AMD CPU speculation... and expert conjecture

[truegenius]

1.) GCN 1.1 can already run x86 SSE4 instruction sets autonomously for GPGPU applications. Those are still not sold as stand alone Scalar processors...they make great vector processors, and can perform scalar operations that are massively parallel (see cryptocoin mining), though they are not good at micro scale scalar operations like a CPU.
-my 2 cents

hurrey, i am only 182 weeks behind in bitcoin , i only have to wait for mere 3.5 years to start mining

forever alone

anyone's 2 cents here on this issue ?

 

[juanrga]

@juan, my challenge to you is to prove that Phi going from AIB to LGA is an APU.

I don't know how to break it to you, LGA Phi is just a Phi chip without an additional CPU. Meaning it's an aborted GPU (Larrabee) and only a GPU in a socket. It's connected via QPI.

This is exactly what I have been saying will happen. APU is just a stop-gap measure. The real solutions will be dGPU (or aborted dGPU like Larrabee successors) with dCPUs or even APUs.

You're not going to see a Xeon Phi as a standalone product. It's a coprocessor. If it were to be an APU, it would not be a coprocessor but the main processor.

This is exactly what I predicted would happen and it just seems like Intel announced it happening sooner than AMD. I said that there are situations where a simple APU wouldn't have enough power. APUs are limited to mid-range solutions because of TDP and die size restraints. A big dCPU and a big dGPU will never, ever fit on an APU. But please don't misunderstand what I'm getting at. Yes at some point you can fit Tahiti class GPU on an APU, but at that time it'd be possible to make another > 350mm^2 GPU, which would destroy the APU.

No. Pay attention to the slide that I gave above

The current discrete card is a "coprocessor" that requires of a CPU (e.g. ordinary Xeon) to run the OS and to feed the data. The future socketed version is a "standalone CPU" that runs both the OS and the compute. I recall again that Intel uses the term CPU to refer to APUs.

The rest of your post was rebutted before and several times.

KL is supposedly Silvermont cores with AVX 3.1 (512bit) extensions. So it's not an APU. It doesn't have a graphical element like the shaders of AMD/NVidia GCN/CUDA cores.

KL utilizes modified Silvermont cores for throughput.

The general definition of APU doesn't requires a GPU. In HSA parlance an APU is TCU+LCU on same die. The TCU can be a GPU, can be a DSP, can be some other throughput accelerator or a combination of them

http://en.wikipedia.org/wiki/Accelerated_processing_unit

The first public use of the term APU

http://scalability.org/?p=90

 

[juanrga]

juan is right and wrong at the same time, yes APU will become the more powerful processing unit. and more cost efficient both in power and production.

yes the apu's will be taking over, AMD will shift all production to an apu, as they get smaller they will soon add cores to both cpu and gpu. and as on kaveri they will work more closely together on same processes, which will increase processing done. fx will not last more than 3-4 more years before they have the apu's just annihilating the separated components.

the only thing i see them continuing is dgpu for gaming, where you will need the dedicated processing power.
the GPU/PPU in kaveri alone is capable of i think i read 3 Tbytes of processing power and nowhere near being utilized fully for what it can do. yeah it's not the greatest for gaming but is a much more powerful processor than being let on at the moment. the software capable of being used on them hasn't hit.

the change is coming guys, and it's going to be a whole new world.

We agree on everything except on dGPU for gaming. You make a very good point, but this is what I think. The principle of locality identified on computing at exascale level also applies to gaming, even if you use the dGPU only for graphics, because the identified 10x barrier of moving data at exascale level remains.

Second, dGPUs for gaming are not designed in isolation. Both AMD and Nvidia reuse designs aimed at compute. Nvidia GeForce cards for gaming are derived from Tesla cards for HPC. Once Nvidia and AMD stop designing and selling cards for compute (both Nvidia and AMD agree that APUs will be much faster), the costs of the new design will be transferred entirely to the gaming cards which, in my opinion, will kill the market. The cards will be so expensive that nobody will purchase them.

This is similar to AMD selling FX Piledriver when the cost of the design was split between Opterons and FX (with most part of the cost being included in the more expensive Opterons). Once AMD stopped developing Steamroller Opterons it also stopped developing Steamroller FX.

The people who is signing the petition to AMD to develop FX Steamroller seems unaware of these questions. Even if AMD did accept the petition and could design a new FX chip in less than one year (which is unlikely) all the cost of the design would be transfered to the FX Steamroller chips, which couldn't be sold so cheap as the Piledriver FX were.

Third, dGPUs for gaming are also attacked by the low end side. Current APUs are killing the low-end dGPUs. When APUs achieve enough performance for most people, dGPUs for gaming will be killed because the enthusiast market will be so small that the development cost of anew design will be prohibitive. I already gave before the link with Intel plans to kill gaming dGPUs for 2015 or so.

It's a good summation Juanrga but I think people here are more concerned (and rightfully so) with what they can buy today or tomorrow for a desktop PC. Those future ultra high end APUs are going to continue to cost a fortune for the next 3-4 years. Those Xeon Phi CPUs are in the $3500-$5000 range. They will be cool but geared towards HPC not gaming.

People want to know what they'll be buying for BF5/BF6/BF7. If they're running a 290x now they'll still likely be buying a discrete for the next 2-4 years.

There's no magic here when it comes to APUs. It's all about transistor counts, density, yields and process nodes. TSMC/GF are just now getting their teeth wet with FinFET. Intel stumbled a bit at the start on their FinFET roll out. They're just now getting the yields high enough to make their massive 15 core server parts with gobs of cache.

We know what NVidia can do with a 7B transistor (Titan), and we know what AMD can do with a 2.4B transistor APU (Kaveri). Neither is going to change drastically without a major die shrink or much higher yields. The 16nm/14nm nodes should make for some interesting parts but how soon those will make it into consumer grade high end APU parts depends on the yields. I've heard rates as low as 10-20% for GF and 20-30% for TSMC for their FinFETs.

The point is that those future ultra high end APUs will be cheaper than a hypothetical CPU+dGPU with same performance. Not even the HPC community (which pays for superexpensive supercomputers) could afford the costs of a hypothetical CPU+dGPU with same performance. This is why AMD, Nvidia, and Intel are designing APUs for those supercomputers. Check the link that I gave above where AMD reports its plan to use a 10TFLOP APU to build the fastest supercomputer.

There is no plans to build it using a traditional CPU + GPU architecture because it is impossible, that is why I call them "hypothetical". Nobody will build them.

I agree that people will continue purchasing GPUs before they are killed. Intel claims will kill them by year 2015 or so. I think 2018 is more realistic.

Your point about transistors was replied before. It is not magic, it is what the laws of physics say. I already explained what the laws say and why the discrete card will be slower. That is why Intel, is migrating from discrete card to socket because the socket version will be offer superior performance. Intel admitted it at SC13.

I also mentioned that there is consensus among all the scientists/engineers on which is the magnitude of the performance wall at exascale level: 10x. That is why Intel, AMD and Nvidia are developing the fastest systems (1000x faster than current supercomputers) around APUs. A hypothetical 1000W CPU+dGPU would be slower than the 300W APU.

I will not spend more time on this discussion, because it is being overly-repetitive. If people here continue believing that a dGPU will be always faster,* they will be shocked about 2018 when Nvidia and AMD will release their future APUs and none dGPU (I repeat they are not designing any dGPU for that level of performance).

* This is similar to when discussing with someone who only studied Newtonian physics and believes that by adding energy an object in motion will be always faster. Newtonian laws are only an approximation to more general laws of physics and those more general laws predict a c-wall. At ordinary performance level of current desktops a high-end dGPU is fastest than an APU. At exascale level the APU is fastest than any dGPU due to the existence of a performance wall.

 

[jdwii]

Well i guess based on secret documents and Intel's word dGPU will be dead soon, who would of guessed that.. I guess my mind can't comprehend how a single piece of silicon will be able to overcome the performance of 2 and how heat and space issues will be the thing of the past, will probably have a 20 Billion transistor APU(with arm of course) built on 5nm in 2020 and will be mass produced.

Amazing a company that could never create anything competitive at all in the GPU market will eliminate the dGPU market, i guess anything's possible and if intel says so i guess it must be true.

 

[jdwii]

Amd llano iGPU built on 32nm, IGPU performance around 5550 A8 3850 cost $140 when it came out

Amd Trinity/Richland built on 32nm, IGPU performance around 6570 cost $140 when it came out

Amd Kaveri Built on 28nm, IGPU performance around 20% better compared to the 6800K around 7750 DDR3 edition performance maybe little less. cost 185$ when it came out
For every release you could always build a better machine with a separate CPU and GPU if anything more so this time around with the A10 7850K

2014 TSMC begins 20nm production , globalfoundries claims 14nm FinFET in the next 2 quarters not sure if they will allow for 1+Billion transistors might be made for only for mobile

 

[Cazalan]

KL is supposedly Silvermont cores with AVX 3.1 (512bit) extensions. So it's not an APU. It doesn't have a graphical element like the shaders of AMD/NVidia GCN/CUDA cores.

KL utilizes modified Silvermont cores for throughput.

The general definition of APU doesn't requires a GPU. In HSA parlance an APU is TCU+LCU on same die. The TCU can be a GPU, can be a DSP, can be some other throughput accelerator or a combination of them

http://en.wikipedia.org/wiki/Accelerated_processing_unit

The first public use of the term APU

http://scalability.org/?p=90

KL is a CPU. There are no accelerators. There is no TCU/LCU like pair. The CPU cores simply execute 512bit x86 vector instructions (AVX3). We don't call FX-8350s APUs because they support AVX or SSE4 do we?

 

[juggernautxtr]

but at the cost of wattage is the point Jdwii, they are going for effinciency and the separated components use more power than the combined. 7770 still uses 85+ watts + 100 watt cpu(750k)=180+ watts, the combined components use 100 watts-.

The evolution of gpu for compute on the gpu side has begun as never before. you will see that they are going to use them more and more for compute cycles than just gaming. Adobe is already writing to the gpu program side as they will achieve better performance from the gpu with their programs.

as for dgpu disappearing, highly and very unlikely, I don't think they will ever be able to get high end gaming from the the apu's, the games will get all the more complicated and the apu graphical side will only become an assistant.
because it will also be busy helping in the compute to feed a Dgpu.

 

[Cazalan]

I will not spend more time on this discussion, because it is being overly-repetitive. If people here continue believing that a dGPU will be always faster,* they will be shocked about 2018 when Nvidia and AMD will release their future APUs and none dGPU (I repeat they are not designing any dGPU for that level of performance).

So they will switch from dGPU to dAPU. The same paradigm will hold where you can plug in an AIB (add-in board) to increase your graphics and compute performance. You're making mountains out of mole hills.

 

[esrever]

I am thinking more along the lines of multiple apu sockets. A pair of 7850k's would be a 8 core gaming beast. I also think it would be of great benefit to any anyone already accustomed to buying 4-6 socket motherboards to scale performance. Especially going forward as more emphasis is place on gpu workloads.

Thats great and all but no software is going to take advantage of it. Also the cross APU communication would be a very expensive technology if they want to have good bandwidth and low latency.

 

[juanrga]

KL is supposedly Silvermont cores with AVX 3.1 (512bit) extensions. So it's not an APU. It doesn't have a graphical element like the shaders of AMD/NVidia GCN/CUDA cores.

KL utilizes modified Silvermont cores for throughput.

The general definition of APU doesn't requires a GPU. In HSA parlance an APU is TCU+LCU on same die. The TCU can be a GPU, can be a DSP, can be some other throughput accelerator or a combination of them

http://en.wikipedia.org/wiki/Accelerated_processing_unit

The first public use of the term APU

http://scalability.org/?p=90

KL is a CPU. There are no accelerators. There is no TCU/LCU like pair. The CPU cores simply execute 512bit x86 vector instructions (AVX3). We don't call FX-8350s APUs because they support AVX or SSE4 do we?

Intel also calls its i7-4770k a CPU when it is not. it is an APU (CPU+GPU).

The existent Phi is a TCU and requires a LCU (e.g. a Xeon CPU) to work. The existent Phi doesn't use CPU cores, but custom x86 cores modified for throughput. The KL Phi also uses custom x86 cores, derived from Silvermont, but substantially modified for throughput.

Executing the OS in a TCU is far from optimal. Thus the KL Phi has to have a LCU inside if it is designed to work on standalone mode.* How Intel exactly achieves this is not still known, but I expect something similar to a big.LITTLE configuration.

* David Kanter disagrees. He claims that the KL Phi will be a TCU and that the OS will have to be run on a separate LCU (e.g. an ordinary Xeon CPU) for efficiency. However Intel is emphasizing that the new KL Phi will work on standalone mode without any need of Xeon CPUs. It is worth mentioning that Kanter's initial speculation about the KL was incorrect

Intel’s small cores, such as the Silvermont core represent another option that seems reasonable, however a deeper analysis shows a fundamental mismatch.

Contrary to his former analysis, the cores in the KL Phi are derived from Silvermont.

but at the cost of wattage is the point Jdwii, they are going for effinciency and the separated components use more power than the combined. 7770 still uses 85+ watts + 100 watt cpu(750k)=180+ watts, the combined components use 100 watts-.

In fact I already mentioned that there is a strong consensus among the experts in the field about the efficiency of APUs at exascale level: 10x more performance than a dGPU with same power consumption.

 

[juanrga]

I am thinking more along the lines of multiple apu sockets. A pair of 7850k's would be a 8 core gaming beast. I also think it would be of great benefit to any anyone already accustomed to buying 4-6 socket motherboards to scale performance. Especially going forward as more emphasis is place on gpu workloads.

A multi-die package of two 7850k looks better than dual-socket.

I think we will see an evolution towards single-socket motherboards even in the high-end.

 

[juanrga]

I claimed here that Kaveri was being made on bulk. Some people disagreed and claimed that Kaveri was being made on SOI.

Then I claimed that AMD was moving from 28nm bulk to 20nm bulk and then to FINFETs on bulk. Again some people disagreed.

Then I claimed that SOI was essentially dead and that the future was FINFET on bulk. I received extensive replies about how SOI was very superior to bulk... but once again my predictions are confirmed.

SOI (both FDSOI and FINFET on SOI) will represent about a 5% of production in next 14/10/7 nm nodes, whereas FINFET on bulk will represent the remaining 95%.

https://www.semiwiki.com/forum/content/3128-soi-future-flop.html

Also IBM (the inventor of SOI) is abandoning its fabs

http://www.extremetech.com/gaming/176286-ibm-is-looking-at-leaving-the-semiconductor-manufacturing-industry

AMD did the right move when migrated from SOI to bulk.

 

[palladin9479]

Amd llano iGPU built on 32nm, IGPU performance around 5550 A8 3850 cost $140 when it came out

Amd Trinity/Richland built on 32nm, IGPU performance around 6570 cost $140 when it came out

Amd Kaveri Built on 28nm, IGPU performance around 20% better compared to the 6800K around 7750 DDR3 edition performance maybe little less. cost 185$ when it came out
For every release you could always build a better machine with a separate CPU and GPU if anything more so this time around with the A10 7850K

2014 TSMC begins 20nm production , globalfoundries claims 14nm FinFET in the next 2 quarters not sure if they will allow for 1+Billion transistors might be made for only for mobile

As someone who's had the first two, I can say with certainty that you are wrong. Llano's weren't the best but had their uses, Trinity / Richland is definitely worth it. The problem is your taking the most expensive of each and grading it as an enthusiast platform which they aren't. It is nearly impossible to find a comparable dGPU + dCPU (era equitable) platform at $120~$140 USD. I know because I've tried, and tried often. One of my hobbies involves building compact multi-purpose systems and in that arena APU's dominate. You have to reach $160 USD for the dGPU + dCPU to start taking over and even then only when you have a larger footprint and more power to work with.

Short Version: Stop assuming the highest marked product sets the value for all subsequent products.

 

[jdwii]

^ During that time you could build a Tri core athlon machine and a 6570HD video card(both of which were 70$), This is around Llano times.

 

[palladin9479]

^ During that time you could build a Tri core athlon machine and a 6570HD video card(both of which were 70$), This is around Llano times.

DT Llano would of been better. Llano was a 32nm quadcore enhanced Phenom II (K10.5) which was superior to the 45nm Phenom II's and their cheaper Athlons. Also the cheap 6570 HD you speak of uses the same memory as the Llano (1GB DD3) and is 40nm. Funny thing is you would of gotten the same performance as the Turks Pro GPU was, quite literally, bolted onto the K10.5 die, all at 32nm.

Your probably thinking the laptop Llano's which had issues due to crappy OEM support and only using DDR3-1333 memory. The later ones were pretty good (3530/3550MX) with DDR3-1600 though.

A8-3870K (all others are die harvests of this)
32nm 3.0 Ghz 4 cores
1MB L2 cache (vs 512kb on the older 45nm Phenom II / Athlon chips) 400:20:8 GPU core config, 600Mhz GPU clock, 128-bit DDR3-1866 max memory support. 100W.

Other versions were 65W with a reduced clock speed and the A6's had a reduced GPU config of 320:16:8 @443Mhz.

 

[jdwii]

Keep in mind i feel like all APU's are great for performance per watt but for performance per dollar its a loser. Again you could get more FPS with a 6570 and a athlon X3 back in the llano days compared to a A8...I'm an Amd fanboy and i even state this as fact. Its really not hard to look at data and the real numbers and then come back with a statement, that's is the main reason why i love this industry there is no such thing as yes and no or maybe its always a Yes or No. A is faster than B based on this data....

 

[palladin9479]

Keep in mind i feel like all APU's are great for performance per watt but for performance per dollar its a loser. Again you could get more FPS with a 6570 and a athlon X3 back in the llano days compared to a A8...I'm an Amd fanboy and i even state this as fact. Its really not hard to look at data and the real numbers and then come back with a statement, that's is the main reason why i love this industry there is no such thing as yes and no or maybe its always a Yes or No. A is faster than B based on this data....

Except Llano DT was a faster CPU then the older Athlon's. 32nm, quadcore and more L2 cache. The Athlon's didn't even have L3 cache so not even that argument is valid.

I'm really wanting to know what models your comparing? The highest clocked Athlon was 3.4Ghz while the highest clocked Llano was 3.0Ghz with the Llano being stupid easy to OC. It would be one thing if you were comparing them against something like a Phenom II X4 960, but against an Athlon X3, not a chance.

Maybe your having a problem remembering what the specs where of the CPU's of that era. Four core, 4MB L2, 32nm, 3.0hz vs Three Core, 1.5MB L2, 45nm.

That's not even a competition.

-=Edit=-

Ok digging for historical prices.
Rana 45nm Athlon II
Athlon II X3 460
3.4Ghz 1.5MB L2 Cache, three core, $87 USD (not 70).

So maybe you were thinking a 450 or 440? Those are 3.2 and 3.0 Ghz respectively, the same as Llano.

Anyhow as someone who actually builds these things, I can testify that APU's are excellent price for performance as long as you stay under $140~150 USD (budget) or going SFF. The A10-7850K is just stupid expensive right now, I'm waiting for the A8-7600 since it's a significantly better CPU.

 

[jdwii]

^^^ Most games only use 3 cores and back in that day it was even more true the extra money being spent on the GPU allowed for higher FPS on games on average


Edit
Also i'm basing my pricing on what i could get at the time on newegg when i had to built 10+ systems per month. I wish they were cheaper per FPS but that was not the case based on benchmarks from tomshardware and anandtech

 

[jdwii]

^ also are you stating i never built these things i already built a Amd A10 7850K and its around 20% slower on average compared to a 7770HD and 750K II x4 when gaming at 1080P based on BF4, AC4, NFS rivals, and even RE5...and couple of other games. Both cost 450$ to make.

I also own a llano laptop and its amazing. I also own a A10 5800K and a A8 llano 3850. So yeah i know what they can do per dollar. And based on living in the united states and using newegg its still better from a FPS standpoint to use a dCPU+dGPU over a APU for gaming.

 

[palladin9479]

^^^ Most games only use 3 cores and back in that day it was even more true the extra money being spent on the GPU allowed for higher FPS on games on average

Go back and reread it. The APU was cheaper then the combo you specified. The Llano has better single threading performance then the Athlon II. This goes back to the kool-aid you were spouting a few pages back about the K10 uArch. Llano is K10.5, it's an enhanced Phenom II. It has better performance, per clock and per watt then the Stars uArch (K10), that makes it better then Athlon II in every metric.

I posted the relevant information, Llano beat out Athlon II X3 + Turks. Now if you want to use a full Phenom II then it gets a bit dicey due to the L3 and which application your using. Llano still had more L2 cache, which is far more important then L3, along with being on 32nm. AMD was very conservative with the clock settings on Llano due to the Turks iGPU on the side. Using K10stat you could get very creative with undervolting / overclocking Llano and it's clock speed would easily match Phenom II x4's.

I've stated, several times now, that the 7850K is way overpriced for it's performance. The A8-7600 is far superior value, $120~130 USD is the targeted release price which puts it well ahead of anything else you could build.

And based on living in the united states and using newegg its still better from a FPS standpoint to use a dCPU+dGPU over a APU for gaming.

That's because your only comparing the absolute best and wondering why it's a poor value. The old 3820 and 3650 were far better values then the 3870K. Though the 5800K was already priced really well at $120 USD and the 6800K is pushing the line at $140 USD.

I'm really wanting to know where the f*ck your getting these magical CPU's priced at $50 USD to bundle with a better dGPU. Decent dGPU's will cost $85~100 USD for the GDDR5 versions which leaves you very little room left for any sort of CPU. So unless you start going north of $160 USD there simple isn't budget space available, not to mention power and physical space. The old "expensive memory" argument never held water as era appropriate memory wasn't any more expensive (1600 for llano, 1866 for Trinity/Richland and 2133 for Kaveri).

 

[Master-flaw]

Micro-centers where it's at...running down there to build an HTPC for someone...
$130 for a 7850K/mobo combo? More than worth it...Best deal they have.

 

[palladin9479]

Micro-centers where it's at...running down there to build an HTPC for someone...
$130 for a 7850K/mobo combo? More than worth it...Best deal they have.

Yeah they offer crazy good prices for walk in's. $130 USD for the 7850K is insane though I worry about the quality of board their pairing it with.

Cardinal rule of building computers, never ever go cheap on PSU or motherboard.

:Edit:

Did some digging on newegg and my god have high speed RAM prices gone down. Currently looking at something like this,
MB: ASRock FM2A88X-ITX+ $89.99
Memory: 8GB GSkill Ares DDR3-2400 CL11 $76.99. The DDR3-2133 memory was CL10 and $74.99 .
APU: A8-7600 @$120~125 (MSRP is $119).

That's ~$287 for the core system and it's all PicoPSU compatible if someone wants to go SFF. I could go cheaper and use a $50 USD board, but I won't do that as those boards tend to be poor quality and suck horribly. Put a nocture or other 65W LP cooler on it and put it inside a M350 for a really small yet powerful computer.

 

[ColinAP]

I claimed here that Kaveri was being made on bulk. Some people disagreed and claimed that Kaveri was being made on SOI.

Then I claimed that AMD was moving from 28nm bulk to 20nm bulk and then to FINFETs on bulk. Again some people disagreed.

Then I claimed that SOI was essentially dead and that the future was FINFET on bulk. I received extensive replies about how SOI was very superior to bulk... but once again my predictions are confirmed.

SOI (both FDSOI and FINFET on SOI) will represent about a 5% of production in next 14/10/7 nm nodes, whereas FINFET on bulk will represent the remaining 95%.

https://www.semiwiki.com/forum/content/3128-soi-future-flop.html

Also IBM (the inventor of SOI) is abandoning its fabs

http://www.extremetech.com/gaming/176286-ibm-is-looking-at-leaving-the-semiconductor-manufacturing-industry

AMD did the right move when migrated from SOI to bulk.

All of these things were available publicly on roadmaps from GF or AMD, you didn't "predict" any of them. Being "right" about all these things just means you've done the sensible thing and believed what the companies (who aren't allowed to lie to their shareholders) said, while the ones who disagreed have fallen foul of thier wishful thinking. But it doesn't make you some kind of technology Nostradamus.

 

[juanrga]

No. My claim about bulk Kaveri wasn't on any roadmap. At contrary, some people took an public GF roadmap and said me that Kaveri was SOI

Some people also linked to this article to prove me wrong

http://www.pcper.com/reviews/Editorial/GLOBALFOUNDRIES-2013-and-Beyond

I recall that after official presentation of Kaveri some people here was still pretending that Kaveri was SOI because one slide from AMD mentioned "SHP".

Same about my claim on 20nm bulk. No roadmap or anything as that. At contrary, some people here used an article that suggested AMD was returning to SOI for 20nm

http://news.softpedia.com/news/AMD-and-GlobalFoundries-Interested-in-FD-SOI-for-the-20-nm-Process-267519.shtml

Idem about my claim about FINFETs on bulk being the future and SOI being 'dead'. No roadmap or anything like that. At contrary, the replies that I received included a dozen of links to articles that supposedly disproved me. This was one of the articles mentioned

http://www.advancedsubstratenews.com/2013/07/globalfoundries-on-cost-vs-performance-for-fd-soi-bulk-and-finfet/

Thus, contrary to your claim, I didn't believe what companies said, but only what passed my expert-eye filter. I am rather sure you can find some intermediate label between "technology Nostradamus" and someone who is right from pure chance. But the label that you can find for me is irrelevant, the relevant concepts are "bulk" and "FINFET".

 

[juanrga]

A pair of recent Kaveri reviews

http://www.reviewstudio.net/1331-amd-kaveri-review-a10-7850k-general-performance

http://www.eurogamer.net/articles/digitalfoundry-2014-amd-a8-7600-kaveri-review