@@Kholaslittlespot1 Don't bother unless you are really enthusiastic about overclocking. I once spent over a month fine tuning my memory timings and voltage to stabilize a better overclock. The whole time it was always on the verge of being stable, but it took forever to find just the right combination of values to actually make it stable. It was a maddening gauntlet of tweaking, testing and analysing seemingly paradoxical results! All for a 1% performance gain in certain memory intensive applications.
@@fnorgen just learn the most important ones that actually have huge impact (like shown in this video) It doesn't have to be at the edge of stability, but sometimes you really can tighten those down (2-3 days of work are worth it)
Agreed. I think Steve from GN said a few months ago they were planning to do an updated version of their RAM timing video from 3 or 4ish years ago. So hopefully that will come out once DDR5 starts to hit the market.
don't worry, as soon as it's all sorted out, DDR5 will invert-flip-randomize it all around again so it's like trying to navigate 11 dimensional M-theory, lol. And if you think that's confusing, wait till "constructor theory" gets involved with the timings set by the CPU, lol.
@@Kholaslittlespot1 It's really simple you just pick a random number then double it then roll a dice and add the two numbers together then divide by Pi. This woks for most of the settings except the ones that need to be double some other setting and should probably not exist since they are just a multiple of something else. Then there is that one that is like how long should I hold data before the capacitance drops and it's like.... You tell me. If for some reason you ever figure all this out your probably wrong since you likely didn't guess how much resistance the memory traces had so all your stability testing was skewed by that. Once your done testing it all you get to enjoy an OS that has been corrupted by unstable RAM for the past month a bluescreened into a dribbling mess so you won't be able to tell if you ever actually got it stable. Then the first hint of HandBrake or any other intensive task will likely bring the whole stack of cards tumbling down until you give up and pick an JDEC or JDEC like speed and decide that having a working machine is more important than worrying about a couple of %
Great info BZ, Tom's Hardware actually wrote an article about this video. Keep it up, always turn to your videos to learn as much as I can when it comes to memory and cpu over clocking.
Amazing that a company which is manufacturing sticks can be so incompetent. RRDS/RRDL/FAW are by far the most impactful DRAM timings these days and those settings are way worse than JEDEC spec.
It doesn't help that most of the manufacturers who make DIMMs are not the people who make the actual memory. I've always find it odd that there is trust in companies to bin and make memory faster when they themselves aren't involved in the production of the ICs. There is only one company you can say sells both high speed kits and is a DRAM fabricator, that being Micron
The memory manufacturer can only program in so many timings into the extra spd space there’s not space for every timing. Everything else is up to motherboard manufacturers to optimize xmp/auto timings for every kit out there. Would be cool if all the best timings could be saved in the spd eprom.
@@nickpowell7432 Yeah for sure, would make stuff less likely to work in one click though due to board, BIOS and IMC differences. What is unforgiveable is explicitly programming a timing which makes the RAM perform worse than JEDEC.
yup i had this corsair 3600 32gb kit, didnt know much about ram when i bought it, constant errors, hiccups, finally set it too 3000 and adjusted timings accordingly and saw 1% lows go up by 50% and average fps up by 7% or so, wonky shit
Sorry if it was mentioned in comments already, but it helped me check what’s actually in my XMP profile (using Thaiphoon) and what is actually applied. My motherboard only applies the primaries. So checking the full XMP profile TFAW was like 44, so I ignored it (went for 16), but its tRC and TRRDS/L were lower than what my motherboard applied. So I just put in what was in the XMP and my Linpack results went up by 6-7% right off the bat. Good option if you don’t want to spend a lot of time on tweaking your ram :)
It's not XMP fault or the MB, these manufactures are just trying to hit high numbers to sell the same RAM for a higher price and to avoid instability since their chip can't handle lower latency. MB should set the timing based on the XMB and not based on MB data since RAM manufactures should give the correct XMP number.
they dont want to do proper binning so they use timings looser then needed for the majority of sticks, to ensure even the worst ones will still work at rated xmp... i use to be friends with a guy, we would buy tray ram, would use multi boards to test ram and dial in a proper spd for higher then rated clocks and flash it to the spd on the sticks, resell on ebay and locally... never had a single return for ram not meeting spec or being flawed out of the box.. mostly had people comiong back wanting more of the same exact stuff to match with what they already got... what i found was that kits rated at jdec and kits rated at the best speeds possible, were almost always just binned sticks, and often more then half a tray would work 100% flawless at the xmp dialed on for the tpp speeds sold... we even sold some ram at higher then any official ram manufacturer was selling them flashed, we dumped the 2 lowest jdec specs from the spd and added 2 more... one that was max speed for what companies were selling then one i dialed in, around 2 perf tiers higher, out of a tray we only got around 25% that would test out flaweless at those specs but... omg did they sell... zero returns... more then a few people upset we didnt have any more and couldnt find trays of ram to buy of that type at the time...(years later he and his bro got ram when the stuff was being dumped by corporations dirt cheap and binned a bunch more and made a much higher profit due to the boom in retro pc building...) we had some fun back then... honestly.. i would like to see better auto-training for set speeds.. i have seen it get better but.. i have also seen a couple gems of boards that if you set the ram to a specific clock, the board would reboot and try dif timings till it could post and boot to windows successfully, from there you had to loosen/tighten things till you got it dialed in.. but.. gods damn thats a nice feature.... once had a buddies system take over 45min to dial itself in but.. 3hrs later we had found the "sweet spot" for the ram, and written down the key timings(both primary and sub timings we tweaked).. its crazy how big a deal even 2ns can make.. from unstable to 100% solid for over a week of leaving p95 blend running using 785% of the system ram(he forgot it running when he went on a work trip and.. nobody goes into his basement office without him there or telling them to.. so.. yeah.. "guess its solid..god damn.. that must have burned some power" LOL..
Every time I watch one of your memory videos it gives me a tremendous urge to tweak my timings more and see how much tighter I can make them lol but my gskill kit has paper thin heatsinks that don't do anything so it always runs really hot and makes me sad
Either iprove the air circulation in your case, or tie a small fan to the memory. My Crucial kit was blue screening as soon as it got hot after tweaking the timings, and a small 5V (Raspberry Pi) fan sitting on top of RAM stopped all the errors :P
If you have RipjawsV then remove the head spreader entirely. It makes the temps better by about 2C in my case. GSkill designed an anti-radiator which makes the temp. worse.
I just found that in Thaiphoon burner you can click the report button on the toolbar and it shows more timings (full report view) - or you can export File > Export to... > Complete HTML report
jep same here 😀 . granted me 300 points in timespy on my 5800x with a 3200C14 kit clocked @ 3600-14-14-14-28-44 plus new subtimings obviously. That is what happens if you try to overclock your xmp settings. silly me :D Edit: since you mentioned it, it is a teamgroup ARGB kit 4x8. tFAW was 38 now its 20
Trashing the timings to relax the setting enhances the stability of the module. The manufacturer wants that because that means more kits sold, not returned nor RMA'ed because of stability issues. It's a move to ensure profits, screwing the end user in the process since most just bother to use XMP only.
Knowing what each timing does do not help with overclocking, cos for the most part you just have to lower everything as much as possible except the frequency
I'm not massively au fait with DRAM tweaking (just XMP them these days), but I'm familiar enough with the ranges to expect from cursory checks with CPU-z that my eyebrows would go up if I saw numbers like those. I'd guess that kits with XMP speeds of 3600, 3200 or even slower from other makers would also clock to 4800MHz with those timings, so these are basically not what they're touted/adertised as and a complete con.
I always shy away from tweaking my memory timings, for no matter how many tutorials and articles i read, i can't seem to understand what exactly each value do, and how they work in relation to the other values, or how to test stability in a relieable and timely manner. Constantly running into conflicting information, or behavior i can't really reproduce on my system doesn't help either. Now i see even some manufacturers don't know what they are doing either, many just seem to loosen the timings, until the memory just doesn't crash at a specific speed on whatever awkward configuration they test it on, programm this into the xmp and call it a day.
"i can't seem to understand what exactly each value do" You don't need to, just remember lower = better and see what works. Having more in-depth info about how the timings interact is useful, but not necessary.
Ryzen DRAM calc, will get you in the ballpark at least. Plus it has a number of useful macros for calculating tertiary and impedance settings. It's not the be all and end all, but it's a good start.
Wish there was a more standard way for dealing with faster frequencies and timings, clearly no one is satisfied with JEDEC You're pretty much at the mercy of the DIMM manufacturer if you want to use XMP as intended, compatibility is still meh and if you want add more memory even buy getting another of the same kit you better pray they haven't swapped IC makes and completely changed up the sub timings.
Gained over 50 marks for the CPU test with a very slight adjustment.. 3466MHZ XMP2.0, to using auto and manually tuning just the CL, tRCD, tRP, tRAS, tRFC. tRFC (607) has to be quite high or I get instability.
I'm unimpressed with the 2x8 Oloy kit I have. Wouldn't run @3600mhz until it got all the way up to 1.4v and it's CAS 18. Trash. Hopefully your kit is better.
@@x8jason8x How are your other timings? My RAM won't go above 3000MT/s no matter what voltage I use, and most of the timings suck, though I do get CL14 (Corsair Vengeance XMP 3000MHz CL16 8Gb Rev B)
@@x8jason8x replaced rev e with more rev e lol. At least it will be better binned they use the better chips on thier own mem kits and sell the rest. What did you get out of the ballistics? I’ve got 3600 ballistics. Rev e is harder to find info for since everyone drools over Samsung b die.
@@nickpowell7432 Yeah I think they are rev e too but way better binned, 16-18-18 @ 1.35v, also 3600. I haven't overclocked these ones in particular, couldn't tell you how low they'd go.
Ballistix 3600 on ASUS/MSI 16-18-18-38-84 and a whole lot ot auto. GIGABYTE seems to pull all the correct timings at 16-18-18-38-58, with tFaw etc set correctly.
The Crucial 3600s try to force 1802mhz, but MSI only allows 1800mhz. So TRFC goes out of whack (Micron B Single rank) on MSI AMD boards. Not sure about ASUS or GIGA.
Nice explanation, that might explain a lot of what I've seen trying different boards and chips in AM4 builds with the same 4 sticks of micron. No Asrock boards yet so that'll have to be next. Tonight a few cans, order some food, open the windows and mess with an A520I AC and a 3500x. Woohoo.
My Z490 Gigabyte mobo tends to set its own secondaries and tertiaries, which it sets mainly according to the "Memory Enhancement Settings" setting. In my experience it sets everything decently tight regardless, but then again that may cause issues with certain kits.
Similar happened when my PC can't boot, on my old ddr3 2400mhz 16gb kit with XMP enabled. It took me more than weeks, to found the problem. It was the secondary timing won't match something with the motherboard. Checked the jdec in CPUZ, AIDA64, I tried to match all timings table in softwares. Then, lucky I tried lowered the TRRD to 7 and CWL to 12, and then boom! the PC booted instantly with xmp 2400mhz in the table software shows TRRD = 8 , CWL =13 for 2400mhz xmp, so weird. I agree company never really explain you the secondary timings.
You need to stress test. XMP is an overclocking profile. People buy xmp kits at high frequencies but most of the time it takes a lot of vccsa voltage to get stable. With Ryzen running 1:1 with RAM the IF can become unstable, also IMC. With Intel like the 10900k more VCCIO and VCCSA can be needed. As both the IMC and the cache connection are affected. Just because the system feels more stable, does not mean the RAM is not erroring randomly. You wont know until you test. I have tested the RAM with the 10900k with HCI/memtest etc and past (4000MT/s C15 Dual Rank 4 x DIMMs). Then errored in large prime95 ffts. Bumping vccio by a small amount fixed this, as well as lowering cache to 44x. I get 15.4k in Time Spy cpu. All games are smooth as silk. All voltages are very safe for 24/7.
I had this issue with my Corsair sticks. They are supposed to be 3600, but performance decreased from 2933 (IIRC). And I'm not good enough at Memory OC to do more than little tinkering with timings. Then again 3200 profile run very well (better than 2933) and I did some further very light OC on that and it is stable. I guess it is more of an issue since I have 4 sticks (I didn't have enough money so I was buying 2x4GB 3 months apart).
XMP imo is just for time-saving anyways. If you are on the high-end, XMP is kinda pointless because you wanna manually set that stuff anyways (else: why bother paying that much money?). And on the lower end / budget range, you are just wasting money paying for an XMP profile. I mean, do you really wanna pay 50 bucks more on to get a higher frequency XMP-profile for the same kit/die? Or would it be better to just buy the identical hardware with different XMP (of a potentially different bin, if you are unlucky) and manually set those numbers?
Love these videos. But I have to admit, probably 90% of computer builders don't go beyond looking at rated specs. More than half probably don't go beyond whether it has RGB.
i use XMP as a starting point then dial things in from there... these 4x16gb TimeTec 3600 sticks run below XMP at 3733 at xmp volts.... the xmp is loose, im actually ok with that in this case, as it would allow it to work with alot of systems that other kits i have delt with wouldnt post in with xmp enabled... (first gen ryzen mostly..) these... are hynix djr die ram.. 210 for my ram kit... well its rated via xmp at 18-22-22-42, im gettin 3733@17-20-20-38-cr1, with a tREFI at over 32k(board max), this is on x299/7940x, so, 4k and up on ram is...not that easy or common.. doable but... honestly not worth the timings i would need to use vs what im getting at 3733.. at 1.3v... im pretty sure i could dial the trfc in a bit but... honestly im pretty happy with the results so far....
So is it better to get a high clock kit with looser timings and clock it down to tighten the timings, or is it better to just get a lower clocked kit with tight timings (for Ryzen)? Like is it better to get a 3600 kit with like 14,15,15,35? Or like a 4000 kit with 18,20,20,40 and clock it down to 3600-3800?
I can't use XMP for the 4x8GB Patriot Viper 4400MHz (19-19-19-39) in the MSI Z490 Godlike MB (i9-10900K) because it crashes Warzone. Same for following the manual OC settings video by AHO, maybe because a few available choices differ from the instructions given by AHO but I've tried everything before giving up.
Do you know what sa and io voltage your motherboard has when you set xmp ? Also the fact that your ram loads into windows with 4 sticks on xmp is impressive.
hmm question, so ive lowered all my timing pretty much as much as possible and ran memtest for 24hours and prime95 for 12hours no crashed or errors, and i used aida, cinebench and timespy to check performance to make sure the timings increased performance along the way. But my question is can a timing ever be to low, i have some timings that i was able to get down to 1 without encountering errors, will that make my memory degrade quicker or something possibly is there any risk to that at all?
Maybe you could help me choose the memory dice: I currently have: Mobo MSI X370 Pro carbon Proc Ryzen 7 1700 Memory F4-3200C14D-16GFX G.SkILL 2x8GB 3200 cl14 mhz samsung b-die (single rank) I don't want to change hardware but needs more ram memory. Due to the fact that the first generation of procesor is sucks when it comes to the memory controller, to keep 3200mhz I just have to replace the set of ram. I am wondering if you want to buy a set F4-3200C14D-32GFX 2x16gb 3200mhz cl14 samsung b-die (dual rank) or buy something else? Or maybe buy crucial ballistix 2x16 3600mhz cl16 and set them to 3200mhz. I know they are sometimes single rank, perhaps micron d and sometimes dual rank micorn e. I do not know what to do, I would like to buy a frame that I will use one day when changing the platform. I wonder if on newer generations of the processor it is at all possible to combine such sets F4-3200C14-16GFX with F4-3200c14-32GFX, i.e. throwing two single rank sticks and two sticks with higher dual rank capacity? I do not know what to do to advise something?
Oh MSI just ignores sub timings, that explains why i had to set the XMP manually on my MSI board when i was testing it, figured it was just cause the speed was over 4000 not like a would keep a b-die kit kit at stock anyway... It was a 4133 viper steel on a msi b550I board, IIRC the auto was worse than what was on the spd chip (i did force 1T like the profile called for) and the bios is just like you expect the IMC to do this? yea right i'll do this instead without realizing it is a APU and will not care
I have a 3200 kit that reduces the FCLK to 1600 to run synchronous... It underclocks the infinity fabric. I had to manually set it to 4000 and 2000 and add some voltage here and there.
my Adata xpg silver 3000 2 sticks does lots more issues for some reason tthey didnt have those issues when i had a r 7 1700 i didnt face those issues only when i used them on 3400 g always crashing and slowing down my system i tried lots of things finaly decided to sell them and replaced them with hyuper x fury 3200 mhz
l cant stop getting surprized every time l watch a video where the latest most expensive piece or top tier hardware has the same issues l have with my old shitty low end PC xD
I now have a 4x8Gig Memory Kit from TForce with Samsung B-Die on them, validated for 4000 Cl18. When I activate D.O.C.P on my MoBo it wont even boot the System (5950x, IF clocked at 1.9GHz atm, wanted to try 2GHz but haven't gotten into it for now). Any suggestions what would cause it to fail to boot? Should I try Manual Voltage override ?
What motherboard are you using? Is that RAM on their QVL? Being it fails to POST I am wondering on compatibility. May not hurt to check your UEFI version on your MB too
@@Frizzy9000 Haven't checked up that yet, running a X570 Dark Hero from Asus, got it in June and installed newest UEFI back then, right before I installed any OS's... Gonna check these suggestions up could be one of the reasons...
Try to slowly raise the ram Frequency from default, that gives you a rough idea if the cpu can handle it because the mobo will set really loose timings to try to boot.
Ive got 2 sets of dual chanel trident z 3200cl14 8×2 single rank And they wont but a single mhz higher than xmp Im using a b550 steel legend and ryzen 5900x Any advice? I tried every possible memory chanel mod single dual and full dimm
This is my problem that I am searching for 2 years of lag in games also my ram is gskill royal xmp 4000 cl17 dual kit xmp works really bad on my aourus z390 master!At first it works xmp on and good to go but after a while I realize massive delays lags everywhere I can't figure out what the hack I spend tons of money change components but not ram ! Than I change the timing manually as in this video after 2 years problem solved.
Mine just crashes often - it runs @ 3200 could be on it gaming for the day then next day bluescreens on startup - or like just now its fine all day then I go on CZcams and Ads are showing with Adblock on -wtf- so disable it - refresh - enable - BSOD - I find using my browser often causes BSOD when overclocking my Ram not sure why, that's your area of expertise I guess. . . Hyper X Predator 3200 - MSI X470 M7AC Ryzen2700X
AIDA64 memory tests wont tell you much outside a general idea if you have gone too far without causing noticed instability, now, running he full bench using a quick report, the photoworxx test at least on my 7940x gives you a much better idea if your fucking shit up or making things better.. hell the overall bench can give you an idea if you made it worse or better.. why i store multi reports as i test tweaks.. very handy tool.. aida's memory bench is useful as a quick "did i make it better or worse in general?" type test but.. yeah i have found it useless beyond a vague idea.. (if you see it jump up or down alot, you get a general idea if your fucking up or not..) anyway, im having fun dialing this new system i build after my x58 board died.. (a 6c12t xeon overclocked to 4.6-4.7 is quite competitive with first gen ryzen in most games faster in several..), this x299 systems been a learning exp.. but also been more fun dialing a system in then i have had for years.... so many are so close to their limit as it is.... this.. ram and cpu have given me a shitload of fun..
The VRMs for the memory are so insignificant it seems rather unlikely to be the cause; are you sure you didn't have the memory controller setup wrong? Some of the early Ryzens needed the termination impedances tweaked or they wouldn't drive the memory correctly.
@@ghoulbuster1 There's too much to go into, but setting RAM terminations, vSoC, and RAM impedances is pretty key to get Ryzen 1000 to go anywhere. Furthermore most termination and impedance stuff is really on a case-by-case basis so there's no real good tutorial out there. That being said you shouldn't expect much, best I managed on the platform (stable) is 3333C14. It just doesn't clock well.
@@IK4MS I see well, I wanted to go 3200 C16 on it since that's a good speed for ryzen but I spent days trying to stabilize it using really loose timings and changing everything and nothing worked. I could try again but I'm sure the problem is the cheap B350 mobo that I used.
This is the case for my Gskill Trident Z (CL14 Samsung B) The tRFC is set to something insane like 500+ with XMP enabled. AMD Ryzen 5900X on an X470 Board (This issue was also present when I was using a 2700X on the same Board)
"The tRFC is set to something insane like 500+ with XMP enabled." That's because it's required to operate properly at ~95c. Lower RFC's only work if you keep the memory cool, which people aren't guaranteed to do. RFC will only affect performance by a handful of %. RRDS/RRDL/FAW can half or double it.
"That's it for this video" _continues for 6 more minutes_
Never change Buildzoid
that's the thing
"put the bin out"
"after this video, it's nearly over"
15 mins later divorce proceedings considered.
"quick video" *18:55*
As someone who only ever set primary timings until recently, I appreciate this video.
Primary timings net me better performance over xmp or dram calculator
I really need to make the effort to learn more about secondary timings
@@Kholaslittlespot1 Don't bother unless you are really enthusiastic about overclocking. I once spent over a month fine tuning my memory timings and voltage to stabilize a better overclock. The whole time it was always on the verge of being stable, but it took forever to find just the right combination of values to actually make it stable. It was a maddening gauntlet of tweaking, testing and analysing seemingly paradoxical results! All for a 1% performance gain in certain memory intensive applications.
@@fnorgen just learn the most important ones that actually have huge impact (like shown in this video)
It doesn't have to be at the edge of stability, but sometimes you really can tighten those down (2-3 days of work are worth it)
So basically the kit manufacturer wanted to push the freq as high as possible because its the big number buyers look at...
Yeah, and this shows why third party benchmarking and reviews are so important.
I would've liked to see a demonstration of what the performance is like with these timings set as you suggest they should be.
I would love to see if it was stable with those settings. My guess is it wouldn't even boot.
Would be great if you'd make a single video explaining all the ddr4 timings. Piecing the info all over the internet is really frustrating.
Agreed. I think Steve from GN said a few months ago they were planning to do an updated version of their RAM timing video from 3 or 4ish years ago. So hopefully that will come out once DDR5 starts to hit the market.
don't worry, as soon as it's all sorted out, DDR5 will invert-flip-randomize it all around again so it's like trying to navigate 11 dimensional M-theory, lol. And if you think that's confusing, wait till "constructor theory" gets involved with the timings set by the CPU, lol.
Yeah I could really do with actually knowing what my secondary timings should be. Right now it feels like I'm stabbing in the dark a bit.
@@Kholaslittlespot1 It's really simple you just pick a random number then double it then roll a dice and add the two numbers together then divide by Pi.
This woks for most of the settings except the ones that need to be double some other setting and should probably not exist since they are just a multiple of something else. Then there is that one that is like how long should I hold data before the capacitance drops and it's like.... You tell me.
If for some reason you ever figure all this out your probably wrong since you likely didn't guess how much resistance the memory traces had so all your stability testing was skewed by that.
Once your done testing it all you get to enjoy an OS that has been corrupted by unstable RAM for the past month a bluescreened into a dribbling mess so you won't be able to tell if you ever actually got it stable.
Then the first hint of HandBrake or any other intensive task will likely bring the whole stack of cards tumbling down until you give up and pick an JDEC or JDEC like speed and decide that having a working machine is more important than worrying about a couple of %
Actually he has said before that it doesn't matter to know what they all do, and that he actually doesn't know what almost none of them actually does.
Ugh… I had already given up on manual timings lol. Back to the grind again…
Great info BZ, Tom's Hardware actually wrote an article about this video. Keep it up, always turn to your videos to learn as much as I can when it comes to memory and cpu over clocking.
Getting XMP to work already is a lottery, this is just a bonus lol
Some context: That's an $850 kit. For 2x8GB.
Omg...
Samsung B-die in a nutshell
@@ghoulbuster1 Uh, you can easily get a high-bin b-die kit at 2x8GB for $120. Usually less. No idea what your reply even means.
@@sandmann521 B-die can cost as low as 90 bucks to as high as 800. Depending on the binning.
Amazing that a company which is manufacturing sticks can be so incompetent. RRDS/RRDL/FAW are by far the most impactful DRAM timings these days and those settings are way worse than JEDEC spec.
tRFC is also very impactful.
It doesn't help that most of the manufacturers who make DIMMs are not the people who make the actual memory.
I've always find it odd that there is trust in companies to bin and make memory faster when they themselves aren't involved in the production of the ICs.
There is only one company you can say sells both high speed kits and is a DRAM fabricator, that being Micron
It's not being incompetent, it's marketing.
The memory manufacturer can only program in so many timings into the extra spd space there’s not space for every timing. Everything else is up to motherboard manufacturers to optimize xmp/auto timings for every kit out there. Would be cool if all the best timings could be saved in the spd eprom.
@@nickpowell7432 Yeah for sure, would make stuff less likely to work in one click though due to board, BIOS and IMC differences.
What is unforgiveable is explicitly programming a timing which makes the RAM perform worse than JEDEC.
yup i had this corsair 3600 32gb kit, didnt know much about ram when i bought it, constant errors, hiccups, finally set it too 3000 and adjusted timings accordingly and saw 1% lows go up by 50% and average fps up by 7% or so, wonky shit
Sorry if it was mentioned in comments already, but it helped me check what’s actually in my XMP profile (using Thaiphoon) and what is actually applied. My motherboard only applies the primaries. So checking the full XMP profile TFAW was like 44, so I ignored it (went for 16), but its tRC and TRRDS/L were lower than what my motherboard applied. So I just put in what was in the XMP and my Linpack results went up by 6-7% right off the bat. Good option if you don’t want to spend a lot of time on tweaking your ram :)
Aida will show you full SPD with few other timings.
It's not XMP fault or the MB, these manufactures are just trying to hit high numbers to sell the same RAM for a higher price and to avoid instability since their chip can't handle lower latency.
MB should set the timing based on the XMB and not based on MB data since RAM manufactures should give the correct XMP number.
they dont want to do proper binning so they use timings looser then needed for the majority of sticks, to ensure even the worst ones will still work at rated xmp...
i use to be friends with a guy, we would buy tray ram, would use multi boards to test ram and dial in a proper spd for higher then rated clocks and flash it to the spd on the sticks, resell on ebay and locally... never had a single return for ram not meeting spec or being flawed out of the box.. mostly had people comiong back wanting more of the same exact stuff to match with what they already got... what i found was that kits rated at jdec and kits rated at the best speeds possible, were almost always just binned sticks, and often more then half a tray would work 100% flawless at the xmp dialed on for the tpp speeds sold...
we even sold some ram at higher then any official ram manufacturer was selling them flashed, we dumped the 2 lowest jdec specs from the spd and added 2 more... one that was max speed for what companies were selling then one i dialed in, around 2 perf tiers higher, out of a tray we only got around 25% that would test out flaweless at those specs but... omg did they sell... zero returns... more then a few people upset we didnt have any more and couldnt find trays of ram to buy of that type at the time...(years later he and his bro got ram when the stuff was being dumped by corporations dirt cheap and binned a bunch more and made a much higher profit due to the boom in retro pc building...)
we had some fun back then...
honestly.. i would like to see better auto-training for set speeds.. i have seen it get better but.. i have also seen a couple gems of boards that if you set the ram to a specific clock, the board would reboot and try dif timings till it could post and boot to windows successfully, from there you had to loosen/tighten things till you got it dialed in.. but.. gods damn thats a nice feature.... once had a buddies system take over 45min to dial itself in but.. 3hrs later we had found the "sweet spot" for the ram, and written down the key timings(both primary and sub timings we tweaked).. its crazy how big a deal even 2ns can make.. from unstable to 100% solid for over a week of leaving p95 blend running using 785% of the system ram(he forgot it running when he went on a work trip and.. nobody goes into his basement office without him there or telling them to.. so.. yeah.. "guess its solid..god damn.. that must have burned some power" LOL..
Every time I watch one of your memory videos it gives me a tremendous urge to tweak my timings more and see how much tighter I can make them lol but my gskill kit has paper thin heatsinks that don't do anything so it always runs really hot and makes me sad
Zip tie a fan to them
Either iprove the air circulation in your case, or tie a small fan to the memory. My Crucial kit was blue screening as soon as it got hot after tweaking the timings, and a small 5V (Raspberry Pi) fan sitting on top of RAM stopped all the errors :P
If you have RipjawsV then remove the head spreader entirely. It makes the temps better by about 2C in my case. GSkill designed an anti-radiator which makes the temp. worse.
@@MasonStormSunny just aesthetical cooling. Trash
@@MasonStormSunny lol wtf 🤣
I just found that in Thaiphoon burner you can click the report button on the toolbar and it shows more timings (full report view) - or you can export File > Export to... > Complete HTML report
broken calculator gang
Waiting patiently for a video rambling about DDR5 and its sky high latencies
jep same here 😀 . granted me 300 points in timespy on my 5800x with a 3200C14 kit clocked @ 3600-14-14-14-28-44 plus new subtimings obviously. That is what happens if you try to overclock your xmp settings. silly me :D
Edit: since you mentioned it, it is a teamgroup ARGB kit 4x8. tFAW was 38 now its 20
You really help me alot I thank you...ur a legend
Trashing the timings to relax the setting enhances the stability of the module. The manufacturer wants that because that means more kits sold, not returned nor RMA'ed because of stability issues. It's a move to ensure profits, screwing the end user in the process since most just bother to use XMP only.
Can anyone recommend a tutorial that explains memory timings and overclock techniques?
Depends what you're trying to do.
@@SL1PSTAR overclocking
@@kamipls6790 🤣 I should of said "depends what you're trying to overclock"
Is this a joke? Almost everything I learned about ram oc'ing is right here. The other 10% is trial and error.
Knowing what each timing does do not help with overclocking, cos for the most part you just have to lower everything as much as possible except the frequency
I'm not massively au fait with DRAM tweaking (just XMP them these days), but I'm familiar enough with the ranges to expect from cursory checks with CPU-z that my eyebrows would go up if I saw numbers like those. I'd guess that kits with XMP speeds of 3600, 3200 or even slower from other makers would also clock to 4800MHz with those timings, so these are basically not what they're touted/adertised as and a complete con.
I always shy away from tweaking my memory timings, for no matter how many tutorials and articles i read, i can't seem to understand what exactly each value do, and how they work in relation to the other values, or how to test stability in a relieable and timely manner.
Constantly running into conflicting information, or behavior i can't really reproduce on my system doesn't help either.
Now i see even some manufacturers don't know what they are doing either, many just seem to loosen the timings, until the memory just doesn't crash at a specific speed on whatever awkward configuration they test it on, programm this into the xmp and call it a day.
"i can't seem to understand what exactly each value do"
You don't need to, just remember lower = better and see what works.
Having more in-depth info about how the timings interact is useful, but not necessary.
Ryzen DRAM calc, will get you in the ballpark at least. Plus it has a number of useful macros for calculating tertiary and impedance settings. It's not the be all and end all, but it's a good start.
@@nathangamble125 read sixtyinsix statement and get out calculator and do the maths and compare to your ram
They know what they're doing.
High frequencies and low CL sell, a low tRRD doesn't.
Great now i gotta check the Adata kit in my Gigaboard...i'd manual set but there's soo many settings idk wtf is what.
Wish there was a more standard way for dealing with faster frequencies and timings, clearly no one is satisfied with JEDEC
You're pretty much at the mercy of the DIMM manufacturer if you want to use XMP as intended, compatibility is still meh and if you want add more memory even buy getting another of the same kit you better pray they haven't swapped IC makes and completely changed up the sub timings.
LMAO @ the implication that companies are meant to do extra work for extra money!
This is "value add", not for us but for them!
Gained over 50 marks for the CPU test with a very slight adjustment.. 3466MHZ XMP2.0, to using auto and manually tuning just the CL, tRCD, tRP, tRAS, tRFC. tRFC (607) has to be quite high or I get instability.
Just bought some OLOY 2 x 16 3200MHz kit and I was surprised it came with Micron E-die chips.
I'm unimpressed with the 2x8 Oloy kit I have. Wouldn't run @3600mhz until it got all the way up to 1.4v and it's CAS 18. Trash. Hopefully your kit is better.
@@x8jason8x How are your other timings?
My RAM won't go above 3000MT/s no matter what voltage I use, and most of the timings suck, though I do get CL14 (Corsair Vengeance XMP 3000MHz CL16 8Gb Rev B)
@@nathangamble125 I couldn't tell you now, I replaced it with a crucial ballistix set.
@@x8jason8x replaced rev e with more rev e lol. At least it will be better binned they use the better chips on thier own mem kits and sell the rest. What did you get out of the ballistics? I’ve got 3600 ballistics. Rev e is harder to find info for since everyone drools over Samsung b die.
@@nickpowell7432 Yeah I think they are rev e too but way better binned, 16-18-18 @ 1.35v, also 3600. I haven't overclocked these ones in particular, couldn't tell you how low they'd go.
thanks, it is extremely useful.
Ballistix 3600 on ASUS/MSI 16-18-18-38-84 and a whole lot ot auto.
GIGABYTE seems to pull all the correct timings at 16-18-18-38-58, with tFaw etc set correctly.
so in this case it's pretty much the other way round. The Timings on the Kit are tighter than the ones that Asus/Msi would automatically set
@@LinkStorm13I love consistency
The Crucial 3600s try to force 1802mhz, but MSI only allows 1800mhz. So TRFC goes out of whack (Micron B Single rank) on MSI AMD boards. Not sure about ASUS or GIGA.
Nice explanation, that might explain a lot of what I've seen trying different boards and chips in AM4 builds with the same 4 sticks of micron. No Asrock boards yet so that'll have to be next.
Tonight a few cans, order some food, open the windows and mess with an A520I AC and a 3500x. Woohoo.
My Z490 Gigabyte mobo tends to set its own secondaries and tertiaries, which it sets mainly according to the "Memory Enhancement Settings" setting.
In my experience it sets everything decently tight regardless, but then again that may cause issues with certain kits.
Me, with G.Skill memory, a Ryzen processor and Asus boad: [Relieved Robert Downey Jr.]
My memory was unstable at XMP settings, I had to tune it manually to make it stable.
Marketing chasing clock speeds ;)
Yeah went from 10400 to 10700 turning XMP off. I also have a XMP II setting that shot it down to 8000.
Similar happened when my PC can't boot, on my old ddr3 2400mhz 16gb kit with XMP enabled.
It took me more than weeks, to found the problem. It was the secondary timing won't match something with the motherboard.
Checked the jdec in CPUZ, AIDA64, I tried to match all timings table in softwares.
Then, lucky I tried lowered the TRRD to 7 and CWL to 12, and then boom! the PC booted instantly with xmp 2400mhz
in the table software shows TRRD = 8 , CWL =13 for 2400mhz xmp, so weird. I agree company never really explain you the secondary timings.
This is eye opening. Thank you!
Cool, we need Secondary and Tertiary timings exposed.
Good video.
Enabling xmp on my pc made it extremely unstable when gaming.Bumped the voltage 1.35 to 1.36 and all good now.
You need to stress test. XMP is an overclocking profile.
People buy xmp kits at high frequencies but most of the time it takes a lot of vccsa voltage to get stable. With Ryzen running 1:1 with RAM the IF can become unstable, also IMC. With Intel like the 10900k more VCCIO and VCCSA can be needed. As both the IMC and the cache connection are affected.
Just because the system feels more stable, does not mean the RAM is not erroring randomly. You wont know until you test.
I have tested the RAM with the 10900k with HCI/memtest etc and past (4000MT/s C15 Dual Rank 4 x DIMMs). Then errored in large prime95 ffts. Bumping vccio by a small amount fixed this, as well as lowering cache to 44x. I get 15.4k in Time Spy cpu. All games are smooth as silk. All voltages are very safe for 24/7.
@@GodKitty677 what would you say is a safe 24/7 ddr voltage for a typical ddr4 kit?
@@MrBrancomusic Depends on the memory IC's. 1.36 volts should be safe, if 1.35 volts is the default.
@@MrBrancomusic On Hynix DJR and Samsung B-Die you can daily 1.5v. Stick to around 1.35v for other IC.
@@DripDripDrip69 I dont know which die my ram is but it says Samsung in spd.
Geil Super Luce RGB.
I guess im fine at 1.36v.
Damn that will be slower than my ancient Sodimms at 3000Mhz 13-14-14-37
I had this issue with my Corsair sticks. They are supposed to be 3600, but performance decreased from 2933 (IIRC). And I'm not good enough at Memory OC to do more than little tinkering with timings. Then again 3200 profile run very well (better than 2933) and I did some further very light OC on that and it is stable. I guess it is more of an issue since I have 4 sticks (I didn't have enough money so I was buying 2x4GB 3 months apart).
Buildzoid and I ARE best friends :)
I guess just manually inputting the rated voltage , timings and frequency instead of trusting xmp or auto settings to figure it out in the bios?
great, now I can't trust XMP anymore :(
setting up new systems just got a whole lot more tedious
I would not worry too much, in most cases you will only notice a difference in benchmarks unless your really looking for it :)
XMP imo is just for time-saving anyways. If you are on the high-end, XMP is kinda pointless because you wanna manually set that stuff anyways (else: why bother paying that much money?). And on the lower end / budget range, you are just wasting money paying for an XMP profile. I mean, do you really wanna pay 50 bucks more on to get a higher frequency XMP-profile for the same kit/die? Or would it be better to just buy the identical hardware with different XMP (of a potentially different bin, if you are unlucky) and manually set those numbers?
The performance loss if more from Gear 2 then XMP, if you're on zen2+ make sure FCLK/UCLK/MCLK are synced, and run gear 1 on RKL upto 3733-
ish
Suggestion/Request: full analysis (Buildzoid style) of the ASUS ROG Crosshair VIII Extreme.
Love these videos. But I have to admit, probably 90% of computer builders don't go beyond looking at rated specs. More than half probably don't go beyond whether it has RGB.
Hence the title of this channel
Like your new background style
Hi, do you have any advice for motherboard ddr5 With i7-12700?
I don’t know a ton about memory kits, but 1.6v struck me as quite high. Most xmp profiles I see only go up to 1.5v
Thanks for not killing my soul with that other bad wallpaper.
Should I leave XMP off before owing RAM?
SO THATS WHY MY GAMES ARE STUTTERING AND ARE SHIT ON XMP ON AND SMOOTH ON XMP OFF? fuck man...
Thats interesting im currently buying new ram and after watching this im so glad I have an Asus Mobo so I guess my XMP should be fine😁🤔
i use XMP as a starting point then dial things in from there...
these 4x16gb TimeTec 3600 sticks run below XMP at 3733 at xmp volts.... the xmp is loose, im actually ok with that in this case, as it would allow it to work with alot of systems that other kits i have delt with wouldnt post in with xmp enabled... (first gen ryzen mostly..)
these... are hynix djr die ram.. 210 for my ram kit... well its rated via xmp at 18-22-22-42, im gettin 3733@17-20-20-38-cr1, with a tREFI at over 32k(board max), this is on x299/7940x, so, 4k and up on ram is...not that easy or common.. doable but... honestly not worth the timings i would need to use vs what im getting at 3733.. at 1.3v... im pretty sure i could dial the trfc in a bit but... honestly im pretty happy with the results so far....
Well they should list timing and ram ic and ic configuration
my 2x16gb 3600mhz c18 Team vulcan Z has the trfc set on 990 when enabling xmp LOL. That's way higher than if the ram were run on JEDEC spec
How is HyperX sticks? Most of my sticks are HyperX C-Die and CJR/DJR...
Takk!
I could never quite nail the TLDR value.
So is it better to get a high clock kit with looser timings and clock it down to tighten the timings, or is it better to just get a lower clocked kit with tight timings (for Ryzen)? Like is it better to get a 3600 kit with like 14,15,15,35? Or like a 4000 kit with 18,20,20,40 and clock it down to 3600-3800?
Whichever has the lowest latency, really. At least when it comes to B-Die/Rev. E.
I could really use some advice on decent timings for the patriot viper blackout 4000 kit. This kit hates xmp anyway.
I can't use XMP for the 4x8GB Patriot Viper 4400MHz (19-19-19-39) in the MSI Z490 Godlike MB (i9-10900K) because it crashes Warzone. Same for following the manual OC settings video by AHO, maybe because a few available choices differ from the instructions given by AHO but I've tried everything before giving up.
Do you know what sa and io voltage your motherboard has when you set xmp ? Also the fact that your ram loads into windows with 4 sticks on xmp is impressive.
I heard the Adata 5000 C19 is the best DJR kit, is it true?
To me the Renegade 5333 C20 looks better on paper.
hmm question, so ive lowered all my timing pretty much as much as possible and ran memtest for 24hours and prime95 for 12hours no crashed or errors, and i used aida, cinebench and timespy to check performance to make sure the timings increased performance along the way.
But my question is can a timing ever be to low, i have some timings that i was able to get down to 1 without encountering errors, will that make my memory degrade quicker or something possibly is there any risk to that at all?
Which timings my dude?
Depending on the ram, some timings can go down to 1 without problems.
You can tell because some are set to 1 when left in AUTO
Maybe you could help me choose the memory dice:
I currently have:
Mobo MSI X370 Pro carbon
Proc Ryzen 7 1700
Memory F4-3200C14D-16GFX G.SkILL 2x8GB 3200 cl14 mhz samsung b-die (single rank)
I don't want to change hardware but needs more ram memory. Due to the fact that the first generation of procesor is sucks when it comes to the memory controller, to keep 3200mhz I just have to replace the set of ram.
I am wondering if you want to buy a set F4-3200C14D-32GFX 2x16gb 3200mhz cl14 samsung b-die (dual rank) or buy something else?
Or maybe buy crucial ballistix 2x16 3600mhz cl16 and set them to 3200mhz. I know they are sometimes single rank, perhaps micron d and sometimes dual rank micorn e.
I do not know what to do, I would like to buy a frame that I will use one day when changing the platform.
I wonder if on newer generations of the processor it is at all possible to combine such sets F4-3200C14-16GFX with F4-3200c14-32GFX, i.e. throwing two single rank sticks and two sticks with higher dual rank capacity?
I do not know what to do to advise something?
Oh MSI just ignores sub timings, that explains why i had to set the XMP manually on my MSI board when i was testing it, figured it was just cause the speed was over 4000
not like a would keep a b-die kit kit at stock anyway...
It was a 4133 viper steel on a msi b550I board, IIRC the auto was worse than what was on the spd chip (i did force 1T like the profile called for) and the bios is just like you expect the IMC to do this? yea right i'll do this instead without realizing it is a APU and will not care
Isn't gear 2 worse than gear 1?
4600 CL22 with JEDEC subtimings gets over 12000 points.
Has anyone done any testing with the team group delta ddr4 kits?
what does DRAM calculator for Ryzen say the best timings could be?
Can I hire you to do timing on my pc ?
I have a 3200 kit that reduces the FCLK to 1600 to run synchronous... It underclocks the infinity fabric. I had to manually set it to 4000 and 2000 and add some voltage here and there.
Should you retrain your memory every boot when ocing?
Does this info also apply to AMD? (X570 Aorus Pro Wi-Fi v1.1, Ryzen 5600x, Corsair RGB Micron E Die 4000MHz cl19 8Gb(4sticks) )
My 3000 c15 Trident z kit won't go past 3000 or lower than c15 :( I'd love to know how to get to 3200 c14 at least.
my Adata xpg silver 3000 2 sticks does lots more issues for some reason tthey didnt have those issues when i had a r 7 1700 i didnt face those issues only when i used them on 3400 g
always crashing and slowing down my system i tried lots of things finaly decided to sell them and replaced them with hyuper x fury 3200 mhz
l cant stop getting surprized every time l watch a video where the latest most expensive piece or top tier hardware has the same issues l have with my old shitty low end PC xD
I now have a 4x8Gig Memory Kit from TForce with Samsung B-Die on them, validated for 4000 Cl18. When I activate D.O.C.P on my MoBo it wont even boot the System (5950x, IF clocked at 1.9GHz atm, wanted to try 2GHz but haven't gotten into it for now). Any suggestions what would cause it to fail to boot? Should I try Manual Voltage override ?
What motherboard are you using? Is that RAM on their QVL? Being it fails to POST I am wondering on compatibility. May not hurt to check your UEFI version on your MB too
@@Frizzy9000 Haven't checked up that yet, running a X570 Dark Hero from Asus, got it in June and installed newest UEFI back then, right before I installed any OS's... Gonna check these suggestions up could be one of the reasons...
Try to slowly raise the ram Frequency from default, that gives you a rough idea if the cpu can handle it because the mobo will set really loose timings to try to boot.
If you really wanted to make a fair comparison force 2666 to run in G2 as well
my ram just died feels amazing :D, instead of having 16gb now it is 64gb :D
Ive got 2 sets of dual chanel trident z 3200cl14 8×2 single rank
And they wont but a single mhz higher than xmp
Im using a b550 steel legend and ryzen 5900x
Any advice?
I tried every possible memory chanel mod single dual and full dimm
keep calm, raise SOC voltage
@@Yasaro007 it was in the wrong dimms
This is my problem that I am searching for 2 years of lag in games also my ram is gskill royal xmp 4000 cl17 dual kit xmp works really bad on my aourus z390 master!At first it works xmp on and good to go but after a while I realize massive delays lags everywhere I can't figure out what the hack I spend tons of money change components but not ram ! Than I change the timing manually as in this video after 2 years problem solved.
my tRFC is 420
This memory gen seems disappointing so far though the same was probably said about DDR4 launch as well.
If my 5800x can run at 2000flck, is it worth trying to push my 3800 CL14 kit to 4000mhz
yes if you can keep the timings or just check and compare performance
Eres la gaver Buildzoid!
I paid for XMP so I'm going to use all the XMP? Uh oh
Could never trust XMP before, just another factor against it and with true proof that it is scuffed and DIY should be wary about it.
Mine just crashes often - it runs @ 3200 could be on it gaming for the day then next day bluescreens on startup - or like just now its fine all day then I go on CZcams and Ads are showing with Adblock on -wtf- so disable it - refresh - enable - BSOD - I find using my browser often causes BSOD when overclocking my Ram not sure why, that's your area of expertise I guess. . . Hyper X Predator 3200 - MSI X470 M7AC Ryzen2700X
I've come to the conclusion that XMP is pretty much ass and it's better to have disabled, unless you get very lucky.
AIDA64 memory tests wont tell you much outside a general idea if you have gone too far without causing noticed instability, now, running he full bench using a quick report, the photoworxx test at least on my 7940x gives you a much better idea if your fucking shit up or making things better.. hell the overall bench can give you an idea if you made it worse or better.. why i store multi reports as i test tweaks.. very handy tool.. aida's memory bench is useful as a quick "did i make it better or worse in general?" type test but.. yeah i have found it useless beyond a vague idea.. (if you see it jump up or down alot, you get a general idea if your fucking up or not..)
anyway, im having fun dialing this new system i build after my x58 board died.. (a 6c12t xeon overclocked to 4.6-4.7 is quite competitive with first gen ryzen in most games faster in several..), this x299 systems been a learning exp.. but also been more fun dialing a system in then i have had for years.... so many are so close to their limit as it is.... this.. ram and cpu have given me a shitload of fun..
There's nothing to wonder here. It's called marketing BS
Those XMP sub timings are just awful.
Some motherboards do not even run XMP profiles. I had a B450 (Asus) that failed to load up a 3200MHz XMP because of shitty VRMs.
Most b450 b350 asus are shit. B550 is kinda
Good
The VRMs for the memory are so insignificant it seems rather unlikely to be the cause; are you sure you didn't have the memory controller setup wrong? Some of the early Ryzens needed the termination impedances tweaked or they wouldn't drive the memory correctly.
@@lucidnonsense942 Mind explained that out?
I have a 1600 and anything higher than 2933 is unstable, nothing I did managed to fix it.
@@ghoulbuster1 There's too much to go into, but setting RAM terminations, vSoC, and RAM impedances is pretty key to get Ryzen 1000 to go anywhere. Furthermore most termination and impedance stuff is really on a case-by-case basis so there's no real good tutorial out there.
That being said you shouldn't expect much, best I managed on the platform (stable) is 3333C14. It just doesn't clock well.
@@IK4MS I see well, I wanted to go 3200 C16 on it since that's a good speed for ryzen but I spent days trying to stabilize it using really loose timings and changing everything and nothing worked.
I could try again but I'm sure the problem is the cheap B350 mobo that I used.
Kudos for the thumbnail, it's lowkey amazing
This is the case for my Gskill Trident Z (CL14 Samsung B) The tRFC is set to something insane like 500+ with XMP enabled. AMD Ryzen 5900X on an X470 Board (This issue was also present when I was using a 2700X on the same Board)
"The tRFC is set to something insane like 500+ with XMP enabled."
That's because it's required to operate properly at ~95c. Lower RFC's only work if you keep the memory cool, which people aren't guaranteed to do.
RFC will only affect performance by a handful of %. RRDS/RRDL/FAW can half or double it.
I would think think that the change from gear 1 to 2 also contributes a lot to this slowdown
nope it's entirely the tRRD timings.
in fact, the before and after is gear 1 command rate 2, versus gear 2 command rate 1, so it's very close to a non-factor if memory serves correctly.
I know noobs more then likely wont know this but isnt this why we manually tune our ram lol.
the thing I'd like most is if they copied across timings is if they copied between BIOS updates. XMP is easy. I avoid BIOS updates it is so boring.
damn
Nice
Still on the trash capture card, praying there's an upgrade at some point.
odd twist of Scott Manley, "check your sub-timings, and Overclock safe", lol.
just run DISM and fix the damn corrupt OS :D