What did you have to do to get 128GB ram working on Zen3 CPU?

[A Computer Guy]

Trying to get 128GB ram to work reliably with my 5950x but was encountering difficulties with WHEA errors 1, 18, and 47.
Reducing frequency helped the situation. Currently testing my 2nd 64GB ECC kit alone for any issues before retrying 128GB.

1
A fatal hardware error has occurred. A record describing the condition is contained in the data section of this event.

18
A fatal hardware error has occurred.
Reported by component: Processor Core
Error Source: Machine Check Exception
Error Type: Bus/Interconnect Error
Processor APIC ID: 0

47
A corrected hardware error has occurred.
Component: Memory
Error Source: Unknown Error Source

It's is possible that my CPU did not escape being damaged from my prior motherboard mishap.

5950x and PBO overheated something or busted custom loop?

If you mean regarding the block reassembly? I just do it by hand being careful not to overtighten them, approximating the torque I needed to remove them. I didn't have any leaks or cracks in the acrylic of any of my blocks so far. I should probably get the right tool for it for the long term...

www.techpowerup.com

 

[68Olds]

I think you need a 4x32GB kit. There are no 64GB sticks on the QVL.

 

[A Computer Guy]

I think you need a 4x32GB kit. There are no 64GB sticks on the QVL.

The configuration is 4x32GB (2 x 64GB kits, each Dual Rank) Also I can verify they are all using the same IC's (physically inspecting the module) and timings in UEFI/BIOS.

 

[68Olds]

I failed my reading comprehension quiz today.

 

[A Computer Guy]

I failed my reading comprehension quiz today.

Assuming my 2nd kit passing testing I'm hoping a bump in DRAM voltage and reduction in frequency will stabilize it but it got me curious to what other users experiences have been trying to max out the ram supported on Ryzen.

 

[TheoneandonlyMrK]

The configuration is 4x32GB (2 x 64GB kits, each Dual Rank) Also I can verify they are all using the same IC's (physically inspecting the module) and timings in UEFI/BIOS.

So 2X32GB kit's and your wondering why it isn't working.

The timings are set on each of your sticks for only 2x that stick not four.

4X 32Gb kit would have the right timings set for four stick's.

And the speeds you are attempting to run them at(setup for 2X32GB) are not possible with 4 sticks.

 

[ir_cow]

Memory' Frequency, gear ratio and motherboard is the main factors. I have no problem with 4x32GB at 3200. Start to do higher and that's when you need to mess with voltages or have a MB that is of higher quality.

Go to 3600-4000 and your playing with gear ratio and SoC voltage.

 

[A Computer Guy]

So 2X32GB kit's and your wondering why it isn't working.

I've mixed kits before without much issue.

The timings are set on each of your sticks for only 2x that stick not four.

4X 32Gb kit would have the right timings set for four stick's.

The previous 4x16 DR configuration I had didn't have this issue. But you have a good point.

And the speeds you are attempting to run them at(setup for 2X32GB) are not possible with 4 sticks.

4x32 DR at DDR4-3200 had been working mostly well except a handful of WHEA errors the most serious being a reboot when the system is left idle for long hours.
Obviously zero WHEA errors would be the goal at any frequency. Reducing frequency to DDR4-6667 alone seems to be working to eliminate WHEA 1 and 18 errors with stray 47's happening rarely.
Other interesting quirk is I can't post these 4 sticks at all unless I setup SOC, CLDO VDDP, VDDG CCD, and VDDG IOD voltages manually first before installing all four.
If someone has already been through this before they might already have tips that is likely to work and stabilize a configuration like this.

for current reference here is a current 64GB only configuration at DRAM 1.2v and at 1.2v 4 sticks does operate aside from the current WHEA issues noted.

 

[billeman]

Running 4 Dual Rank DIMMS is a pain, I can just run 1833(3633)C16 with that setup and even then for full stability I run it at 1800(3600)C16.
And indeed with higher voltages on SOC , VDDP , VDDG and a little higher VDIMM than the DIMM's spec.

 

[A Computer Guy]

Running 4 Dual Rank DIMMS is a pain, I can just run 1833(3633)C16 with that setup and even then for full stability I run it at 1800(3600)C16.
And indeed with higher voltages on SOC , VDDP , VDDG.

What voltages do you use? I can't really push he frequency on these kits (2x32) at all but 1.35v does allow me to push the timings down significantly to respectable CL16 @DDR4-3200. After verifying my 2nd kit didn't give me a dead stick, my next step was to see how they respond to 1.35v in 4x32 but I suspect I might need other voltage tweaks assuming I'm not dealing with the impossible.

 

[ir_cow]

Set that bad boy to 2T. GDM is nice, but not easy on the IMC with 4x dual rank

 

[TheoneandonlyMrK]

I've mixed kits before without much issue.

The previous 4x16 DR configuration I had didn't have this issue. But you have a good point.

4x32 DR at DDR4-3200 had been working mostly well except a handful of WHEA errors the most serious being a reboot when the system is left idle for long hours.
Obviously zero WHEA errors would be the goal at any frequency. Reducing frequency to DDR4-6667 alone seems to be working to eliminate WHEA 1 and 18 errors with stray 47's happening rarely.
Other interesting quirk is I can't post these 4 sticks at all unless I setup SOC, CLDO VDDP, VDDG CCD, and VDDG IOD voltages manually first before installing all four.
If someone has already been through this before they might already have tips that is likely to work and stabilize a configuration like this.

for current reference here is a current 64GB only configuration at DRAM 1.2v and at 1.2v 4 sticks does operate aside from the current WHEA issues noted.

View attachment 270090

You asked why, I told you why, you do you though.

Your integrated memory controller is the limiter that and piss poor configuration.

It doesn't take the same Trfc to run four stick's as it does two is one example.

I have run mixed pairs too, this is how I know what I told you.

As a base find the timings of an actual quad set, set looser versions of those and add more voltage as required but be realistic, the load on the IMC is at max so under load it will fail if pushed to fast.

 

[A Computer Guy]

You asked why, I told you why, you do you though.

Your integrated memory controller is the limiter that and piss poor configuration.

It doesn't take the same Trfc to run four stick's as it does two is one example.

I have run mixed pairs too, this is how I know what I told you.

As a base find the timings of an actual quad set, set looser versions of those and add more voltage as required but be realistic, the load on the IMC is at max so under load it will fail if pushed to fast.

I guess I have been lucky because in my previous endeavors Trfc was never a problem for me running 4xSR or 4xDR modules.

 

[TheoneandonlyMrK]

I guess I have been lucky because in my previous endeavors Trfc was never a problem for me running 4xSR or 4xDR modules.

I didn't mean to imply Trfc causes a problem with 4 sticks per say, just setting it up too low for four stick's would, it can often be a auto set option but some dual channel kits do actually specify it in which case running two dual kits can then have the kind of issues you mentioned whea ,memory related glitches but often only occasionally, when adequately warmed and loaded up in use for example, which is why I mentioned it.

Yours look automatically set ATM.

And I gave it as an example, dig around there's a lot of memory related timing settings, that you are trying to run mostly configured for two sticks but with four, that's why I suggested looking up quad kit tunings.

 

[A Computer Guy]

I didn't mean to imply Trfc causes a problem with 4 sticks per say, just setting it up too low for four stick's would, it can often be a auto set option but some dual channel kits do actually specify it in which case running two dual kits can then have the kind of issues you mentioned whea ,memory related glitches but often only occasionally, when adequately warmed and loaded up in use for example, which is why I mentioned it.

That makes sense.

Yours look automatically set ATM.

Yep, it's just on auto.

And I gave it as an example, dig around there's a lot of memory related timing settings, that you are trying to run mostly configured for two sticks but with four, that's why I suggested looking up quad kit tunings.

I'll have to do some digging. I know there is at least one subtiming that ends up changing automatically once 4 dimms are installed but maybe it's not enough. I might just have to loosen them up more and shake them down one by one if I can't find any quad kit examples that closely match my kit.

 

[ir_cow]

It too bad I don't have a quick way to setup 128GB on a 5800X. My test benches are being used for other things right now.

 

[A Computer Guy]

My 2nd kit alone appears to be ok. Memtest86 4 passes no issue. OCCT testing with my 2nd kit passes sse and avx2 tests for 1 hour each with no WHEA.

On lineup for tonight 128GB tests...(individually then combinations of)

• Speeds
  • from 3200, 2933, 2667, 2400
• 1.35v
  • because the ram sticks previously showed signs of improvement with higher voltage
• GDM=disable, 2T
  • suggestion by user
• Loosen subtimings
  • suggestion by user and UEFI tries to tweak subtimings automatically when 4 of these particular dimms are installed

In 4x32 config OCCT typically reveals the WHEA errors fairly quickly (so far) except the one that causes reboot after the system goes idle for a long period of time. (4 to 6 hrs)

 

[Mussels]

64GB worked pretty much out of the box for me, i ran 2x32 + 2x8 for testing some benchies at one point and that worked fine too.

Most you should need to do is raise the SoC voltage


WHEA errors aren't the ram, that's the memory controller - which is where the SoC voltage comes in

 

[hckngrtfakt]

64GB worked pretty much out of the box for me, i ran 2x32 + 2x8 for testing some benchies at one point and that worked fine too.

Most you should need to do is raise the SoC voltage


WHEA errors aren't the ram, that's the memory controller - which is where the SoC voltage comes in

Same for me.

Even with the added load to the IMC from the 4 sticks, and setting the GDM to ON (from auto)
my system is beyond stable with the default AMP profile and 64gb (4x16).

However setting the GDM to auto causes random restarts during startup, and setting it to OFF the system refuses to post.

 

[billeman]

What voltages do you use? I can't really push he frequency on these kits (2x32) at all but 1.35v does allow me to push the timings down significantly to respectable CL16 @DDR4-3200. After verifying my 2nd kit didn't give me a dead stick, my next step was to see how they respond to 1.35v in 4x32 but I suspect I might need other voltage tweaks assuming I'm not dealing with the impossible.

VSOC 1.125 (I think this one helped a lot to attain 3600+), VDDG 1.0V, VDDP 0.9V, VDIMM 1.37 (instead of 1.35V XMP default).
I was actually amazed that it is stable now at 3600 because previously I had to run 3200 (wanting absolute stability because this will be my new VMWare ESXi server).
BTW , running gear down mode , not 2T.

 

[A Computer Guy]

VSOC 1.125 (I think this one helped a lot to attain 3600+), VDDG 1.0V, VDDP 0.9V, VDIMM 1.37 (instead of 1.35V XMP default).

If you haven't already can you tick the box on adjusted voltages on my survey?

I was actually amazed that it is stable now at 3600 because previously I had to run 3200 (wanting absolute stability because this will be my new VMWare ESXi server).
BTW , running gear down mode , not 2T.

I was able to get though some tests last night. At 3200 2T seemed to help a lot (no 1 or 18 errors at all) but I still got a single 47 after a few hours. Reverting changes and applying 1.35v dimm so far seems to be working. After an hour of OCCT (ram peaked at 56c, only 2c higher than at 1.2v), working with some light workloads, and letting the system idle overnight, I haven't gotten any error or spontaneous reboot. Perhaps my next test tonight will be to reset DRAM voltage to 1.2 and increase VSOC and see how that works out.

 

[A Computer Guy]

Update:

So for 2 days only raising 1.35v dram resolved the WHEA errors except when I left the PC idle overnight last night. (just a few 18, and handful of 47 errors)
Increased soc to 1.125v for today.

Tonight trying the combo of 2T, 1.35v dram, and 1.125 soc.

 

[A Computer Guy]

It seems I can't get around the idle time reboot problem, every time when I lock the workstation and walk away for a few hours. Even at 2T and 1.2v soc, reducing speed, and relaxing timings more it makes no difference. I have given up on getting 128GB working for now. I'll probably open a ticket with asrock to see if they can do anything for compatibility.

 

[Deleted member 185158]

Have you a fan to put on the dimms? Maybe get them running cooler, get less errors?
Ok

 

[A Computer Guy]

Have you a fan to put on the dimms? Maybe get them running cooler, get less errors?
Ok
View attachment 270888

I'm at a point where I am not getting errors anymore (well the errors seem to occur only right before the reboot). Just reboots are occurring after leaving the machine idle for a few hours. I can do 2hr OCCT runs, 3hr zip jobs, 3hr handbrake jobs no problem. Even after thermal cycling the ram from 30c (now cold winter room) to 55c between runs. I just realized I left the new ram kit in the machine. I'll try swapping it out with the previous kit, maybe the reboot issue is simply with the new kit.