What kit do you have? It sounds like you have the G.Skill 32 GB (2x 16GB) 4400 MHz 17-18-18-38 1.50v (Samsung B-Die, dual-rank) kit, yes?
Last edited:
Intel 13th Gen CPU's
- Thread starter RCHeliguy
- Start date
Sascha Reynders
"Fast & slow girl, the best of both worlds"
Yep, that's the one, part number F4-4400C17D-32GVK.
Nice. I can run it at 4200 MHz 16-16-16-32 and 4000 MHz 14-14-14-28 Gear 1 on my 12900KS, that's with tuned subtimings (secondary, tertiary, etc.) @ 1.56v. I can boot 4266 and 4300 MHz but not stable, not even with RAM @ 1.65v and VCCSA and VDDQ in the 1.400v-1.500v range. Not the RAM's fault, it's almost certainly my IMC's.
On your 13th gen, are you running stock XMP DRAM voltage (1.50v)? I'm trying to see how far this kit can go either with improved timings @ 4400 MHz or improved frequency with stock timings or both but I can't because of my 12900KS's IMC being the limiting factor beyond 4200 MHz.
Have you tuned the subtimings? Tuning subtimings, especially on B-die kits, can bring large performance gains.
Although you have the Ripjaws version (I believe not as good heatsink and no temp sensor), there should still be lots of headroom left in that kit.
Last edited:
Sascha Reynders
"Fast & slow girl, the best of both worlds"
4000-14-14-14-28 is pretty impressive
mine wouldn't run 4000-14-15-15-30, but then again: I didn't raise voltage over stock XMP. On my 12900K I was running VCCSA at 1.25 and VDDQ at 1.30.Yep, stock XMP voltage at 1.50 on the 13900KS as well, although it reads 1.51 in HWInfo64. VCCSA and VDDQ on "Auto" are 1.36 and 1.35 respectively. Any attempt to lower those voltages manually results in instant BSOD's when launching a CPU stress test in OCCT. Guess the KS likes it hot
Primaries are stock XMP except tRAS, secondaries and tertiaries are tightened up:
I could play around a bit more with tREFi but I'm not sure I would gain much from going any higher ? Haven't tried tightening the primaries yet.
The Ripjaws do have temp sensors actually, that's how I know this kit runs hot as hell without active cooling
I picked up a cheap-ish dual-fan RAM cooler on AliExpress and now they max out at around 30c while gaming and 40c under heavy memory testing. Before that, the sticks would easily hit high 40's to low 50's while stress testing...that's why my tREFi is still at 40000. The heatsinks are not that bad either: I've also got a kit of 4x8gb Trident Z RGB 4000CL15 that runs at least as hot. But those sticks were impossible to run at their rated speed in Gear 1 on my daisy-chain topology board (on the 12900K that is, didn't try them on the 13900KS).
Last edited:
Nice! You got some good tiings there. I wasn't sure if you were aware of subtimings and their impact but looks like you know what you're doing.
You can almost certainly raise tREFI to max (65xxx or whatever) but at already 40000, I doubt you'll get any meaningful performance benefits.
B-die is safe daily at 1.60v, probably even 1.62-1.65 but 1.60 almost certainly based on many people running those voltages, daily, for years. I wouldn't worry about raising DRAM voltage to 1.6 or, at least 1.57-ish unless temps become a concern. There are some b-die kits that run 1.55v @ stock XMP such as the 2x 16GB 4000 MHz 14-15-15 kit.
In order to achieve my speeds/timings - 4200 MHz, 16-16-16-32, full timing tuned - I have to run DRAM 1.56v, VCCSA 1.37v, VDDQ 1.40v. Most people say VCCSA is daily-safe up to 1.4v with VDDQ higher at 1.45v-1.50v.
My previous DR B-die kit, the 3600 MHz 16-16-16-36 1.35v kit, could do the exact same speeds & timings as my current kit, the only difference being it needed 1.6v to do so (current kit needs 1.56v). I ran that for a year @ 1.60v, never an issue.
If you like tinkering (for not much performance gain, lol), you should up DRAM voltage to 1.6v and see if you can run 4533 MHz or 4500 MHz @ fully stock XMP timings. Prob need to raise VCCSA & VDDQ - maybe bring them up to 1.4v and 1.4-1.45v respectively. Then, if it works, you can try lowering timings along with primaries. You may even be able to do something like 4500 MHz 16-17-17. If you can't increase frequency, then try lowering primaries to 16-16-16-33 (or 16-16-16-36) or 16-17-17-xx.
Your speeds and timings are good so you'll probably barely get any performance gains, just more for fun tinkering.
What sorts of bandwidth & latency are you getting according to Aida64 and IntelMLC?
You can almost certainly raise tREFI to max (65xxx or whatever) but at already 40000, I doubt you'll get any meaningful performance benefits.
B-die is safe daily at 1.60v, probably even 1.62-1.65 but 1.60 almost certainly based on many people running those voltages, daily, for years. I wouldn't worry about raising DRAM voltage to 1.6 or, at least 1.57-ish unless temps become a concern. There are some b-die kits that run 1.55v @ stock XMP such as the 2x 16GB 4000 MHz 14-15-15 kit.
In order to achieve my speeds/timings - 4200 MHz, 16-16-16-32, full timing tuned - I have to run DRAM 1.56v, VCCSA 1.37v, VDDQ 1.40v. Most people say VCCSA is daily-safe up to 1.4v with VDDQ higher at 1.45v-1.50v.
My previous DR B-die kit, the 3600 MHz 16-16-16-36 1.35v kit, could do the exact same speeds & timings as my current kit, the only difference being it needed 1.6v to do so (current kit needs 1.56v). I ran that for a year @ 1.60v, never an issue.
If you like tinkering (for not much performance gain, lol), you should up DRAM voltage to 1.6v and see if you can run 4533 MHz or 4500 MHz @ fully stock XMP timings. Prob need to raise VCCSA & VDDQ - maybe bring them up to 1.4v and 1.4-1.45v respectively. Then, if it works, you can try lowering timings along with primaries. You may even be able to do something like 4500 MHz 16-17-17. If you can't increase frequency, then try lowering primaries to 16-16-16-33 (or 16-16-16-36) or 16-17-17-xx.
Your speeds and timings are good so you'll probably barely get any performance gains, just more for fun tinkering.
What sorts of bandwidth & latency are you getting according to Aida64 and IntelMLC?
Last edited:
Sascha Reynders
"Fast & slow girl, the best of both worlds"
Thanks ! But I wouldn't go as far as to say I know what I'm doing
I mainly watched a lot of Buildzoid's YT videos on DDR4 timings and overclocking, and then used this guide:MemTestHelper/DDR4 OC Guide.md at oc-guide · integralfx/MemTestHelper
C# WPF to automate HCI MemTest. Contribute to integralfx/MemTestHelper development by creating an account on GitHub.
github.com
I might try some day when I find the time, but for now I'm happy with the performance I'm getting. At the moment I'm mostly focussing on configuring all of my sims to run at stable 90fps at all times on my Reverb G2.
My trial version of Aida64 has expired, so I can't check that. But I've ran both Intel MLC v3.9a and v3.10, not sure how to interpret the huge difference in latency between both versions though...especially since the v3.10 results are in the same ballpark (latency wise) as my old 7700K running Corsair Vengeance 2x8gb 3200CL16 at bone stock XMP settings...
Intel MLC v3.9a results:
Intel MLC v3.10 results:
I know veeeery little about RAM overclocking, but that nice guide looks like an excellent primer.
Incidentally, it also basically confirmed my long-held suspicion that my Asus mobo killed my CPU (or at least damaged its IMC), with excessive VCCIO/VCCSA. The galling thing is that I wasn't even attempting a real overclock, but simply selecting the XMP profile for my DIMMs. (As I've said here before, the response from Asus is one of the reasons they are at the back of the queue for all future tech purchases in my family.)
Sascha Reynders
"Fast & slow girl, the best of both worlds"
It's a pretty good guide yeah, on my last 3 CPUs (9700K, 12900K, 13900KS) I could just put in the tight timings and use the tRFC table to get my memory stable. Except for tREFi, that was always a bit of trial and error depending on temps.
What kind of voltages are we talking about here ? VCCIO is no longer a thing on 12th and 13th Gen, but on my 9700K I remember running it at 1.30-1.35v without issues. And VCCSA at 1.25-1.30v as well, depending on the OC.
It's a few years ago now so I can't remember the numbers (might have 'em written down tho) but one of the two was deeply into the yellow range, maybe one tick away from the red/purple danger zone, and the other was well beyond the threshold for the danger zone.
Ahh, in fact I just found the old pictures: turns out both were purple - IO was set to 1.328 and reading back as 1.3125, SA set to 1.24 and reading back as 1.25. These do sound "small" to me, and are no higher than you mention yourself, but are apparently not advisable for a 7700K...
I didn't notice at first (a week or so) and didn't look at the details in the BIOS until issues surfaced, because I was doing something that should have been a "set and forget" operation, or so I thought. Was pretty cross when I found out, because I had experienced various weird USB issues soon after enabling XMP, which only "mostly" went away when I reverted to the base JEDEC profile on the DIMMs. (So maybe it wasn't the IMC that was damaged but part of the IO circuitry...)
Sascha Reynders
"Fast & slow girl, the best of both worlds"
Weird, IMHO those voltages shouldn't have been a problem, even if they were in the purple. I ran my 7700K at 4.8-4.9 all-core with similar -if not higher- IO and SA voltages for a couple of years and never had any issues, that was on a Z170 ROG Maximus Ranger and later on a Z270 ROG Maximus Code board . In fact, it's still running strong in my work-from-home PC. Maybe your CPU had an exceptionally weak IMC, or you had a motherboard with a lower quality VRM ? Who knows...anyways, not cool Asus support just blew you off on something that sounds like a genuine problem to me.
Hard to know, maybe I just got unlucky. (The board wasn't bargain basement - Z170A - but again, maybe bad luck.)
Thinking back though, I reckon the only reason I ended up even looking into the voltages was because after some heavy google sessions to investigate the weird issues I was having (inc. USB and on-board audio problems) I tripped over some threads with other Asus customers discovering that selecting XMP for their RAM had put voltages into the danger zone, which caused a bunch of really odd symptoms and in some cases permanent damage.
Sascha Reynders
"Fast & slow girl, the best of both worlds"
Z170-A should have been good enough, like you say: probably bad luck...
I must say: that Z270 board I mentioned also gave me problems, blew out the USB3 controller for no apparent reason. Didn't bother to RMA it though 'cos I still had the Z170 lying around, which is still running the 7700K as we speak by the way. But that's been the only problem I've had with Asus out of the last 6 boards I bought from them.
Sorry to necro bump... But I am just wondering... Did anyone here use or need the upgraded contact plate around their CPU? I have seen mixed suggestions regarding them. Mixed results too.
Makes me wonder if I should do same...
I'd say it depends on how heavy your cooler is or rather how much your mobo is flexing right now.
And how often you push the cooling to its Limit.
If you're only simracing with 50-80W and have a light watercooler on the cpu, I don't see a reason to upgrade it.
But if you're rendering a few hours per week and have a Noctua D15 hanging on the mobo, then I'd do the upgrade.
And how often you push the cooling to its Limit.
If you're only simracing with 50-80W and have a light watercooler on the cpu, I don't see a reason to upgrade it.
But if you're rendering a few hours per week and have a Noctua D15 hanging on the mobo, then I'd do the upgrade.
No rendering, but I have a Deepcool AK620 I will be mounting. Should I just order this?
If you change the cooler anyway or are just building the system, then I'd definitely get the frame!
Especially with that big fat cooler
I thought you might have your system running already and were thinking about taking off the cooler + CPU, putting the frame in place and remounting everything.
Last edited:
