Question DOCP profile stops working after going from 2 to 4 sticks of identical RAM

[pepijndevos]

I have built a PC with the following parts:

• AMD Ryzen 9 7950X - Processor4.5 GHz (5.7 GHz) - 16-cores - 32 threads - 80 MB
• ASUS TUF GAMING X670E-PLUS - MoederbordATX - Socket AM5 - AMD X670 - DDR5 - Realtek S1220
• Kingston FURY Renegade - GeheugenDDR5 - 32 GB: 2 x 16 GB - 288-PIN - 6400 MHz - CL3

With this configuration DOCP at 6400 MHz was unstable, but 6000 MHz worked well. However, my work turned out to require additional RAM, so I ordered another kit of the exact same RAM from the same store. However, now not even the 6000 MHz profile is stable.

This makes some amount of sense in retrospect, since the additional capacitive loading of the extra RAM sticks means the CPU has to work harder to drive the data lines. I don't think this has anything to do with the commonly parroted matching of RAM kits beyond that YES a kit was tested with that amount of sticks, and adding more sticks causes a different loading on the output drivers in the CPU.

Anyway, I have two questions:

What makes that the original kit didn't work with the 6400 MHz profile? Is this a CPU issue, a motherboard issue, a RAM issue? The CPU only claims to support up to 5800 MHz so maybe 6000 MHz was already a lucky stretch. But then should I expect 4 sticks to work at 5800 MHz? How big of a factor is motherboard signal integrity?

How can I manually tune the RAM to make a custom profile that's somewhere between the 4800 MHz and 6000 MHz profile? Can I just start from the 6000 MHz profile and keep decreasing the frequency until it becomes stable, or are there other important parameters that need to be set in lockstep?

I'm coming at this with my intuition as an electrical engineer, with little to no experience in overclocking. I would kind of expect that the cycle counts and voltages from the 6000 MHz profile would be fine for a lower frequency, but that there might be other timing parameters that I'd need to decrease for 4 RAM sticks due to the slower rise/fall times. But there are a near infinite amount of unexplained abbreviations in the timing section, so I'm a bit lost.

p.s. I have recently upgraded my BIOS after the whole ASUS voltage debacle went down

 

[kanewolf]

I have built a PC with the following parts:

• AMD Ryzen 9 7950X - Processor4.5 GHz (5.7 GHz) - 16-cores - 32 threads - 80 MB
• ASUS TUF GAMING X670E-PLUS - MoederbordATX - Socket AM5 - AMD X670 - DDR5 - Realtek S1220
• Kingston FURY Renegade - GeheugenDDR5 - 32 GB: 2 x 16 GB - 288-PIN - 6400 MHz - CL3

With this configuration DOCP at 6400 MHz was unstable, but 6000 MHz worked well. However, my work turned out to require additional RAM, so I ordered another kit of the exact same RAM from the same store. However, now not even the 6000 MHz profile is stable.

This makes some amount of sense in retrospect, since the additional capacitive loading of the extra RAM sticks means the CPU has to work harder to drive the data lines. I don't think this has anything to do with the commonly parroted matching of RAM kits beyond that YES a kit was tested with that amount of sticks, and adding more sticks causes a different loading on the output drivers in the CPU.

Anyway, I have two questions:

What makes that the original kit didn't work with the 6400 MHz profile? Is this a CPU issue, a motherboard issue, a RAM issue? The CPU only claims to support up to 5800 MHz so maybe 6000 MHz was already a lucky stretch. But then should I expect 4 sticks to work at 5800 MHz? How big of a factor is motherboard signal integrity?

How can I manually tune the RAM to make a custom profile that's somewhere between the 4800 MHz and 6000 MHz profile? Can I just start from the 6000 MHz profile and keep decreasing the frequency until it becomes stable, or are there other important parameters that need to be set in lockstep?

I'm coming at this with my intuition as an electrical engineer, with little to no experience in overclocking. I would kind of expect that the cycle counts and voltages from the 6000 MHz profile would be fine for a lower frequency, but that there might be other timing parameters that I'd need to decrease for 4 RAM sticks due to the slower rise/fall times. But there are a near infinite amount of unexplained abbreviations in the timing section, so I'm a bit lost.

p.s. I have recently upgraded my BIOS after the whole ASUS voltage debacle went down

Not surprising. It may be the same name, but there is no guarantee that it is "identical". RAM is sold in sets for a reason. The manufacturer guarantees that the SET will work at the advertised specs. Mix and match RAM is a gamble. You might be able to manually adjust memory voltage up a few thousandths. But the WORK of getting the memory to cooperate is now your responsibility.

 

[pepijndevos]

We agree that adding more RAM changes the performance, except you come at it from PC building experience and I come at it from electrical engineering. To me it seems unlikely that different units have significantly different performance, but adding more RAM just fundamentally makes it slower. In EE we call this fan-out.

Would you expect that if I use two RAM sticks from DIFFERENT kits, that performance would be as bad as with 4 sticks from those same kits? Put differently is it the fact that they are matched, or is it the amount of sticks?

Do individual RAM kits come with custom DOCP profiles, or are they matched in the sense that they are binned to have similar timings within the kit?

Anyway, I'm more interested to learn how to do the work of getting them to cooperate.

The 6000 MHz profile runs the sticks at 1.3V IIRC, while the 6400 MHz profile runs them at 1.4V or something like that. I assume it's safe to up the voltage to whatever is in the 6400 MHz profile, but keep the rest of the settings like the 6000 MHz profile?

Up to actually damaging the chip, is there a downside to higher voltage other than increased power usage?

 

[USAFRet]

To me it seems unlikely that different units have significantly different performance,

Not unlikely at all.
In fact, very common.


Lets say the baseline and +/- tolerance is "100, +/- 5"
So, anywhere between 95 and 105 passes the test.
For an individual stick, or sold set.

Lets say your original RAM is at 103.
The new RAM is at 96.
Both sets, individually, still within manufacturers tolerances.

But they won't play well together.


We had this in jets quite often.
That particular FCC will NOT work in jet 105. But works well in all the others.

 

[kanewolf]

Anyway, I'm more interested to learn how to do the work of getting them to cooperate.

There are lots of threads on this board about manually overclocking RAM. That is all you are doing. It is a LOT of trial and error. Clear CMOS and try a slightly different RAM voltage or clock or some other RAM parameter...

 

[Misgar]

@pepijndevos. Do individual RAM kits come with custom DOCP profiles?

The answer is yes (except they're normally known as XMP or EXPO profiles, not DOCP which is terminology used in Asus BIOS).

XMP profiles apply to Intel CPUs.

EXPO profiles apply to AMD CPUs.

It can be difficult to find suitable RAM with an EXPO profile.

Most XMP RAM will run in an AMD rig, but you may have to adjust memory timing settings for stability.

If a memory kit contains XMP and EXPO profiles, the two profiles may contain subtly different timings at the same frequency.

I acquired a bulk lot of 8GB DDR3 DIMMs from Kingston and Corsair. At first I thought the two batches were identical, but closer inspection of SPD values in AIDA64 showed subtle variations between DIMMs.

In many DIMMs, the SPD contained more than 20 different memory timings, covering a wide range of frequencies, divided roughly equally into normal and XMP sections.

To further complicate matters, the programmed frequencies often differed between DIMMs, with clock rates slightly higher or lower than "normal" values, plus different primary, secondary and tertiary timings.

It was possible to fit four 8GB Corsair DIMMs in a motherboard and find three subtly different SPD timings at the same frequency.

In some combinations, the BIOS selected a glorious mixture of timings pulled out of several different DIMMs, making the best of a bad job.

If you mix kits of DIMMs, there's no guarantee they'll contain the same memory chips. One kit might use SK Hynix, another kit might use Samsung B-die. SPD values may vary between kits too. Different binning adds to variations.

You can check the memory chips used with Thaiphoon Burner.

Four DIMMs can be harder to overclock than two DIMMs.

Dual-rank DIMMs have different characteristics to single-rank DIMMs.

Mixed DIMMs can be harder to overclock.

If you're prepared to manually configure memory timings, instead of invoking DOCP (with your fingers crossed), you may achieve more success than automatic overclocking.

If four DIMMs don't work at DOCP 6,000, try DOCP 5,800 or DOCP 5,600. Alternatively relax the Primary timings after selecting a particular DOCP value.

The more memory voltage you apply with XMP or manually, the greater the eventual risk of electro-migration damage.
https://www.reddit.com/r/overclocking/comments/t4r35b/ddr5_max_safe_voltage_247/

On a final note, I purchased 2x32GB (64GB total) with my 7950X rig and I'm running it at 4,800MT/s for stability. My apps do not benefit more than a few percent from overclocked RAM, making it a pointless exercise.