AMD's Future Chips & SoC's: News, Info & Rumours.

[juanrga]

I understand so, no change in IPC got it!

Not IPC change at core level. Minimal total IPC change from faster IMC.

And what is the IPC change in Zen+? +3%? -2%?

 

[Gon Freecss]

I understand so, no change in IPC got it!

Not IPC change at core level. Minimal total IPC change from faster IMC.

And what is the IPC change in Zen+? +3%? -2%?

At the core level? Zero.

 

[jdwii]

I just think the 2700X should be priced at the non K 8700 model as it competes more with that. Pricing the 2700X at the 8700K is just suicide

 

[juanrga]

OMG, The number of sites that cannot do basic math is increasing: techradar, guru3D, VideoCardz, wccftech, TweakTown, TechPowerUp, el chapuzas informático and others cannot do basic math! The 2700X wasn't 14% faster than 1800X in applications neither 4% faster in games. Only a non-technical site as Forges is obtaining the correct percentages from the graphs in the review.

How can I trust now anything those tech sites measure in their reviews, when they fail to do basic math?

 

[goldstone77]

OMG, The number of sites that cannot do basic math is increasing: Guru3D, videocardz, WCCFTECH, Tweaktown and others cannot do basic math! The 2700X wasn't 14% faster than 1800X in applications neither 4% faster in games. Only a non-technical site as Forges is obtaining the correct percentages from the graphs in the review.

How can I trust now anything those tech sites measure in their reviews, when they fail to basic math?

I think you and Gon Freecss should email the editors, and eagerly wait for their replies! Don't forget to post their responses for all of us to read! <eating popcorn in anticipation>

 

[Gon Freecss]

@juanrga;

Wanted to ask. How does the Zen architecture perform with the AMD compiler compared to the Skylake architecture with the Intel compiler? Also, how does it perform when you specifically code for both architectures? On average.

 

[goldstone77]

Tech Talk 4 - Nvidia GPP, Ryzen 2700X Benchmarks, Intel Pumping Multithreaded Games.
AdoredTV
Published on Mar 27, 2018
A roundup and analysis of some tech news from the past week.
https://www.youtube.com/watch?v=WYvOhJsO71I&feature=youtu.be
Including the percentages or percentage points however you like it!

 

[juanrga]

@juanrga;

Wanted to ask. How does the Zen architecture perform with the AMD compiler compared to the Skylake architecture with the Intel compiler? Also, how does it perform when you specifically code for both architectures? On average.

When running GCC with medium-level optimization flags, the 32-core Zen is 39% faster than the 22-core Broadwell in base score. When using Intel and AMD compilers, the performance gap is reduced to 15%.

The Intel compiler is better and extracts more performance from the hardware.


Broadwell has only 256bit SIMD units. Skylake Xeons have 512bit units. Using specific code for Skylake Xeon can increase several server/HPC workloads by huge amounts such as 50% or 80% faster than Broadwell Xeons.

https://www.anandtech.com/show/11615/intels-data-center-event-live-blog-830am-pt

 

[Gon Freecss]

@juanrga;

Wanted to ask. How does the Zen architecture perform with the AMD compiler compared to the Skylake architecture with the Intel compiler? Also, how does it perform when you specifically code for both architectures? On average.

When running GCC with medium-level optimization flags, the 32-core Zen is 39% faster than the 22-core Broadwell in base score. When using Intel and AMD compilers, the performance gap is reduced to 15%.

The Intel compiler is better and extracts more performance from the hardware.


Broadwell has only 256bit SIMD units. Skylake Xeons have 512bit units. Using specific code for Skylake Xeon can increase several server/HPC workloads by huge amounts such as 50% or 80% faster than Broadwell Xeons.

https://www.anandtech.com/show/11615/intels-data-center-event-live-blog-830am-pt

Doesn't this show the Zen architecture being slow? >45% more cores, yet only ~15% more performance.

 

[goldstone77]

@juanrga;

Wanted to ask. How does the Zen architecture perform with the AMD compiler compared to the Skylake architecture with the Intel compiler? Also, how does it perform when you specifically code for both architectures? On average.

When running GCC with medium-level optimization flags, the 32-core Zen is 39% faster than the 22-core Broadwell in base score. When using Intel and AMD compilers, the performance gap is reduced to 15%.

The Intel compiler is better and extracts more performance from the hardware.


Broadwell has only 256bit SIMD units. Skylake Xeons have 512bit units. Using specific code for Skylake Xeon can increase several server/HPC workloads by huge amounts such as 50% or 80% faster than Broadwell Xeons.

https://www.anandtech.com/show/11615/intels-data-center-event-live-blog-830am-pt

Doesn't this show the Zen architecture being slow? >45% more cores, yet only ~15% more performance.

Is this because Intel has better single thread performance? If so that's amazing, because no one has ever heard that before!

 

[8350rocks]

@juanrga;

Wanted to ask. How does the Zen architecture perform with the AMD compiler compared to the Skylake architecture with the Intel compiler? Also, how does it perform when you specifically code for both architectures? On average.

When running GCC with medium-level optimization flags, the 32-core Zen is 39% faster than the 22-core Broadwell in base score. When using Intel and AMD compilers, the performance gap is reduced to 15%.

The Intel compiler is better and extracts more performance from the hardware.


Broadwell has only 256bit SIMD units. Skylake Xeons have 512bit units. Using specific code for Skylake Xeon can increase several server/HPC workloads by huge amounts such as 50% or 80% faster than Broadwell Xeons.

https://www.anandtech.com/show/11615/intels-data-center-event-live-blog-830am-pt

Doesn't this show the Zen architecture being slow? >45% more cores, yet only ~15% more performance.

Actually that depends entirely upon the workload.

If you notice, SPECfp peak and base are ~50% faster on EPYC

SPECInt peak and base are approximately 20% faster on EPYC

In Data center, the typical usage case trumps single core IPC most of the time.

Not to mention we have no idea if these SPEC benchmarks still included libquantum and the other rubbish they recently removed because it unfairly skewed benchmarks toward Intel in benchmark scenarios.

 

[juanrga]

Not to mention we have no idea if these SPEC benchmarks still included libquantum and the other rubbish they recently removed because it unfairly skewed benchmarks toward Intel in benchmark scenarios.

The AMD compiler also cheats libquantum. And the Intel compiler cheats libquantum on both AMD and Intel hardware.

 

[jdwii]

https://wccftech.com/amd-ryzen-7-2700x-user-publishes-benchmarks-overclocking-results-ahead-of-launch/

4.3Ghz OC with a H90 and using 3600 MHz DDR4 memory

 

[8350rocks]

Not to mention we have no idea if these SPEC benchmarks still included libquantum and the other rubbish they recently removed because it unfairly skewed benchmarks toward Intel in benchmark scenarios.

The AMD compiler also cheats libquantum. And the Intel compiler cheats libquantum on both AMD and Intel hardware.

There is no AMD compiler, only GCC

 

[aldaia]

Not to mention we have no idea if these SPEC benchmarks still included libquantum and the other rubbish they recently removed because it unfairly skewed benchmarks toward Intel in benchmark scenarios.

The AMD compiler also cheats libquantum. And the Intel compiler cheats libquantum on both AMD and Intel hardware.

There is no AMD compiler, only GCC

There is also LLVM compiler. AMD AOCC compiler is just LLVM tuned for Ryzen.

EDIT: I understood you were talking in general, but you are right in those evaluations neither LLVM nor its AMD flavour AOCC are used.

 

[jdwii]

Just read that Amd epic processors will be used with yahoo japan nice win

 

[juanrga]

Not to mention we have no idea if these SPEC benchmarks still included libquantum and the other rubbish they recently removed because it unfairly skewed benchmarks toward Intel in benchmark scenarios.

The AMD compiler also cheats libquantum. And the Intel compiler cheats libquantum on both AMD and Intel hardware.

There is no AMD compiler, only GCC

I just gave above a table with SPEC scores for Intel compiler, GCC, and AMD compiler. The table contains the older AMD compiler, which was based in the Open64 code-base. The new AMD compiler is a fork of LLVM/Clang.

https://www.phoronix.com/scan.php?page=article&item=amd-ryzen-aocc&num=1

 

[aldaia]

@juanrga;

Wanted to ask. How does the Zen architecture perform with the AMD compiler compared to the Skylake architecture with the Intel compiler? Also, how does it perform when you specifically code for both architectures? On average.

When running GCC with medium-level optimization flags, the 32-core Zen is 39% faster than the 22-core Broadwell in base score. When using Intel and AMD compilers, the performance gap is reduced to 15%.

The Intel compiler is better and extracts more performance from the hardware.


Broadwell has only 256bit SIMD units. Skylake Xeons have 512bit units. Using specific code for Skylake Xeon can increase several server/HPC workloads by huge amounts such as 50% or 80% faster than Broadwell Xeons.

https://www.anandtech.com/show/11615/intels-data-center-event-live-blog-830am-pt

Doesn't this show the Zen architecture being slow? >45% more cores, yet only ~15% more performance.

Actually that depends entirely upon the workload.

If you notice, SPECfp peak and base are ~50% faster on EPYC

SPECInt peak and base are approximately 20% faster on EPYC

In Data center, the typical usage case trumps single core IPC most of the time.

Not to mention we have no idea if these SPEC benchmarks still included libquantum and the other rubbish they recently removed because it unfairly skewed benchmarks toward Intel in benchmark scenarios.

Its spec 2006, so yes, they still include libquantum.

However some official SPEC numbers show better results for EPYC 7601.

SPECfp®_rate2006 = 2070
SPECfp_rate_base2006 = 1800
https://www.spec.org/cpu2006/results/res2018q1/cpu2006-20171215-51339.html

SPECint®_rate2006 = 2500
SPECint_rate_base2006 = 2260
https://www.spec.org/cpu2006/results/res2017q4/cpu2006-20171211-51086.html

Those results are still done using the old AMD compiler. Latest version Open64 4.5.2.1 was released in 2013.
I have no idea how the newer AMD compiler (AOCC 1.1) will fare.

 

[aldaia]

Not to mention we have no idea if these SPEC benchmarks still included libquantum and the other rubbish they recently removed because it unfairly skewed benchmarks toward Intel in benchmark scenarios.

The AMD compiler also cheats libquantum. And the Intel compiler cheats libquantum on both AMD and Intel hardware.

There is no AMD compiler, only GCC

I just gave above a table with SPEC scores for Intel compiler, GCC, and AMD compiler. The table contains the older AMD compiler, which was based in the Open64 code-base. The new AMD compiler is a fork of LLVM/Clang.

https://www.phoronix.com/scan.php?page=article&item=amd-ryzen-aocc&num=1

Current AOCC version is 1.1
https://www.phoronix.com/scan.php?page=article&item=amd-aocc-11&num=1

Unfortunately no spec results to compare with old compiler.

 

[jdwii]

So the preorders went up for Ryzen 2000 series and guess what the 2700X is only $329.99 and yes for that price I would say it’s a nice option also the 2700 was $299.99.

I just said pricing the 2700X at the same price as the 8700K is ballsy

 

[juanrga]

Using HFR review as reference I expect the next average performance for Pinnacle Ridge:

2700X ~ 312
2600X ~ 246
2600 ~ 228

 

[-Fran-]

Using HFR review as reference I expect the next average performance for Pinnacle Ridge:

2700X ~ 312
2600X ~ 246
2600 ~ 228

So ~10% increase in performance in all-average? Sounds about right.

Decomposing that, it should be 5% clock, 4% memory shenanigans and 1% HOPES, TEARS AND DREAMS!

But in all seriousness, even being a bit overly pessimistic, 10% sounds too good even by just the clocks and memory bumps.

 

[jdwii]

10% bump and if they price the 2700x at 329.99$ it will be competitive and a option

CR15 2700X OC 4.3Ghz 3600mhz memory
Single core 181

https://youtu.be/A6IBeVTwb3E?t=88

that is impressive i remember my 4790K at 4.7ghz was only 11 points higher then that.

 

[-Fran-]

10% bump and if they price the 2700x at 329.99$ it will be competitive and a option

CR15 2700X OC 4.3Ghz 3600mhz memory
Single core 181

https://youtu.be/A6IBeVTwb3E?t=88

that is impressive i remember my 4790K at 4.7ghz was only 11 points higher then that.

That is an interesting point to note, actually.

Might be the time for me to officially jump ship back to AMD. My i7 2700K has been amazing and hands down my best CPU of all times, but it deserves a rest. Plus, 16GB RAM is getting too tight for me... Ugh... Stupid RAM prices!

Cheers!

 

[goldstone77]

Those single thread scores are heading into the right direction. They still need to get a handle on internal latency though. Ram prices are expected to easy up at the end of the year to the beginning of next year.