[Sweclockers] AMD Zen coming in Q3 2016, will be on 14 nm

[cbn]

I never paid attention to Intel's base clock since on my desktop I always had sufficient cooling. For my next laptop upgrade, I will pay extra $100-200 to get a chip with a much higher base clock. The CPU temperature does not exceed 94C, which is way below 105C maximum the 3635QM is rated at.

What programs get you to 94C?

 

[cbn]

If you read Notebookcheck wrt max load data on even the best Intel laptops, when loading the CPU and GPU to 99% similtaneously, you will almost always get CPU or GPU throttling or both. For example, the 970 SLI Aorus X7 will for sure throttle the i7 under such a scenario. The few laptops that might cope better are 9-12 lbs bricks with 1.7-2.2 inch thickness and 1.5 hour or less battery life. I don't consider those products 'laptops'.

If the heavier 9 to 12 lb laptops cope better then I'm thinking the throttling we are seeing on the lighter laptops is cooling related. (Makes sense to me considering the OEMs would be more prone to compromise on a laptop cooler destined for a lighter weight laptop than a desktop cooler.)

With that mentioned, I just don't think the desktop APUs should be throttling as as much as they do.

 

[Abwx]

With that mentioned, I just don't think the desktop APUs should be throttling as as much as they do.

Not with Prime 95 and Furmark simultaneously, there s no simultaneous application + game that will stress the gear as much as thoses two power virus, often Prime 95 alone will exhaust most of the TDP.

 

[Idontcare]

However, I am still concerned about these two things:

1. The price enthusiasts (including myself) are willing to pay for large iGPUs.This especially when the processor appears to be largely price sensitive. (It just seems that the culture of enthusiast desktop is against large iGPU for the most part. Folks like having choice of not only the exact GPU they use (via video card), but also the ability to sell the GPU and upgrade it).

2. How much CPU throttling will exist in stock applications (during iGPU load) if the FM3 socket is only rated 95 watts and the processors come with 95 watt coolers?

Can't speak to #2, but #1 is a sales & marketing agents dream come true with 2.5D/3D chips because they can mix-and-match to create a veritable SKU soup of CPU and iGPU capabilities.

I know people here tend to view sales & marketing as some anonymous dastardly villain, but really they are supposed to figure out how best to position a product (binning, clockspeed, power, integrated features, etc) with the customer base. And giving them the option of tailoring SKUs all the more selectively for the spectrum of customers out there is just a win in my book.

Let the engineers develop the best CPU cores. Let them develop the best GPU cores. But don't task them with knowing what the customer wants or needs when it comes to the balance of CPU cores and GPU cores, or power consumption, or ASP.

Leave that to the folks who have the data from market research to know best what re-integrated SKUs to field in the market.

Is that including quantum effects?

Quantum effects are more related to the gate oxide thickness, and that more or less got cut way back with the transition to high-k and metal gates. It isn't gone, but it isn't the devil that it was made out to be in anticipation of SiOx gates scaling below 1nm thicknesses.

On the flip side, there are technologies that depend on tunneling, flash NAND being the most prominent in the consumer arena.

 

[Idontcare]

Not with Prime 95 and Furmark simultaneously, there s no simultaneous application + game that will stress the gear as much as thoses two power virus, often Prime 95 alone will exhaust most of the TDP.

Prime95 is not a power virus. It is a legitimate stand-alone application used to find new Mersenne Prime numbers.

If a SoC has troubles running Prime95 within its rated TDP then it has problems with being rated at that TDP.

Furmark on the other hand is definitely a synthetic power-virus program that bears no reflection on the practical uses of one's SoC. Anyone can short a circuit to ground and accomplish zero useful work in the process, furmark scores no points there.

 

[cbn]

Can't speak to #2, but #1 is a sales & marketing agents dream come true with 2.5D/3D chips because they can mix-and-match to create a veritable SKU soup of CPU and iGPU capabilities.

I know people here tend to view sales & marketing as some anonymous dastardly villain, but really they are supposed to figure out how best to position a product (binning, clockspeed, power, integrated features, etc) with the customer base. And giving them the option of tailoring SKUs all the more selectively for the spectrum of customers out there is just a win in my book.

Let the engineers develop the best CPU cores. Let them develop the best GPU cores. But don't task them with knowing what the customer wants or needs when it comes to the balance of CPU cores and GPU cores, or power consumption, or ASP

Yes, I think the interposer and being able to mix in different dies is pretty awesome myself.

As I mentioned, I just question how many people will want a large iGPU integrated via interposer on desktop. (Even if the TDP of the socket were increased tremendously for a really large GPU via interposer, I have to think most people would want their separate video cards. It's just how people do things and use their computers, and to break that would require the silicon seller to provide a very large discount IMO.)

For a laptop, on the other hand, I think having the choice of small, medium, large, very large GPU die via interposer would be ideal.

 

[elemein]

Yes, I think the interposer and being able to mix in different dies is pretty awesome myself.

As I mentioned, I just question how many people will want a large iGPU integrated via interposer on desktop. (Even if the TDP of the socket were increased tremendously for a really large GPU via interposer, I have to think most people would want their separate video cards. It's just how people do things and use their computers, and to break that would require the silicon seller to provide a very large discount IMO.)

For a laptop, on the other hand, I think having the choice of small, medium, large, very large GPU die via interposer would be ideal.

Hey there!

So er... Noob question but, what is an interposer? Googling doesnt seem to be turning up the greatest results 😱

 

[Abwx]

Prime95 is not a power virus. It is a legitimate stand-alone application used to find new Mersenne Prime numbers.

If a SoC has troubles running Prime95 within its rated TDP then it has problems with being rated at that TDP.

Furmark on the other hand is definitely a synthetic power-virus program that bears no reflection on the practical uses of one's SoC. Anyone can short a circuit to ground and accomplish zero useful work in the process, furmark scores no points there.

When i said power virus i didnt imply that it was doing nothing usefull but that it use more power than Linpack for instance, besides Furmark is not a power virus strictly speaking, it s a rendering of a picture that is a fur, hence the name, and there s a score that give the perf.

 

[cbn]

Hey there!

So er... Noob question but, what is an interposer? Googling doesnt seem to be turning up the greatest results 😱

Hi elemein,

"Interposer" was something IDC originally brought up in post #124 (along with 2.5D and TSV)

Basically this is a system that allows two or more silicon dies to communicate to each other. (see below for sample diagram):

http://www.pcper.com/category/tags/380x

There are two classes of true 3D chips which are being developed today. The first is known as 2½D where a so-called silicon interposer is created. The interposer does not contain any active transistors, only interconnect (and perhaps decoupling capacitors), thus avoiding the issue of threshold shift mentioned above. The chips are attached to the interposer by flipping them so that the active chips do not require any TSVs to be created. True 3D chips have TSVs going through active chips and, in the future, have potential to be stacked several die high (first for low-power memories where the heat and power distribution issues are less critical).

The discussion of these interposers, 2.5D chips, and TSVs (in this thread) has been used in the context of AMD and GF being able to create a specific CPU die and a specific GPU die(s) and then connecting via this interposer. This would greatly increase freedom to create numerous SKUs by allowing mixing an matching various GPU sizes with the CPU die. This, in contrast, to the current system where AMD and GF have only one relatively large and costly die with the cpu and fixed sized GPU (512 sp) integrated.

 

[cbn]

Ran Furmark stress test with Prime 95 on my Pentium G3258 (4.3 Ghz cpu overclock with all voltages set on auto. iGPU at stock speed. stock cooler.) for over 20 minutes.

CPU speed was rock solid 4.3 Ghz the entire time.

Now granted the stock cooler that comes with these G3258 processors is rated at 95 watts, but I still think the APUs mentioned in this thread are cpu throttling more than they should.

 

[cbn]

Furmark on the other hand is definitely a synthetic power-virus program that bears no reflection on the practical uses of one's SoC. Anyone can short a circuit to ground and accomplish zero useful work in the process, furmark scores no points there.

I've read graphics drivers are throttling Furmark though.

Example: http://www.anandtech.com/show/4008/nvidias-geforce-gtx-580/17

As for FurMark, due to NVIDIA’s power throttling we’ve had to get a bit creative. FurMark is throttled to the point where the GTX 580 registers 360W, thanks to a roughly 40% reduction in performance under FurMark. As a result for the GTX 580 we’ve swapped out FurMark for another program that generates a comparable load, Program X. At this point we’re going to decline to name the program, as should NVIDIA throttle it we may be hard pressed to determine if and when this happened.

 

[Enigmoid]

My 3635QM throttles to 2.35Ghz when running distributed computing on the CPU and the laptop dGPU at the same time. The GPU doesn't throttle when the CPU and GPU are 99% loaded simultaneously.

In fact, if I only run PrimeGrid/Seti@Home on the CPU, I am often in the 3.19-3.3Ghz range, but as soon as I add the GPU into the mix for other projects, my CPU speed on all 8 threads falls below the 2.4Ghz base clock.

I never paid attention to Intel's base clock since on my desktop I always had sufficient cooling. For my next laptop upgrade, I will pay extra $100-200 to get a chip with a much higher base clock. The CPU temperature does not exceed 94C, which is way below 105C maximum the 3635QM is rated at.

In practice that means for my laptop usage a 2.8Ghz Intel i7 with a 4.0Ghz Turbo would not be just 18% faster than a 2.2Ghz i7 that with a 3.4Ghz Turbo, but actually 27%!

The programs I run aren't even synthetic like Furmark/Prime95. If I tried those, it would be disastrous. I bet my 3635 would drop to 1.8-2Ghz.

If you read Notebookcheck wrt max load data on even the best Intel laptops, when loading the CPU and GPU to 99% similtaneously, you will almost always get CPU or GPU throttling or both. For example, the 970 SLI Aorus X7 will for sure throttle the i7 under such a scenario. The few laptops that might cope better are 9-12 lbs bricks with 1.7-2.2 inch thickness and 1.5 hour or less battery life. I don't consider those products 'laptops'.

My laptop 3630qm will run prime 95 at 3.2 ghz solid.

Running furmark on the dgpu (660m OC to 1085 mhz) results in a steady CPU speed of 2.4 ghz and gpu of 1085 mhz. Temps are around 80 degrees.

lenovo y580.

Honestly I think on my laptop its a power or bios limitation.

 

[elemein]

Hi elemein,

"Interposer" was something IDC originally brought up in post #124 (along with 2.5D and TSV)

Basically this is a system that allows two or more silicon dies to communicate to each other. (see below for sample diagram):

http://www.pcper.com/category/tags/380x

The discussion of these interposers, 2.5D chips, and TSVs (in this thread) has been used in the context of AMD and GF being able to create a specific CPU die and a specific GPU die(s) and then connecting via this interposer. This would greatly increase freedom to create numerous SKUs by allowing mixing an matching various GPU sizes with the CPU die. This, in contrast, to the current system where AMD and GF have only one relatively large and costly die with the cpu and fixed sized GPU (512 sp) integrated.

Why would an interposer be used instead of a point-to-point interconnect? Seems like an interposer would just waste more silicon and insulate heat. I can understand as a foundation for several chips, but if there isn't many chips why not just point to point?

 

[Idontcare]

Why would an interposer be used instead of a point-to-point interconnect? Seems like an interposer would just waste more silicon and insulate heat. I can understand as a foundation for several chips, but if there isn't many chips why not just point to point?

It is all about performance, cost, complexity, etc. And not just for the specific IC or package in question, rather for the entire supplier eco-system that is involved in creating a final product for the end-user.

Any argument you would make in support of creating a monolithic SOC, reducing the chipset count, in a product today is an argument you would make in support of creating a re-integrated SOC in a product come 4-5 years from now.

 

[AtenRa]

Ran Furmark stress test with Prime 95 on my Pentium G3258 (4.3 Ghz cpu overclock with all voltages set on auto. iGPU at stock speed. stock cooler.) for over 20 minutes.

CPU speed was rock solid 4.3 Ghz the entire time.

Now granted the stock cooler that comes with these G3258 processors is rated at 95 watts, but I still think the APUs mentioned in this thread are cpu throttling more than they should.

None of the APUs are throttling down with only Prime 95 running, you have to engage the iGPU at the same time in order for the CPU cores to throttle. We have said this 1000 times already, when you only use the CPU it doesnt throttle.

 

[Abwx]

Ran Furmark stress test with Prime 95 on my Pentium G3258 (4.3 Ghz cpu overclock with all voltages set on auto. iGPU at stock speed. stock cooler.) for over 20 minutes.

CPU speed was rock solid 4.3 Ghz the entire time.

.

Yes, it can do so, and for a reason, there s no AVX (and no AES) on thoses Pentium, at equal frequency it s about half the throughput of an i3 or of an AMD Kaveri, so that s really not a performance, if i can say so given the half throughput...

 

[Fjodor2001]

http://www.kitguru.net/components/c...skylake-technologies-new-512-bit-fpu-rumours/

http://wccftech.com/amd-zen-architecture-release-schedule-revealed-rolled-server-market/

WccfTech reports citing its own sources that with “Zen” AMD will return to its traditional practice of introducing server processors powered by the latest micro-architectures first and then follow with chips for client PCs. There are no exact schedules given, but taking into account relatively slow ramp up of server processors and platforms by server makers, if AMD wants to find its new Opteron chips in 2016 servers in more or less significant quantities, it will have to introduce its new CPUs in the first half of the year.

If that is true, when when can we expect Zen based client CPUs? If it's much later than the server CPUs, don't they risk losing competitiveness?

Also, it's interesting that Intel is doing the opposite and introducing new uArch in server CPUs after mobile/laptop (and to some extent desktop).

 

[witeken]

If it's much later than the server CPUs, don't they risk losing competitiveness?

How can you lose something that you don't have?

 

[Tuna-Fish]

If that is true, when when can we expect Zen based client CPUs? If it's much later than the server CPUs, don't they risk losing competitiveness?

They have no client competitiveness to lose. If they can make a CPU that can actually sell in the server market at a decent ASP, it's a no-brainer for them to sell every chip into servers until they saturate the market. They will simply make more money that way.

 

[Fjodor2001]

How can you lose something that you don't have?

Ok, so you already have benchmarks from Zen based CPUs? Otherwise how can you make such a claim... 🙄

 

[ShintaiDK]

AMDs server marketshare is close to nonexistant. And tis only going downwards whatever is left.

 

[Fjodor2001]

AMDs server marketshare is close to nonexistant. And tis only going downwards whatever is left.

And your point with that wild-ass-guess-with-no-evidence-to-back-it-up-statement was exactly what? 😵

Do you also have benchmarks of AMD's Zen 14 nm server CPUs?

 

[monstercameron]

And your point with that wild-ass-guess-with-no-evidence-to-back-it-up-statement was exactly what? 😵

Do you also have benchmarks of AMD's Zen 14 nm server CPUs?

You pay shintae attention when it comes to negative declarations? He also said that the consoles won't use apus and we know how that turned out.

 

[JDG1980]

If that is true, when when can we expect Zen based client CPUs? If it's much later than the server CPUs, don't they risk losing competitiveness?

Also, it's interesting that Intel is doing the opposite and introducing new uArch in server CPUs after mobile/laptop (and to some extent desktop).

Server CPUs are where the big profit margins are. If Zen is a competitive architecture, then focusing on regaining lost server market share first makes sense - especially if yields at 14nm are low at first. In my opinion, this is actually good news - it means that AMD is optimistic about Zen's performance.

 

[ShintaiDK]

You pay shintae attention when it comes to negative declarations? He also said that the consoles won't use apus and we know how that turned out.

Never any faith lost is there. Zen will fix everything for AMD. Not to mention K12 that is their prime focus unlike Zen.

So they will go from what, a couple of % to 25% when Zen is released? AMDs CPU revenue will what? Quadrouple in a year?

All made on a discount R&D budget that keep getting cut.

Then reality called...

Ever since K8, AMD have wanted a comeback in servers. Only to fail badly every time. At the expense of mobile and desktop as well. What they couldnt do the last ~10 years with much bigger R&D budgets will suddenly happen in 2016/2017?