[Ichirou]

What happens in TM5 1usmus when you try pushing the VDIMM higher, all else held equal?

 

[Sixteen9863]

What happens in TM5 1usmus when you try pushing the VDIMM higher, all else held equal?

Errors starting in cycle 2 when set to 1.5V

Almost instantly threw an error 2 when set to 1.55V.

For reference, the above timings passed 6c at 1.45V and 450% HCI memtest.

Finished testing the 15-16-16-16-31 timings at 1.45V shown in the OP as well, passed 600% memtest but gave error 4 on the second cycle of TM5 1usmus v3. Going to try 17 RCD now but I think 15-17-17-17-32 is already worse than almost every example I've seen with this RAM.

 

[Ichirou]

If you've changed nothing besides raising VDIMM, then I think your kit's just poorly binned.
B-die should never get worse by increasing VDIMM alone outside of overheating.

 

[Sixteen9863]

If you've changed nothing besides raising VDIMM, then I think your kit's just poorly binned.
B-die should never get worse by increasing VDIMM alone outside of overheating.

That's what I thought as well, just wanted to see if there was something I was missing (or just some hopium to ease my pain of wasting 10+ hours/day for a week 😂)

Would it be worth trying to swap the sticks into the opposite slots?

 

[Ichirou]

That's what I thought as well, just wanted to see if there was something I was missing (or just some hopium to ease my pain of wasting 10+ hours/day for a week 😂)

Would it be worth trying to swap the sticks into the opposite slots?

You can try it, but I highly doubt it would change anything.
Again, unless heat is an issue, you should not get errors from simply raising VDIMM and changing nothing else.
Most B-die scales just fine all the way until 2.1V, subject to motherboard compatibility.
But there were certain batches and older batches that simply had poor scaling or awful timings no matter what you did.

 

[bscool]

Just my input and I didn't read everything. Your ram doesn't have temp sensors and you put a fan farther away from them. My guess it is temp related. Even my best bin(gksill 4000c14-15-15) when they hit in the 38c to 40c range error with tight timings.

And I doubt you are keeping yours under 40c. I have a 2000rpm Phanetk t30 right over mine to keep them under 40c at 1.55v.

 

[Sixteen9863]

Just my input and I didn't read everything. Your ram doesn't have temp sensors and you put a fan farther away from them. My guess it is temp related. Even my best bin(gksill 4000c14-15-15) when they hit in the 38c to 40c range error with tight timings.

And I doubt you are keeping yours under 40c. I have a 2000rpm Phanetk t30 right over mine to keep them under 40c at 1.55v.

Hey, thanks for sharing your thoughts. The fan is quite close to the RAM (maybe 2cm/1"?) but it's closer to one end than the other rather than sitting on top of the sticks. The case fan configuration should also be pretty good (2x 140mm Noctua intakes on the front and a 120mm Noctua rear exhaust, on top of the top 120mm fan pointed at the RAM). I thought it would be okay with everything blasting at 100%, especially when there's no heavy CPU or GPU workload to add heat to the case. Room is also climate controlled at around 20-22 degrees Celsius. Didn't think it would be worse than people using tiny 60mm-80mm fans on top of their sticks, or dangling 120mm fans with zipties/resting them on their GPU, esp. since it's blowing in fresh air through the mesh top of the case. I also thought if it was a temperature related issue, it would manifest over time and not instantly upon commencing the test.

I really do wish there was a temperature sensor on these Patriot kits, even thought about buying a 2pin probe but not sure it's worth the effort if I can't even cool it with overkill fan speeds that sound like a jet turbine. Wanted to be sure about the heat, so I tried again with the case open on my desk, removed the front CPU fan covering the RAM sticks and set all the fans to 100%:

At 1.55V it still fails instantly. 1.5V failed on the second cycle again, so no improvement apparently. I guess I just don't see how it could be running that hot when other people seem to be pushing even 1.6-1.7V no problem, not sure what else I could do.

 

[des2k...]

tRAS & tRC don't scale with voltage on these sticks, try higher values

Maybe one of the sticks are causing issues, so you may need to test them individually.

On my side they also don't like too cold or too hot, at lower voltage with a fan they fail vs no fan. They also need high soc voltage to pass testing.

I didn't read all your post, but did you load xmp & drop the IF/Mem to 3600- 3800 to test your imc/vddg for stability ?

Unless your dealing with some bad sticks, these bdie kits are solid and need a good cpu/imc for 3800cl14-4000cl16. They are one side so 1.6v will not heat them fast enough to crash the test early.

Here's mine, if it helps,

 

[bscool]

Like @des2k... mentioned it could be 1 or 2 weaker dims. Or even a dim slot that is weak. On DDR5 MB a weak slot is a big issue. But most MB will have weaker slots. And you have to test 1 dim and slot at a time and most wont do it because it is so time consuming.

Errors that happen within a minute or two are not from temps but if you are get them after 10 to 15 minutes plus as an example there is a higher chance it could be temp related. I know nothing about AMD mem OC so I cant comment on timings.

 

[Sixteen9863]

tRAS & tRC don't scale with voltage on these sticks, try higher values

Maybe one of the sticks are causing issues, so you may need to test them individually.

On my side they also don't like too cold or too hot, at lower voltage with a fan they fail vs no fan. They also need high soc voltage to pass testing.

I didn't read all your post, but did you load xmp & drop the IF/Mem to 3600- 3800 to test your imc/vddg for stability ?

Unless your dealing with some bad sticks, these bdie kits are solid and need a good cpu/imc for 3800cl14-4000cl16. They are one side so 1.6v will not heat them fast enough to crash the test early.

Here's mine, if it helps,
View attachment 2596399

I initially tested with very loose timings (like 18-22-22-22-44, something like that) and 3800/1900 on Prime95 Large FFTs + Unigine Heaven or OCCT VRAM to make sure it was stable before proceeding. It is, as long as I don't push the VDIMM past 1.45V. I have tried everything from fans at full speed to a gentle fan curve with the RAM fan off, no luck. I have very little headroom to tighten secondaries as well, probably because of the voltage being capped at 1.45V. Can't get anything close to what you have on your setup for the secondaries.

Like @des2k... mentioned it could be 1 or 2 weaker dims. Or even a dim slot that is weak. On DDR5 MB a weak slot is a big issue. But most MB will have weaker slots. And you have to test 1 dim and slot at a time and most wont do it because it is so time consuming.

Also you be surprised what it takes to cool the dims especially when they packed tight like using 4 dims or 2 dims in Apex or 2 dim MB. Not much room for air to get in there. Obviously error that happen within a minute or two are not from temps but if you are get them after 10 to 15 minutes plus as an example it is a good chance it is temp related. I know nothing about AMD mem OC so I cant comment on timings.

Funnily enough, I switched the sticks to opposite DIMMs and TM5 1usmus didn't error out until almost the end of 6 cycles, as opposed to failing within the first 2 cycles, so maybe there's something to that. Since I have four sticks though, I don't see how I could even avoid a weak DIMM slot if I could identify it?

 

[bscool]

You cant avoid it but you don't waste weeks or month trying to get something working when you know what the issue/limit is. Most don't test and keep banging their head against the wall.

Example if you test and find a slot or dim can only do 3733 at said timings you either except that or get another MB or more ram to bin.

I know this from doing it wondering why I can run something others can and it ends up being a MB or ram.

 

[Sixteen9863]

You cant avoid it but you don't waste weeks or month trying to get something working when you know what the issue/limit is. Most don't test and keep banging their head against the wall.

Example if you test and find a slot or dim can only do 3733 at said timings you either except that or get another MB or more ram to bin.

I know this from doing it wondering why I can run something others can and it ends up being a MB or ram.

Definitely suffering from some degree of sunk cost fallacy on this one, but I would like to understand better what the issue is. I've seen Veii and some other members talking about tweaking terminal resistances and CAD BUS settings so just making sure I'm not missing anything. Fwiw, I have tried most of the suggestions I've seen and I've pulled pretty much lever I can by now, so I'll probably just tighten the secondaries as much as I can here and call it a day.

I am kinda interested to see how performance would differ if I ran 3600/14 instead of 3800/15 with **** timings, though...

 

[Martin v r]

 

[Martin v r]

1,56v remember to set rtt nom and park
it is 4x8 Patriot 4400 CL19

 

[des2k...]

I initially tested with very loose timings (like 18-22-22-22-44, something like that) and 3800/1900 on Prime95 Large FFTs + Unigine Heaven or OCCT VRAM to make sure it was stable before proceeding. It is, as long as I don't push the VDIMM past 1.45V. I have tried everything from fans at full speed to a gentle fan curve with the RAM fan off, no luck. I have very little headroom to tighten secondaries as well, probably because of the voltage being capped at 1.45V. Can't get anything close to what you have on your setup for the secondaries.



Funnily enough, I switched the sticks to opposite DIMMs and TM5 1usmus didn't error out until almost the end of 6 cycles, as opposed to failing within the first 2 cycles, so maybe there's something to that. Since I have four sticks though, I don't see how I could even avoid a weak DIMM slot if I could identify it?

It was suggested on this site before for daisy chain boards with 4dimms.

You can test the sticks individually. I think same voltage and see which pass with lower tCL, higher speed.
These would be the stronger sticks and they would go on the weaker mobo dimms (usually A1-B1).

The weaker sticks (higher tCL, lower speed per voltage) would go on the best mobo dimms, usually A2-B2.

Never had to do this as my 4 sticks are close enough and the IMC limit on Amd.

 

[mongoled]

@Sixteen9863
As others have suggested, you need to check the sticks one by one to see what they tolerate when run individually.

The info given by des2k above me is golden.

For the record I am one of those users running high vdimm with 4 x 8GB sticks.

One thing I did not see you mention (or I missed it) do you know what PCB your modules are sitting on ?

As they should hopefully all be on an A2 pcb...

 

[Ichirou]

It was suggested on this site before for daisy chain boards with 4dimms.

You can test the sticks individually. I think same voltage and see which pass with lower tCL, higher speed.
These would be the stronger sticks and they would go on the weaker mobo dimms (usually A1-B1).

The weaker sticks (higher tCL, lower speed per voltage) would go on the best mobo dimms, usually A2-B2.

Never had to do this as my 4 sticks are close enough and the IMC limit on Amd.

Or the exact same settings, but lower voltage each time, in a test like TM5 to see at what point they throw errors or successfully pass. That’s much more consistent and easier to test. That’s what I do.

 

[Sixteen9863]

1,56v remember to set rtt nom and park
it is 4x8 Patriot 4400 CL19

Thanks, tried various RTT settings but didn't seem to make much difference for voltage. I was trying to get a very modest tWR/tRTP of 20/10 to pass TM5 yesterday to no avail, going to try @Bloax 's suggested settings of 6/3/5 and 40/20/30/20 with low ProcODT and AddrCmdSetup of 56 and see if that helps.

It was suggested on this site before for daisy chain boards with 4dimms.

You can test the sticks individually. I think same voltage and see which pass with lower tCL, higher speed.
These would be the stronger sticks and they would go on the weaker mobo dimms (usually A1-B1).

The weaker sticks (higher tCL, lower speed per voltage) would go on the best mobo dimms, usually A2-B2.

Never had to do this as my 4 sticks are close enough and the IMC limit on Amd.

@Sixteen9863
As others have suggested, you need to check the sticks one by one to see what they tolerate when run individually.

The info given by des2k above me is golden.

For the record I am one of those users running high vdimm with 4 x 8GB sticks.

One thing I did not see you mention (or I missed it) do you know what PCB your modules are sitting on ?

As they should hopefully all be on an A2 pcb...

Or the exact same settings, but lower voltage each time, in a test like TM5 to see at what point they throw errors or successfully pass. That’s much more consistent and easier to test. That’s what I do.

Almost want to do this for science but it's actually my girlfriend's PC and I think I'm already testing the limits of her patience 😂 If the wall is just bottom percentile sticks that won't scale with voltage, I think I'll take @bscool 's advice and cut my losses here. It seems all of you with the same kits haven't had any issues going 1.55-1.60V+ so my situation has to be an outlier.

Does it matter if the sticks from each kit go into the same channel or not? I always thought it was better to keep them together for some reason.

According to buildzoid, it's an A2 PCB (with chips on the left and right).

 

[Ichirou]

Thanks, tried various RTT settings but didn't seem to make much difference for voltage. I was trying to get a very modest tWR/tRTP of 20/10 to pass TM5 yesterday to no avail, going to try @Bloax 's suggested settings of 6/3/5 and 40/20/30/20 with low ProcODT and AddrCmdSetup of 56 and see if that helps.






Almost want to do this for science but it's actually my girlfriend's PC and I think I'm already testing the limits of her patience 😂 If the wall is just bottom percentile sticks that won't scale with voltage, I think I'll take @bscool 's advice and cut my losses here. It seems all of you with the same kits haven't had any issues going 1.55-1.60V+ so my situation has to be an outlier.

Does it matter if the sticks from each kit go into the same channel or not? I always thought it was better to keep them together for some reason.

According to buildzoid, it's an A2 PCB (with chips on the left and right).

If you're merging kits, you want each kit to be in a separate channel. Which kit for which channel is up to you to test.