Zen SuperPI 32M 5G/4G Challenge

[unityofsaints]

@Massman Can we get the OP fixed? The tables don't work since the migration to communities.hwbot.org

[websmile]

Massman left hwbot at the end of last year... please contact or tag richba5tard

[flanker]

Any news about WinXp mod to Ryzen? Did you tried it guys?

[unityofsaints]

It's a shame so many of the images in this thread got lost with the forum migration Lots of good info is now inaccessible.

[Max-Hardware_Numb3rs]

Guys, who is up for another 4/5GHz SPI32M  round ? 

[flanker]

Im for it  Maybe Jordan also can join.

[KaRtA]

I still have my board and chip. Board isn’t the best clocking though  

[Max-Hardware_Numb3rs]

I'll start, Win 7 Sp1 fresh install, updated, no other tweaks than pagefile set to 200 on C:\ and 0 in the other drives 

This is my starting point now for the OS, before touching the hardware again I will spend time here.

[Max-Hardware_Numb3rs]

I have a weird issue, it seems that sometimes the system boot with bad performance, I have to restart many times to get a proper score, doest happens to you as well ?

[KaRtA]

Yes. Happens to me too. Lose a few seconds in efficiency. 

 

I know why. You will work it out pretty quick. 

Edited February 4, 2019 by KaRtA

[Max-Hardware_Numb3rs]

I'm talking about a lot more, even minutes, when I boot at 4GHz I have 3 random levels of performance, if I run the 16K test I can have around 0.189, 0.172 or 0.109 as best result, obviously the "109" gave me the best scores in the 1M and 32M, if I boot and in the 16K I have the lower ones, everything is slow, so I have to restart until I'm in the "109" session.

This apply mainly to the SPI, I noticed that only a few other benchmarks may have a very small performance decrease when the system is scoring low with SPI.

Ryzen is fun, but there are some very mysterious thing around this plattform ?

[KaRtA]

If you don’t set affinity then your score will be rubbish. 

You need to sort a Waza out or you are wasting time too. The score can vary a fair bit without it and I even noticed up to a 1 second variance using differnt cores too. 

Ive got other platforms on the bench at the moment prepping for a review so I can’t do a 4ghz just yet for you to compare. 

[flanker]

Do you have set some of performance BIAS for Superpi?

[Max-Hardware_Numb3rs]

I tried a lot of things, CH6, CH7, affinity, priority, bias, maxmem, prefetch, superfetch, waza (OCX and manual), paging, registry tweaks, os stripped, tight timings, loose timings, SMT on/off,  and so on and on...

Probably I'm missing something that is right under my nose and I can't see it, but with this platform I'm going crazy.

This is what I mean by bad restart:

 

"normal" run, sometimes I can get it to 8m32s

 

With waza:

aza 

 

[KaRtA]

That looks like you have too low maxmemm to me. 600 is too low.

When I run below 900mb I have very bad performance as it runs out of memory. Try 1000-1200 and see if the problem goes.

[Max-Hardware_Numb3rs]

Here is set to 4000, i tried from 600 and up, I've spent days and weeks with this 2700X and all the possible combinations of settings, probably I'm just cursed on this benchmarks ?

[Max-Hardware_Numb3rs]

I share here some test that I'm doing, it may help others that are looking to improve their scores.

I decided to start from scratch, I think it's always a good thing to do when I need a better understanding of the situation.

As OS base:

• Win 7 SP1 updated, not stripped
• All drivers installed
• Maxmem set to 4000
• Large system cache enabled 
• 200MB paging on drive C (0 in the other partitions)
• Power plan profile set to high performance
• Classic theme and visual effects set for best performance
• Indexing disabled 

So nothing special, some of the common windows tweaks (that I will check individually if are really worth or not later on)

As for the hardware:

2700X @ 4GHz and the Asus CH7 with the memory profile on Stilt's 3200MHz Fast (just to load something better than stock, but a good base actually)

Tested all 16 cores with affinity and picked the best run (8m45s.893):

After that, I pushed the memory to 3600MHz and I got 8s512 reduction (best run) with the core #10 

Then I set the timings to 11 10 10 (same tRAS, tRC and sub): 2s981 gain 

Now I switched to my own custom memory profile: 1s200 gain

I'm currently doing test with waza, but so far I wasn't able to get a gain, mostly a decrease in performance, am I doing something wrong ? probably yes, but I will find out eventually.

To be clear, I'm not asking you to share your tweaks (if you don't want to), but just to spice a bit the competition in a platform that seems forgotten by the community, and imho very funny to play with since there's no CB or other annoying issues.

 

[Guest]

Maybe those tweaks will be re-discovered. At 4g there is already like 20s to gain from private tweaks.

[Max-Hardware_Numb3rs]

Yes, that’s what I’m going after, I’m stubborn, I will find out sooner or later ?

[KaRtA]

I think the simple response as a tweak is the topography on the Crosshair hurts it’s speed. 

 

Possibly need to use Gigabyte or Asrock for the boost. Without those boards though, it’s hard to tell, plus they both have cold issues as far as I know. 

[Guest]

Zeneffect got 8m 20s or so with an Asus B350 board, so that can't be it.

[newlife]

I still need to get some decent memory but I'll join in with you guys when I get some free time

[Max-Hardware_Numb3rs]

I have various boards here, like a Gigabyte X470 Gaming 7,  I think I went to -96 once with the cascade without issues, it may be worth a try, tonight I try a 4GHz there and we’ll see. 

[Max-Hardware_Numb3rs]

I tried with the Gigabyte Aorus Gaming 7 X470, but unfortunately there's no improvement there, plus the Crosshair have a better memory management and voltages, I can reach higer clocks with tighter timings, so the board it's not the solution.

I tried a 2200G with the CH7, not bad considering it's a Gen1 Ryzen.

[flanker]

interesting, how the small Raven Ridge is close to Pinnacle Ridge (better IPC, higher cache) !