• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

5800X no OC crashes AIDA64

Associate
Joined
29 Jun 2009
Posts
549
Hi @MrPils
First I tested SOC/IOD/CCD/VDDP at 1100/1050/900/900. CPU NB LLC on mode 1 (most aggressive) did not spike the SOC voltage at all under load and didnt go over 1.1v at all in either Ryzen master or HWiNFO64... you were correct as I expected.
I then went about testing IF1900/DDR3800
Reset to default settings, disabled XMP and increased SOC/IOD/CCD/VDDP to 1200/1150/975/975. CPN NB LLC on mode 1
Booted into Windows fine and ran a few quick tests like Membench default. SOC/IOD/CCD/VDDP all showing correct
Next lowered voltages to 1175/1125/950/950, rebooted and tested. All good
Started settings timings - TRRDS/TRRDS/TFAW/TWTRS/TWTRL to 4/4/16/4/8 booted into windows and tested. All good. I noticed gear down mode was enabled
Disabled GDM... Windows ran into problems. 02 code on reboot and had to reset bios. Luckily I had started saving a profile to test IF 1900.

Should I continue making changes to primary latencies with GDM enabled with IF1900?

At the moment I just booted into Windows with IF1800/DDR3600 and CAS14 timings, 1100 SOC

Not surprising regards gear down mode as you're running 32GB. Carry on with GDM enabled and see where you get to, 1T command rate is not worth raising TRRD/TWTR timings for. You can do a quick check when youre done with GDM enabled tuning to see how much of a hit you'd have to take on your chipset latencies if you were to go with 1T cmd @ 3800mhz, I'm betting it wouldnt be worth it though :)
 
Soldato
OP
Joined
6 May 2009
Posts
19,923
Not surprising regards gear down mode as you're running 32GB. Carry on with GDM enabled and see where you get to, 1T command rate is not worth raising TRRD/TWTR timings for. You can do a quick check when youre done with GDM enabled tuning to see how much of a hit you'd have to take on your chipset latencies if you were to go with 1T cmd @ 3800mhz, I'm betting it wouldnt be worth it though :)
Doesnt GDM allow memory to run at half speed sometimes? So if enabled and it kicks in would get a huge drop in performance
Wouldnt that be worse in the long run than having 3600Mhz / 1800IF, tighter timings and no GDM kicking in

I can of course try for 1900IF and keeping TRRD/TWTR the same whilist leaving GDM enabled then tune the main timings

edit - gdm only works with even timings doesnt it? Would this rule out cas15 and either fall back to cas16 or drop to cas14 & crash
 
Last edited:
Associate
Joined
29 Jun 2009
Posts
549
Yes GDM only works with even numbered CAS Latencies, though you should still be at CL14 anyway so I don't think that will matter much. Youll likely need a combination of looser tRCDRD and GDM enabled to get 3800mhz stable.

GDM doesnt make memory half speed, its half way between 1T and 2T command rates. It allows for 2 wait states on either the read or write operation (cant remember which) while allowing the other to run at 1 wait state so its quite literally 1.5T command rate. Going from 1T to 2T command rate was never the end of the world, so the effect of going from 1T to 1.5T is less of a hit than that but it does have a disproportionate increase in stability compared to the loss in performance. This means its often worth the small loss in performance in exchange for higher speeds and/or tighter timings when using multi rank memory configurations.
 
Soldato
OP
Joined
6 May 2009
Posts
19,923
Gave it a try and seesm stable at 3800Mhz / 1900IF but does require 1.5V DRAM in bios which shows 1.528V in HWinfo64.
1.48V in bios resulted in errors running DRAM tests / memtest.

I also ran the Extreme1@anta777 config in TestMem5 v0.12 for 15minutes. No errors and didnt feel too hot to the touch

A slight increase in membench scores but wouldnt say a huge difference over 3600Mhz 14/14/14/28-1T

uDm5VuP.png

q7JTV5a.png

lLqY6mg.png
 
Soldato
Joined
29 May 2005
Posts
4,899
No its normal. Single CCD chips have half write throughput and L3 cache is slower (because its bigger) on Vermeer cpus overall
makes sense regarding the writes. tho, I was talking about the L1 L2 cache. they are slower on Vermeer. thats single CCD again
 
Last edited:
Soldato
OP
Joined
6 May 2009
Posts
19,923
It’s like that dual rank single rank business.
Sorry, not sure if you are confirming or still saying its wrong :confused:

Comparing to posts on overclock.net, everything is normal. https://www.overclock.net/threads/o...24-7-memory-stability-thread.1628751/page-247

If anyone else with a 5800X wants to post AIDA64 cache & memory benchmark scores, feel free to prove me wrong

Just passed 3 hours of AIDA64 stability testing too with a max temp of 85c so think im good on the current settings
Thanks again @MrPils for all your help, this is addictive stuff and no doubt will try other things with the config... maybe something like CAS12 3400. Though not sure itll be worth the effort
 
Associate
Joined
29 Jun 2009
Posts
549
Sorry, not sure if you are confirming or still saying its wrong :confused:

Comparing to posts on overclock.net, everything is normal. https://www.overclock.net/threads/o...24-7-memory-stability-thread.1628751/page-247

If anyone else with a 5800X wants to post AIDA64 cache & memory benchmark scores, feel free to prove me wrong

Just passed 3 hours of AIDA64 stability testing too with a max temp of 85c so think im good on the current settings
Thanks again @MrPils for all your help, this is addictive stuff and no doubt will try other things with the config... maybe something like CAS12 3400. Though not sure itll be worth the effort

You're welcome dude - and yes, very addictive too. There's far more options with memory to get good performance than there are with CPUs these days, it makes a change having things to tinker with.

It also seems you have attracted your very own personal troll...I can't help but think that's what's going on now. Your AIDA scores are perfectly fine, especially the latency - below 54ns is great :)
 
Soldato
OP
Joined
6 May 2009
Posts
19,923
L3 cache does look like either a bug in AIDA64 or a BIOS issue.
Comparing to a screenshot of a Ryzen 5 3600 with the same RAM, L3 cache was 563 / 491 / 574

You're welcome dude - and yes, very addictive too. There's far more options with memory to get good performance than there are with CPUs these days, it makes a change having things to tinker with.

It also seems you have attracted your very own personal troll...I can't help but think that's what's going on now. Your AIDA scores are perfectly fine, especially the latency - below 54ns is great :)

:D who doesnt like their own troll. L3 cache is probably incorrect, as above
 
Last edited:
Caporegime
Joined
17 Mar 2012
Posts
47,648
Location
ARC-L1, Stanton System
Some motherboards are a bit dumb when it comes to PBO, some just allow the CPU to run away with volts if the CPU needs it or not, which 9 times out of 10 on a 5800X it doesn't.

So it just runs too high volts too hot and you don't gain anything from it, at best more often than not it can result in lower performance.

Set PBO to advanced, make sure: (PTT 142 Watts - TDC 95 - EDC 140) are unchanged, set the curve optimiser symbol or value to Negative and play with the values, so 5, 7, 10, 12, 15.... and maybe with a boost off-set, +50Mhz, +100Mhz... that is the way to tune a 5800X and you will get results.
 
Back
Top Bottom