Can someone help me with getting my ram stable?

[Erazem]

Hi,

I'm having problems with ram stability in some very specific scenarios, to be more precise while using handbrake to transcode 4k video to av1 I'd will eventually get a system crash with a whea 46 error (supposedly related to memory) in the event log. Only happens with handbrake transcoding 4k video to av1. Everything else is stable (can pass hours off occt cpu/ram tests, test mem with extreme anta profiles/1usmus profiles, hci MemTest with well over 300% coverage. I know I'm missing something here so. System specs are in my profile but just in case, I have a 5600x, gigabyte b550m gigabyte b550m aorus elite and 2x16GB ddr4 g skill f4-3200C16-16GTZKW. And am running them at these settings (all confirmed to be stable by corecycler (with prime95, ycruncher, aida64), occt, ...):

ppt - 115W
tdc - 125A
edc - 125A
boost override +125 (which somehow makes the max cpu speed 4775MHz

curve offsets:

c1 -2
c2 -20
c3 -5 (best core)
c4 -24
c5 -24
c6 -12

temps are around 60-70 (it's summer and room temps can get to 30 degrees C so...)

the ram timings/frequency/.. set by the board when xmp is enabled:

those suggested by dram calculator for zen as safe are:

and those in the xmp profile are:

I've tried increasing the dram voltage to 1.38V and soc voltage to 1.08V, neither of which helped.

Any ideas? Or did I just lose the silicon lottery and that's it?

I'd be happy with it being stable at 3200 and xmp timings, but if somehow I could get it a bit higher (3400?) than that it would be even better, but increasing the frequecy even to like 3333 results in crashes all over the place so I dobut that's going to happen.

 

[ShrimpBrime]

I would try command rate at 2T instead of 1T and leave the rest alone where you had it/like it. Then do some testing.

GL!

 

[Erazem]

I would try command rate at 2T instead of 1T and leave the rest alone where you had it/like it. Then do some testing.

GL!

Thanks, I'll change it to 2t next time I reboot my computer, I hope it helps.

Do you think it's the cpu memory controller not being that good, the mother board or the ram?

Because I'm going to get a new power supply (honestly the 600W lc power one has lasted a lot longer than I expected it would since everyone said lc power is crap, but it's time to upgrade it because I'm planning on doing some other upgrades too which I doubt it can handle + when I bought the current case a few weeks ago I had to get an extension cable for the 8 pin 12V mother board power cable) and I was thinking of getting a 5900x (or maybe a 5950x both have gotten pretty cheap these days and I care more about multithread performance than single thread), 2x32gb ddr4 3600 (in the next month or two) and a 7900gre (or whatever will have a decent price/performance then, preferably amd because imo nvidia cards are overpriced for what they offer and I don't like the whole 12GB vram in all at least somewhat sanely priced cards) in autumn (to hot for gaming now anyway). But if I'm already buying a new cpu and ram maybe I'd be better off going for am5 (perhaps a 9900x) since it's going to be faster and more upgradable so I'm not sure not sure which is better.

Sticking with am4 is cheaper, I don't have to replace the mother board (which is kinda annoying to do lol), while am5 is more expensive (I'm especially bothered by motherboard prices because if I want one with what I want (decent overclocking potential, 4 nvme slots, 4+ sata ports) it also means paying for things I don't need/want like wifi), faster, can take 2x48gb ddr5 and more upgradeable.

The choice is even harder since I don't know what is limiting the memory speed/latencies/stability. If it's the motherboard than I'm better off just going to am5, but if it's the memory/cpu than the new cpu and ram might not have any issues with any of that.

 

[ShrimpBrime]

I'd say the board could be the issue for a couple reasons, mainly the chipset and tier. And bios revision.

When you do upgrade, just remember more memory is for redundancy and less performance. Typically it's harder to run XMP profiles with a lot of memory installed. Purchase what you need only.

I made the jump to DDR5 shortly after release, which is nearly 2 years ago now. So if you upgrade, I think a platform switch to DDR5 would be wiser, opinion of course. So when you are ready, perhaps make a build suggestion thread so you can get ideas and options/opinions.

 

[ir_cow]

Need to start at square one. Do you know if it's stable without your tweaks applied?

If it passes memory stress test, preceded to do some adjustments one at a time. Ryzen Calculator is just a suggestion. It isn't perfect because not all CPUs, Memory and motherboards are of equal quality.

 

[Erazem]

I'd say the board could be the issue for a couple reasons, mainly the chipset and tier. And bios revision

I have the latest bios/agesa. And yeah I know, it's not the best board/chipset available.

When you do upgrade, just remember more memory is for redundancy and less performance. Typically it's harder to run XMP profiles with a lot of memory installed. Purchase what you need only.

I'm running into memory constrained scenarios relatively often (virtual machines (running just 3 with 8GB for each of them and oops 24GB is gone already), 7zip compression (the highest levels of compression use like 20GB+ while compressing) + I'm one of those people that have 1000+ tabs (I think my record is above 3000) open in several browsers (yes I'm totally going to read all of them and no I won't use bookmarks ).. even on my phone the tab count is usually aka 100+).

Need to start at square one. Do you know if it's stable without your tweaks applied?

Huh, good question, I was using the stock cooler for years and was basically thermaly constrained because the cpu would hit 95°C easily, buying the cooler I have now (Thermalright Phantom Spirit 120 SE) was basically like getting a new cpu because suddenly I was no longer thermaly constrained and the cpu could actually sustain all core/thread loads at 4400-4600MHz (depending on application) which got me started on this whole tweaking thing.

All I know for certain is that it is stable at my current settings doing everything except transcoding 4k videos to av1 using handbrake. No, wait, that's too simplified.

There's one movie (some Iron Man 2 4k hdr hevc rip) that will make it crash, I'm not even sure if others would too, because I have transcoded other 4k hdr hevc movies to av1 without any crashes.

 

[ShrimpBrime]

I have the latest bios/agesa. And yeah I know, it's not the best board/chipset available.

I'm running into memory constrained scenarios relatively often (virtual machines (running just 3 with 8GB for each of them and oops 24GB is gone already), 7zip compression (the highest levels of compression use like 20GB+ while compressing) + I'm one of those people that have 1000+ tabs (I think my record is above 3000) open in several browsers (yes I'm totally going to read all of them and no I won't use bookmarks ).. even on my phone the tab count is usually aka 100+).


Huh, good question, I was using the stock cooler for years and was basically thermaly constrained because the cpu would hit 95°C easily, buying the cooler I have now (Thermalright Phantom Spirit 120 SE) was basically like getting a new cpu because suddenly I was no longer thermaly constrained and the cpu could actually sustain all core/thread loads at 4400-4600MHz (depending on application) which got me started on this whole tweaking thing.

All I know for certain is that it is stable at my current settings doing everything except transcoding 4k videos to av1 using handbrake. No, wait, that's too simplified.

There's one movie (some Iron Man 2 4k hdr hevc rip) that will make it crash, I'm not even sure if others would too, because I have transcoded other 4k hdr hevc movies to av1 without any crashes.

I'm completely opposite. I OCD at like 4 tabs. At 10 I may sometimes randomly close the browser. Phone or PC. But a few lists of bookmarks save I don't dare to try and count. 3/4 of the time can't find what I need and just re-google search some shit. Lol.

 

[Erazem]

I would try command rate at 2T instead of 1T and leave the rest alone where you had it/like it. Then do some testing.

GL!

Ok, I changed the command rate to 2T (which also made me have to disable gear down mode for it to stick) and got the following:

Tried transcoding the problematic movie and it crashed in a few second afterwards (usually it takes hours), so I changed it back to gear down mode on, command rate 1T:

Any other ideas?

 

[ShrimpBrime]

Ok, I changed the command rate to 2T (which also made me have to disable gear down mode for it to stick) and got the following:

View attachment 353990

Tried transcoding the problematic movie and it crashed in a few second afterwards (usually it takes hours), so I changed it back to gear down mode on, command rate 1T:

View attachment 353991

Any other ideas?

What an interesting outcome. I didn't expect that, generally T2 brings better stability.

Gear down just uncouples the memory controller from the DRAM frequency and also allows for compatibility of odd number cas latencies.

If you default the system, does it remain stable that way?

 

[ir_cow]

So are you manually entering all those values or is it what the motherboard does by default with XMP profile is enabled?

I ask because some of those values look extra low for it to be auto unless you also have a low latency mode enabled as well.

 

[GerKNG]

your Curve Optimizer is not stable.

all confirmed to be stable by corecycler (with prime95, ycruncher, aida64), occt

i can run -30 all core in all of these stresstests for months but not idle or play games without random crashes and system lockups until i lower it to -8.
Curve Optimizer is always an issue at idle, low load, transients. never at high load scenarios.

 

[Erazem]

What an interesting outcome. I didn't expect that, generally T2 brings better stability.

Gear down just uncouples the memory controller from the DRAM frequency and also allows for compatibility of odd number cas latencies.

If you default the system, does it remain stable that way?

Right now I'm back at the original settings. Which are stable (except for that one transcode afaik). I tried setting some very high timings yesterday:

and it completed 1h of occt memory test (85% memory, sse, threads auto), 1h occt (85% memory, avx2, threads auto), a few runs of testmem 5 1usmus_v3 and 1 run of testmem 5 Extreme1@anta777

I tried even higher timings (I don't remember what exactly but something like tCL 24, not sure about the rest, but all very high) and mclk 1700, fclk 1700, uclk 1700 + very high voltage for memory and it didn't boot so I had to reset the cmos.

I think it's safe to say I'm not getting mclk 1700, fclk 1700, uclk 1700 with this memory/cpu/motherboard no matter what.

But I think that if I raise the timings slightly it might end up being stable at mclk 3200, not sure will try on next reboot.

So are you manually entering all those values or is it what the motherboard does by default with XMP profile is enabled?

I ask because some of those values look extra low for it to be auto unless you also have a low latency mode enabled as well.

The timings from the pictures posted above are what the motherboard sets when I enable xmp. Only thing that was changed by hand is the soc voltage and in the experiment with the gear down mode/command rate the gear down mode and command rate.

your Curve Optimizer is not stable.

i can run -30 all core in all of these stresstests for months but not idle or play games without random crashes and system lockups until i lower it to -8.
Curve Optimizer is always an issue at idle, low load, transients. never at high load scenarios.

But it's not crashing at idle/low load/tranisients, it's crashing when transcoding a certain movie from 4k hevc hdr to av1. I can be idle, post online, ... all the time with no crashes. As I've said before, the only way I can get it to crash is if I open handbrake and start transcoding some Iron Man 2 4k hevc hdr rip to av1. Everything else doesn't crash.

your Curve Optimizer is not stable.

i can run -30 all core in all of these stresstests for months but not idle or play games without random crashes and system lockups until i lower it to -8.
Curve Optimizer is always an issue at idle, low load, transients. never at high load scenarios.

Just to add, I did get crashes at idle and the computer wouldn't shut down (the os would turn off but the fans remained spinning) previously when the curve offsets weren't stable. In fact the computer not shutting down was the last symptom to go away (I assume it has something to do with core0 being more important than the others for some things).

 

[GerKNG]

and you get crashes while transcoding this certain movie after a cmos reset as well?

 

[Erazem]

and you get crashes while transcoding this certain movie after a cmos reset as well?

Unfortunately yes

I think I should probably relax the timings a bit, just not sure which ones are most likely to solve the issue with minimal per formance impact?

Maybe 18:20:20:40?

Or should I try uncupling the fclk/uclk/mclk and try changing each to see if it's the cpu or the memory, and if it happens to be the memory is there any point where it makes sense (aka the increase in the fclk/uclk frequency makes up for the extra memory access latency due to having them uncupled)?

 

[GerKNG]

you get crashes without any overclock and XMP? sounds like faulty hardware then.

 

[sneekypeet]

I hate to be that guy, but Im looking at that PSU as a possible culprit.

 

[Zach_01]

Have you tried it without negative V/F Curve (CurveOptimizer) at all?
Also did you check if Windows Event Viewer is stored any errors on those crashes? (error ID)

 

[Erazem]

It's always error 46, which is supposed to be related to memory..

I've tried the same settings except with cas set to 18 and it still crashed.

Now since some of you have suggested that it could be the curve offsets and because I was the one who asked for help and thus ignoring suggestions would be kinda dumb (I mean why ask for suggestions if you're going to ignore them?) the next test will be to reduce them by a lot and see what happens. Something like 0, -10, -2, -10, - 10, -4 (original was -2, -20, -5, -24, -24, -12). But that will have to wait for tomorrow, it's almost time to sleep and it's hot enough in the room without a computer acting as a space heater.

you get crashes without any overclock and XMP? sounds like faulty hardware then.

Could be, but I think it's all out of warranty (afaik?) and I don't have any other am4 cpu/motherboard nor any ddr4 to test that theory.

I hate to be that guy, but Im looking at that PSU as a possible culprit.

That's getting replaced soon (in a few weeks) in any case (because it's ~10 years old (and I'd rather replace it before it dies in a way that will damage something else), probably too weak to handle any more mayor upgrades (especially gpus/cpus since the higher end gpus can use a lot of power as can cpus with lots of cores) and because since I have a new case with the psu at the bottom I had to use an extension cable for the 8 pin 12V motherboard power connector which isn't really an ideal solution).

But let's first see if the much less aggressive curve offsets help. Since that requires just a change in settings and no new hardware.

 

[Zach_01]

Also, is this a new system? Did this Event ID 46 appeared from the early days of the system?
Googling this I found that this may be caused by CPU OC or CPU undervolt or even faulty CPU.

 

[mechtech]

A reason why I like 3200 jedec. If xmp gives issues, still get 3200 with no issues, but loose timings, which is better than 2133 jedec.................

 

[Erazem]

A reason why I like 3200 jedec. If xmp gives issues, still get 3200 with no issues, but loose timings, which is better than 2133 jedec.................

View attachment 354293

Also, is this a new system? Did this Event ID 46 appeared from the early days of the system?
Googling this I found that this may be caused by CPU OC or CPU undervolt or even faulty CPU.

The cpu/motherboard/memory were bought in 2021

Anyway, I tried with the curve offsets disabled and it still crashed