more support for 3DNow! from major games devs than they've had with attempts at forcing APU "compute" / HSA specific code optimizations on a primarily disinterested public.

Since I am uk based I liked your twisting of the ebuyer

prices. The amd rig has win7 pro and the Intel one had win8 with bing.

But again you also ignored the cheaper a8 7600 rig with proper win8 for £320;http://www.ebuyer.com/662329-zoostorm-desktop-pc-7260-0039

It gets better if you bother to actually look at this model:http://www.ebuyer.com/662337-zoostorm-desktop-pc-7260-3017

It's the same model without is for £240. Add £70 for win8 that means it costs £310.

Here is an a10-5700 zoostorm for £235:
http://www.amazon.co.uk/gp/aw/d/B00...storm+a10&dpPl=1&dp ID=41AQL5j3miL&ref=plSrch

Has win8 with bing.

Plus using your logic of going cheap why would you be getting a pentium dual core to last a few years anyway?

Plus for games like lol,dota and Minecraft I know plenty of people running them fine on the apu graphics. Remember most comparisons of the apus are to core i3 CPUs which have faster igps than Pentium's. Plus Intel gaming drivers are improving but are still a bit meh.

The thing is what is stupid is that none of the a8 7600 zoostorms use win8 with bing meaning the extra cost is the windows license.

With Zoostorm, the price delta is much smaller than what I have seen with Intel vs. AMD using Dell, HP, Lenovo on Newegg.

I suspect this has to do with Zoostorm buying in much smaller quantities than the large OEMs. (ie, Zoostorm is basically paying closer to a DIY price on both the AMD and Intel processors).

 

[DrMrLordX]

AMD CPU business shrank 70% since 2011, and they forecast a 15% drop in revenue, focused on the CPU business. I expect that 15% drop in total revenue to reflect in 20-25% drop in their CPU business, and even more than that on the CMT chips.

And that will probably represent a total shutdown of Piledriver-era chip sales. Lots of Piledriver buyers, for whatever reason, don't want Kaveri/Kaveri refresh so they'll either buy Intel or just sit on their current chip and wait.

The cat cores are much more competitive in terms of costs than the CMT cores.

If you're talking Richland/Piledriver, yes.

If you're talking Kaveri/Steamroller . . . maybe. Yeah stuff like the FX-7500 is more expensive than an E1 Jaguar, which is why those chips haven't replaced the aging and slow dual-core cat parts. The A4-7300 is closer in cost just for the chip, but cost-of-integration is higher (the listed TDP for the 7300 is also absurd. 65W???) so there hasn't been a migration to 1M Kaveri for the AiO desktop.

If you're talking a 1M Carrizo replacement for 1M Kaveri . . . we don't know yet. Remember these parts are full-on SoC so total cost of integration will be lower than for APUs like 1M Kaveri, and more on par with the cat parts. Also bear in mind that HDL libraries are going to cause die shrinkage across the board, mitigated somewhat by all the extra transistors on Carrizo (for whatever reason). They taped out 1M Kaveri parts, so I'm quite certain we'll see 1M Carrizo.

I'd expect 1M Carrizo to bottom out at around 6-8W TDP and be small enough to be cost-competititve with E1/E2 cat cores. The Excavator module will be clocked around 2-2.5 ghz and easily outperform those old parts, and will probably carry 128-256 shaders.

If Su is on the ball, we'll see those in AiO units as they phase out the cat parts early this year. Then, in 2016 . . .

I think AMD isn't extending their cat line up because Zen will replace it, not because they don't have plans for that market.

Exactly. Some Zen variant is going to replace everything, in my opinion. It is the only uarch that AMD has revealed as moving to the Samsung/GF 14nm process. It would be the logical successor to the cat cores AND Excavator.

So why are there no high clocked dual core cat chips?

They don't scale well voltage-wise. Have you seen how hot an E1-2500 can get inside an AiO chassis? It's possible to push one over 60C with common software. A 2 ghz Jaguar dual-core would require better cooling for the entire unit, which costs $$$. It would also require extra binning and validation, which would increase the price of the units with higher clockspeeds.

Carrizo won't work as a replacement for the current cat cores because the die size is too big. Yes, it will have the performance, but not the low cost to manufacture.

I severely doubt that. Do you think a Carrizo variant of the A4-7300 is going to be all that large? I expect a chip like that, possibly with more shaders, will replace cat cores in AiOs. Or at least, it ought to.

AMD needs to bump off the E1/E2 cat cores, 'cause guess what, here come the 14nm Intel chips.

 

[cbn]

Carrizo won't work as a replacement for the current cat cores because the die size is too big. Yes, it will have the performance, but not the low cost to manufacture.

I severely doubt that. Do you think a Carrizo variant of the A4-7300 is going to be all that large? I expect a chip like that, possibly with more shaders, will replace cat cores in AiOs. Or at least, it ought to.

A carrizo variant of the A4-7300 will be based on the same 245mm2 die as all the other Carrizo SKUs. The major difference is that it will have a lot more silicon disabled.

In contrast, cat soc on 28nm is ~101mm2.

 

[cbn]

So why are there no high clocked dual core cat chips?

They don't scale well voltage-wise. Have you seen how hot an E1-2500 can get inside an AiO chassis? It's possible to push one over 60C with common software. A 2 ghz Jaguar dual-core would require better cooling for the entire unit, which costs $$$. It would also require extra binning and validation, which would increase the price of the units with higher clockspeeds.

The dual core cat core chips should scale the same voltage-wise as any quad core cat chip.

Fact is the Athlon 5350 @ 25 watt TDP has four cores at 2.05 Ghz and the iGPU@ 600Mhz.

This compared to the Sempron 2650 @ 25 watt TDP with two cores @ 1.45 Ghz and iGPU @ 400 Mhz.

Furthermore, the reviews I have seen with Sempron 2650 report a much lower power consumption than the other AM1 SOCs.

 

[ElFenix]

That would be a pretty significant advantage for AMD, to bring their large and confused line of processors down to two chips: Kaveri for socketed desktop, and Carrizo for BGA desktop/mobile. Of course that sort of ignores the tablet space but let's not talk about that, we'll start another contra revenue spat.

i thought carrizo itself was 2 chips - carrizo-m being cat cores and carrizo being construction cores.

any richland/trinity floating around has to be new old stock. i doubt amd is making those anymore. they finally just went oos at my local microcenter.

amd is clearly still making vishera as the e parts wouldn't just be relabels of stuff that's already been shipped.

 

[cbn]

i thought carrizo itself was 2 chips - carrizo-m being cat cores and carrizo being construction cores.

Carrizo-L is the name of the chip with Puma+ cores.

 

[mrmt]

Exactly. Some Zen variant is going to replace everything, in my opinion. It is the only uarch that AMD has revealed as moving to the Samsung/GF 14nm process. It would be the logical successor to the cat cores AND Excavator.

This. This is exactly what I don't expect from Zen. For this to be true Zen would have to scale from tablets to high end workstations in terms of performance, and on top of that be low cost enough for AMD semi-custom business. It would have a market span larger than Broadwell. How AMD would be able to accomplish that with a R&D budget several times smaller than Intel's is beyond my wildest expectations.

I think Zen will have to compromise in lots of places, especially on performance, if it is to be low cost enough for the semi-custom market. I really doubt they will go much further than the bottom of the barrel market brackets they are selling now.

 

[cbn]

They taped out 1M Kaveri parts, so I'm quite certain we'll see 1M Carrizo.

AMD never taped out a 1M Kaveri, the A6-7400K dual core SKU is based on the full 245mm2 die with one module and 256 streasm processors disabled. Furthermore, there is 1MB of cache disabled in the remaining active module.

P.S. The same was true for the Trinity and Richland dual cores. In some cases 256 stream processors were disabled (out of a possible 384) along with one module to make SKUs like the A4-5300 and A4-6300.

 

[DrMrLordX]

A carrizo variant of the A4-7300 will be based on the same 245mm2 die as all the other Carrizo SKUs. The major difference is that it will have a lot more silicon disabled.

We'll see. Regardless, harvested dies can be sold below cost since it's recovery of failed parts. AMD has done it before, and they can do it again. Personally I expect them to tape out a 1M design (my bad if I mistakenly thought the 1M Kaveris were designed to be that way).

The dual core cat core chips should scale the same voltage-wise as any quad core cat chip.

You can't put a 5350 in an AiO box with the same overall cooling setup as an E1-2500 machine (15W TDP). The cooling solutions used in that bargin box are barely capable of keeping the E1 running at acceptable temperatures. Heck, the ones I've used before idle at something like 57-58C. Ridiculous.

Why the Sempron 2650 has a 25W TDP is anyone's guess, unless it's just AMD playing fast and loose with numbers somewhere.

What I am trying to tell you is, from real-world experience seeing how these E1s behave under any kind of load given the included cooling solutions, that trying to increase the clockspeed by nearly %50 is probably going to be a deal-breaker except for some very nicely-binned chips. TDP numbers are great and all, but they don't tell the whole story.

Furthermore, the reviews I have seen with Sempron 2650 report a much lower power consumption than the other AM1 SOCs.

Right, but how do they fare against E1 and E2 chips? And, more importantly, if you bumped them up to 2 ghz, how would they look then?

i thought carrizo itself was 2 chips - carrizo-m being cat cores and carrizo being construction cores.

Carrizo-L actually, but someone else pointed that out already. See below for my "conspiracy" theory.

any richland/trinity floating around has to be new old stock. i doubt amd is making those anymore. they finally just went oos at my local microcenter.

Gosh I hope so.

amd is clearly still making vishera as the e parts wouldn't just be relabels of stuff that's already been shipped.

Correct, I suspect that the E chips and probably the 8310 are coming off newer wafers. AMD is still taking delivery of 32nm wafers (ignoring, for the time being, the possible disappearance of $400 mil worth of wafers to who-knows-where-and-for-what-reason) so they're going to do something with them. And then there's the 8370 . . .

Carrizo-L is the name of the chip with Puma+ cores.

Right. Except the last time I saw any solid information about Carrizo-L was awhile back. Will Carrizo-L really be cat cores with the Carrizo name attached to them?

Part of me thinks that it will, since semi-custom/embedded customers probably like the cat cores (ex: MS, Sony), so AMD had better find a way to keep improving the driving force behind their new source of revenue. Another part of me suspects that the Puma+ core thing was a rumor, and that AMD is cutting corners by taping out a smaller variant of 2M Carrizo as Carrizo-L to replace the cat cores in all non-semi-custom jobs where they go currently. Does this mean that AMD will be trying to push CMT chips onto their semi-custom buyers? Maybe. If the performance and low power usage is there, they will probably not mind too much.

There is also the possibility that they will continue to develop future cat core variants that just haven't made it onto any publicly-displayed roadmaps specifically for their semi-custom buyers. I just don't think AMD's R&D dept has the funds to do all that work simultaneously.

This. This is exactly what I don't expect from Zen. For this to be true Zen would have to scale from tablets to high end workstations in terms of performance, and on top of that be low cost enough for AMD semi-custom business. It would have a market span larger than Broadwell. How AMD would be able to accomplish that with a R&D budget several times smaller than Intel's is beyond my wildest expectations.

It is a question of what costs AMD the most in terms of R&D, and how many compromises they are willing to make. If the "main" Zen product is an 8C(8M?)/16T part aimed at high-end users and servers with a 95W TDP, then, theoretically speaking, they could just cut down the design into 4C, 2C, and even 1C parts to meet different market needs. That would mostly shut them out of tablet space, but a 1C Zen should have a TDP of, what, ~12W? Maybe less? It would be doable. If Zen can outperform Puma+ (assuming it even sees completion) and Excavator at any given clockspeed up whatever clockspeed Zen launches at (3 ghz? 4 ghz?), then semi-custom buyers will probably want to see it in their next purchase rather than a cat core.

Just taping out a bunch of cut-down Zen chips at equal or lower clockspeeds probably costs less than continuing development of 3+ processor lines simultaneously. It would be like the k8 days all over again.

I think Zen will have to compromise in lots of places, especially on performance, if it is to be low cost enough for the semi-custom market. I really doubt they will go much further than the bottom of the barrel market brackets they are selling now.

Undoubtedly. They will probably hack it down into smaller versions of the 8C/8M part we're seeing on roadmaps. Then they'll start tacking on iGPUs as necessary, even if they have to do an MCM job.

edit: Here is the most recent news I've seen on Carrizo-L:

http://www.extremetech.com/extreme/...arrizo-l-are-both-28nm-mobile-only-processors

Make of that what you will. It looks like Carrizo-L will be using the existing Puma core from Beema/Mullins. Or, at least, that's what AMD is saying. Considering how close they are to launch and considering the fact that they have demoed some hardware, that is probably true.

So, no Puma+. Just Puma.

That sort of plays into my theory that they're punting on the cat cores, and sort of doesn't. At the very least, it shows that AMD has no updated cat cores on the roadmap to speak of, since Carrizo-L != Puma+.

Also interesting to note that AMD is now claiming Carrizo will go all the way down to 10W, rather than 12W.

 

[Blitzvogel]

This. This is exactly what I don't expect from Zen. For this to be true Zen would have to scale from tablets to high end workstations in terms of performance, and on top of that be low cost enough for AMD semi-custom business. It would have a market span larger than Broadwell. How AMD would be able to accomplish that with a R&D budget several times smaller than Intel's is beyond my wildest expectations.

I think Zen will have to compromise in lots of places, especially on performance, if it is to be low cost enough for the semi-custom market. I really doubt they will go much further than the bottom of the barrel market brackets they are selling now.

Moar cores? But one must remember that Zen is about ARM - AMD x86 compatibility at least from a socket standpoint. Perhaps the mobile/tablet platforms will be x86 + ARM on one die for software and OS compatibility in some sort of BIG.Little fashion or AMD could be going for broke with some way of efficiently translating x86 instructions to ARM and vice versa. Transmeta managed to do something similar. Laptops and desktops would strictly be x86 where it is king. AMD could go either x86 or ARM or even further flesh out x86 - ARM interoperability on servers.

From an x86 standpoint though, AMD's best bet would be something like Broadwell where the design focus is on efficiency and mobile computing which can then grown into variants with more cores to suit other platforms. I also don't see why AMD couldn't go BIG.Little with a couple cat cores + a couple Zen cores either but AMD is in such a tough spot unit they are up and producing a new core on a new process tech.

Depending on your point of view, AMD hasn't released a "big core" since the Phenom II days. CMT is a cluster**** considering the state of multithreaded software, and each integer core doesn't compete with Stars unless it's ramped up to high clocks. A Stars core with AVX2 on 14nm would be a great start.

 

[cbn]

We'll see. Regardless, harvested dies can be sold below cost since it's recovery of failed parts. AMD has done it before, and they can do it again. Personally I expect them to tape out a 1M design (my bad if I mistakenly thought the 1M Kaveris were designed to be that way).

In some cases they are failed parts, but I do not believe all the disabled silicon has a defect in it.

Example, maybe a particular die has a defect in one of the modules, but the entire iGPU is good. However, to make a balanced 1M part that fits into AMD's product segmentation plan 50% to 67% of the iGPU is disabled as well as 1MB of cache in the remaining active module.

Why the Sempron 2650 has a 25W TDP is anyone's guess, unless it's just AMD playing fast and loose with numbers somewhere.

Pretty sure the downclocking of Sempron 2650's dual cat cores and iGPU is done for product segmentation, just like the all the 1M Richland and Kaveri parts have the vast majority of their iGPU disabled as well.

 

[cbn]

What I am trying to tell you is, from real-world experience seeing how these E1s behave under any kind of load given the included cooling solutions, that trying to increase the clockspeed by nearly %50 is probably going to be a deal-breaker except for some very nicely-binned chips. TDP numbers are great and all, but they don't tell the whole story.

These E1 and E2 Kabini are the low TDP (15 watt) cat core chips.

What I am referring to are the higher TDP (25 watt) cat core chips that in some cases overlap with the FM2 dual cores in certain performance metrics.

 

[Shivansps]

They dont overlap, the best AM1 cpu is just a little bit better than the current gen worse FM2+ APU, the 7300, and the AM1 is worse on ST, on IGP the best AM1 can not beat the worse on 1150, lets not talk about the 7300.

And the AM1/5350 and 7300/FM2 build are very close in price, i whould take the 7300 or the Celeron 1150 ANY DAY over the 5350...

The only exception is ITX, and thats bring me back to my original point, why FM2 fails so badly at ITX?

 

[DrMrLordX]

These E1 and E2 Kabini are the low TDP (15 watt) cat core chips.

What I am referring to are the higher TDP (25 watt) cat core chips that in some cases overlap with the FM2 dual cores in certain performance metrics.

Well, again, we're talking $300-$400 desktop AiO and all that, right? The 25W chips don't have a snowball's chance in hell of getting in there, regardless of whether they're cat cores or otherwise.

What I'm trying to say is, Carrizo (not just Carrizo-L) has a good chance to operate at 10-15W TDP which is the space currently occupied by junky E1/E2 cat chips. AMD can move Carrizo into that space. They would be absolutely insane not to do it. I'll bet you there will be price parity between them and the 15w cat chips.

With our luck, they'll move Carrizo-L in there which would still be a step up, but not as big of one.

There is no need to try and shoehorn 25W cat chips into that space. The cooling solutions preferred by OEMs simply will not support them.

 

[cbn]

Well, again, we're talking $300-$400 desktop AiO and all that, right? The 25W chips don't have a snowball's chance in hell of getting in there, regardless of whether they're cat cores or otherwise.

Here is the listing for $300 to $400 AIO at Newegg:

http://www.newegg.com/Product/Produc...0023501%204019

I see a lot of low power chips in this listing, but I wonder if this is primarily due to AIO chassis not being able to handle higher power chips or simply due to fact the processors in these low price machines need to be low cost?

I suspect it has to do more with processor price because there are plenty of AIOs with 35 watt and 54 watt processors in the higher price brackets:

$400 to $500 price bracket: http://www.newegg.com/Product/Produc...0023501%204020 (Some of these machines have 35 watt Intel processors)

$500 to $750 price bracket: http://www.newegg.com/Product/Produc...01+4021&Page=3 (Some of these machines have 54 watt Intel processors)

 

[TurtleCrusher]

Well, again, we're talking $300-$400 desktop AiO and all that, right? The 25W chips don't have a snowball's chance in hell of getting in there, regardless of whether they're cat cores or otherwise.

What I'm trying to say is, Carrizo (not just Carrizo-L) has a good chance to operate at 10-15W TDP which is the space currently occupied by junky E1/E2 cat chips. AMD can move Carrizo into that space. They would be absolutely insane not to do it. I'll bet you there will be price parity between them and the 15w cat chips.

With our luck, they'll move Carrizo-L in there which would still be a step up, but not as big of one.

There is no need to try and shoehorn 25W cat chips into that space. The cooling solutions preferred by OEMs simply will not support them.

Those super slim desktop AIOs used to be powered with 35w C2D/i3 mobile cpus just fine, so 25w isn't unreasonable. My laptop has a 25w p8700 and a 20w GPU and it doesn't really get that warm while gaming. Why a 25w APU couldn't be in an AIO is beyond ridiculous. It can.

 

[mrmt]

It is a question of what costs AMD the most in terms of R&D, and how many compromises they are willing to make. If the "main" Zen product is an 8C(8M?)/16T part aimed at high-end users and servers with a 95W TDP, then, theoretically speaking, they could just cut down the design into 4C, 2C, and even 1C parts to meet different market needs.

I don't think Zen will be remotely aimed at high end users at all. It is built on a process highly tuned for SoCs and whatever the scope of their new CPU design it must be low cost enough for semi-custom. This level of performance simply isn't compatible with semi-custom, as it must need attractive costs in the first place. Of course they can throw more hardware at the problem in the form of moar cores, but this approach isn't working for them.

Undoubtedly. They will probably hack it down into smaller versions of the 8C/8M part we're seeing on roadmaps. Then they'll start tacking on iGPUs as necessary, even if they have to do an MCM job.

That's not the compromise I'm talking about. I think Zen performance per core won't be much distant from the cat cores. Closer to cat levels than to Skylake levels, that would be my forecast. You can't have Skylake-like performance without paying for Skylake-like R&D and without having Skylake-like manufacturing costs, and the two are something AMD cannot afford today.

Moar cores? But one must remember that Zen is about ARM - AMD x86 compatibility at least from a socket standpoint.

What we say here in this forums is that we want moar cores, but what we don't say is that we want moar cores in a given SKUs but clocked as high as in SKUs with less cores. And that's why moar cores might work on servers but certainly won't work on consumer market at any level.

AMD today has plenty of 16 core choices on the server market, if there was a remotely prospective market for them they would bring those chips for the consumer market (nobody buys them for servers anyway).

Perhaps the mobile/tablet platforms will be x86 + ARM on one die for software and OS compatibility in some sort of BIG.Little fashion or AMD could be going for broke with some way of efficiently translating x86 instructions to ARM and vice versa. Transmeta managed to do something similar. Laptops and desktops would strictly be x86 where it is king. AMD could go either x86 or ARM or even further flesh out x86 - ARM interoperability on servers.

Transmeta tried this approach and failed, because once Intel switched to power optimized designs software approach couldn't beat dedicated hardware approach and Transmeta went down. I also don't believe Zen will have any complex, fancy thing like big.little or any kind of emulation as it would put even more strain on AMD software guys. They are already behind with their pet vaporware of the day, HSA, I don't think they would stretch themselves to the point of building x86 emulation. And given the reputation they have on the market, nobody would give them a chance. And why would you want big.LITTLE outside a battery constrained environment?

I think Zen will be "boring", plain vanilla beefed up cat CMP cores and that's it.

Depending on your point of view, AMD hasn't released a "big core" since the Phenom II days. CMT is a cluster**** considering the state of multithreaded software, and each integer core doesn't compete with Stars unless it's ramped up to high clocks. A Stars core with AVX2 on 14nm would be a great start.

I'm of the vision that AMD market screwed up and called the CMT cluster a core and a CMT core a module, so Bulldozer was a big core design, just an atrocious one.

 

[cbn]

Some more info from Anandtech:

http://www.anandtech.com/show/8968/star-swarm-directx-12-amd-apu-performance

So although DirectX 12 is not going to bring the same kind of massive performance improvements to iGPUs that we’ve seen with dGPUs

Going by the info from this the article, it appears AMD quad core CPUs could close the gap with Intel Core i3 provided the image quality settings are dialed up (via larger dGPUs). At lower detail settings (expected with large iGPUs or small dGPUs ) the gap between Intel Ciore i3 and AMD quad core would remain close to the same.

So based on that, I think this further supports AMD shipping future big core APUs with much smaller iGPUs being a good idea. (This for the lower cost to manufacture and thus lower potential retail price to us as consumers).

 

[Blitzvogel]

What we say here in this forums is that we want moar cores, but what we don't say is that we want moar cores in a given SKUs but clocked as high as in SKUs with less cores. And that's why moar cores might work on servers but certainly won't work on consumer market at any level.

Don't worry, I always treat moar cores as purely a joke, accept in GPUs, where it makes sense since the software and hardware economics allow it to work pretty efficiently.

It's like using 8 jet engines of 5,000 lbs thrust (Bulldozer) each versus 4 of 10,000 lbs thrust (Sandy Bridge and beyond). There are trade offs of going with a bunch of weak engines versus a couple very powerful ones, of which usually the favored solution is fewer more powerful engines when production, economics, and general application suitability are concerned until you aim way too high and hit a practical design limit when things get scaled up.

Anytime I can make an analogy with aircraft and computers, I take advantage of it :awe:

 

[DrMrLordX]

Those super slim desktop AIOs used to be powered with 35w C2D/i3 mobile cpus just fine, so 25w isn't unreasonable. My laptop has a 25w p8700 and a 20w GPU and it doesn't really get that warm while gaming. Why a 25w APU couldn't be in an AIO is beyond ridiculous. It can.

Don't get me wrong, I'd love to see the average ~$400 AiO selling at Home Depot/Staples packing an i3 mobile or even a 5350.

Or, a few months down the line, a 35W Carrizo. That would be awesome.

I just don't see it happening in a product like the HP 18-5110.

At least that product has dropped to $340 on the HP store and $360 at the Office Depot site. It may be a sign that they're trying to clear out that crap for something a bit newer. But I have used an 18-5110, I'll probably be stuck using one again sometime in the future, and I can tell you that whatever thermal solution they've chosen for that particular AiO chassis is not impressive.

I've also done time on a Gateway One. It's a tad better (E-350 vs E1-2500) but the thermals are correspondingly worse. Gateway refuses to acknowledge the existence of such a thing on their us.gateway.come site which is funny. Now the only one which shows up is an A4-based model which is a step up in price and performance from an E-350-based POS. So clearly someone, somewhere, at some price point is willing to put something decent in an AiO chassis thesedays using AMD chips, albeit at an ~$130 premium over where the crappy dual-core cat units are sitting right now, or an $80 premium over where the same machines were about 6-8 months ago.

If your average consumer that wants to buy a bargain desktop box walks into Best Buy, Staples, or wherever and finds AMD machines equipped with 35W Carrizo quads with full-on shader complement and the whole shebang at ~$400 then I say hell yes, do it HP/Gateway/Asus/whoeverelseislistening. But they will be looking at a better chassis design/cooling solution than what is in the 18-5110 (and similar) which probably means lower margins for the OEM just on that account.

I don't think Zen will be remotely aimed at high end users at all. It is built on a process highly tuned for SoCs and whatever the scope of their new CPU design it must be low cost enough for semi-custom. This level of performance simply isn't compatible with semi-custom, as it must need attractive costs in the first place. Of course they can throw more hardware at the problem in the form of moar cores, but this approach isn't working for them.

It's an 8C/8M chip, though. Who else but server/workstation crowds and eccentrics (read: oddballs that liked FX chips or LGA2011 users) want that kind of thread parallelism? It's a 95W chip specced to handle 16 threads, for crying out loud. That's getting into SPARC T1/T2 territory of ludicrosity.

I think Zen performance per core won't be much distant from the cat cores.

If AMD pulls a Bulldozer-level of fail then yeah maybe so. I'm thinking/hoping that, clock per clock, it will at least be on par with Excavator, and will just suffer from low-ish clockspeeds to start. See I don't think that 14nm process mandates poor IPC, it just screams low clockspeeds.

I don't think we'll see IPC on the order of Skylake either, but hey it's got 8C/8M for a reason.

Some more info from Anandtech:

Going by the info from this the article, it appears AMD quad core CPUs could close the gap with Intel Core i3 provided the image quality settings are dialed up (via larger dGPUs). At lower detail settings (expected with large iGPUs or small dGPUs ) the gap between Intel Ciore i3 and AMD quad core would remain close to the same.

Where'd you get that idea? First of all, this is an early preview of one possible use of the DX12 API. There is no indication of what, if anything, this particular piece of software has to do with mature DX12 software titles.

Secondly, in low quality mode (which is the mode where you seem to think the A10/A8 receive the least improvement), you're seeing a ~%58 increase in performance for both the A10 and A8 chips. The Intel chip experiences a boost of "only" ~24%. The improvements for the AMD APUs may not be as impressive at lower detail levels where the software is more CPU-limited, but it still heavily favors the AMD chips in terms of overall improvement. The gap does not, as you claim, stay the same. AMD gains across the board.

So based on that, I think this further supports AMD shipping future big core APUs with much smaller iGPUs being a good idea. (This for the lower cost to manufacture and thus lower potential retail price to us as consumers).

According to the data you linked in your post, AMD seems to be doing just fine with what they've got already. Here's another tidbit: does that bit of software do anything with Intel's or AMD's iGPUs as far as compute goes? I'm guessing no. Guess what happens once people start utilizing that, quite possibly through the DX12 API? You can bet your custard that Intel will prod MS into making sure DX12 is thusly-capable to make sure that Broadwell and Skylake perform at their best. They need product differentiation to get Haswell owners on board with the next generation(s) of CPU. AMD's already there, just waiting for software like that to go mainstream.

 

[cbn]

The improvements for the AMD APUs may not be as impressive at lower detail levels where the software is more CPU-limited, but it still heavily favors the AMD chips in terms of overall improvement.

Yes, DX12 favors AMD because the CMT processors have ~better MT, but weaker ST than Core i3 (and DX12 helps spread the load out to better make use of AMD's MT advantage).

However, the greater the graphics load, the greater AMD benefits compared to Intel.

So if mature game titles mirror these results to any degree, we have to think about what is a better value? Reduce iGPU size and make the big core processors more affordable? This leaves room in the budget for greater dGPU (and the potential to further maximize AMD's benefit from DX12)

Besides that, I just don't believe the average consumer wants the big iGPUs. What I'll bet is in greater demand is a good CPU with modest iGPU at a lower price.

 

[cbn]

Here's another tidbit: does that bit of software do anything with Intel's or AMD's iGPUs as far as compute goes? I'm guessing no. Guess what happens once people start utilizing that, quite possibly through the DX12 API? You can bet your custard that Intel will prod MS into making sure DX12 is thusly-capable to make sure that Broadwell and Skylake perform at their best. They need product differentiation to get Haswell owners on board with the next generation(s) of CPU. AMD's already there, just waiting for software like that to go mainstream.

I do think that is interesting, but until that day comes I think AMD should back off adding so much iGPU (re: It is just too costly for them to use 512sp iGPU relative to what they could do with less iGPU).

Basically, this iGPU size issue appears to me to be one of opportunity cost.

 

[cbn]

Secondly, in low quality mode (which is the mode where you seem to think the A10/A8 receive the least improvement), you're seeing a ~%58 increase in performance for both the A10 and A8 chips. The Intel chip experiences a boost of "only" ~24%.

But the Core i3 is still 40.5% faster overall in low quality mode. That is a pretty large gap.

This compared to the Medium quality and Extreme quality where the Core i3 was less than 9% and 4% faster respectively.

 

[mrmt]

It's an 8C/8M chip, though. Who else but server/workstation crowds and eccentrics (read: oddballs that liked FX chips or LGA2011 users) want that kind of thread parallelism? It's a 95W chip specced to handle 16 threads, for crying out loud. That's getting into SPARC T1/T2 territory of ludicrosity.

If Zen is indeed 8C on the consumer market that means AMD is again trying to compensate with more cores the weak performance of each individual core, but on the overall landscape of the server market, things get much bleaker.

Assuming that AMD is indeed going with SMT, what's the point of 95W 16 thread chips for servers when Intel is fielding either 36 threads chips or 16 highly clocked threads? That's right, there's no point, Zen is DOA by all standards on servers. This is against Intel 22nm line, once Xeon 14nm arrives the comparisons will get much worse.

 

[cbn]

If your average consumer that wants to buy a bargain desktop box walks into Best Buy, Staples, or wherever and finds AMD machines equipped with 35W Carrizo quads with full-on shader complement and the whole shebang at ~$400 then I say hell yes, do it HP/Gateway/Asus/whoeverelseislistening. But they will be looking at a better chassis design/cooling solution than what is in the 18-5110 (and similar) which probably means lower margins for the OEM just on that account.

21.5" chassis appears to be sweet spot for 35 watt chips.

Maybe if a person wanted to put a 35 watt chip inside a 18.5" chassis a fancier cooler would be required? Not sure, but I do know a 35 watt 245mm2 Carrizo will be more expensive than a 101mm2 cat core chip.

With that mentioned, I can't imagine why too many folks would want 18.5" over 21.5"? Isn't 21.5" pretty much the most common commodity panel size for desktop monitors?

EDIT: According to Newegg --> http://www.newegg.com/Product/Produc...bop=And&Page=1, the smaller 1366 x 768 panels (as found in HP 18-5110) are indeed cheaper than the 21.5" 1080p but the difference is not great.