• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Can someone know what cause low single score and high latency?

pushing the IMC too hard can result in the error correction going nuts which kills the performance
Id be interested to see what it does at 3600 1:1 your fsb ratio is pretty dam high it might be hurting you
 
Post zentimings please.

Edit: If zentimings are the same as last, try the following: vsoc 1.12, ccd 0.94, vddg 1.04, vddp 0.9. Running these timings that high isn't usually required at 1900 IF and steals from the pwr-budget.

You could try flat 14, tras 28, trc 40, trfc 240, see if performance drops?

For comparison I run 4000cl16 flat, 288trfc, gdm on and I get 52.2ns and 60k, 32k, 55k på aida. My ramkit is a shit-bin. I run the voltages described above, but slightly lower on ccd and vddp.
zen.png


can u share your 4000/2000 timming?
 
reinstall window, single score is still not I want. How to get 280 r15 and 700 cpuz? any tips?
View attachment 216860

Did you try the suggestions earlier to try using PBO only without the OC switcher and using Cine R23? You still haven't determined whether your chip just doesn't have the ST performance, or if the OC switcher is messing with your ST scores.

685 should be 4925-4950MHz effective clock. If you're not at 5.0GHz or better, effective clock not "core clock", on the core that the bench uses (usually Core 0), you will not see 700. That's like blindly hoping for 1700 R23 ST when your chip doesn't have that potential. Hit those boost clocks reliably in HWInfo, then we'll talk about the scores you want. Use Snapshot Polling in HWInfo settings.

And Cinebench is terrible at gauging memory performance. Run a few times through the Membench app integrated into DRAM Calculator and see what time you get on Easy and Normal. Third tab at the top of the app.

 
for me, 3800/1900 on my 5900x is the limit, even with higher VSOC, everything over 3800/1900 gets me lower scores
 
for me, 3800/1900 on my 5900x is the limit, even with higher VSOC, everything over 3800/1900 gets me lower scores
That's cause you're overvolting the VSOC. Increase VSOC too much and you'll see lower single/multi boost frequencies.
 
That's cause you're overvolting the VSOC. Increase VSOC too much and you'll see lower single/multi boost frequencies.
you missed the "EVEN" in the sentence. Even on stock VSOC the score gets lower if i breake the 1900 barrier.
 
you missed the "EVEN" in the sentence. Even on stock VSOC the score gets lower if i breake the 1900 barrier.
What is your stock vsoc? Could be iod aswell. My experience is that higher fclk than 1900 requires higher vsoc and iod, but often lower ccd and vddp. I have only OC to 5600X to 2000+ fclk, but it worked the same on them both.
 
Pro tip... OP when you screenshot, select the window you want to screenshot and press Ctrl + Alt + PrtSc then just paste it here with Ctrl + V instead of posting your entire desktop that people then have to click on and zoom in on text. Better for privacy too only sharing the SS from the focused window. Plus then you don't have a dumb local SS file lingering on your system.
1631642420081.png
 
Last edited:
Does it pass tm5 extreme? Dual rank is harder to do it, but my cpu can do 4000/2000 but have error with tm5.
 
Does it pass tm5 extreme? Dual rank is harder to do it, but my cpu can do 4000/2000 but have error with tm5.
Yeah, did 25 without errors. I have another setting with slightly better timings, but it required a bit more dram voltage and barely made a difference to performance. Doing CL15 2T or 14 with GDM is impossible unless I run ram at 1.55V+ which makes it overheat.
 
You may want to try to lower the CPU LLC in the BIOS to improve single core score a bit.
 
Using llc2 now.
I add in attachment my BIOS settings.
I have a 5900X on an Asus Crosshair VIII Formula, and I am using the power plan Ultimate Performance from this thread here
I usually am in the 685-690 range for the single core score in CPU-Z, don't know if you can find here anything else that may help.
 

Attachments

Last edited:
I add in attachment my BIOS settings.
I have a 5900X on an Asus Crosshair VIII Formula, and I am using the power plan Ultimate Performance from this thread here
I usually am in the 685-690 range for the single core score in CPU-Z, don't know if you can find here anything else that may help.
can't download any link in that thread? can u upload the ultimate power file? thanks.
 
can't download any link in that thread? can u upload the ultimate power file? thanks.
Here it is. All credits should go to ManniX-ITA for this.
 

Attachments

You can run dual rank at 2000 1:1?

Impressive.
It can post and do aida64 mem bench but cinebench got a performance hit. No idea what cause it coz previous owner can do it fine.
 
Does it pass tm5 extreme? Dual rank is harder to do it, but my cpu can do 4000/2000 but have error with tm5.
If you use GDM passing tm5 is much easier with lower voltage, but you loose a bit of performance, and you need CL, CWL, RTP, WR to be an even number. 2T can be an option, not as stable as GDM, but much easier to stabilize than 1T and slightly faster than GDM it seems.

What errorcode in TM5 do you get?
 
Performance losses at 4000/2000. What do I need to do?
 
Performance losses at 4000/2000. What do I need to do?
How is core speed at load on 3800 vs 4000? Higher ns i aida vs 3800? If soc/iod is too low it increases latency, if it's too high it can eat into you powerbudget.
 
Core speed is the same as 3800. 2ms latency reduction with 4000/2000 but multicore score got nerfed.
 
Core speed is the same as 3800. 2ms latency reduction with 4000/2000 but multicore score got nerfed.
In linpack I get the same with 4000 and 3800, in SOTTR I get about 5 more fps with 4000cl16 vs 3800cl15. Aida and other benches are better with 4000.
 
Back
Top