8 PACK MEMORY RANGE GROWING: SAY HELLO TO 8 PACK RIPPED EDITION & 32GB KITS!!!

Sure. Black values on the left are 'calculate fast'
Compare timings are mine which I set on the right in red

I just copied the fast timings and set tRCDRD to 14 and put DRAM voltage on 1.45v.
Checking now I appear to have missed a couple

2aK1ltB.png

MrPils. Wow, that's a lot of useful info, thanks. To be honest I just wanted to see if the fast timings worked (mainly CAS14)
I dont want to corrupt my drive / OS but do have it imaged onto another SSD in case it fails
Is 1.45v needed / safer than using the recommented 1.40v max?

Just getting back into this sort of overclocking since building a Ryzen machine. My guess is that better timings will come into play more with a faster CPU like the 5900X i'm waiting for. Currently using a 3600, but its 'only' 4.2Ghz and seems way faster than my older intel at 4.5Ghz. The reason will no doubt be down to higher clocked memory and other advances, not just CPU

Thanks.
 
Hey, just wondering if you have any clue on this behaviour. I've used a more aggressive fan profile and dropped temps and 16/16/16/16/32/48/288 is now perfectly stable (with tWCL 14) at 3600mhz

I then follow your procedure and go to turn CL, tCWL, tRAS TRC and TRFC down 1 (so 15/13/31/47/282) and it refuses the CL change but accepts the other timing changes. Thinking it might have a problem with C15, I tried C14 and 12/30/46/276 and it dumps out and puts me on a very poorly trained C15/15/15/15/36/51/374 profile back at 2133mhz.

As you've seen, i'd probably be tempted to up the voltage a bit, or tune down the other timings but leave the CL and primary's at 16 but I'd love to know what you think

If you have Gear Down Mode enabled you can only use even Cas latencies. Switch that off and try 1T command rate (or 2T if that wont work) and then youll be able to use odd numbered Cas latencies as well as even. Test it carefully if you go to 2T command as that can give a noticeable performance drop.

Sure. Black values on the left are 'calculate fast'
Compare timings are mine which I set on the right in red

I just copied the fast timings and set tRCDRD to 14 and put DRAM voltage on 1.45v.
Checking now I appear to have missed a couple

MrPils. Wow, that's a lot of useful info, thanks. To be honest I just wanted to see if the fast timings worked (mainly CAS14)
I dont want to corrupt my drive / OS but do have it imaged onto another SSD in case it fails
Is 1.45v needed / safer than using the recommented 1.40v max?

Just getting back into this sort of overclocking since building a Ryzen machine. My guess is that better timings will come into play more with a faster CPU like the 5900X i'm waiting for. Currently using a 3600, but its 'only' 4.2Ghz and seems way faster than my older intel at 4.5Ghz. The reason will no doubt be down to higher clocked memory and other advances, not just CPU

Looks like some of the timing sets have been improved in 1usmus's tool then, it used to spew out some proper random stuff. If you're worried about corruption just dont touch tRAS or TRFC timings, not much chance with the rest unless you drop them all too far too fast. You'll hit instability before corruption if you test properly :).

1.45v is fine as long as your modules don't get too hot. The extra voltage introduces more heat and B-Die is sensitive to heat when tuning timings down. You'll find looping tests will work at first then as the test goes on and the modules warm up you'll get single errors popping up. I test with a fan over the modules then remove it for a final pass of stability checking when I'm happy with where I'm at. You'll find previously passed tests fail if you need some memory cooling. Your choice then is either to reduce voltage and increase timings or sort out some kind of airflow. They wont need much, just a light breeze is plenty. B-Die is fine up to 1.5v daily as long as you don't let it overheat. Each memory module is different as to where its heat tolerance is and how hot it gets - many variables in play here: chip quality, heatspreader quality, thermal interface, case temp and airflow etc so you need to test for your personal rig.
 
If you have Gear Down Mode enabled you can only use even Cas latencies. Switch that off and try 1T command rate (or 2T if that wont work) and then youll be able to use odd numbered Cas latencies as well as even. Test it carefully if you go to 2T command as that can give a noticeable performance drop.
Cheer's. Had forgotten about geardown and odd numbers. C14 seemed to need too much voltage to even boot, and it really liked 288 tRFC but I've managed this with your tips, at 1.41v, which nets 112.56 on the membench default test and is stable for 250% HCI pro.

6sBYKkl.png
 
Cheer's. Had forgotten about geardown and odd numbers. C14 seemed to need too much voltage to even boot, and it really liked 288 tRFC but I've managed this with your tips, at 1.41v, which nets 112.56 on the membench default test and is stable for 250% HCI pro.

Nice work dude, that's a sweet tune for 4 sticks and a good time for 3600mhz. You can find with 4 sticks (dual ranks) that some secondary and tertiary timings really don't want to move any lower without force. The trick is not to get caught up trying to make things work and to pick off everything that will work first. Everything is well within safety bounds for voltages there and getting much more out of it would likely take a lot more work and some voltage increases that could end up giving you heat issues - in other words you're at the point of diminishing returns. When tuning memory becomes an obsession its worth going further, until then just enjoy what you've achieved! :)
 
Nice work dude,...... When tuning memory becomes an obsession its worth going further, until then just enjoy what you've achieved! :)

Thanks very much, all down to your tips in this thread :) . If you ever have spare time and the inclination, I'm betting you could write a new general memory guide sticky for the forum that isn't 9-17 yeas old, mainly taken from the posts you've already written here.
 
Thanks very much, all down to your tips in this thread :) . If you ever have spare time and the inclination, I'm betting you could write a new general memory guide sticky for the forum that isn't 9-17 yeas old, mainly taken from the posts you've already written here.

It's not so much a question of spare time and inclination as something like that could be written over a period of time. Its more that I've never been asked to do it and I'm not sure I would be suitable to do so anyway. The big issue with someone who already knows a subject writing a "start from scratch" guide is that they already know the subject they're writing about. I would not be able to write a guide that bypasses assumptions, mistakes and pitfalls faced by a new to memory tuning user because I just wouldn't think of them. The problem with knowing what to do is that you forget a lot of what not to do. I'm totally ok with my posts being compiled/rewritten by someone who has used it to get going and as you can tell from this thread am happy to share knowledge and help set people off in the right direction when they are stuck or go wrong. The issue is that my starting point for writing such a guide would likely be too technical and somewhat blind to how overwhelming a task it can appear at first to be. I do repeat myself a lot in this thread describing the same method in various different ways as coming from a more technical standpoint on memory settings and setup although what I write is correct, to someone who is not sure of themselves or new to memory settings it is not always clear what I am saying (which is fine and something I expect). In a conversational format you can get past this, in a guide format its not so simple.

I also only have first hand experience with B-Die on AMD and older Intel platforms (X99 mainly), advice I give on other memory sticks and platforms is always based off this knowledge but I definitely cannot say I have the same depth of knowledge of Micron or Hynix sticks and certainly wouldn't claim to know much about modern Intel dual channel platforms. I also fear this knowledge will likely only be relevant for a couple more years or so, I'll be starting on a blank slate figuring out the relationships between timings and particular idiosyncrasies of different DDR5 memory chips myself in the not so distant future.

I'm totally happy for anything I've posted here to be used in any format at all - its all public information and offered freely, but for sure I think the flow of a guide needs to be written by someone who has recently achieved rather than someone who already knows otherwise its just not relevant to the intended userbase. Does that make sense at all?

Anyway, if the mods here want a new sticky made I'll happily help out directly with such a project if needed and obviously give permission for any of the content of my posts to be used, but I sincerely think it would need to be a collaborative effort from users of various technical levels for such a guide to be useful.
 
Does that make sense at all?

Yeah for sure, you know the limitations of your knowledge better than anyone. Just thought I'd mention it, if it struck a chord with you. I was aware you were repeating yourself to different people and its also obviously tough given the nature of the beast with so many variables. I mainly say that because I ended up copy and pasting various bits of your posts into a little, slightly disjointed but ultimately helpful guide and printed it off to have in front of me to follow as i was testing, which was easier and saved time delay checking up timing relationships and advice on my phone as I'm in the bios :D

Anyway, if the mods here want a new sticky made I'll happily help out directly with such a project if needed and obviously give permission for any of the content of my posts to be used, but I sincerely think it would need to be a collaborative effort from users of various technical levels for such a guide to be useful.

Fair point, I might see what I can cobble together in a bit of free time in the next month or so and if it achieves anything comprehensible. Might have to go elsewhere for some E-die knowledge (main variant that perform's well I think) but I think it's do-able.
 
Yeah for sure, you know the limitations of your knowledge better than anyone. Just thought I'd mention it, if it struck a chord with you. I was aware you were repeating yourself to different people and its also obviously tough given the nature of the beast with so many variables. I mainly say that because I ended up copy and pasting various bits of your posts into a little, slightly disjointed but ultimately helpful guide and printed it off to have in front of me to follow as i was testing, which was easier and saved time delay checking up timing relationships and advice on my phone as I'm in the bios :D
Fair point, I might see what I can cobble together in a bit of free time in the next month or so and if it achieves anything comprehensible.

You would be far better suited to that than me as you have just run through the process yourself from a clean slate so to speak (but I would be happy to contribute/proof read/collaborate at whatever points would be helpful). The main issue any guide has is explaining the unknown to individuals in a way they understand, everyone has different ways of understanding what they're told so what is clear to one person is not to another even if they are at a similar level of knowledge. If I sat down and wrote something I thought was comprehensive it would be more instructions than guide. Generic instructions don't teach why things work, just what things work and the user is therefore unable to tune their system individually - the Ryzen timing calculator takes that approach and I dislike it immensely for it. Users just become reliant on instructions instead of learning why and being able to figure out improvements or limits for themselves. While its great for the ego to be treated like a tome of knowledge, it gets old fast and eventually the tome disappears leaving very little lasting knowledge behind. Nobody learns if spoon fed.

Might have to go elsewhere for some E-die knowledge (main variant that perform's well I think) but I think it's do-able.

The big difference with Micron and Hynix is both need much higher TRCD, TRP, TRC and TRFC values. For example with two single sided sticks where 14-15-12-1T @ 3600mhz+ is fine on B-die you would be looking at 14-17-16-GDM with Micron and 14-18-18-1T with Hynix. TRC is usually around 50 or more and TRFC almost always needs to be 500+ as well, basically double B-Die. Where these style chips are useful is easy access to really high mhz or running high capacities without stressing the memory controller and therefore requiring high voltages. If you wanted to run 4 x 16gb or even 4 x 32gb sticks for example either would be an excellent choice. In performance terms you can do fairly well on Intel as it just keeps scaling with more mhz, but AMD cannot make the best of these style of chips while going off 1:1 ratio Infinity fabric gives such a penalty. Hynix doesn't like voltage so much, you don't get much scaling above 1.45v where B-Die scales all the way to 1.6v give or take before you need to start limiting how much memory windows sees to avoid blue screens. Micron sticks will handle voltage but require some messing around with reducing vTT, behaviour which starts to show itself even as low as 1.42v in some cases...and each stick's vTT behaviour is different meaning getting two (or four) with similar vTT scaling requirements can be almost impossible.

B-Die is a really simple beast - keep it cool, keep it below 1.55v daily (preferably 1.45v to 1.5v for some safety margin), set your tRRD's, tFAW and tWTR's to 4-4-16-4-8 and start increasing mhz until you reach your goal. Then you just keep lowering the main timings according to formulas until its unstable. tRAS going below "normal" allows tRC to go lower than other kits and gives a wider amount of choice for tRFC too, 1T is not an issue (but not always fastest with dual ranks or 4 sticks) and secondary and tertiary timings will just keep going down and down. If you save a bios profile as you make changes the only danger you face is going too low on tRAS or tFRC and corrupting an install, but if you use a bootable pendrive for testing its not really much of an issue. When you find the limits they're usually quite hard limits whereas with other chips there's far more grey area where you can get yourself into trouble by trying to force stability.
 
Nice setup, definitely don't go back to XMP! In particular you have done well with your tRRD, tFAW and tWTR timings. These in combination with Command Rate 1 will give you excellent gains in gaming irrespective of your other timings.

There's a few extras you could get there for free most likely:

1) tRCDWR can always go to 8 with most B-Die, its just a thing. I've not seen this fail to work yet.
2) Likewise tRP can usually go down to tCL - 2 at already stable settings, again this is usually free and just a thing it does.
3) tRAS goes below the usual formula with B-Die, reduce it slowly though as you dont want to corrupt your hdd. The idea is to find the point where it takes longer to train at post or wont quite load into windows then go up again from there. If you aren't careful you can corrupt your OS, so don't try and make it stable, consider it a hard limit for your particular chips when you find it.
4) With tRP and tRAS reduced you can then reduce tRC to the total of tRP + tRAS.
5) Once tRC is at minimum make tRFC a multiple of tRC and find the low limit - you'll be moving down in big chunks so it will no post / no boot when you find it. Don't try going below 200 unless you are happy risking your OS, most kits end up around the 220 to 250 mark. Once you have the limit check if your board wants to have it bang on a multiple or within a +-2 range, so say your limit was 250 you would test 248 to 252 and see if a particular specific setting works better than the others. Exact multiples should technically be optimal but often aren't. Use SuperPi and AIDA64 for this.
6) As you already have Gear Down Mode disabled you can try CL13 - some boards just don't like it and wont post. With CL if you can get into windows with the memory warmed up you can usually stabilise it with a small amount of extra memory voltage. If you need 0.05v or higher don't, its not worth it. If it no posts forget it, it's not happening.

The final bits you can try and tweak are the tertiary timings. SC at the end stands for Same Channel, these will all usually go down really low if only using two sticks but dual rank can limit that depending on the motherboard. SD means Same Dimm, dual rank will limit you here so these wont go down as far as single rank setups can go. DD means Different Dimm and is for swaps between channels - if your memory trace layout is good on the board you can go really low here (ITX boards can usually go down to 1). Tertiarys help with latency but AIDA doesnt really pick it up so check performance with SuperPi.

If you want more mhz you will most likely need to go to tRCDRD 15, you will need to gain around 100mhz to see any benefit compared to running tRCDRD 14 though. This timing is big for performance.

Testmem5 is good for testing if you know your cpu, cache and fabric are already stable - it only hits the dimms and will save you time compared to testing using mixed workload stress apps.

Much of the above is a repetition of my earlier posts, but is laid out specific to where you are at right now so may help a bit more than just referring to generic B-Die tweaking instructions.

Still trying to digest this as there's a lot of learning. Thanks :)

I havent altered any specific settings but have set DRAM voltage from 1.45 to 1.42. I'll give Testmem5 a go to see if its stable at 1.42v

Ix36O0W.png
 
Still trying to digest this as there's a lot of learning
This was the best I managed on the similarly binned 3200/14 kit with a ryzen 3600 but maybe this will give you an idea of what to further tighten but every kit is different so YMMV.

198.png
 
This was the best I managed on the similarly binned 3200/14 kit with a ryzen 3600 but maybe this will give you an idea of what to further tighten but every kit is different so YMMV.

198.png
Thanks. Mine is the 3600 cas16 kit however. Should that make much of a difference? Guess so as its totally different ram
 
Still trying to digest this as there's a lot of learning. Thanks :)
I havent altered any specific settings but have set DRAM voltage from 1.45 to 1.42. I'll give Testmem5 a go to see if its stable at 1.42v

Finding your minimal voltage is a good idea, it will not affect the tweaks I gave you above as none of them will require more voltage than your memory needs to be stable already.

You have the advantage that you've already got your memory running nicely, so if you do want to get a bit more out of your memory setup just take it slowly, go one step at a time and take a while over it. I numbered each hint so you can tackle them separately, so for example next time you switch on set only tRCDRW to 8, save and exit and go about your day as normal. Once you are happy that's stable (after a day or two of normal usage or after stress testing, either is fine) you can think about moving to step 2 and changing tRP to 12. You don't have to do it all at once. :)
Steps 3, 4 and 5 you have already performed once when you achieved your current results, all you are looking to do to maximise your work is push tRAS lower 1 increment at a time until you find instability. If you apply the same slow and steady approach of testing/using your pc for a day or two as you increment the timing down you'll safely find your tRAS limit (21 is the minimum possible, all my kits run 22 just fine right up to 4000mhz but wont do 21 at all). Once tRAS is minimised, tRAS + tRP will give you your optimal tRC value for step 4 (as this is formulaic you don't need to step down to it you can simply type in the result and go). For example if you end up getting tRAS down to 24 and tRP 12 is stable, your new tRC value is 36.
Based on that tRC 36 you are looking at tRFC values of either 240 (5x36) or 276 (6x36) as optimal for step 5. You know 252 is stable already (so therefore 276 will be as well), so give 240 a try - I would be shocked if this did not work @ 3600mhz. I wouldn't bother trying 204 (4x36) as that's too low for the vast majority of B-Die without high volts. Once you find out if 240 is a stable do some quick benchmark testing of 238,239,241 and 242 to see if any tuning is of benefit to your board.

Save bios profiles as you go so that if you get a post failure you can be back up and running quickly.

Step 6 is....optimistic. I would not expect it to work, but you may get lucky. I've seen it work before but not often, less so on 16gb sticks. I would expect it not to post or to blue screen - you are likely to need significant extra voltage to get it to work in this case, so honestly don't get your hopes up there.

Setting tRCDRD 15 could be considered step 7, you can then try going up in mhz incrementally but bear in mind you will need to get to 3733mhz before you see a noticeable benefit from increasing tRCDRD. This may require messing with infinity fabric voltages and additional stress testing related to that so its only something I would do if you find yourself bored one day. If you have CL13 working don't try it, its totally not worth it.

Above all take your time, this is something you can do a bit at a time as the fancy takes you. There's no need to rush into it as something you must get done ASAP. Your system is already running fast, this is just tweaking out the final couple of settings to get it perfectly tuned. Its not necessary, its optimal. :)

Thanks. Mine is the 3600 cas16 kit however. Should that make much of a difference? Guess so as its totally different ram

Yours is a 2 x 16gb dual rank kit running command rate 1T and his is a 2 x 8gb kit single rank kit with Gear Down Mode enabled, so yes you're correct they're totally different. Don't try copying timings from other users with random memory kits, work with your own platform and get to know your own hardware. Copying others memory timings teaches you nothing when it doesn't work. :)
 
Last edited:
Thanks Pils
I noticed the DRAM speed is set at 3800Mhz instead of 3600Mhz which is probably why the time is lower. I presume here FCLK is 1900Mhz
3800Mhz does seems to work with FCLK at 1900 but only CAS 16.
I'll do some more tweaking tonight
 
Thanks Pils
I noticed the DRAM speed is set at 3800Mhz instead of 3600Mhz which is probably why the time is lower. I presume here FCLK is 1900Mhz
3800Mhz does seems to work with FCLK at 1900 but only CAS 16.
I'll do some more tweaking tonight
Yeah FCLK 1900 will make big inroads into latency - the membench test baked into the Ryzen checker is almost entirely latency based. Different apps behave in different ways so you need to performance check based on your real world use cases not just in benchmarks. You will almost certainly find that tRCDRD is your limiting timing for stability at 1900FCLK/3800MCLK so build your timings based around the lowest tRCDRD you can get. 16-15-12 will be faster than 15-16-12 in the vast majority of applications.

Above all though have fun, if it gets tedious stop tweaking and use the PC for a bit and come back to it later. Its much easier when its play instead of work, once frustration sets in its time to stop for a bit :)
 
I just bought me 2 sets of this. memory heaven.

TEAM GROUP RIPPED EDITION 16GB (2X8GB) DDR4 PC4-28800C14 3600MHZ DUAL CHANNEL KIT - BLACK/

Nice.
Is there any reason for going 4x8GB over 2x16GB? Timings on that are slightly better but it works out more expensive and can't use the other 2 slots in future (i doubt 64GB RAM will be needed any time soon mind)

Does Ryzen like having all 4 slots used?
 
Nice.
Is there any reason for going 4x8GB over 2x16GB? Timings on that are slightly better but it works out more expensive and can't use the other 2 slots in future (i doubt 64GB RAM will be needed any time soon mind)

Does Ryzen like having all 4 slots used?
If you get a board that can clock really well with 4 slots then its worth doing it from purely an overclocking/tweaking perspective and ignoring the value and upgrade angle - he has a fancy pants X570 Aorus Extreme. :) When a manufacturer has put the time and expense into getting the slots clocking high when fully populated then 4 x 8GB single ranked sticks can almost reach the higher speeds and tighter timings they're known for in pairs. Most boards though have the second pair of slots just tacked on as an afterthought, all the work goes into the first two slots as most people will be running 2 x 8gb and will never use the second pair anyway. In most cases 2 x 16gb will clock better as it will avoid using the (usually) hampered second pair of slots.
 
Back
Top Bottom