scan.co.uk advised us to lower the memory to 3200MHz
3200MT/s is the supported speed quoted by CPU World for the 5900X:-
https://www.cpu-world.com/CPUs/Zen/AMD-Ryzen 9 5900X.html
Memory controller |
Memory channels (total): 2
Supported memory: DDR4-3200
ECC supported: Yes [1] |
Anything faster than 3200MT/s is down to the Silicon Lottery. In many systems, it's possible to achieve higher speeds
with only two DIMMs fitted, but when you run 4 DIMMs, the additional load on the IMCs (integrated Memory Controllers) often requires a significant XMP reduction.
scan are messing me around saying it's memory when no errors are found on any of the memory when mentest86 is done in slots 2 and 4
Testing one DIMM per channel (in slots 2 and 4) at high XMP/EXPO/DOCP speeds is the "ideal" scenario, because the IMCs are not heavily loaded. With a really fast memory kit and just 2 DIMMs, some people push for a 4000MT/s overclock with 1.4V.
https://www.reddit.com/r/overclocki...w_stableworth_it_is_to_oc_a_5900x_to_4748ghz/
When you add a second pair of DIMMs to any system (AMD or Intel) you're making a big change. The increased loading on the memory bus often requires a lower XMP/EXPO/DOCP setting. It's a bit like adding a heavy trailer to the back of a small car. Performance will suffer.
When you invoke automatic XMP/EXPO/DOCP overclocks on a fast set of DIMMs, it's a bit like tuning an engine to Stage 3. Success is not always guaranteed. Push a system too far and something could fail.
In the case of a 5900X, as
@thestryker says, it's worth testing the 4 DIMMs at JEDEC base speed 2133 or 2400MT/s, i.e. with XMP/EXPO/DOCP) switched off.
I'd expect MemTest86 to pass all tests with 4 DIMMs at JEDEC, the base speed designed to ensure maximum RAM stability on all systems.
https://www.jedec.org/category/technology-focus-area/main-memory-ddr-sdram
Memory: Corsair DDR 4 3600MHz 128 GB
When Scan sold you a system with 128GB 3600MT/s RAM fitted, they provided the option to tune your system to "Stage 3".
Did Scan actually set the BIOS to 3600MT/s with 4 DIMMs fitted and guarantee stable operation at this speed
in writing? Or was the RAM speed set lower?
128GB memory kits (32GB per DIMM) can be more difficult to overclock than 64GB kits featuring the same number of DIMMs (4 x 16GB). 32GB DIMMS often have twice the number of memory chips (dies) compared to 16GB DIMMs, which increases IMC bus loading.
With 4 x 32GB you've got a pretty much 'worse case' scenario. Heavily populated 32GB DIMMs with lots of dies and 2 DIMMs per IMC (4 DIMMs total). Not ideal for fast speeds.
On a DDR5 system, people recommend 2 x 48GB (or 2 x 64GB if/when available) for the greatest chance of overclock success. Fiting 4 x 32GB or 4 x 48GB DDR5 is equally likely to cause speed problems.
In such cases, the normal advice is to try manual overclocks, raising the RAM speed in small increments above JEDEC, e,g. 2600/2666MT/s, 2800/2933MT/s, 3000MT/s, etc. Perform a full MemTest86 run at each RAM speed and aim for zero errors. Any errors, even one, reduce the RAM speed by one step or increase CAS/CL by 1 clock cycle.
Remember to set the DDR4 supply to 1.35V when manually overclocking. The JEDEC default for DDR4 is 1.2V at 2133/2400MT/s. I advise caution if manually setting 1.4V and above, in attempt to find stable overclocks. Too many Volts are bad for the IMC.
https://en.wikipedia.org/wiki/DDR4_SDRAM
If you search the forums, you'll find lots of people experiencing problems with 4 DIMMs in standard Intel/AMD desktop systems. The normal advice is to reduce RAM overclock speeds with 4 DIMMs to achieve stability.
I have several Xeons servers with quad channel IMCs and 8 DIMM slots all populated. What I'd really like is a Threadripper with quad or 8-channel memory, but for now I'll stick with my 3800X and 7950X.
Best of luck.