mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2020-11-01, 01:33   #1
Ethan (EO)
 
Ethan (EO)'s Avatar
 
"Ethan O'Connor"
Oct 2002
GIMPS since Jan 1996

2×32×5 Posts
Thumbs up ECC Confirmed Effective on Ryzen 7 3800xt / x570 Motherboard / Windows 10 20H2

ECC is now available on mainstream platforms, the memory itself is quite reasonably priced, and it works :)

I recently built my first AMD PC in well over a decade - I picked up an Asus x570 motherboard planning to put a 5xxx chip in it in November, and bought an open box Ryzen 7 3800xt on ebay for $290 to work with until then.

I populated two memory slots with Micron MTA18ADF4G72AZ-2G6 modules, spec'd as 32GB Dual Rank DDR4-2666 CL20 EUDIMMs. These were about $160/each on Amazon.

Although specified as DDR4-2666 CL20, the modules I received were populated with DRAM qualified for 3200CL19, and they are currently running happily without issue at 3733 CL16 with temperatures around 45C - a nice bonus of the ECC modules is temperature sensing.

ANYWAY, back to subject. I deliberately made the memory unstable by dropping the memory voltage until I started seeing ECC errors thrown in memtest86+, booted into Windows, and completed a DC-LL assignment of 56690237. Over the course of the run, Windows registered over 2000 WHEA events, corresponding to corrected ECC hits, but no errors were thrown in prime95 and the residue matched - great!

By lowering the CPU voltage carefully I was able to confirm that ECC is working for the L3 cache as well - ECC errors there are logged separately from main memory errors.

On a system like this, using prime95 for memory stability testing accordingly means running a stress test or assignment and watching the WHEA count since prime95 doesn't sense any error.



As an aside - the Lian Li LanCool II Mesh Perfomance is the best case I've ever used at keeping GPU temps under control. Just leagues ahead. And the case is physically a delight to work with as well. Highly recommended for continuous high wattage work.

Last fiddled with by Ethan (EO) on 2020-11-01 at 01:36
Ethan (EO) is offline   Reply With Quote
Old 2020-11-01, 02:16   #2
moebius
 
moebius's Avatar
 
Jul 2009
Germany

461 Posts
Default

I don't think that you will be completely happy with the 3200 CL19, although ECC Ram is of course more reliable, I have a Ryzen 3800X with 3000 Cl 18 memory modules, which also runs stable at 3200 Mhz (no error with Prime95 so far), CL 14 RAM would probably bring you 30% more speed at PRP.
moebius is offline   Reply With Quote
Old 2020-11-01, 02:24   #3
Ethan (EO)
 
Ethan (EO)'s Avatar
 
"Ethan O'Connor"
Oct 2002
GIMPS since Jan 1996

2×32×5 Posts
Default

Quote:
Originally Posted by moebius View Post
I don't think that you will be completely happy with the 3200 CL19, although ECC Ram is of course more reliable, I have a Ryzen 3800X with 3000 Cl 18 memory modules, which also runs stable at 3200 Mhz (no error with Prime95 so far), CL 14 RAM would probably bring you 30% more speed at PRP.
Oh, it's running at 3733 CL16-20-16-16 with no ECC errors. I had to lower the voltage at that speed to get it to throw errors. It will run at 3400 CL14 but for my primary application it performs faster with the higher clock.

I sort of took a bet that these modules would perform well beyond their specs, because the DRAM chips this dense are are very new and running on a process Micron has really dialed in.
Ethan (EO) is offline   Reply With Quote
Old 2020-11-01, 06:32   #4
phillipsjk
 
Nov 2019

5×13 Posts
Default

Nice work.

I should look into how to check the logs for ECC errors on my own systems.

I have been migrating to ECC RAM since deciding to store my data on ZFS with single redundancy (double redundancy planed for the offsite backup). At that point my greatest fear was a bitflip in RAM corrupting my data on disk.

BTW, ZFS works as well. My initial array was a bunch of hand-me-down disks: half of which were "free" because they were pulled out of service due to data errors (but took a reformat with no errors). 2-3 drives failed (one at a time) with no data loss. Being able to check data integrity (within hours) after a move is nice.

I think I may go with mirrored XFS for my "workstation": and trust the drive firmware to detect data errors at the sector level. I would trust that more than BTRFS available on GNU/Linux without licensing issues (plus it would be faster anyway). (My server with ZFS is FreeBSD.)
phillipsjk is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
New confirmed pi(10^27), pi(10^28) prime counting function records kwalisch Computer Science & Computational Number Theory 34 2020-10-29 10:04
Ryzen help Prime95 Hardware 9 2018-05-14 04:06
PIV Effective Equivalent curiosities... petrw1 Software 0 2009-12-05 04:41
What to do after my DCs are confirmed? edorajh Software 2 2003-12-30 15:28
Effective way to generate prime numbers (infinitive) Evgeny Dolgov Math 1 2003-12-08 09:25

All times are UTC. The time now is 23:54.

Thu Dec 3 23:54:14 UTC 2020 up 20:05, 1 user, load averages: 1.15, 1.26, 1.24

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.