Up until today i only had 2x32GB sticks. But everything has been running just great. Occasional crashes, but I’ve been playing with the PBO enhancements levels. Nothing else. XMP on at 6400 working great.
I had an impulse buy and filled the rest of the slots
So… the new RAM arrives (Fully matched set obv). I get it installed, load into bios, all detected, load optimized, set XMP, F10.
Nothing, for about 5mins, then the ‘failed boot’ error message. So I load back into bios, reload optimized. Nope. Tried a few different things:-
PBO at 70/level 1
PBO default
PBO off
XMP at 6000/5800/5400
Low latency
High bandwidth
Both
Updated bios to 14/08/24 F31
Found ALL the new settings that have been added.
Tried the new setting for training algo, and tweaks for 6000+MT/s
(Reloaded Defaults each time it failed)
Right now I have the BIOS at optimized defaults. Nothing else. All 128GB are recognized. All still registered as fully matched sets (just 4 weeks apart in the build times) Its just running at 3600MT/s.
I have also experiences this and chalked it up partly to ryzennmemory controller, but also motherboard topology. You can search YT for “ram daisy chain” and one of the search results should be called “rambling about motherboard memory layouts”.
Slots 2 and 4 generally perform better than 1 and 3 when using 2 sticks, then performance degrades further when populating all 4 sticks. Something to do with electrical interference.
Not all motherboards have this problem but it’s a very common trace layout and there isnt really a way to tell which ones have it when shopping around.
What @EniGmA1987 said. With matched quads from 5200 up often it’s been necessary to manually find the right voltages and termination settings rather than rely on training. Not sure if AGESA 1.2.0.0a changes that.
I’m aware of one or two people that’ve gotten 2x32+2x32 conditionally stable at 6000 but it’s taken them a lot of tinkering and maintenance as not all AGESA versions will do it. Since XMP at 5400 doesn’t work you’ll need to back off until POST is possible and then start looking for pathways to tune up and tighten timings.
FWIW I didn’t have trouble pushing a matched quad up to DDR5-4800 with slack timings on Zen 4 but the JEDEC 4800 profile wouldn’t POST. It’d also boot 5200 slack but wasn’t stable and, based on DIMM temperatures at 4800 under memory stress, I opted not to try to stabilize 5200+. Maybe if I get a chance to revisit the cooling.
Stub reflections necessarily occur any time the number of DIMMs exceeds the number of channels. Since (m)ATX boards are usually quad DIMM and dual channel most them have this limitation. Some tweakery can be done to shorten the stubs and reduce reflections somewhat but the only complete solution is to drop to dual DIMM, dual channel as typical of ITX boards.
The desktop ATX exceptions I’m aware of are the Aorus Tachyons. Haven’t looked particularly hard, so there might be other similar boards.
The system POSTs with 5200/5400 but fails the stress-ng test. I have no clue about memory timings. Not a foggyist.
I have saved a BIOS profile for this so i can tinker a bit more in the future. Tonight I will set the system off running MemTest.
You could either manually configure clocks and timings to make it work, or try to jump into a 4x48GB kit, for some reason those seems to be able to work with ~5200MHz XMP/EXPO out of the box way easier than 32gb UDIMM kits.
Yup. Technically everything over 3600 with 4 dimms is an overclock, so at 5000 or even 4800 you’d want to stability test properly to make sure it’s stable. 1h stress-ng is a good start but not enough to make sure.
There’s not too many tools on Linux sadly. The memory of guys are mostly on windows. Stress-ng is good, as well as any y-cruncher stress test that shows high memory load (the cli tool shows it), and there are bootable memtest86(+) binaries.
On windows testmem5, HCI memtest and karhu work well. The latter is paid though.
The only way to do it proper is not use EXPO but start but slowly up the want speed as i asume this is not bought as a set or supported by the mobo
I had similar issues for me it turned out to be the bios not the latest on my mobo
Especially 4 dimm sets are an absolute pain to get them working on higher speeds, as hardly any set really runs out of the box
Unless your extremely lucky
So the only way to try to get a higher speed running is start stepping up with the speed in your bios and to be honest don’t be surprised it does not work at all ( Especially with 4 dimms)
I returned the extra 2 sticks. I can’t even get JEDEC speeds to work with all 4 populated.
But during my journey i found AHOC has a video giving his timings for Hynix RAM. So I have tried those, and so far i have validated them running just fine at 6200MT/s.
