AMD RyZen CPU Architecture for 2017

AMD OC utility:

ZwY05FK.jpg


Also 1800x apparently got to 5.1 with LN2.
 
This is just Cinebench, but we can already estimate Zen IPC based on these results.

Core i7-6900K: 8 cores (16 threads), base = 3.2 GHz, turbo = 3.7 GHz
Ryzen 1800X: 8 cores (16 threads), base = 3.6 GHz, turbo = 4.0 GHz.

In single-threaded case Ryzen has 8.1% clock advantage (assuming both reach max turbo, since only one of eight cores are used).
In multi-threaded case Ryzen has 12.5% clock advantage (assuming base clocks as Cinebench fully taxes all cores).

Ryzen is tied in single-threaded case. Ryzen has 1 - 1.0 / 1.081 = 7.5% lower IPC in this case.
Ryzen leads by 9% in multi-threaded case. Ryzen has 1 - 1.09 / 1.125 = 3.1% lower IPC in this case.

In Cinebench, Zen core IPC seems to be roughly equal to Haswell and significantly above Ivy Bridge. Haswell vs Broadwell vs Ivy vs Sandy Cinebench comparison here: http://www.anandtech.com/show/9482/intel-broadwell-pt2-overclocking-ipc/3. However the 5775C Broadwell model has EDRAM, so the comparison of Broadwell vs others is not 100% valid.

If Ryzen can match Haswell in IPC in other tests as well, the clock advantage means that it 1800X will beat i7-6900K in most use cases. But 6900K has quad channel memory controller and full rate AVX, so it will certainly beat Ryzen in HPC tests and memory bound cases. Cinebench R15 doesn't use AVX and I doubt it is memory bandwidth bound. We need to see more benchmarks to know how many use cases are affected by these architectural differences.
 
Last edited:
So it might be that higher clocks in the fully loaded 8 cores case are the reasons Ryzen peforms better in MT scenarios vs 6900k
 
mmmh anyone know if there are documents like Intel® 64 and IA-32 Architectures Software Developer Manuals for those AMD CPU yet ?
(And where to find them ^^)
I can't wait for benchmarks but those should help me wrap my head around their capabilities.

It seems Family 17H documentation isn't available yet :(
 
Last edited:
If Ryzen can match Haswell in IPC in other tests as well, the clock advantage means that it 1800X will beat i7-6900K in most use cases. But 6900K has quad channel memory controller and full rate AVX, so it will certainly beat Ryzen in HPC tests and memory bound cases. Cinebench R15 doesn't use AVX and I doubt it is memory bandwidth bound. We need to see more benchmarks to know how many use cases are affected by these architectural differences.

Just for saying, i have a quad channel memory based Intel ( 4930K), and if the bandwith number are astonishing ( C9 latency ) compared to "standard" dual channel consumer parts.. it is only visible on memory benchmark.. i will not win 1fps in games due to that, nor on professional softwares... AVX "full speed " is another question.

So it might be that higher clocks in the fully loaded 8 cores case are the reasons Ryzen peforms better in MT scenarios vs 6900k

A little bit of both.. but it seems the Turbo of AMD ( senseMi) do a better job than the 6900K at maintain higher clock speed with full cores. But in reality clock for clock, it seems that both architecture, Intel ( whatever is the generation of Intel, who have mostly only increase clock speed, not much IPC performance ) and Zen are mostly equal ...

The difference between both are really tiny. if you was just seen scores, you can believe this 2 are the same processors. ( who is indeed a good news for AMD for a first Zen iteration ).
 
Last edited:
6900k turbo is 4.0ghz with turbo boost 3.
Good to know, since Intel official specs and Wikipedia do not mention anything about 4.0 GHz max turbo.

https://ark.intel.com/products/94196/Intel-Core-i7-6900K-Processor-20M-Cache-up-to-3_70-GHz
https://en.wikipedia.org/wiki/Broadwell_(microarchitecture)

Found article about turbo boost 3. Seems to require special driver + mobo support. I wonder whether AMD had enabled turbo boost 3 in that particular single thread comparison.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2
People are also reporting all core turbo for 6900k in Cinebench is around 3.5GHz.
That's surprising. That 140W TDP seems to be doing wonders. Of course Cinebench doesn't use AVX at all, so higher clocks are possible. Heavy AVX usage can even drop clocks below base values.

But we don't know how AMDs turbo works yet. It might be that Ryzen 1800X is also above base clocks in multi-threaded benchmark. IIRC there's also an (optional) turbo mode in Ryzen that clocks the CPU higher than max turbo if TDP/thermal headroom allows it. Unfortunately this means that we need to wait for more benchmarks to draw good conclusions about IPC.
 
Last edited:
Good to know, since Intel official specs and Wikipedia do not mention anything about 4.0 GHz max turbo.

https://ark.intel.com/products/94196/Intel-Core-i7-6900K-Processor-20M-Cache-up-to-3_70-GHz
https://en.wikipedia.org/wiki/Broadwell_(microarchitecture)

Found article about turbo boost 3. Seems to require special driver + mobo support. I wonder whether AMD had enabled turbo boost 3 in that particular single thread comparison.
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

...


Here is confirmation 6900K was running with 4GHz Turbo Boost 3.0


http://images.anandtech.com/galleries/5485/AMD Ryzen Tech Day - Lisa Su Keynote-33.jpg
http://images.anandtech.com/galleries/5485/AMD Ryzen Tech Day - Lisa Su Keynote-33.jpg
click on link if you want to see it as it was LARGE image :)
 
Last edited:
Interesting snippets from this Linus video
  • The XFR is a 100Mhz boost above the boost clock if cooling permits.
  • The 6900K vs 1800X Cinebench was 3.2Ghz vs 3.6Ghz respectively
 
Back
Top