Cache design is a mixture of tradeoffs for die area, power consumption, aliasing, associativity, coherency, latency and bandwidth, and you can sometimes make one aspect a lot worse, but in such a way that it allows another aspect to get much better. It's a good example of when Amdahl's law should be used. Bulldozer's cache was infamously crap, and it's possible that that's because they prioritised lowering latency over everything else, rather than actually going for ensuring, for example, that the cache was likely to have the required data in it in the first place.
![[Image: rr3qqve2ommy.jpg]](https://i.redd.it/rr3qqve2ommy.jpg)
Seems like Windows 10 has improved its scheduler to help alleviate the Ryzen issues, not sure what this will mean for the Dolphin Benchmark though. I'm not getting mine until next month though.
(03-21-2017, 12:10 PM)Chaoslux Wrote: [ -> ]Seems like Windows 10 has improved its scheduler to help alleviate the Ryzen issues, not sure what this will mean for the Dolphin Benchmark though. I'm not getting mine until next month though.
I'm not so sure because they're reporting that no other games that they tested with are seeing any performance benefits:
https://twitter.com/BitsAndChipsEng/status/843874881049649158
Wanted to share something with you all. I switched motherboards and the BIOS in my new board has options for using less core such as 6 (3x3) 4 (2x2) 4 (4x0) 2 (1x1) 2 (2x0). I'm assuming these options are letting me choose to have the cores run from one CCX or 2. I could be wrong, but what else would it mean?
So.....I ran the benchmark on the 4x0 setting and the 2x2 setting.
![[Image: BfmSiuZ.png]](http://i.imgur.com/BfmSiuZ.png)
2x2
![[Image: xZzSUSd.png]](http://i.imgur.com/xZzSUSd.png)
4x0
Either I am misunderstanding this setting and it means something else, or CCX has no real effect on the Dolphin benchmark. There's a few second difference but that means nothing because I ran the test 3 times on each and got slightly different results each time.
1700 @ 3.8 GHz by the way.
No, that setting is doing exactly what you think it's doing.
Other than possible Linux performance, it seems that the first generation Zen architecture simply does not include the emulation-performance-boost-voodoo that Intel has implemented since Haswell (I'm being somewhat tongue-in-cheek here).
In particular, the difference in per-GHz performance between Sandy and Ivy Bridge is not all that far removed from the difference in per-GHz performance between Ivy Bridge and Ryzen. In other words, Ryzen has similar increase in performance over Ivy Bridge that Ivy has over Sandy.
BTW, I've made an edit to the benchmark thread and spreadsheet where I'm also now marking 3.0GHz and 4.0GHz submissions with inverted colors for easier IPC comparisons.
(03-24-2017, 06:35 PM)Nintendo Maniac 64 Wrote: [ -> ]No, that setting is doing exactly what you think it's doing.
Other than possible Linux performance, it seems that the first generation Zen architecture simply does not include the emulation-performance-boost-voodoo that Intel has implemented since Haswell (I'm being somewhat tongue-in-cheek here).
In particular, the difference in per-GHz performance between Sandy and Ivy Bridge is not all that far removed from the difference in per-GHz performance between Ivy Bridge and Ryzen. In other words, Ryzen has similar increase in performance over Ivy Bridge that Ivy has over Sandy.
BTW, I've made an edit to the benchmark thread and spreadsheet where I'm also now marking 3.0GHz and 4.0GHz submissions with inverted colors for easier IPC comparisons.
I mean in all fairness to Ryzen it handles the games well enough. I don't know about 100% perfectly nor does it match Intel. But if someone wants to build an AMD rig and also emulate with Dolphin it's extremely viable.
To be honest, I'm much more interested in Raven Ridge (both desktop and mobile variants).
(03-26-2017, 08:10 AM)DarkHacker Wrote: [ -> ]https://www.reddit.com/r/Amd/comments/61ccp5/ryzen_7_397ghz_vs_7700k_5ghz_retest_with_faster/
Interesting results with high frequency memories. (it seems too good to be true, IMHO)
It might be, but people have been telling me for a few weeks that Ryzen performs better and better with higher frequency memory.
Unless I can find someone to donate me a higher frequency kit I'm not testing it lol. I have 2400 and I would upgrade but not with the prices inflated so much how they are now. My kit cost $60 less than a year ago and $105 today.