I had one system up on 4011 for a while but ran into stability issues among other things. Since the 4008 BIOS was running stable on the other system and that system has now become the most stable system I backeleveled the 4011 to 4008. But I realized later that that system's stability erosion had more to do with the cpu overclock. Backed the cpu down and it is again stable so the 4011 BIOS was not to blame. I need to get both of them up on 4011 and see if I can get stability at > 3200 CL14 which is where I am at now.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

The system was not stable at 3400.

Had a blue screen after some time. Was also almost too good to be true :)

Now I'm at 3333, lets see how that works.

The fact that its able to boot at these settings is something new to my setup :)

---

There are so many variables with regard to memory tuning that stability takes weeks if not months to arrive at. I get my best information from the Ram Calculator and Memory Stability threads over at Overclock.net forums. Along with the thread for my current motherboards I have been able to reach stability at 3200 and CL14 Safe timings. I had stability before on earlier BIOS' at 3333 CL14 Fast timings but that degraded with later BIOS'. Maybe the 4011 BIOS will allow me to get back to 3333 or even 3400.

I have also started to read the Crosshair VII Hero thread since I think that is the motherboard I am going to buy for the Windows 7 computers upgrade. I see lots of reports of 3533 being stable on that motherboard along with the 2700X.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Yes, there are many many variables, and I just don't understand the technology well enough to do educated adjustments.

I'll take the standard timings, and push the frequency as high as it will go, and leave it at that (for now).

3333 crashed after about 45 minutes.

But 3266 does seem more stable. Its at about 1h45 mins now, so improved, but not definitively reliable yet.

Time will tell. But 4011 is definately an improvement on my system. With 4009 I never reached 10 minutes without crashes at 3200.

---

Well I spent a good portion of the afternoon trying to get 3400 running. I actually did several runs of PassMark on both Safe and Fast settings and they passed wonderfully. Finally decided to run a half hour of stressapptest and it threw 47 errors in the first 5 minutes. Dummy. Should have started with stressapptest first and then if good moved on to PassMark.

At least I profiled what if/any improvements in system performance was due solely to memory clocking. Just as has been posted, there is negligible improvement in system performance when moving one step up in memory clocks once you get to 3200 memory speed. Diminishing returns.

So I backed it down to 3333Mhz CL14 Fast settings and got back to crunching. Should have started there instead of jumping two whole steps in clocking. Wasted 3 hours of crunch time that I won't get back.

Will see if that is still BOINC stable as my previous 3200 CL14 Fast settings. I was stable at this speed in earlier BIOS'

Good to see it come back again.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

These new 4008/4011 BIOS respond well to just selecting the memory strap and leaving everything at Auto. I learned that trying to set EVERY LITTLE sub-timing that the RAM Calculator suggests just doesn't work anymore with these new BIOS'

Best is to select one level less D.O.C.P. setting than the actual RAM memory XMP rating and then choosing the memory speed you actually want in the Memory clock speed. That sets up the memory with better optimized secondary sub-timings than the actual XMP setting does which is too loose.

Then just set the primary timings that the RAM Calculator suggests like 14-14-14-14-28-42 and the tRFC setting. Then just leave everything else on Auto. These new BIOS' seem to be more stable with Geardown enabled which makes the command rate more like 1.5T instead of 1T. Also my memory seems to like 48 ohms better than the 53.3 ohms that was previously suggested and currently in the Calculator. That is a common observation in the forums for Samsung B-die memory.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Thanks for your comments.

I will see if Gear Down Mode (GDM) could perhaps make my RAM stable at higher speed, but I'll have to measure real world performance to see whats fastest (3200 vs. eg 3400 w. GDM).

48 Ohms have helped me in the past, so that'll have to be tried out as well, allthough that would purely be for reliability.

The system crashed at 3266 after many hours, so now I'm trying 3200, still with standard settings for now.

---

My 3333 CL14 Fast crashed in the night so I put it back to 3200 CL14 and went back to bed. Today I will go back and try again 3333 CL14 Fast with more memory and SoC voltage.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I was reading some articles about finding a good memtester that would run in Windows.

Finding that memtest64, wouldnt find errors, even with settings that eventually crashed, I tried to read up on it.

I found a guy promoting "RAM test" (by Karhu software). A program that it seems is basically made for testing RAM on Ryzen CPU's (will work on others too)
The downside was, it costs 10$ (9.99). After some time thinking about it, I decided to try it, and bought a license. Its not that much money anyway.

And it has proven to be a good decision. I can run the program at 3200 for an hour (and more) without errors, but at 3266 I get an error from the program within 20 seconds (at a setting where the system could run 12+ hours yesterday).
You can set the program to stop on errors or keep running. And you can set it to beep on eroors.

It has helped me very much in finding reliable settings. I found that setting back GDM to auto did nothing, the system is still stable (perhaps its on pr. default at 3200), but the ProcODT has to be set at 48 ohm on my system, higher than that I get memory errors.

I like the simplicity of it, and I like how fast it detects errors, at least on my system, when I push things too much.

---

I have been experimenting a little with Ryzen DRAM Calculator, and its so called "safe" settings.

I made settings for 3533, 3400, and 3200, all with the safe calculations.

3533 didn't boot. And I honestly didn't expect it to.

3400 did boot, but windows crashed loading the desktop.

Tried 3333 with the 3400 settings, and it booted and started correctly. Running RAM tester gave me an error within 1 minute.

3200 with safe settings, booted fine, but RAM tester threw an error after about 30 seconds.

3200 with bios defaults, but with 48 ohm ProcODT, runs fine. Default settings are a lot more relaxed, as far as I can see.

So, I guess I'll stick to 3200 at (almost) default for now, which was really also what I was hoping for.

My only problem with this is that the system should be able to run at the safe settings (at least at 3200), as they make sure that the specs of the actual ram is maintained. So the bios / agesa is not doing what it should.

---

That ram tester from Karhu Software now seems to be the tester of choice among the most active forum posters. As you stated, it finds errors the fastest and most consistently.

I wish I could find the equivalent for Linux. Google Stressapptest is the only solution for Linux I have found. If it passes that then I just have to put it into production load and hope for the best.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Well, then it seems that I, almost by chance, have found a good tool for the job.

It does work well, its nice to have a tool that this reliably finds errors relatively fast.

It does make experimenting and troubleshooting much easier.

---

That software only has shown up recently in the last six months or so on the forums. It got the nod from ASUS Raja and The Stilt as the best ram tester. Previously from the 2017 Ryzen launch, HCI Memtest was the popular choice. It still is in major use and is on the list of preferred RAM testers. Everyone that has used the Karhu Software tester says it is well worth the €10.

Here is the original link great-new-memory-stability-tester-ram-test.html

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Hey Keith,

So i tried that new software on my faulty computer and it came back with an positive error. So i guess now the trick is to pull all the ram stick out and run them 1 at a time to see which stick is the bad one?? Too bad it can't tell me which one is the faulty one so I don't have to test individually. Well at least it was able to find it. Now for a full day of testing

Z

---

Hi Zalster. Were you testing your RAM overclocking or just the sticks at stock JEDEC 2133? If the fault showed up with the RAM overclocked, then test again with the RAM backed off a bit from the existing overclock or have the system boot with defaults so the RAM tests at stock 2133. That would be the way to prove a stick has a hardware fault and not just from being pushed a bit too far with the RAM overclock.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Well, they are OC but at 2.8GHz when the error was detected. So now I am testing each stick individually to see if I can find out which stick is the problem.

---

Also just wanted to point out, that a simple memory error can be covered up or nullified by Windows own built-in memory error correction functions. They are known as WHEA errors and can be mostly benign in nature. IOW, you wouldn't see the error cause an application error. But if the memory error causes the system to write to a protected memory location because a bit flip changed the memory address that is more serious and almost always crashes the machine. That would be time to ask the vendor for a RMA replacement because the memory hard failed.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Found the faulty stick... I'm pretty sure it's out of warranty. But I'll check just to make sure. Now, trying to get a hold of 1 or 2 sticks of RAM that match. I doubt they sell single stick but might see 2 sticks. Guess I'll see... Thanks Keith..

Edit...

Hahahaha.... I bought them 5/18/2017 Almost a year to the day....

Edit 2..

I submit the RMA anyway. Will see what they say....

---

If memory is sold as a kit, whether two or four sticks, vendors usually want the entire kit back in a RMA, even if only one stick is bad. Let us know how you make out with the RMA.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

That will suck, that's 8 sticks all together...

---