AMD RyZen CPU Architecture for 2017

[Alexko]

Oh noes, i'll try to brace myself again for the pollution of this architecture thread with RAM settings and game run results ..

There are SKU-specific threads for that purpose, and I for one welcome @Clukos's RAM benchmarks, as I'm eager to see whether (and to what degree) Ryzen remains sensitive to RAM settings now that its RAM and cache latencies have significantly improved, yet not to the point of matching Skylake's.

 

[entity279]

There are SKU-specific threads for that purpose.

That's precisely my point. I'd welcome the benchmarks too otherwise

 

[hoom]

https://www.anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600

 

[Malo]

https://www.anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600

There's some discussion at AT forums about the review and it not mirroring other sites. AT specifically states they have all latest Spectre/Meltdown patches applied, which could be why Intel is showing as they are compared to other reviews. Also regarding memory speeds and that AT uses stock settings, which are higher on Ryzen, especially the new platform.

 

[DavidGraham]

There's some discussion at AT forums about the review and it not mirroring other sites. AT specifically states they have all latest Spectre/Meltdown patches applied, which could be why Intel is showing as they are compared to other reviews. Also regarding memory speeds and that AT uses stock settings, which are higher on Ryzen, especially the new platform.

All other sites are testing with latest patches, Anand numbers are a true anomaly since they seem to have Intel scores so low even compared to competition, and even compared to AMD's own testing, I truly wonder how this got past their attention before they published the review.

 

[Malo]

https://www.gamersnexus.net/hwrevie...x-review-game-streaming-cpu-benchmarks-memory

The vcore on CPU and motherboard and overall power draw for Zen+ looks fantastic for the same speed. That's a huge difference at 4Ghz stable. So whilst it may not have achieved much overall core speed increase, it's definitely able to boost much better due to much lower voltage requires to hit those speeds and therefore better temps to keep it sustained.

 

[Malo]

All other sites are testing with latest patches, Anand numbers are a true anomaly since they seem to have Intel scores so low even compared to competition, and even compared to AMD's own testing, I truly wonder how this got past their attention before they published the review.

Yeah stock cooler and stock JEDEC RAM speed on the Intel isn't really showing its capabilities but that's how they do their tests. And against stock AMD on Zen+, the Ryzen is looking great.

 

[DavidGraham]

Yeah stock cooler and stock JEDEC RAM speed on the Intel isn't really showing its capabilities

This isn't about stock coolers or RAM, all Intel CPUs (6700K, 7700K, 8700K) are showing massively lackluster performance. Something no other outlet has been able to reproduce, even with stock hardware.

On a side note, why would AMD publish test results with a GTX 1080 and not a Vega 64?

 

[kalelovil]

This isn't the first time Anandtech's game bench results have been odd.
Their Coffee Lake review showed the i5-8400 often clearly beating the i7-8700k.
https://www.anandtech.com/show/1185...e-lake-review-8700k-and-8400-initial-numbers/

 

[HMBR]

regarding Anandtech's gaming results; in their defense they are one of the few running stock ram speeds, and being clear about using all the meltdown/spectre fixes and windows updates, so who knows... but I think it's more due to their selection/methods or something;

but even ignoring their gaming results and looking at many difference sources it's clear that in most games the 2700x is looking a lot closer to the Intel CPUs; no doubt older less optimized CPU heavy titles might still show a big advantage for Intel,

but I also would mention this result from "audacity", running a 2011 encoder Ryzen was far behind, now with a recent update Intel is much faster and Ryzen is a lot closer...
https://www.pcper.com/reviews/Proce...view-Zen-Matures/Media-Encoding-and-Rendering

Intel still holds a good advantage in h265 encoding
https://static.techspot.com/articles-info/1613/bench/HandBrake.png

but I think overall things are looking good for AMD, the 2700x beats the 8700K for MT, is not far from the 7820X and with a significant price advantage;
gaming looks better on the 2700x than Skylake-X and it's not massively slower than the 8700K on average...

I think the hardware.fr averages are a good representation of that
https://www.hardware.fr/articles/974-19/indices-performance.html

applications they show only a 3% deficit compared to the 7820x; while the 8700K is over 11% behind the 2700x

as for gaming (their games selection is a little outdated... but...) the 2700x is 8.5% behind the 8700K,

 

[hoom]

So in summary: much as expected a clock buff, fixes some low hanging fruit, improves perf/W.
2600X looks like a pretty nice sweet-spot.


Interesting architectural points:
Same die size despite higher density of the process, thats a min-effort/cost thing I guess.
Means they are leaving some performance improvement on the table? Should at least give better thermals & less leakage.

Improved cache latencies: L1 -1clock, L2 from 17clocks to 11, initial design was 12 but not mature for original Ryzen launch, hit the 12 with Threadripper & Vegan but improved again for Zen+.
Has there been any good explanation for the extra latency step in the Zen L2? Lower latency at 64 & 128KB is some kind of flow-on from the larger L1 instruction cache?

 

[Gubbi]

All other sites are testing with latest patches, Anand numbers are a true anomaly since they seem to have Intel scores so low even compared to competition

All other sites have their 8700Ks massively exceeding its 95W TDP, implying all-core max-turbo (multi-core enhancement ON). No review, AFAICT, have their 8700K running withing TDP spec (AT's 8700K is at 110W).

Cheers

 

[DavidGraham]

All other sites have their 8700Ks massively exceeding its 95W TDP, implying all-core max-turbo (multi-core enhancement ON). No review, AFAICT, have their 8700K running withing TDP spec (AT's 8700K is at 110W).

I don't think so, reviewers are aware of this from the last round of reviews, they avoid it not to cause debates. GamersNexus, TechSpot, PCPer, PCGamesHardware, ComputerBase .. all of these sites don't enable all-core max-turbo yet they have massively different results to AanandTech.

 

[fellix]

Has there been any good explanation for the extra latency step in the Zen L2? Lower latency at 64 & 128KB is some kind of flow-on from the larger L1 instruction cache?

Pointer chasing results looks fine, with the expected swing at the L3 boundary:

CPU#000 Type: "AMD Ryzen 7 2700X Eight-Core Processor"
Stride:       8   16   32   64  128  256  512   1K
       2 KB   4    4    4    4    4    4    4    4
       3 KB   4    4    4    4    4    4    4    4
       4 KB   4    4    4    4    4    4    4    4
       6 KB   4    4    4    4    4    4    4    4
       8 KB   4    4    4    4    4    4    4    4
      12 KB   4    4    4    4    4    4    4    4
      16 KB   4    4    4    4    4    4    4    4
      24 KB   4    4    4    4    4    4    4    4
      32 KB   4    4    4    4    4    4    4    4
      48 KB   7    7    8   12   12   10   12   12
      64 KB   8    9   10   12   12   11   12   12
      96 KB  10   10   10   12   12   12   12   12
     128 KB  10   10   11   12   12   12   12   12
     192 KB  11   11   11   12   12   12   12   12
     256 KB  11   11   12   12   12   12   12   12
     384 KB  13   13   12   14   14   12   14   16
     512 KB  15   14   13   15   14   16   17   16
     768 KB  25   26   26   34   39   42   41   43
    1024 KB  27   31   33   35   39   45   46   43
    1536 KB  34   34   36   38   41   46   46   46
    2048 KB  37   35   38   39   43   47   46   46
    3072 KB  38   38   39   40   44   46   47   46
    4096 KB  35   36   37   39   42   45   46   45
    6144 KB  38   38   39   39   43   46   46   47
    8192 KB  43   51   66   69   94  104   85  102
   12288 KB  54   70  104  168  276  280  282  281
   16384 KB  54   72  107  175  286  288  289  290
   24576 KB  55   72  107  175  286  289  290  291
   32768 KB  54   72  107  175  286  288  290  291

 

[hoom]

Cool, that looks like it should

Why does Anand latency graphs have the step then?
https://www.anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600/3

 

[Silent_Buddha]

Does anyone know if the new Ryzen+ CPUs have better support for 4 channel memory? The 1600x that I have does well with certain memory as long as it is limited to 2 channels, but completely falls apart with 4 channels. I'm wonder if the new processors do better with 4 channels of memory. If it does I might pick one up. Otherwise no reasons to replace the 1600x until Ryzen 2.

Regards,
SB

 

[kalelovil]

A cursory glance at Asus motherboard documents shows high frequency 4 DIMM support for Ryzen 2nd generation actually regressing compared to 1st generation.
http://dlcdnet.asus.com/pub/ASUS/mb...9.2120714594.1524297732-1327312654.1521109423
http://dlcdnet.asus.com/pub/ASUS/mb...5.2120714594.1524297732-1327312654.1521109423

(Perhaps Asus have not fully tested 2nd generation memory support yet and are focusing on 2 DIMM setups first, rather than a real regression).

 

[entity279]

(Perhaps Asus have not fully tested 2nd generation memory support yet and are focusing on 2 DIMM setups first, rather than a real regression).

Pretty sure that's the answer.

If my Skylake-X + QVL dual sided RAM means anything though, buying a memmory in the MB's list doesn't guarantee that the kit works at XMP settings. I've tried 2 kits thus far and no go

Likely ppl with single sided kits/lower number of sticks do better thought

 

[Clukos]

Stock 2700X (only tuning memory), this thing is pretty beastly:

 

[CSI PC]

Stock 2700X (only tuning memory), this thing is pretty beastly:

Let us know how the voltage/thermals pans out when you decide to either push all cores to higher fixed frequency or general overclocking.
The core voltage as standard seems pretty nice relative to previous Ryzen generation.