X670E Taichi Carrara struggling to run 4x16GB
Printed From: ASRock.com
Category: Technical Support
Forum Name: AMD Motherboards
Forum Description: Question about ASRock AMD motherboards
URL: https://forum.asrock.com/forum_posts.asp?TID=61423
Printed Date: 27 Nov 2024 at 3:42pm
Software Version: Web Wiz Forums 12.04 - http://www.webwizforums.com
Topic: X670E Taichi Carrara struggling to run 4x16GB
Posted By: Dieg
Subject: X670E Taichi Carrara struggling to run 4x16GB
Date Posted: 08 Sep 2024 at 2:10am
|
Hello, Following a recent CPU upgrade from Ryzen 9 7950X to Ryzen 9 9950X, a few days ago I also upgraded RAM from 4x16GB G.Skill Trident Z5 Neo 6000MT/s CL30 (F5-6000J3038F16GX2-TZ5NR) to G.Skill Trident Z5 Royal Neo 6000MT/s CL28 (F5-6000J2836G16GX2-TR5NS) and while the older sticks ran flawlessly using their EXPO profile, the new ones aren't posting even when reducing frequency from 6000MT/s down to 5200MT/s. I'm greatly confused, because the main difference between those sticks are primary timings and it's the first time I encountered such problems on this platform since November 2022 (back when BIOS was super buggy and RAM support was very bad). I was aware of problems with 4 sticks due to experiences of my acquaintances, but they exclusively occured at higher frequencies, way beyond 6000MT/s. In order to deal with this, I flashed my BIOS from 3.06 to 3.06.PB02 [Beta] to no joy, as memory training fails and motherboard reboots the PC and follows up by setting RAM speeds to 3600MT/s. This motherboard has been with me for many years now, and back when I had Ryzen 9 7950X it even managed to run aforementioned old sticks at 6200MT/s with tightened timings without a sweat, so I suspect that either Ryzen 9 9950X has faulty IMC, the RAM sticks themselves could be faulty, or the BIOS is not yet ready to handle them. Does anyone else run their X670E Taichi Carrara in configuration with 4 RAM sticks and could share their experiences? |
Replies:
Posted By: NDRE28
Date Posted: 08 Sep 2024 at 12:37pm
|
Hello! 4x16GB DDR5 RAM configuration is problematic. If possible, return the 4x16GB kit and buy a 2x32GB kit. PS: Also, I really doubt that there is any problem with your CPU. |
Posted By: eccential
Date Posted: 08 Sep 2024 at 10:26pm
|
Agreed with @NDRE28 Two DIMMs per channel for non-registered & non-buffered DDR5 has been a big no-no from the get go. So much so, I don't even want motherboards with two DIMM slots per channel, if I can avoid them. There are 48GB DDR5 DIMMs these days, so you're able to get 96GB with just two DIMMs. I imagine we'll eventually get 64GB DIMMs, also. You must've gotten super lucky if your first set worked, but neither AMD nor Intel ever officially supported such, so you really can't complain to them if they don't work. |
Posted By: Dieg
Date Posted: 08 Sep 2024 at 10:40pm
|
Thanks for your answers. I guess it was dumb luck after all. Unfortunately, I'm stuck with these for the time being as I already sold my previous two kits. They are still in the same household, but it would be rude to demand that family member returns them, leaving them stranded for about 2 weeks without RAM.  On top of that, I'm unable to find 2x32GB kits of Royal Neo in any of the local European stores. It seems that the vendors ordered and imported only 2x16GB variants. I would rather not go through importing them myself to spare me the headache, as I did it once with different hardware and couriers decided to play basketball with the parcel.That said, I'm going to be stubborn and try to stabilize them myself to maybe get them up and running at 4x16GB 6000MT/s. I probed the ground a bit and they are able to POST and BOOT using EXPO profile if ProcODT and RTT are set manually, but they crash as soon as the load gets high. That gives me a small glimmer of hope that it comes down to bus termination values and timings + AGESA/BIOS updates. I'm currently in the process of testing individual kits to see if they aren't throwing any errors with EXPO profile turned on in order to eliminate possibility that one of the kits is faulty. One little problem that I found is that they are listed in G.Skill's QVL for X670E Taichi Carrara, but not on ASRock's QVL for that motherboard. God knows which version of BIOS G.Skill used in their tests. From my personal experience when I still had Ryzen 9 7950X, the BIOS-related stability improved greatly in the initial revisions. For example v1.04 couldn't POST with 2 sticks at all; not even EXPO.  Around 1.11 Beta that took a complete turn around and 4 sticks could be run with EXPO profile, and also OC to 6200MT/s. But then it went downhill sometime around v1.28 as that's when my old sticks could no longer boot at 6200MT/s without loosening timings. That was solely BIOS-related, because tighter timings worked just fine if I rolled BIOS version back. They also ran rock solid 24/7, which makes me wonder what exactly is the catch with the new kit being unable to repeat those feats.Fast forward to v3.05 and upgrading to Ryzen 9 9950X, I was able to run the two old kits at 4x16GB 6000MT/s with tightened timings, and then I bought and installed these... a complete catastrophe, to put it lightly. I'm not able to roll back to previous, "proven" versions of BIOS as they don't support my CPU. I hope that history repeats and updates for Ryzen 9000 series. I will report back with my findings in a few days. |
Posted By: Dieg
Date Posted: 07 Oct 2024 at 9:57pm
|
I had a few hectic weeks and couldn't find time to do a proper writeup. I ended up sending those Royal kits back and ordering 2x32GB G.Skill 6000MT/s CL30-40-40 (F5-6000J3040G32GX2-TZ5NR) instead. They are currently running at 6200MT/s CL30-38-38 without any voltage bumps and I'm more than happy with them. Royal kits were a pain in the butt to get them to reliably boot in 4x16GB 6000MT/s CL28 configuration, but in the end it was possible thanks to manual impedance settings. Alas, this is how it ended as it was time to send them back to the seller. The most I got out of them were 14 minutes in Karhu Ramtest on Ryzen 9 7950X3D with 9950X and 7950X crashing at around 10 minutes. Lowering SoC down to 1.2V increased stability on all of these CPUs. I hadn't tried throwing more voltage at the problem, but I doubt it would have solved anything. |
Posted By: NDRE28
Date Posted: 07 Oct 2024 at 10:33pm
Yeah. I agree. I am happy that your RAM problem is solved now. |
Dieg wrote:Posted By: DemonAk
Date Posted: 19 Oct 2024 at 4:30am
|
Try reset all settings and set in bios: dram freq: 5200, 5400, 5600, 5800*, 6000* dram timings: all auto DFE Read Training = Enabled Gear Down Mode = Enabled Once stability is found, you can turn these back to Enabled: Memory Context Restore = Disabled Power Down Mode = Disabled voltages: VDDIO = VDD = VDDQ = 1.36 VSOC = 1.22 or 1.25 VDDP = auto VDDG CCD/IOD = auto VDD Misc = auto Infinity Fabric = 2000 UCLK = MEMCLK bus configuration: ProcOdt = 48 ProcCaDs = 30 ProcDqDs = 34.3 DramDQDs = 34 RTT_NOM_RD = 4 RTT_NOM_WR = 4 RTT_WR = 2 RTT_PARK = 5 DQS_RTT_PARK = 5 go to AMD-->CBS-->DDR Options-->DDR Training Options and set: RX2D_TrainOpt = manual TX2D_TrainOpt = manual TX DFE Taps = 2 taps RX2D voltage step size (2^n) = 1 TX2D voltage step size (2^n) = 1 RX DFE Taps = 3 taps Memory training time will be 3:40 minutes each reboot until you enable these settings Memory Context Restore = Enabled Power Down Mode = Enabled |