RANT: I HATE THE INTEL 13th GEN MEMORY CONTROLLER
ฝัง
- เผยแพร่เมื่อ 3 ต.ค. 2024
- My Patreon: / buildzoid
Teespring: teespring.com/...
Bandcamp: machineforscre...
Twitch: / buildzoid
Twitter: / buildzoid1
Facebook: / actuallyhardcoreovercl...
#overclocking #ddr5 #intel
I grew my hair back after I quit memory overclocking as a hobby. Went from a Norwood 3 to Norwood 1. My blood pressure is down to manageable levels too.
Can't believe I've been throwing my money down the drain on Propecia and Minoxidil when quitting overclocking would have been the ultimate hair loss cure all along!
…aaand you instead decided to meme on or (much worse even) buy bored ape yacht club NFTs.
No stress like moe, more happy life!
Same🎉🎉 i feel free now :)
why worry over cheap Taiwanese and Chinese crap lol.
8 motherboards
3 CPUs
3 Hynix A-die memory kits
Absolute hell with every single combination if you decide to push them.
Are there even eight 2 dim z790 boards because I don't think this type of overclock makes sense on 4 dim board. Z690 unify could work also but still from 690 lineup
@@metroplextitan8816 no, there are only 4
@@metroplextitan8816 the 4dimmers are hell at 7200
Z790 Apex
@@ActuallyHardcoreOverclocking I know that extremely well when I built my wife's pc tforce 7200 wouldn't even post on z790 hero let alone pass stress tests
You're doing it wrong. You're supposed to run a couple of tests for 2 minutes, then blame software for being buggy when it crashes!
The average PC enthusiast does seem to assume that if it POSTs, it's stable.
@@GhostMotley and if it doesn't, the BIOS is crap and needs a hotfix, the hardware sux or [at best] they lost the silicone lottery 🤣
@@ChrispyNut I see people all over Twitter/Reddit/other forums claiming they are running DDR5-8400+ stable, some even claim DDR5-9000+ stable.
Would suspect they are likely lying or they aren't stress testing the RAM hard enough or long enough.
@@GhostMotley Considering most people don't even realize memory in general can have silent memory errors and even attempt to correct it many times a day...
I'm gonna guess most people don't understand what perfect system stability looks like and how to test it
I think Intel is binning too much for the 13900ks. No memory worked until I switched to the ks from the k.
Finally a realistic reaction to unstable memory overclocking.
Crazy man yells at a piece of silicon for an hour (love it, though)
... and instantly enjoys it
It's a therapy session. I feel him. I remember when I tried to overclock DDR4 on X99 when it came out. Lost my sanity for months. When I upgraded from that system to the Z390, I didn't even want to overclock the memory. I just set the XMP and forget it...
Man, I've been trying for 4 days to get DDR5 7200 working on my 13900k on a MSI Z690a.
I've been experiencing the exact same insanity as you where after dialing in sweet spot voltages, VST will increase and increase in stable duration until all the sudden it hits a brick wall.
Then I change 0.05V and it fails instantly.
Im starting to believe that DDR5 on 13th Gen has numerous local minima of voltages which sort of work. Which is hell for optimization.
In other words 1.35 cpuVDD2 and 1.45 cpu VDDQ work well, but then nearby voltages don't work at all.
But then 1.41cpu VDD2 and 1.53VDDQ also work pretty well, but 1.42 or 1.4 throws errors in seconds.
Have you tried increasing Vpp voltage? I was wondering if that has an impact on stability. I hit a wall with 4x16gb corsair 5600 vengeance on a msi z690 torpedo ekx with 13700k at 5600. Finally managed to get to 5866 by upping Vpp on ram and passed testmem5
@@darkon81c @darkon81c Yeah man, trust me I tried everything.
I basically scoured Buildzoid's entire z690 z790 videos and tried everything I saw he ever touched, and then touched stuff that he didn't touch.
Vccin, Vpp, PLL voltages. Raised cpu vcore, reduced ring clock, tried different LLC settings.
Later I even got crazy and started messing with DRAM voltage switching frequencies and like I mentioned, I found a couple random sweet spots which could pass 2 or 3 iterations of VST with switching frequency raised.
About the only thing I didn't try was changing cpu mounting pressure or cooler mounting pressure, but I'm just done with the headache of testing and validating whether or not it actually does anything or if I just randomly get lucky.
Like BZ said, it would randomly be semi stable(pass VST for a few iterations, and do so again when I try again without restarting), then after a few restarts or the next day I would load the EXACT SAME settings and they would just not work.
Update: On 7000MHz, just had it pass 100 iterations of VST, and 400% hci memtest.
Then restarted and it crashed instantly when I tried it the next morning.
@@0sm1um76MSI z690a is a trash board
@@0sm1um76had vst run for 3 hrs. Then had prime 95 go for 24hrs on both cpu and memory stress tests. Then I updated my bios and nothing is stable
This is exactly what i've found. It's is basically impossible to say that a 13Th gen platform is stable.
it's random and got me crazy for months...
i've tested
7 13900k
3 13900ks
3 PSU
4 MB
and.... 9 ddr5 kit.
A 10900k with CL14 memory is much snappier than any DDR5 kit. You can not only see it but feel it. lol. I went from 45ns with a 10900k to 64 ns with this DDR5 Kit with the 13700KF. My benchmarks went down instead of up with DDR4 CL14 memory 3400 (interconnect at full speed stable) where as the DDR5 with the 13700KF I am stuck at half the interconnect speed causing much higher latency.
@@benjy117 it's always the same thing.
a new tech in the early stage "aka for early adopters" is overpriced and not good at all at least for 2/3 gen.
Ddr5 is a scam. CPU memory controller can't handle it. A waste of money.
@@benjy117Tech Yes City seems to confirm this in his latest videos. He says he's noticing weird lag in Windows and his video editor with 13900k that wasn't there with 10900k.
its not @@benjy117
I'm glad I'm not the only one that was going crazy trying to achieve higher memory DDR5 speeds.
I also experienced the randomness of DDR5 stability.
Great video's love the deep dive, one of the few channels that actually have street cred with components. I can tell your pissed when the "kermit" voice rears its head - its all good. Thanks for the great content!
"rant" in the title of a one-hour video.
Man, this is gonna be good
"what the alignment of the sun and moon is right now" LMAO!!! 😆🤣
This is where the fun begins
DDR5 8000 stable is about as real as 2000-2133 FCLK stable on Zen 3. People just don't stress test very thoroughly.
ON non-APU and not 5500 specifically of all Zen 3's. Zen 3 APU's can hit 2300 FCLK stable unlike CPU's - somewhere in 1900-2000 range
About as stable as 1700 FCLK on Zen1
Zen3 APUs can shoot beyond 2500 stable that just needs dumb levels of VSoC. Zen1 can almost touch 2000 but it has to be very certain RAM, board (not every cheap bucket fits) and lucky silicon
@@previousslayer 2500 FCLK with iGPU running or iGPU disabled?
It's not cursed. It's just RNG on it's stability. Every time you ask it to do something it rolls a d200 and if it lands on 200 it crashes.
As an Intel engineer I can confirm this is correct. The reason we were stuck on 14nm for so long was because we couldn't find smaller dice.
@@DigitalJedi Nice
Love your content. Friend and myself both have almost identical setups and we maxed out at 7800 with the tightest timings we could manage on a z790 maximus hero. 8000 was impossible to stabilize. Same experience. Personally I was never able to get anything above 7400 stable until I purchased a 13900ks. I suppose the memory controller on the ks is indirectly binned because of the seeming correlation between the p core SP score and the strength of the memory controller, but who really knows. I also had to purchase custom heat sinks for the ram to keep it cool enough to remain stable with high trefi timing.
thank you for this post. im running a i7 13700k (with hero z790) and stable at @7400, but @7600 it falls of a cliff. my ram is simply too fancy for my cpu chip. Z5 trident, its rated at 7800 at 1.45v 36... i guess there is maybe some tiny room left on the timings. but the frequency seems borked at these speeds. cant run any faster than this. Seems like what you say is happening. I read another update from a person that tested allot 13900k chips. They found 1/15 runs 7800 or 7800+. So 1 in 15!!! I guess ks chips are nicer. Lol. Ill be fine at 7400 for now. Just happy its stable.
@@Ks-zz9lh yeah absolutely. You'll gain more from tightening the timings than increasing the frequency but 7800 is possible with a very very select handful of motherboards. The ONLY board I know of that consistently hits 8000+ is the z790 apex.
I'd also HIGHLY recommend you do NOT run an I9 wide open on stress tests like linpak. I degraded my own 13900k screwing around and I found out. Didn't take months, or weeks or even days. It took 5 hours running aida64 extreme at 290w and from that point forward, I had to use 1.35v on the vcore to stabilize the CPU @ 5.6ghz where as before it was stable at 1.28v. Almost NOBODY is going to tell you this, but truly, watch the the wattage at extended periods. Strongly recommend you keep it under 300w
@@Unobserved07 holy moly that is a horror story of its own..
I stay away from p95, linpack, etc.. those ultra heavy things I'd never see in real life usage.. but that quick deterioration is crazy considering the voltage you mentioned to start isn't even that high. Does gaming with a 13900 ever get that high wattage? If you overclock and are at like 1.3v on an overclock, and play games for like 8 hours... that gonna degrade?
By restraining it, would that mean keeping your overclock settings but just limiting its power limit settings to like 253w?
@@Unobserved07so are you saying simply running 300+ W for extended periods by itself degrades the CPU? Even if the temps are kept in control? Or is it just because 300W usually means insane heat?
I knew going DDR5 this early was gonna be a hassle. 13700K with vipers 4000CL19 at 4200CL16.
How tight did you make the tertiaries? If I play with any of them, it ruins my reg in bf4 😒
What I've found is that you can find meta-stable VCCSA/IMC VDD2 settings that will randomly break and that a truly stable configuration will be completely different. E.g (1.2 VCCSA/1.5VDD2 metastable, but 1.0 VCCSA/1.4VDD2 stable @7200 dual-rank)
I've had similar issues with my 13900ks with mounting pressure. I put on a contact frame and oh man..... i had to remount 5-6 times with different levels of torque on the frame and cooler to get decent timings and the stability can change day by day. absolute nightmare and i can't even tell if the IMC bin on my CPU is trash, my MB is trash or my ram is trash because i'm getting random errors no matter what I try
I’m curious as to what speed that matters at? I’m not an ocerclocker I just want to know for reference and buying decisions. You say right timings but do you mean at any speed or 7200 plus?
@@nathanwhite617 7200 is very very difficult to again in my experience with a 4 dimm motherboard.
If you're not buying a board specifically meant for ram OCing(more than $250), 7200 will take you days probably.
That being said, 7000 could be easy or hard depending on silicon lottery luck.
I don't know this for sure(this is speculation on my part having used 1 cpu, 2 ram kits, and 2 boards), but based on my experience I feel 6800 should be attainable for most systems as long as you get M die.
YES HOW THE HELL DO WE FIGURE THIS OUT? My main dev machine kept having random crashes of all types of software and I looked into it - spend all this time testing RAM and I wasn't even trying to overclock it Just run it at 3600. I ran 128 GB fine at 4400 but when the degradation finally reached a critical point I had to wipe everything. Now I can't even reinstall anything until I figure out why I'm getting errors on a fresh installed windows (from a clean windowsUSB made another computer)
I moved out away from my kids so I could build this and now I'm f****** stuck in the water. And I don't have the money to just replace the motherboard CPU and RAM and then sort it out later
Thank you for being the guinea pig for arrow lake
I know this is kinda old but how tight did you end up mounting the contact frame?
Intel memory controler's haunted.
I wonder if the intel 13900k and 14900k issues cropping up are related. I saw someone post that they often see discoloration on the mem controller contact pads likely csused by heat.
Framechasers is going to react to this no doubt
I shoulda made it longer to waste more of his time then
@@ActuallyHardcoreOverclocking Holy based
@@ActuallyHardcoreOverclocking yeah f that guy
you mean try learn anything, he is scammer unstability king
🙃🙃🙃🙃🙃 love u zoid
You’re quite correct. 8000 was totally broken on my 690 Dark as well. The 790 Apex will get you there though. It’s inexplicably good. I don’t know what ASUS did to engineer it but it’s basically sorcery.
Does z790 rog strix E. Stand a chance against 790Apex?
@@princekittipon6510No shot. And part of it is poor BIOS support. My Hero has BIOS 0008 and 0032. It’s no Apex, but these two revisions make it a respectable offering.
@@princekittipon6510 Nope. Apex is an 2-dimm board. Strix E is a 4-dimm board. Not even close in comparison.
@@OnyxReXtNIt’s not the main factor. 4-DIMM is fine if you know what you’re doing. It’s not a majorly different experience when you’re using a competent sample. Some people get absolutely godawful cursed PCBs and I feel for them. The Hero sample I got is an absolute monster tbh. It makes the 690 Tachyon I tested look foolish. Compared to my old 690 Dark it’s sort of a toss up.
@@OnyxReXtN i have g skill 6400 32-39 A die. And i dont see any difference between 8000and 6400. Why are you guys overclocking again? Im not being sarcastic. Just curious
I feel so vindicated, I'm on DDR4 but I've been experiencing the exact same thing pushing 2xDR 4133 Gear 1 on my 13700K. Stability tests seem to be completely random:
VST will pass for an hour and TM5 will fail in 4 minutes on the same settings
TM5 will pass 32 cycles of Absolut and VST will BSOD in 4 seconds on the same settings
TM5 will pass 32 cycles of Absolut and HCI will get 1 error at 1500%
No combinations of voltages, RTTs, etc. achieve 100% stability like I'm used to
I thought I was just doing something wrong, but I'm so glad I'm not alone. I'm ready to just daily a 99% stable system instead of 100%. Stability just seems too random to dream of 100%!
Lower to 4000mhz
I was having the same problem with 12700k + 2 DR 4266mhz... Just lowered 1 bin (4133mhz) and it is okay now... Try to lower 1 bin.
Exactly Same issue with 13700k, DR S8B, 4300/4266. You are not alone.
I think this video is pretty helpful, for starters I now know that it's not worth to try stabilize anything above 7600.
Damn. Imagine paying top dollar for not just pretty expensive CPUs, but especially expensive RAM and MB just to be met with perpetual instability. Wow
Yeah, DDR5 and 13th gen are cursed.
13900k with Z790 Apex here, currently trying to get 7000 stable, because everything above 7200 causes issues 💀
And that with G.Skill 7600 sticks.
So close to selling my Apex and going back to DDR4. And my MC SP of 70 probably doesn't help here either, but still...
Yeah think a lot of it is just silicon lottery of the IMC. I bought a second 13900KS and after my first wasn't stable at 6800Mhz DR 2x32GB in my Z790 Apex. My new KS is able to do 7200Mhz but it's unstable randomly like in this video :
Thank you. That video pushed me into the right direction. Tried 3 days to get my 7800 mhz a-dies to work. Same experience with random success, fails etc wirhout changing anything. Then tried your adviced and went down to 6600mhz but with cl30-38-38-28. Reduced latency massively compared to my old 6.6/c32 corsair kit, raised bandwith by 10% and it is reproduceable rockstable. 14700kf/z790-e strix/TG Delta 7800 2x16
My last experience trying to get high RAM speeds going ended with a corrupted BIOS unable to even post anymore. To my luck it was dual BIOS and i was able to flash it back but at that point i knew i dont want any more of that...
@@pcoverthink Some Gigabyte DDR3 boards just had a tendency to corrupt BIOS. Which is probably why at the time gigabyte had dual bios on pretty much every board. I haven't heard of it or seen it on any other board.
madness is not when you do that same thing over and over and expect different results, madness is when you do the same thing over and over and do get different results! thanks intel!
this actually make me remember the bad 12900k / ks vs the great 12600k that you had in some vid
Based and overclocking pilled
The problem is data corruption, otherwise I wouldn't mind.
Sounds like intel has cut out internal error correction stuff to get higher max speed out of the controller, while AMD has internal error correction that end up capping max speed? (like amd does on their infinifabric also)
And then intel has ended up with some design issue that sometimes creates random errors, that then just get passed on to the cpu.
I’m running 8200 on a 13900 KS .Z790 tachyon board.Run memtest for hours y-cruncher .All pass no problems.I also had a 13900K cpu same board and memory sticks was absolutely terrible experience same problem’s he has in this video .13900K biscuits score was 88.13900KS biscuit score 98+.Night and day difference.
8000 here same board easy really 50ns.
I had the same problems, it was due to the temperatures - adies simply close at a certain temperature and do what they want , I have lowered my temperatures below 40 degrees and all the problems have disappeared. - i use karhu (10.000%) & ycruncher vst @8000mhz and its stable with z790 apex
you also have an Apex. Which is the one board I haven't tested.
the z790 apex , and the evga 790 kingpin can both do this - are only limited by heat, the ram kit is rather less the problem - try to lower the temperature. good luck and have fun :)
i had no problems with the heat on my mdie kit but adies are bitchy.
@@ActuallyHardcoreOverclocking I think it would be genuinely interesting if you just set up a little small water loop just for memory on your bench.
I think it would be worthwhile for the amount of time and energy you spend doing this lately. Doesn't have to be fancy or expensive. Just a cheap 4dimm block, some spreaders, a 120mm rad and a pump/res (and obligatory tubing/fittings).
Just a thought. You will still run into trouble of course with the god forsaken ddr5 memory controller, but it is much more manageable until the limits are really pushed.
@@newdawn005 you can't even hit more than 7000 with your m-Die so potentially you didn't even run in to the trouble BZ is ranting about.
@@OMGJL mdie ? i use 7600 gskil adie for-> 8400mhz+
I'm simple man - I see buildzoid ranting for an hour? I say bring it on!!!
What about creating some kind of device which will go and brute force with every possible or limited amount of tests and then output you information which configuration will work stable - yeah, I know it would take ages to test, but you could test multiple configurations at the same time or find best one with your PC.
People have talked about this forever. I recall some crude attempts but it's really just not practical. It would need to be tuned to work with specific bios, able detect boot failures then manually reset or even clear cmos. Then it would need to interpret the stress test output while simultaneously able to reset Windows if it locks up. That's assuming there's minimal logic and it's just adjusting one arbitrary value at a time which is only so helpful anyway.
So yeah, feel free to tackle that if you want.
Imo a text adventure kinda app that you run on your phone and follow instructions that it gives while OCing would be more feasible than that
Apparently the Final Fantasy XIV (specifically with VR mod in mind) is a potentially nice memory benchmark (it seems to scale with cpu and ram speed very well). Just throwing it out there, I haven't tested yet if it's just latency or bandwidth matters too.
Thing with the vr mod is dropping fps below double the refresh rate of the headset causes severe stutters so even a small difference can act like a pass/fail test there.
Final Fantasy XIV is not benchmark is way of life
Love seeing this game everywhere
Thanks for the video. Glad that I listened to it on my phone as my not 3 month old system (13900k Z790 Aorus Master with 7200Mhz ram XMP without tinkering) blue screened even during it. I've lowered it now down to 7000MHz lets see what happens ... wish me infinite luck, I will probably need it ^^
he tells you to lower it down to 6k fyi if you didn't get the memo :)
We recently designed an AMD 7940HS system and run it at 45w, 54w, and 65watts and plugged in the 48GB DDR4-5600 DIMMs to get 96GB of memory. The memory runs at lower than 70c max temp and works great.
We then designed an Intel 13th Gen i9-13900H system running at 45w, 55w, and 64w (speed doesn't matter regarding the problem) and plugged in those same 48GB DDR5-3600 DIMMs into it and they run at over 100c, even if we slow the CPU wattage to 45w. In fact, it was over 120c and we did a BIOS update that lowered it closer to 100c.
Plugging 32GB DDR5-5600 DIMMs into the Intel motherboard lowers the temps to unver 70c. So for some reason, the BIOS is programming some register combination weirdly due to the timings is reads out of the 48GB DIMMs and just cooks the DIMMs.
Thank you for this video as we are going to use Mem TweakIt to read the register settings with the 32GB DIMMs installed and program the registers identically on the 48GB DIMMs and see if the temps come down
Considering the data rate of ddr5 its very possible the motherboard manufacturers are doing an abysmal job on the board layout. Do motherboards usually spec highest ddr rate? Or provide data on their own certification of those paths?
Well, I've been saying pushing limits is waste of time since ages, because you see marginal gains if at all and it's ludicrously hard to stabilize it. I never get greedy, do to a point where it's fast to do vs where I get most of the performance gains over stock. As result, never felt like wasting time nor felt like leaving meaningful performance gains on the table, because if gains are like 1% - that's literally test to test variance - a margin of error worth gains.
Sure, for competitive overclocking the perspective is obviously different, but alse stability is not really relevant further than completing a benchmark and recording the scores.
The socket mounting shit is a nightmare. Its bad enough that you have to think about and see how tight the cpu heatsink and all is for even temps, now you have thermalright or grizzly socket replacement thingies to make it flat, but those need specific torque and stuff.. now memory controller is affected by pressure.. christ
Hey Mate! Thanks so much for this video! I've been pulling my hair out with my Z790 Aorus Master and i9-13900k. I've been running Corsair 6200 C36 kit 2x16GB for some time. XMP was working fine and I had it OC-ed to stable 6600 34-39-39-38 @1.35V.
It's been running like this for half a year. Now it suddenly stopped working, shortly after installing newest firmware for the mobo (to fix vulnerability). The kit still works at JEDEC spec, but the XMP profile or anything fails mem training. I've been trying to downgrade uefi, but it didn't help. Kit doesn't work outside JEDEC spec. I am now trying G.Skill 7800 36-46-46-125 @ 1.45V. I suspected it might not work and I was correct. The XMP runs Windows only, even memtest fails. I am now trying to tune it at 7200 but I still struggle to get it working stable.
Your video has explained all my issues with system working stable and then after reboot failing same tests it has passed. It is so goddamn inconsistent. It runs for hours, all tests done. Then after power cycle it crashes first minute of memtest!
Thanks for your constant work!
Holy shit man, I've learned more stuff about PC memory from this video alone than from 3 years at computer science...
I'm not surprised at all Lol. I have 3 brother whom all have a Computer Science degree yet they're quite oblivious to any of this.
You cant defeat silicon lottery.
I think what you are dealing with is the ECC it self, the system may be producing hundreds to thousands of errors but as you know there is no way to you know in the way the ECC is implemented, it corrects them but some may pass or originate in a way that it can not correct them as it's a bit more limited than the traditional ECC modules, maybe 1% pass of them pass at any random time? This would require days of testing to reach a good measure. I read a year or two ago that the real reason for the ECC exist in DDR5 consumer modules it that the frequencies were so high, that it were naturally producing a lot more errors than on DDR4 and below. Pushing it to 8000 Mhz may push it even extreme that's goes beyond what the ECC can do for you. In such way that errors may be happening all the time at the background and at any random time some just goes through.
Is it possible to disable ECC on DDR5 for stability testing, then re-enable it once target performance is achieved?
@@We_Are_I_Am Sorry for the long period to this answer, i forgot it. As far as i know, no, there is no way to disable it as it works as a complete and independent chip in the DDR5 memory that's targeted to the consumer. Servers may have with their full ECC support from the modules but they usually works at way lower speeds.
I will just leave it here:
7800X3D+6000MHz+AnyAM5board>Intel
I can attest DDR5 stability is very random on Intel 13th Gen at XMP. I tried 4 different kits ranging from 6000 to 7200 and even tried some manual tweaks and downclocks across 4 different Asus Z790 mobos and all kits not totally stable as random WHEA or BSOD in OCCT Large Data Set Variable Test. Never happens when on auto which gimps RAM to 4800 in Gear 2.
DDR4 at XMOP and tight timings is fully stable. Even DDR4 4000 at Gear 1 is fully stable across last CPU and 4300 CL16 almost or maybe even fully stable with 1.33 VCCSA passing all tests so far.
DDR5 stability is a mess on Intel 13th Gen.
Boot - to - boot stability is something that needs to be confirmed with ddr5, and extreme cases confirming it even with cmos reset to apply the same profile.
Signal integrity changes from boot to boot/cold boot, and all of those little things on an electrical level change. I experienced the same thing trying to stabilize simply XMP. Pass Tm5, Vst, Vt3, 2.5b, over and over. Shut down and boot the next day? Error.
The closest ive been so far is by actually reducing the voltages below xmp, and that yielded a series of boot, pass, shut down, wait hours, boot, pass. 3 times over.
BZ, one last hail mary with a z790 Apex, I'm sure you can get 8200+ with that board.
It seems I have no problem listening to rants if creators aren't acting like they're not ranting.
Buildzoid, what's your thoughts 13gen with ddr4? I'm running a 13700kf with 3200 Cl14 Samsung b-die and it's actually and using your videos has helped heaps to make the system faster 👍🦅
feels bad man thx for your hard work
Now that is some quality ranting I haven't seen from BZ in a while. Noice!
Oh holy hell. You actually did it; appreciate the video. Sorry for your headache? I’m the guy on Reddit that was disagreeing with you 😅
This must be one of the best video's on YT ever,
this kinda **** is rather relatable
I kept crashing at 7600MHz until I bought and installed Thermal Right contact frame, no issues anymore for 2 months.
Intel's locking mechanism for the cpu is trash.
I have been this salty. I've wanted to snap my 10900K in half while trying to get 4500CL16 DR working.
It's haunted with the ghosts of sacrificed Intel engineers
Your desktop background has me dying xD
Only MB I have been able to get 8000 stable is Z790 Apex and Z790 Tachyon. Z690 (Not 790) Tachyon did 8000, but tRCD and tRP had to be 55 or it simply would not pass y-cruncher. If your using a 4-slot MB...good luck! Probably never going to happen unless A) A lot of CPU SA/VDD2/TX voltage or B) the next CPU comes out with a better IMC. The sweet spot for my CPU is SA 1.25-1.3 - TX/VDD2 1.4V. This is good up to 8400. Might be just the limit of my IMC with daily safe voltages.
Seems for the Z790 Apex, contact frame or stock LM makes no difference for max mem freq for me. I top out at the same.
Z790 tachyon easy 8000 50 ns stable all day long.
The Asus Z790 APEX is amazing. Just got this board and set it up last night. Got my Gskill 7800 xmp kit stable at standard xmp settings. Haven’t had time to try to push it to 8000mhz. I’ll probably just leave it at 7800. I have it paired with a 13900ks. My SP score floats between 108-115 depending on the mood it’s in on boot up. p core sp score of 118.
@@hadleys.4869 you can do 8000. Set xmp
7800, change to 8000 leave all auto. On my z790 tachyon vdd2 is 1.47 and your mc might be 1.45 enough. Change dram voltage 1.52 and 1.5. Enable high band, low latency, low trip. Go to timing and trefi 65535 and right below 255. Change timing to 38-48-48-48. Make sure you do with clear comos and train mameory by disable fast boot. Couple restarts and shut down than go back to auto. Test and play. This is stable simple 8000 with 54 ns latency. It’s daily if you sre not gonna water. I am on water now so I am 8200
@@facegamefps6725 Yeah....Getting advice on YT is like getting advice at a bar. Might has well take a hammer to the computer. @Hadley S. go to a real forum for advice.
@@theodanielwollff why you care…?I tune my own and I can help people if I want to..matter of fact I will post video so z790 tachyon owners can do it easily. I will buy you hammer and you can destroy your own
I feel you my man.
I coudn't make your timings (HWU video and your "Easily improve your DDR5 XMP's performance", both) run on my 6200 Patriot set.
And at the same time it is rock stable on 7000 with stock 7000 Patriot timings.
Now I'm thinking about droping it to 6000 and trying again...
His timings are extremely aggressive. Just leave everything default except for tREFI (don’t increase past ~15,000).
TRDRDsg = min floor 8
TWRWRsg = min floor 16
Leave the rest default. There’s no tangible difference in performance.
@@bronsondixon4747
Lowered it to 6800 and tried to tune a lil. Basically moved all timings halfway from stock to buildzoids. tREFI 32000
Stable. Will work from here.
Well if you just don't want to put in the the work to make it stable just say so -Some random redditor
I see people saying 8000 is possible, but who knows what they consider stable. Maybe they are seeing OC records and not realizing they are on Ln2 or something.
Meanwhile I'm sitting here with my 3200 DDR4 never experienced any blue screen on this PC. I think I will always just listen to hardcore overclocking as it seems scarily hard to do. Thanks for the content.
ddr4 is easy if u have the right memory chips
Its free real estate
@@bigcazza5260 have dual rank B-Die with 13700k, and am experiencing the same behaviour as BZ @ both 4300/4266.
20 Loop of yCruncher VST and 3 loop of TM5, all passes, and HCI crash at around 100%, next power cycle might show OC failure.
Upping or lowering SA,VDDQ might allow booting and passing VST again, and another power cycle with no setting changes it pack up again.
@@OMGJL are u locking down iols?
@@bigcazza5260 nah, to isolate the issue, I was running 18-22-22-42-2T everything else auto and E core off.
I have Messing with SA and VDDQ voltage both ranges from 1.35 to 1.45 in different combinations.
I am quite certain it's not memory training nor memory itself related but Intel IMC, because at one stage it passed yCruncher VST 20 cycles, and tm5 extreme1, without restarting HCI fails around 100% (1 error each on 4 different thread) and with no setting change it won't boot the next day.
VDDQ TX sweet spots extremely hard for me. I get 100% stability on all test, even TM5 for 48hrs, but cold booting fails. Too high and itll fail test in less than 20 minutes and too low and it wont cold boot. I wonder if the socket heating up afffects the CPU socket mount preasure making cold booting fussy.
had the same issue on a 14700k gigabyte build!!
I have Intel 13700K, Asus ROG Strix z790-e, DDR5 Kingston Fury Renegate 6400 CL32. I have no overclock experience. I spent much time to overclock the memory reading guides. Run the same problem: one day tests are passed OK, next day errors come. Tired and returned to the XMP profile. Stability is much more important than additional percents of perfomance. Overall system perfomance is great anyway.
It's like an unstable FCLK ratio on Ryzen 1xxx-2xxx
I remember gettings random seconds of stutters in games / training instabilities which lead to errors.
Complete waste of time, it's just easier to lower the clocks
Have you tried a contact frame for 13th Gen? Kind of need it to ensure solid contact?
I totally feel your pain. Got exactly the same issue with DR 2x32GB at 7200C30 on apex z790. Some reboots no VST or MT5 issues other times it just errors instantly. I think the training on the edge is super inconsistent when it comes to ODT but on the apex there's no way to see what the values being set are :(
Does it run okay at like 7000C34? I’ve been considering changing up motherboards to a 2 DIMM.
I’m running Z790 Strix E with 2x32GB at 6600C32 everything totally stable.
My goal is 7200C34 on an Apex but now you’ve got me doubting that.
@@bronsondixon4747 I'm sure you'd be in with a chance at getting 7200C34 but not sure it's worth an upgrade. I think it's just luck whether you have an IMC that can do the high speeds. No amount of tweaking can get you that much of a bump in extra Mhz other than switching out for another CPU (which I've done once already as 6800 wasn't even possible).
@@djangoofyork yeah, I’m probably not going to change it out.
Was considering doing an all out build with the 13900KS + Apex but I’ll just wait until Raptor Lake refresh.
is everyone locking down iol and rtl? mine will change to unstable 1 in 5 boots if i dont lock it down. im at the limit of my ddr4 tho so different memory
Hi Buildzoid and community! Sorry for an off topic question, but because this videos is about strange behavior of Intel mc it reminded me about the other strange thing with AMD Zen 4.
I've been able to oc and stabilize my hynix m die 5600 36-38-38-38-80-118 to 6200 32-37-37-37-28-40 with tight secondary and tertiari timings (tighter or same as in your hynix guide), but I have a huge difference in read/copy and write speed in Aida64 now. 95 Gb/s write and 64 Gb/s read/copy to be presice. My CPU is 7600X, Mb is Asus Rog Strix B650E-E.
I've read almost all forums and guides, but without success. Several people said I should try tuning my Proc ODT or Rtt settings or MEM VDD(Q) (currently it's set to 1.25), but it didn't help (or I'm just really bad).
Do you have any tips or ideas? Thank you in advance!
AIDA is just a bad benchmark. The numbers it produces don't mean much of anything.
@@ActuallyHardcoreOverclocking Wow that was quick! Thanks a lot!
So everything is fine? Is there any benchmark I can run to verify my overclocking is meaningful? Because Aida didn't show any difference in read speed between 6000 with auto timings and 6200 with tuned ones.
@@ruslan_riazantsev y-cruncher is a good indicator on terms of compute memory performance and frame time stability in games tends to see considerable improvement from tighter memory timings, so running game benchmarks is also a good way to check. For games, you probably won't see a big difference in average FPS, but 1% lows should be a little better.
@@hellraserfleshlight Thank you, it's a useful tip. I'll definitely try it!
I got told by a random on Chinese forum saying that the random instability is caused by IMC overheating.
So CPU doesn't have temp sensor on IMC, and you won't know if it's over heating. HCI/yC VST will stress IMC pretty hard and overheats IMC -> instability.
I too experience the randomness on 13700k + z690 Pro A + dual rank B-Die above 4000, but I don't want to delid the damn thing and get a direct die block just to test out if this is the case. I can settle on 5.5Ghz + 4000c14... with all that effort and money, it would probably end up with something like 5.6/5.7Ghz + 4266c15/c16, I cbf...
Buildzoid maybe it's worth for you taking a look into this, you have much more hardware resource than me and have more enthusiast on it. I hope you can see my comment.
I think your problem is DDR5 itself. They have pushed the tech beyond reason bettig on internal ecc to fix their physics-limitted shoddy-engineered shitware
Is it possible that this is a side effect of it being ecc?
Guys, this video is from the future, it would happen sometime in April 12, 2023.
My 8,000 speed GSkill DDR5 memory kit is working on my ASUS Apex Z790 Mobo, pared with the 13900K. Didn’t have to do anything but have the latest Bios and apply XMP profile and that’s it. I do have a Fan mounted over the top of the 2 dim sticks running at 1,000RPM. I have a Gen 5 NVME also that can get quite toasty too so I mounted another fan blowing on that even though it has a double sided heatsink. I think I’m going to eventually get me a Chiller and start water blocking the CPU and GPU on any future PC builds. It’s the way to go unless these chip makers start delivering cooler running CPUs.
I can understand the frustration but the question is: is consumer hardware even designed to withstand sadistic stress tests for hours upon hours?
I mean at what point is the testing complete overkill and how does it translate to actual real world workloads?
If you want 100% stability for a mission critical system, you would use JEDEC ECC RAM and not XMP overclock shenanigans.
E.g.: on my z690 Hero (12700K) I can boot DDR5 6600 XMP no problem. I can game, work in Blender whatever.
Yet Y-cruncher fails within 5 minutes.
I reckon, most average joes that put to a system together kick it into XMP, start working with it and unless they actually encounter BSODs or crashes, will never even know that even their bog standard XMP setting was never stable to begin with. I worked with 5600 XMP for well over a year with no issues at all. I never tested it. No clue whether it would have error'd on Y-Cruncher or TM5 or whatever.
Considering that fact that we cannot actually prove stability on any overclock (test it for 10hrs, it might fail at 12 or 24 hrs), the question is: at what point in testing is the stability within acceptable margins?
How big is the probability to actually lose data due to memory corruption (though shouldn't the file system's error correction be able to cope with bit flips?).
I had a similar issue with my old z690 and 12900k... changed out my board haven't had this issue since... one day it would be fine next time i did tweaking id get errors on the same settings... went with the unify x from msi and all my random instabilities went away now i can leave test mem 5... and prime 95 small ffts on for long peroids... before it would bsod on most stress tests or throw up errors.. honestly i think its a motherboard issue i could throw insane voltages at it and still couldn't get the ram fully stable... or it wouldn't throw up errors for a week then they would return... i was pulling my hair out just like you are haha... its worth you trying the apex z790 or unify z790... (if you are already one one ignore me) the unify x boards from msi are great for getting ram stable on intel.
Also your trefi seems pretty high on air cooling. Other people running daily 262K usually put waterblock on mem, because trefi is very temp sensitive. On auto z790 apex uses 32767 trefi maybe that is reasonable trefi for any mem kit that run with stock cooler.
65K is doable with a fan
@@a_void depend on memory clock too, as i testing, from my result, higher memclk usually dont need high trefi, even from 32767 to 262K is not much difference at 8400
His timings are nuts, that’s the crux of the issue.
tFAW 16
tREFI 123456
Both of those are incredibly aggressive.
His tRAS and a few other timings are incredibly aggressive and some of these bring negligible benefit but add instability.
@@speedfastestsolo Sorry, I was talking about tREFI at 8000MHz which I am currently running on. There was about a 0.3ns difference in latency in the Intel MLC latency test dropping tREFI from 110K to 65K at 8000 so not really noticeable and it helped me eliminate errors in tm5 80 cycles 1usmus.
@@bronsondixon4747 Also tRTP, tWR (tWRPDEN), tRDPDEN, tWRRD_sg and tWRRD_dg can cause "random" instability if they're set too tight or wrong in my experience.
I can run tFAW 16 at 8000 stable but maybe I'm lucky.
Maybe it's time for some fast ECC to be created. It's possible the transistor count is so packed that electrons are flipping bits. Worth noting that even servers with ECC still have random crashes.
Yeah, my question exactly, ironically good ECC ram seems to be better suited for overclocking.
In *theory*
DDR5 already has some on die error correction due to the high speed which causes the occasional bit flip. It's not as robust as ECC but I think it's the reason why we see the inconsistency. We might hit occasions were too many bits flip at once and the on die error correction can't fix it. Also I think the socket/dimm interface isn't good enough for consistent signally at high speeds. You see on GPUs and other high speed memory technologies; they are soldered and very close to the core die to ensure strong signalling.
@@neomoz yeah, the ECC only protects the data in the chip, not if it moves to or from the CPU and that could cause inconsistency but still odd how it sometimes works.
Edit: Grammer
This is so ralatable. I’ve spent last 3 weeks trying to get 7600 stable with my g skill 7800 kit + 13900KS + Asus Z790-E (4 dimms). I’m not sure if anything above 7000 is possible on that board tho 😢
4 dimm slots is hard for anything over 7600... I don't think i've ever heard of anyone doing 7800+ on any board that isn't 2 dimm slots. Just crazy that you need a $700 board to make $280 memory work.
Just get 2 dimm board and you can do 8000. I never had issues with my ks. My k toped at 7800
@@MooKyTig MSI Z790i Edge is 2 dimm and can do 8000Mhz. Its mini itx board, but yeah lot cheaper than Apex.
OH FUCK BZ RANT VIDEO, LETS GO
I couldn't even get DDR5 32GB 6000 kit to work on my 12 series 12700k a year ago.
It boots XMP but is unstable in games, Gave up and run it at 4800 stock, Waste of money.
Lol same thing for me but ryzen 7700x instead
Atleast some one tested that. And let us know.
On the LGA1700 socket being really finnicky, I thought I was going insane when I first built my ADL system. Unify X, 12700k, 5600MTs Samsung ICs, and an EK waterblock. I messed around with memory and got something stable at 6200 but CPU thermals were just unacceptable. 90c cinebench at stock. I went out and got a torque wrench and tightened the water block to exactly the specified torque (off the top of my head I think it was 0.6Nm) and suddenly it doesn't boot. JEDEC settings? Didn't boot. I wound up having to loosen the water block and finally it worked, but strangely enough once it got cold outside it stopped booting again, but if I tried a few times (and let the CPU warm up from a few failed attempts) it would run rock solid after that.
Later I installed a BIOS update and suddenly my CPU (stock everything except LLC) was using like 80mV less and my thermals were finally normal. Currently I have an issue where I cant set turbo frequency by core usage. I can set an all core frequency and I can set a per core max, but I cant do 2 at 52 and 8 at 51 or anything like that. Might be a new BIOS to fix it but I just don't feel like it anymore.
I don't understand one thing. Why don't people install the latest bios more often, especially when it comes to memory stability? I'm on a more dated platform on Z490. I have a 10850K and some samsung B-die ram. The system would refuse to boot at anything but XMP and even then some blue screens were in order. After flashing the latest bios, not only was it rock solid stable, I could tighten many timings dropping the latency by about 15ns total. Again, while being rock solid stable.
@@FOGoticus When I was doing the overclocking I made sure I always had the latest BIOS, but now that I have my daily settings dialed in it's kind of a hassle to migrate my settings to the next BIOS version, especially considering that it probably doesn't make anything faster. Sure I could update my bios and then spend 10 minutes setting all the v/f points and memory timings, but even if I can get it to turbo a single core to 52x, that's a
How would DDR5 with ddr4 ish frequencies but tightly tuned timings run?
Would it scale like that?
for example on ryzen 7000, instead of tuning 6000cl32 whatever, how about 3600cl12?(14?)
is that doable? especially when flck runs 1:1.
No, I think BZ mentioned that CAS latency is 26 at its lowest for DDR5
I mean you just stick in 6600MT/s sticks in, enable XMP and call it a day 😂
It's fast enough out of the box that you probably don't need to bother overclock it. Maybe you spend 20 mins and try to loosen the timings a bit.
Remember, over clocking and over powering a cpu is different. There is a upper limit to just how much power can be pushed through a cpu, motherboard and the socket at any given time. By default the i913900k runs at over 300watts continuously until the max temp is reached of 100DegC, being absolute madness, and it exceeds the max published ratings. As this is done by default and there are no warnings, nothing, possible legal action may be pending, as default settings should be safe.
I truly think it's the memory sticks itself and not the memory controller. I've managed to get similar configurations like in the 6400 video(s) with sub timings that are slightly looser than those at less than 1.3V on Hynix M die, and it passed TM5, Ycruncher, ect. The voltages are super picky, but I firmly believe you are using much more voltage than is needed to stabalize a few sub timings which are too tight, such as tRRD_S, tWWRD, and tWRPRE. (tWWRDs in my experience will ALWAYS be unstable if the sg and dg timings are no less than 35 apart, as weird as it sounds).
It's also been found by quite a few people that Z690 boards have no issues with Hynix A die running at XMP and higher (Assuming they're 8 layer pcbs), which would mean that the boards wouldn't have ram issues. I've spent weeks on testing DDR5, and truly believe it's the sticks that are limiting us, not the memory controller.
tRRDL&S at 15/8 still errors in HCI memtest.
@@ActuallyHardcoreOverclocking I've personally found that tRRD_L at 4 is fine, but S to need to go no lower than 8 to be fine. The tWRPRE also seems to be fine no lower than 44. It's mostly tWRRDs that gave me issues. I couldn't get them stable at anything besides 75/45 for sg/dg
@@Zhunter5000RDL will auto-correct to 8 and the tWRRDs you should ignore because the actual IMC stress comes from the eventual tWTRS/L value that it’ll calculate from tCWL. The issue being that (headache time) not all board samples use the correct calculation to get tWTRS/L. S/L like to be in a 1:4 ratio. Min is 4/16 and many board samples will train into 8/32 (tWRRDs will result from the tCWL). Basically just set tWTRS 6 and L 24 (or 8/32 if needed). Between the EVGA boards and ASUS boards there’s a tick difference of 2 cycles when it ends up calculating the tWTRS/L. You gain nothing from setting the tertiary timing though.
Is any A-die kit such as 6400 cl32 fine for high overclocking or do you need the 7600mhz kits?
@@Matias-wu6xp 6400C32 A-die = 🐐
HU: Here's a 17 minute video on DDR5 scaling and Tuning for 13th Gen!
Me: Ugh, maybe later... too long for a DDR5 tuning guide.
BZ: One hour ranting video on 13th Gen Memory/Memory controller!
Me: Ok just let me grab dinner and a couple of beers!
Note: not a dig at HU... they're absolutely awesome as well.
Need to sell this z790 junk and have some quiet R & R with Sapphire Rapids. Quad channel ECC R-Dimm stress testing shouldn't take long or anything...
I feel your pains, I have a z790 Strix with a I9 13th Gen & 4 DDR5 TForce 6200 chips. It will run all 4 with no XMP at 4800 no issues. But if I try XMP1 wont even post, XMP 2 will post and run for awhile and randomly 10 min or 10 hours later will become unstable lol. I'd be happy if I could just get the 6200 but haven't had time to adjust with it. Why add 4 slots if the manufactory isn't going to make them usable... "oh but one day 5 years from now it might work"
I feel bad for you, but let me drop a couple thoughts:
- scientifically speaking, a terrible result is still a result. If that effort shows this gen memory controller is crap in the worst random way, is still data that might help others dealing with this very problem, some will be giving up and sparing that misery, but maybe others will count with that as a base to research. Yours might end being a very miserable, expensive tiny step that was part of some actual result. Maybe. Hopefully.
- these long rants mean a world for many of us. Honest to god (our FSM). Here you discuss the topic giving context that sometimes is just gold. I have enjoyed this hour long video.
Regarding that 5% you can gain, I totally get that is a poor investment, but you know sometimes is just the challenge what makes it good even being almost pointless. On the other side, from 100 to 105 can be minutiae and hard to even notice for the average Joe, but maybe the last 5% from the mem, another final 5% from the cpu and another one from the GPU, can add to be a noticeable improvement. Maybe for example the min 1% framerate goes up a couple notches and some stuttering goes away.
I just stopped the video to play a couple hours PUGB with a friend, from a computer I got in 2014 and is now at mom's. Any 5% is noticeable here.
Did you try all the doom codes?
oems running at dirt slow stock speeds makes a little more sense now.
i appreciate Buildzoids honesty
Just get the APEX-board and all your problems will be solved.
I've been looking at a 13900ks and Apex just so I can OC it since I want to use this system for a long time, but seeing this and the HU video on scaling I'm really questioning it :/
I went with a 13900KS and a Z790 Apex this gen and I feel it's a nice upgrade from a 9900K. I usually upgrade once every 3 to 5 years.
AMD has won me over this round. Sure, you can try your luck ordering multiple 13900K/KS to tip silicon and IMC lottery in your favor, put it on custom water to control the heat, disable e-cores to run 6.0GHz+ all-core, shell out even more cash for top tier DDR5, and then spend days or weeks dialing everything in to get that 5-15% victory over a 7950X3D (but still lose by 20% in some games where the V-cache rips). Or.... you can buy a 7950X3D, a $70 air cooler, grab whatever kit of semi-decent EXPO 6000MHz DDR5 is currently on sale, and haphazardly slap it together and end up within 5-15% of an expertly tuned 13th gen system.
Don't get me wrong, I've loved OC'ing for the 20+ years I've been doing it, but the absolute hassle of min-maxing 13th gen versus what AM5 just does out of the box makes 13th gen seem very silly and masochistic to me.
@bulldzoid, i know this is very dumb but that's why we are here.
So how good or bad would be true ECC memory be to overclock? My rational is that very good ECC memory should beat non ECC if all is equal (what will not be) be higher overclockable because of the extra die and ECC feature, so it can easily correct single bit flips and even double bit flips with extra latency, also ECC memory makes it easy for the os to count errors without needing to stress test and long term, so did you try overclocking ECC ram and how well did it work?
Can you rule out cosmic radiation causing errors in the memory at 1000% coverage? How long does that take? 6 hours? I mean, we know that cosmic radiation can cause errors with non-OC memory at any point, even elections were influenced like that.