Jump to content
HWBOT Community Forums

Zen SuperPI 32M 5G/4G Challenge


Massman

Recommended Posts

yep davis was a tad faster than this on the giga which was faster back then. This was one of the fastest 4 gig results back then. looks like 5000 series does it easily.

 

Now ill have to break it down see if cpu model impacts results or we don't need the highest cache cpu to play the game. I imagine at your experience level you tested 2x8 also loud so coldbug was bad regardless i gather?

8-14-1.jpg

Edited by chew*
Link to comment
Share on other sites

10 hours ago, l0ud_sil3nc3 said:

Here's a few 4Ghz runs for comparison 2100 IF vs 2033 IF both 1:1

Also tried this cpu out on cold but -120c cb really kills the clocks :(

Wait, 2033 is much faster than 2100?

Is that the slow down some people talk about?

I guess you don't want to disclose your timings, but if you run latest beta of ZenTimings, could you spot some differences in non user-configurable timings? I don't have 5000 series to check myself.

Is it strap related? For example, is it the same if you use the lower strap and clock the bclk up or is it just some internal strap that makes things waaay slacker?

Edited by I.nfraR.ed
Link to comment
Share on other sites

15 hours ago, I.nfraR.ed said:

Wait, 2033 is much faster than 2100?

Is that the slow down some people talk about?

I guess you don't want to disclose your timings, but if you run latest beta of ZenTimings, could you spot some differences in non user-configurable timings? I don't have 5000 series to check myself.

Is it strap related? For example, is it the same if you use the lower strap and clock the bclk up or is it just some internal strap that makes things waaay slacker?

I think this is reason of current agesa, but. Still I think 1800 MHz FCLOCK cpould be a little quicker than 2033

Link to comment
Share on other sites

On 11/27/2020 at 12:02 PM, chew* said:

yep davis was a tad faster than this on the giga which was faster back then. This was one of the fastest 4 gig results back then. looks like 5000 series does it easily.

 

Now ill have to break it down see if cpu model impacts results or we don't need the highest cache cpu to play the game. I imagine at your experience level you tested 2x8 also loud so coldbug was bad regardless i gather?

8-14-1.jpg

Hey @chew* haven't had a chance to actually test 2x8GB yet, but it's on the chopping block tonight :)

I only have a 5950x so I can't comment on the lower core count cpu's but I imagine having more cache can only be a good thing just like Zen 2. CB is shit on the two samples I tried at any fclk.

On 11/27/2020 at 1:11 PM, I.nfraR.ed said:

Wait, 2033 is much faster than 2100?

Is that the slow down some people talk about?

I guess you don't want to disclose your timings, but if you run latest beta of ZenTimings, could you spot some differences in non user-configurable timings? I don't have 5000 series to check myself.

Is it strap related? For example, is it the same if you use the lower strap and clock the bclk up or is it just some internal strap that makes things waaay slacker?

@I.nfraR.ed it seems like it's way slower at both 4/5Ghz on this end. I can even train 2133/4266 but not stable. I'm not sure I know about this slow down, but simply running pi 1M you can see the difference, as well as checking with aida64 as all memory operations are slower at 2100 IF vs 2033 IF.

Will try the latest beta of ZT and compare, the copy I had wouldn't open so I'm glad to see you it's been updated for Vermeer.

As for bclk the Aqua doesn't seem to like bclk, or could simply be user error :D I don't know if it's memory divider but even 1900/3800 seems super fast as well.

 

  • Thanks 2
Link to comment
Share on other sites

probably error correction feature in cpu causing slowdown.

bclk on asrock x370 was limted to like 110 previously have to kill all devices no nvme best to physically unplug the bluetooth/wifi from board as it knocks it out will probably break it. pci e gen 1 etc etc all the AMD bclk tricks will be needed for it to maybe work.

 

 

Edited by chew*
  • Thanks 2
Link to comment
Share on other sites

Today are two good Superpi32M results by AMD.

First 32M (for short time AMD WR) 4m 51.494s by @gnasherrr

https://hwbot.org/submission/4622720_gnasherrr_superpi___32m_ryzen_9_5950x_4min_51sec_494ms

 

Later today @safedisk did current AMD WR in 32M 4m 45.918s on Ryzen 9 5900X

https://hwbot.org/submission/4622833_safedisk_superpi___32m_ryzen_9_5900x_4min_45sec_918ms

image_id_2432154.jpeg

image_id_2432157.jpeg

  • Like 2
Link to comment
Share on other sites

I've tried all the tweaks from the old Bulldozer Conditioner (yes, I've disassembled it some time ago and know exactly what it does), also some of my own test tweaks I was working on for the previous zen generations. If anything, one of the registers makes it extremely slow and the whole system gets laggy.

Most of the registers are undocumented, the BKDG for 15h has more information than the document for 17h, but there's probably no parity between all the features and a specific bit in a register on 15h might be something else on 17h. Also Zen doesn't support Trailing Bit Manipulation (TBM) instruction set or at least the matching bit is marked as "Reserved" and that's one of the significant tweaks from BDC.

Others don't really improve the performance in SuperPi. They had some positive effect on Gen1 last time I tried, but it seems all performance bias tweaks aren't applicable on Vermeer, perhaps nothing significant to tweak anymore with this gen.

My modded XP is still working with Vermeer, but is slower than untuned Windows 10. Wazza doesn't help either, so I guess it is game over for XP. I've noticed it was slow with 3000G. It was fine with first gen though, perhaps with Zen+, too.

Back on Windows 10 now.

PS: Untweaked Windows 7 is also slower for me.

PS2: At least now I know which register kills the performance completely and will try to tune the old perf bias options.

Edited by I.nfraR.ed
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...