Corsair RAM connection issue
Corsair RAM connection issue
I've reviewed the RAM specs and confirmed it's a Micron B die with XMP settings at 16,19,19,36. At 3800MHz it should perform well, though you might notice some performance differences compared to the 18,22,22,42 configuration. It's still a solid choice if you need reliability and stability. Let me know if you want more details!
Which rev. B? 8Gb or 16Gb? If it's 8Gb, that stuff is terrible and I'd be surprised if you can get it stable at 3800MT/s. If it's 16Gb, then you'll be getting mostly mediocre B die timings with high tRAS, tRC, tRFC, and sub timings of mediocre. I'm assuming this is a Ryzen 5000 based system? If so, this is my usual RAM OC methodology as well as some roughly expected timings. Make sure you're on an OS without any important data on it before doing RAM OC, this can corrupt OS files. If this is 16Gb Rev.B, set vDIMM to 1.5V and SOC voltage to 1.15V. Tune your primary timings (tCL, tRCDWR, tRCDRD, tRP, tRAS, and tRC) Start with tCL, lower it until the system stops booting (on Ryzen systems tCL, as well as a couple other timings, can only be set to even values when GDM is enabled which if you're new to RAM OC you want to be enabled). Once it stops booting, start raising it until you can go into Windows and run a memory intensive benchmark, my favorite is Y cruncher 1B. It's helpful to have your CPU locked to a specific frequency so you can compare your performance numbers and make sure you aren't getting performance regressions. Once it runs, move on to the next timing. Continue all the way though, though you want to set tRAS and tRC at the same time. tRAS is technically a real timing, but only barely, it only affects performance given its relationship to tRC. Set tRC to tRAS + tRP. tRCDWR will likely do 8 (as low as that timing can go) as it does with almost every memory IC, though I forget if 16Gb Rev.B is one of them. tCL will likely be at 16, tRCDRD will likely be at 16 or 17, tRP will likely be at 16, and tRAS will likely be at 36-40, though I have very little experience with 16Gb Rev.B so your mileage will vary. Stress test This step is optional, but I find it tends to save a bit of time in the long run. Eventually you will need to have to get it memtest stable, so if you can eliminate the primary timings as the possible source of instability, that'll save you effort when you have to eventually figure out what is causing it to be just barely unstable. You can memtest every single timing, but that takes forever. You can just memtest at the very end, but then you're just taking shots in the dark for what the faulty timing is. Doing a stress test now is a decent middle ground. I'd aim for 300% in HCI memtest, but there are a ton of different memory stress tests that work just as well. Tune your secondary timings First thing you should do is set tFAW to 16, if that timing is any higher then it will effectively make tRRD_S and tWTR_S irrelevant, and those timings really do affect performance a lot. Roughly do what you were doing with the primary timings, but on the secondaries, though the order is slightly different. You want to do the short timings before the long timings (tRRD_S and tWTR_S before tRRD_L and tWTR_L), and the long timings will always be greater than or equal to the short timings. tCWL will be equal to or one tick less than tCL (I.E. if tCL is 16, tCWL will be either 16 or 14). I'd expect short timings of 4, long timings of 4 and 8 respectively, tWR of 10, tRTP of 10, and tRDWR of 10, though again, don't have a ton of experience so your mileage will vary and they can be a fair bit off. tRFC is a weird timing on Micron memory. It doesn't tighten up much if at all, so while it does have an impact on performance, it's hard to stabilize and will always end up 600+. I'd personally just set it to 650 and call it a day, leaving tRFC2 and 4 on auto. If this was B die or something else, then I'd mess with it, but on Rev.B you won't get much of anywhere. Stress test again Tune your turnaround timings. Again, do what you did on the other timings, though keep in mind only some timings will likely apply. If you've got 2 DIMMs of single rank memory, the SD and DD timings won't apply. If you've got 2 DIMMs of dual rank memory, the DD timings won't apply (though it is customary to set these to the same as the SD timings). If you've got 4 DIMMs of single rank memory, the SD timings won't apply (though again, customary to set these the same as DD timings). For 4 DIMMs of dual rank DIMMs (my condolences), all timings will apply. tRDRDSCL and tWRWRSCL will likely be 2s, tWRWRSC and tRDRDSC will be 1, and I genuinely have no idea how tWRRD will line up. Final Stress test. Do 1000% memtest (leave it overnight), maybe OCCT if you want, Linpack Xtreme for 30ish passes, and for at least one of these tests have a GPU stress test run at the same time. Memory is temperature sensitive and the GPU outputs a ton of heat, so you want to make sure the GPU doesn't cook the memory to the point where it starts spitting out errors. If it does, you can either redo the overclocking with less voltage, or point a fan at your memory.
The information provided doesn't indicate any change based on the processor model mentioned. It simply states your request without further context.
Essentially everything remains consistent. The memory controllers for the Ryzen 5000 and Ryzen 3000 series are identical, though the Ryzen 3000 chips varied more in their maximum stable Infinity Fabric clock speeds. I’m not claiming your processor can’t reach 1900MHz FCLK, but many units cap around 1866MHz reliably. To verify performance stability, I’d test both 1866 and 1900 MHz settings—locking the CPU to 4.0GHz at 1.25V mid-level LLC should provide reliable results for comparison. My 5900X at 1900MHz (3800MT/s DRAM) would finish a benchmark in roughly 26 seconds. At 1933MHz it would exceed five minutes. While higher speeds are possible, confirming stability is key to avoid performance drops. According to Thaiphoon Burner, your 16Gb Rev.B RAM is solid. The figures above should reflect real-world accuracy. Another point: ZenTimings aids in understanding auto-timing behavior, letting you adjust settings and confirm BIOS changes take effect. Some AMD boards include auto-recovery, which can reset RAM settings if it detects issues, potentially causing a boot without warning. This feature is useful for getting things up and running but can be frustrating if you need precise BIOS tuning.
You can adjust the CPU settings in Ryzen Master to 4.0, but it won't change the actual timings your board uses. The timings you mentioned are fixed by the hardware and won't change just because you set a different value. Your board handles memory timing failures gracefully, so it won't reboot. You're doing great with your testing tools—using Pi, Memtest5, and others is a solid approach.
I wouldn't use it. Over the years I've faced many problems with Ryzen Master and it just doesn't hold up beyond an hour before needing removal. You should aim for minimal background activity in the OS during stress tests so you can push RAM usage to its limits. Ideally, Ryzen Master isn't needed at all. Setting the memory timings in BIOS is straightforward—just change them during a reboot. A solid alternative is to reset the CMOS header on the clear pin, which avoids repeatedly touching those pins. At least this shows it's not just me who deal with faulty MSI boards and broken auto-recovery. It's not flawless on my X570 Master or Taichi, especially the latter, which sometimes recovered but not always. It works in some cases. SuperPI isn't my top choice either; it takes a long time to run benchmarks, while Y Cruncher 1B can show stability in just a minute. Memtest5 is a good option for stress testing.
Thanks, but I'm struggling to keep going. I can't lower the trc at all—every lock-up is slow and I'm struggling with the screwdriver. I've got a new mobo, CPU, cooler, and GPU coming soon, so I'll have to tackle this again soon. I'm going to try loose timings or XMP/docp. It's getting frustrating; I almost snapped a PlayStation over my knee before. Best to leave this alone.