Testing performance boosts on recent Samsung 32GB modules (possibly B-Die)
Testing performance boosts on recent Samsung 32GB modules (possibly B-Die)
Hi guys! I'm trying to overclock my RAM. My configuration: Ryzen 5900X, Gigabyte M550M Aorus Pro-P, 4 32BG-sticks Samsung DDR4-3200 These are strange sticks: https://semiconductor.samsung.com/emea/d...43bb2-cwe/ I've found in datasheets only *AB1-CWE version (A-die): https://up-shop.org/mpattachment/file/download/id/232/ Only mention of *BB2-CWE: https://www.reddit.com/r/overclocking/co...bdie_ddr4/ (not A/M-die, but new 16-Gbit dual-rank B-Die?) In asynchronous mode I was able to overclock FCLK to 1967 MHz. In synchronous mode PC can't boot on 1900 MHz MEMCLK and higher. With 1867 MHz I have errors in Prime95 after 10 minutes. What needs to be re-configured? Settings: VDram 1.41V, VSoC 1.12V, VDDP 0.9V, CCD 0.95V, IOD 1.05V, PLL 1.8V (default) Timings: tCL 18, tRCDRD 22, tRCDWR 8, tRP 20, tCL 38 Subtimings: auto ProcODT: 48 Ohm CAD_BUS: 40-24-24-24 I can't attach a screenshot of ZenTimings because I'm a Linux user and it doesn't run from USB with WinPE.
You should avoid running at high speeds with the Ryzen 5000 when using a 4 dual rank module. The maximum speed you'll achieve is around 3733MT/s, which is the best you can realistically expect. Adjusting the SOC voltage within the sweet spot of 1.1 to 1.2V might help stabilize performance. If you really need 128GB of RAM, don't try to overclock it—full stability testing can take days and the gains aren't worth it.
And if you just make a few easy tweaks without hitting the limits, aim for around 3600 or 3466. This way you'll find it much easier to adjust and stabilize, since you won't need to be super precise with the em value. Rams will be less sensitive, so room temperature changes won't cause big errors. After that, focus on tuning the subtimings like trrd-l/s, trfc, sc, sd, dd. These should make a noticeable difference. Set tfaw to 1 because it depends on trrd-s. For primaries, keep them a bit more relaxed to reduce volatility—mainly for tcl and tcwl. The rest don’t seem to help much with stability. Leave those a bit loose so you can still get stable performance at 1.35 or 1.4v, but go tighter when pushing above 2800. Even a small change like 0.05v can cause instability. This should get you about 95% of the performance you need, with less hassle overall and better stability.
3733 MT/s meets my needs and boosts performance noticeably thanks to Infinity Fabric overclocking. Prime 95 fits perfectly for testing—beyond just stress tests, it handles GIMPS P-1 Factorization, which efficiently uses 128 GB and helps spot issues. I’d like to know: are these drives truly the next-gen B-Die? Do they have overclocking performance data available?
They're referred to as B die, yet that doesn't imply any connection to Samsung 8Gb B die. Based on what I've observed, they perform quite poorly, though they aren't extremely widespread—they appear to be limited to OEM exclusives. It seems you might be battling Samsung's OEM PCB instead, which is even worse, particularly with 8Gb B die, rather than the actual memory chip. In reality, they were released late enough in the DDR4 era that no one really valued their launch (compared to Micron's solid 16Gb Rev. B and Hynix's reliable 16Gb C die, let alone the top-tier 8Gb B die), making it nearly impossible to find overclocking data. For context, memory makers assign a letter to each revision starting at M and progressing through A, B, C, etc., beginning with the first change in density or revision. I can identify at least three distinct Samsung B dies, but only one truly stands out as decent. It's not like they retire names for being unremarkable. Please let me know so we can get a notification.
I've performed small FFTs in prime95 for more than 5 hours without issues, then experienced a crash because of memory problems during Windows boot. Running tests can be frustrating.
you can check your setup by testing different configurations and seeing how they perform. It’s good to know what works without any problems, and based on your experience, adjusting settings shouldn’t affect gameplay much. Just keep an eye on stability and performance across various loads.
I rely on ROTTR [built813 <-- important, or any prior versions] for stability issues; you'll notice artifacts or crashes within 2-5 minutes. Yes, I could run p95 and memtest86 for hours with my old Corsair "Vengeance" RAM without errors, but it was still unreliable and produced random white errors or crashes. Nothing like this occurs with the Trident Z I currently have.
The best option is a DDR4 Samsung 8Gb B die with the 3200 CL14-14-14 speed. The other Samsung B dies I recall are 16Gb DDR4 (the one popular users mention) and 16Gb DDR5. I think a DDR3 model was also released, though I might be mistaken. I've completed more than 24 hours of stress tests and only got one error with 32GB RAM (I'll share the screenshot if I find it). With 128GB, that same issue might appear after up to four days. Small FFTs usually don't stress memory stability, while large FFTs are the real test for it.