New Zen microarchitecture details

[Abwx]

The only gripe I can see right now is that the cooler looks worse than Wraith that the ES Ryzen has (but on the other hand it was OCed so it probably had higher TDP than a retail chip would have).

6900K use Intel s stock cooler.

As for Ryzen how could it be overcloscked s since it will have higher base frequency than 3.4GHz..?.

And btw, its TDP is displayed by AMD in the demos, and we know that the chip is slightly overvolted, so that s yet another mistake of yours...

You realize that you are claiming AMD "overclocked" a chip that had no official specs on the event, right? Also AMD claimed that the lowest base for Ryzen at launch will be 3.4Ghz so that is a guaranteed clock.

Nothing is overclocked, and we know it, it s just the same usual suspects repeating ad nauseam what perhaps suit their agenda, i mean, whoever did read a little know that base frequencies will be higher than 3.4, so whoever come with the "overclocking argument" is just aknowledging that he s willfully trying to mislead the eventual unsuspecting readers...

 

[Crumpet]

i7-6900K machine they used for the other demo is shown in this video at around 2 minutes mark. Maybe somebody could identify what motherboard is used (some ROG line board, so quite expensive thingy for rich kids)? The only gripe I can see right now is that the cooler looks worse than Wraith that the ES Ryzen has (but on the other hand it was OCed so it probably had higher TDP than a retail chip would have).

That's from the gaming demo though, so in theory the Blender machine could have a worse motherboard. That line of thought is more in the conspiracy theory field though, IMHO.

That's the Asus ROG Strix X99

 

[sirmo]

There is really nothing shady about that Blender demo. AMD even let people checkout the computers in question and released the sample files for people to run independently. I really don't think there is any shenanigans here. These guys may cherry pick stuff or omit information to make a product look better, but there is nothing of importance being omitted here.

We should still wait for official benchmarks, but as far as vendor teased benchmarks go, this is about the best transparency I've seen from a vendor.

 

[Sven_eng]

Turbo is designed to LOWER CLOCKS for better efficiency and only allow higher clocks on a very limited number of cores and only if everything else (power draw TDP temps etc)allows for it.
Basically the test was done with ryzen being "overclocked" (fixed clocks at highest all core turbo) and the 6900k being throttled by turbo and an undisclosed motherboard which may or may not have had an even stricter TDP setting then normal.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

I think you are mistaken about what actual overclocking is. Instead of the "undisclosed motherboard which may or may not have had an even stricter TDP setting then normal." You should instead be looking at it as "an undisclosed motherboard which may not be overclocking the CPU by default, unlike those which members of the press used for broadwell E".

 

[bjt2]

bjt2, you are forgetting that there will be benchmarks/workloads that will be stressing a stronger points of intel design, namely higher FP throughput (FMA/AVX). In those benchmarks Zen will likely perform like IB core. For other workloads I also expect near BDW-E levels which will be tremendous achievement.

Zen can do (256 bit wise) 1 FADD + 1 FMUL or 1 FMAC
SKL can do (256 bit wise) 1 FADD+ 1 FMUL or 2 FMAC

In 128 bit code Zen has double throughput.

Only on 256 FMAC code INTEL has advantage. On 256 bit not FMAC code they are on par, but INTEL shares ports with int units and Zen not.
So they are superior in highly optimized FMAC code... What can be? BLAS? Ok, but that is RAM limited. Matlab has autoparallelization for vector code (built in in BLAS linpack or lapack libraries), but this auto parallelization is turned off in add, mul and other simple code, because it's RAM limited...
So only difficult calculations can benefit from 256 bit... They have only 2 FP pipes... AMD 4. And the DIV and SQRT are 128 bit only even on SKL... And AMD have four pipes in that case...

If you think of blender, for instance, a ray tracing algorithm for sure requires also divisions and SQRTs... In this case AMD will shine...

 

[bjt2]

Those tests only show Throughput, we dont know exactly ZENs IPC from 8C 16T tests, unless MT and SMT scaling is the same for both CPUs.

I define IPC of a whole core as the total number of instructions per clock, on all threads that run on that core. And many agree with me. When i want to stress IPC of a single thread, i will say ST IPC...

 

[imported_jjj]

What apps that require heavy lifting from the CPU, aren't scaling yet to many cores? There is LAME but that's not heavy lifting anymore and maybe some that use the GPU might not. Some games but if many cores become popular, games will start to use more cores too.
Don't quite get the focus on single core perf nowadays, even in reviews.

 

[PPB]

Market size is the critical parameter. 4c8t w/o gpu is a pretty small market nowadays.

That is a phallacy. There are no consumer oriented 4c/8t cpus without igp being sold right now to evaluate the TAM for a zen 4c/8t, either native die or not.

You only have e3 xeons or an old sb or ib bas3d i7 on 2011 which dont really count because first one is server or enterprise oriented, the second one is in a hedt platform that is so cost inefficient compared to your regular 115x i7 that it doesnt make sense. On the other hand a igpu less 4c8t from amd can be placed at e3 prices that the unlocked feature will by itself entice more people to buy it if performance is there compared to i7s on 115x


Sent from my XT1040 using Tapatalk

 

[Abwx]

In 128 bit code Zen has double throughput.


Only on 256 FMAC code INTEL has advantage. On 256 bit not FMAC code they are on par, but INTEL shares ports with int units and Zen not.

They have only 2 FP pipes... AMD 4. And the DIV and SQRT are 128 bit only even on SKL... And AMD have four pipes in that case...

If you think of blender, for instance, a ray tracing algorithm for sure requires also divisions and SQRTs... In this case AMD will shine...

http://pc.watch.impress.co.jp/img/pcw/docs/1039/035/ZEN-Architecture.png

http://pc.watch.impress.co.jp/img/pcw/docs/1039/035/Skylake Architecture Block Diagram.png

 

[Dresdenboy]

Turbo is designed to LOWER CLOCKS for better efficiency and only allow higher clocks on a very limited number of cores and only if everything else (power draw TDP temps etc)allows for it.
Basically the test was done with ryzen being "overclocked" (fixed clocks at highest all core turbo) and the 6900k being throttled by turbo and an undisclosed motherboard which may or may not have had an even stricter TDP setting then normal.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

You seem to mix power management for lower power consumption with turbo/boost on multi core MPUs exploiting thermal headroom for increased frequencies.

Very limited number of cores like 4 on DT quadcores, or 8 on i7-6900K?

 

[TheELF]

You seem to mix power management for lower power consumption with turbo/boost on multi core MPUs exploiting thermal headroom for increased frequencies.

Very limited number of cores like 4 on DT quadcores, or 8 on i7-6900K?

Read the link I gave from the anandtech review, intel isn't even saying anymore,all they say is "light workload" which could mean anything.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

For the other people commenting the 6900k has an turbo 2 clock of 3.7 which means that all it's cores are rated to be able to run at 3.7 ,with default settings and default turbo it will never run all it's cores at 3.7 if you change bios settings to make it run at 3.7 on all cores it's basically overclocking since you will gain performance.

Also the 6950x has 25Mb cache while the 6900k "only" has 20Mb and it has been debated that blender is very sensitive towards cache.

 

[Mechanical Man]

Turbo is designed to LOWER CLOCKS for better efficiency and only allow higher clocks on a very limited number of cores and only if everything else (power draw TDP temps etc)allows for it.
Basically the test was done with ryzen being "overclocked" (fixed clocks at highest all core turbo) and the 6900k being throttled by turbo and an undisclosed motherboard which may or may not have had an even stricter TDP setting then normal.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

I have understood that 6900k all core state is 3,5GHz. Also, there is also data about 3,6/3,9 Ryzen, so it was not definately "overclocked" anywhere.

 

[bjt2]

http://pc.watch.impress.co.jp/img/pcw/docs/1039/035/ZEN-Architecture.png

http://pc.watch.impress.co.jp/img/pcw/docs/1039/035/Skylake Architecture Block Diagram.png

I know SKL architecture. It can perform 2x256 FP and 3 vecint and on that graph it's even not clear if they are 256 bit. I was talking of FP. I don't care of vecint.

 

[bjt2]

Read the link I gave from the anandtech review, intel isn't even saying anymore,all they say is "light workload" which could mean anything.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

For the other people commenting the 6900k has an turbo 2 clock of 3.7 which means that all it's cores are rated to be able to run at 3.7 ,with default settings and default turbo it will never run all it's cores at 3.7 if you change bios settings to make it run at 3.7 on all cores it's basically overclocking since you will gain performance.

Also the 6950x has 25Mb cache while the 6900k "only" has 20Mb and it has been debated that blender is very sensitive towards cache.

6950 locked at 3.5GHz and 2 core disabled obtained exact blender score of the 6900k at new horizon...

 

[realibrad]

AMD has never said they will surpass Intel with Zen. They set a goal of 40% better IPC (which they now exceeded)

people at the start of this thread were happy with haswell performance.

Not sure what your point is. People were wondering if AMD might. Hype train gonna hype and AMD is not at fault. Personally I'm 99% sure that my next box will be a Zen system.

I'm assuming you believe that I feel it's a letdown but that would be wrong. I just feel it clears things up for some.

 

[realibrad]

Turbo is designed to LOWER CLOCKS for better efficiency and only allow higher clocks on a very limited number of cores and only if everything else (power draw TDP temps etc)allows for it.
Basically the test was done with ryzen being "overclocked" (fixed clocks at highest all core turbo) and the 6900k being throttled by turbo and an undisclosed motherboard which may or may not have had an even stricter TDP setting then normal.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

Read the link I gave from the anandtech review, intel isn't even saying anymore,all they say is "light workload" which could mean anything.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

For the other people commenting the 6900k has an turbo 2 clock of 3.7 which means that all it's cores are rated to be able to run at 3.7 ,with default settings and default turbo it will never run all it's cores at 3.7 if you change bios settings to make it run at 3.7 on all cores it's basically overclocking since you will gain performance.

Also the 6950x has 25Mb cache while the 6900k "only" has 20Mb and it has been debated that blender is very sensitive towards cache.

Did you read this part?

"Turbo Boost 2.0 is what Intel calls its maximum Turbo or ‘peak’ frequency. So in the case of the i7-6950X, the base frequency is 3.0 GHz and the Turbo Boost 2.0 frequency is 3.5 GHz. The CPU will use that frequency when light workloads are in play and decrease the frequency of the cores as the load increases in order to keep the power consumption more consistent. Turbo Boost 2.0 frequencies are advertised alongside the CPU on the box - TBM3 will be slightly different and not advertised.

TBM3, in a nutshell, will boost the frequency of a single CPU core when a single-threaded program is being used."

Each chip is rated to run a base clock, and a turbo clock. The base clock is a minimum performance when all cores are being used. The boost clock is a boost from the base if conditions allow for the chip to do more. That means, if the chip is running at max tdp, it will not boost to 3.7, but will run at its base. I think you are confusing what TBM3 does, which is to ramp up a single core when a single thread is being used. The chip as far as I know will not run any of its cores below the base, but ill shut off cores not in use. That is very different than downclocking.

 

[cytg111]

..Also the 6950x has 25Mb cache while the 6900k "only" has 20Mb and it has been debated that blender is very sensitive towards cache.

To be fair, same amount of cache per core, not like i5 vs i7 difference.
6950x vs 6900k, one core, same speed, same memory configuration and I would expect identical performance in everything.

edit : as I can see everyone else has pointed out with tests to match. ignore me.

 

[Doom2pro]

OMG please someone publish an official review. I can't take it anymore!

 

[Abwx]

It can perform 2x256 FP and 3 vecint and on that graph it's even not clear if they are 256 bit. I was talking of FP. I don't care of vecint.

Who said that i was talking of VecInt set apart you..?..

One FPMUL in port 0 and one (FPMUL or a FPADD) in the port 1.

At the bottom of the page :

http://www.hardware.fr/news/29-08-2016/

The base clock is a minimum performance when all cores are being used. The boost clock is a boost from the base if conditions allow for the chip to do more. .

It boost at 3.5 with all cores in about any condition, including Prime 95 if the FFT is not too small :

http://www.hardware.fr/articles/946-4/overclocking-consommation.html

Par défaut l'i7-6900K fonctionne à 3.5 GHz

 

[inf64]

Also the 6950x has 25Mb cache while the 6900k "only" has 20Mb and it has been debated that blender is very sensitive towards cache.

Dude are you for real? There are other 6900K owners on overclock.net forum who get exactly the same score to the portion of a second as the New Horizon 6900K. I cannot believe you are trying to disprove something that was so easy to verify and has been done since , numerous times. Please , get back to reality ASAP and stop the denial already.

 

[realibrad]

Who said that i was talking of VecInt set apart you..?..

One FPMUL in port 0 and one (FPMUL or a FPADD) in the port 1.

At the bottom of the page :

http://www.hardware.fr/news/29-08-2016/



It boost at 3.5 with all cores in about any condition, including Prime 95 if the FFT is not too small :

http://www.hardware.fr/articles/946-4/overclocking-consommation.html

Not sure what you are correcting and or clarifying.

 

[dark zero]

Not sure what your point is. People were wondering if AMD might. Hype train gonna hype and AMD is not at fault. Personally I'm 99% sure that my next box will be a Zen system.

I'm assuming you believe that I feel it's a letdown but that would be wrong. I just feel it clears things up for some.

Is the hype is not real, AMD would endinng having a big problem. But if ended to be real in most aspects... AMD is back to the game and for real this time.

 

[Dresdenboy]

Zen can do (256 bit wise) 1 FADD + 1 FMUL or 1 FMAC
SKL can do (256 bit wise) 1 FADD+ 1 FMUL or 2 FMAC

In 128 bit code Zen has double throughput.

Only on 256 FMAC code INTEL has advantage. On 256 bit not FMAC code they are on par, but INTEL shares ports with int units and Zen not.
So they are superior in highly optimized FMAC code... What can be? BLAS? Ok, but that is RAM limited. Matlab has autoparallelization for vector code (built in in BLAS linpack or lapack libraries), but this auto parallelization is turned off in add, mul and other simple code, because it's RAM limited...
So only difficult calculations can benefit from 256 bit... They have only 2 FP pipes... AMD 4. And the DIV and SQRT are 128 bit only even on SKL... And AMD have four pipes in that case...

If you think of blender, for instance, a ray tracing algorithm for sure requires also divisions and SQRTs... In this case AMD will shine...

Interestingly Canard PC measured SR's VFMADD132PD throughput to be 2x BR's throughput, or the same as KL (2/cycle).
SQRTPD gets executed on just 1 unit, DIVPD, too, while VDIVPD has a bit better throughput.

Edit:
I just found out about two new articles by Hiroshige Goto (auto translated to Googlish):
The integer unit of AMD's next generation CPU "ZEN" is completely different from the Bulldozer series
Floating point / SIMD unit of AMD next generation CPU "ZEN" of relatively mature design

 

[realibrad]

Is the hype is not real, AMD would endinng having a big problem. But if ended to be real in most aspects... AMD is back to the game and for real this time.

I think the are in the game, just not going to beat Intel from that quote.

 

[dark zero]

I think the are in the game, just not going to beat Intel from that quote.

Obviously, but at least AMD will wake Intel up to stop to being a loosey goosey and start to finally bring improvements.