Jump to content
HWBOT Community Forums

Zen SuperPI 32M 5G/4G Challenge


Massman

Recommended Posts

Couple of questions for you guys in regards to X370 Hero (Posting here since this seems to be the most active thread in the sub forum)

 

1) Where did you guys get the 7304 bios. Latest on Asus site is 7201.

 

 

2) Does this board still work well when taking Zen2 cold or more advisable to buy the X470-X570 ones.

 

Trying to save a few bucks here by buying the older board.

Link to comment
Share on other sites

10 hours ago, rtsurfer said:

Couple of questions for you guys in regards to X370 Hero (Posting here since this seems to be the most active thread in the sub forum)

 

1) Where did you guys get the 7304 bios. Latest on Asus site is 7201.

 

 

2) Does this board still work well when taking Zen2 cold or more advisable to buy the X470-X570 ones.

 

Trying to save a few bucks here by buying the older board.

1) look here: https://rog.asus.com/forum/showthread.php?112279-X370-X470-AGESA-1003AB-Bioses

2) no comment

  • Like 1
Link to comment
Share on other sites

17 hours ago, rtsurfer said:

Couple of questions for you guys in regards to X370 Hero (Posting here since this seems to be the most active thread in the sub forum)

 

1) Where did you guys get the 7304 bios. Latest on Asus site is 7201.

 

 

2) Does this board still work well when taking Zen2 cold or more advisable to buy the X470-X570 ones.

 

Trying to save a few bucks here by buying the older board.

1. See above, but I got it from a friend.

2. I have hopefully a LN2 fill due tomorrow. Will let everyone know once I have had a try. At this stage at -60 on the SS I am in a world of pain though, it's just in that zone of making training a full pain in the ass.

 

Board is still very capable once you work it all out, only negative is you can't go nuts with A2 PCB DIMMs like the X570 have allowed. My board isn't happy past 4000C12-1t or 2t so running high mem speed and low FCLK for avoiding cold issues isn't possible. 4000C12 with 1000FCLK is stupid slow.

  • Like 1
Link to comment
Share on other sites

2 hours ago, KaRtA said:

LN2 delivery delayed until Monday now. Hopefully finish work early to play again then.

 

@Hardware_Numb3rs do you have a 5ghz run saved? would like to compare for scaling between the different setups.

I went directly for the max clock, I have another screen with another core at 5800, but 2 second slower, I didn't try much experiments or efficiency, with the CBB at -27 the LN2 waste was enormous, I think I burned almost 20L and like 5 hours, now there's a new official bios, I will have the CH8 next week to play with, really hope to get rid of the issue.

Link to comment
Share on other sites

Had a go last night full out, but efficiency seemed a little off I think. Bad mount didn't help max clock I don't think but close enough.

 

https://hwbot.org/submission/4221311_

 

Similar results on cold

No CB @ 1467 FCLK

-180 @ 1500 FCLK

-150 @ 1600 FCLK

-120 @ 1733 FCLK

 

Reboots required torching to -38 to train and boot again.

 

Few more little things to try including a better mount but not sure this CPU has the legs. Nearly there though.

  • Like 1
Link to comment
Share on other sites

The sub 6 is tricky, as going up with the frequency the efficiency get worse, I think we need a 5850+ capable CPU or some OS tweak not yet discovered ? 

Once I have my 3900X and the CH8 I’ll try again, in the meantime I want to build Win7 from scratch, but for spi is also difficult to tune, messing too much with the os damage the score.

 

  • Like 1
Link to comment
Share on other sites

On 8/20/2019 at 11:57 AM, Hardware_Numb3rs said:

in the meantime I want to build Win7 from scratch, but for spi is also difficult to tune, messing too much with the os damage the score.

Tell me about it :D

I've experimented with some registers and there's a positive effect in Pi performance, but I'm far for completing this research, since there is no public BKDG for family 17h :( I was able to gain 3 seconds at 4GHz fixed frequency.

I'm also working on providing frequency and VID control for Zen2 in ZenStates app. Have a partial success and this is what commands I have found so far by trial and error. Setting P-State MSRs just doesn't work, so sending commands to the SMU seems to be the only option right now.

ID Name Note
0x1 TestMessage  
0x2 GetSmuVersion  
0x24 EnableOverclocking Forces base clock and manual overclock mode.
0x25 DisableOverclocking  
0x26 SetOverclockFreqAllCore Sets all core frequency, EnableOverclocking first.
0x27 SetOverclockFreqPerCore Always sets core #0, probably needs additional parameters. EnableOverclocking first.
0x28 SetOverclockVid Alters the VID (in HEX). EnableOverclocking first.
0x29 SetBoostLimitFreqAllCores Probably sets fmax
0x2B ? Sets maximum boost frequency
0x2C GetOverclockCap ?
0x2F ? With multi manually set to 40x, sets the multi to 39.50x
Edited by I.nfraR.ed
  • Like 1
  • Thanks 2
Link to comment
Share on other sites

  • 3 months later...
  • 2 weeks later...

Ok so as far as efficiency  goes take it with a grain of salt. Untuned win10. However I ran multiple times to get consistent results so we can at least compare apples to apples,. All timings and speeds identical. The takeaway is only that ryzen + 2700x in either the gigabyte or x570 boards is slower than ryzen 3000 or the 3900x cache is giving it an advantage. I'll verify if vendor or x570 in general and test a 3800x later. Excuse the lack luster mem clocks and timings. The hardware im working on and with right now can not be beaten on voltage wise because of what i'm working on. In fact I probably shouldn't even be "benching pi" but it falls into the grey area of me testing 2700x operation in x570 boards so that's my story and I am sticking to it. I may grab ahold of an x470 board just to verify some things, I will definitely look into this more and get the correct OS going again for this ;)

 

 

32m 2700x.jpg

32m 3900x.jpg

  • Like 1
Link to comment
Share on other sites

the only non displayable bugs at the time were 3400 mem speeds. my imc was always crappy.

Heres 3200 for comparison with DR bdie. hypothetically based on the scaling a 3600 imc would have put me ballpark 8-11/8-12

 

8-17-1 DR.jpg

Edited by chew*
  • Thanks 3
Link to comment
Share on other sites

3 minutes ago, Bones said:

Good to see you tinkering again Chew, I need to start up again myself and see what I can come up with across a few sockets.

yah i'm interested if an 1800x can boot and if its faster or slower in this board and vice versa r3000 in x370.

we may have a situation here where r1 was 3 steps forward R+ took 3 steps back and R2 took 2 steps forward at least as far as 32m pi goes.

Edited by chew*
Link to comment
Share on other sites

just checking efficiency per board. C8H on chopping block next. No luck with the 32m w7 yet, injected drivers into my old os still no usb :( 

Only difference in the timings is the boards preferred default odt everything else identical. Looks very close.

 

32m 3900x 2.jpg

32m 3900x 3.jpg

Edited by chew*
Link to comment
Share on other sites

10 hours ago, l0ud_sil3nc3 said:

Quick test of W7 on the Aqua, no waza, no tweaks just trying to balance max fabric clock vs. frequency.

 

No tools are working in w7 so this was direct boot from bios

 

@chew* for usb, just use a pcie add in usb card, all working with this method plus a ps2 splitter.

 

 

6.49.719.PNG

that aqua is an expensive mofo…..

 yah zens sending me a card that's native to xp and 7.

zen has my windows oc tool and winring file for r1. Its possible that it might still work for bumping manual voltage and multi in 7.

getting bored running prime and hci 24/7. Need some 32m love to break up the monotony.

Once I have finished all the boring crap i'm working on I can put the screws to this hardware but for now all results will be done @ realistic 24/7 conservative voltages as I need the hardware to stay reliable and consistent till I am done.

 

Edited by chew*
Link to comment
Share on other sites

OK so It took a little while to figure it out in w10 probably because I never bother due to 10 being slow but none the less it's repeatable in either OS.

So first the comparison. taichi vs taichi.

The bios is worth nothing time wise.

The tighter timings are worth 2 seconds at best on a good run.

vddr is 1.44 which is my artificially limited max for the time being it also passed a quick 360% hci run lol.

Ill even post the settings for this run.

you figure out how I pulled 6 secs out of my ass and you will be efficient ;) 

Looking forward to getting back on w7

 

32m 3900x 3.jpg

found effeciency.jpg

found effeciency 2.jpg

Edited by chew*
  • Like 1
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...