First time overclocking Ram please help

[Ebony Falcon]

Hi guys, 

so yesterday I upgraded to a 32gb dual rank 2 dimm kit with Samsung Bdie Moduals. There sold as a b die kit from ocuk and I verified and the spd tab in cpuid 

xmp is 3600 16-16-16-39

team group dark pro 8 pack edition 

 

xmp performance was underwhelming as expected 65ns in aida 64 latency test.

 

however I got this set to get into overclocking 

 

I have a 13600kf and a z790 tomahawk board.

 

I know didaly quat about ram overclocking.

 

where do I start ? 
 

I was going to start by going to 1.4v on the dram voltage and just pushing frequency up one notch at a time untill it won’t boot. Take it back 1 step and test with occt for a hour to make sure it’s not erroring imidiatly.

 

if no errors what do I do next try lowering primary’s one at a time ? Or start dropping the trcf 20 at a time to see how far that goes ?

not really sure what I’m doing and would appreciate some help  

 

[RONOTHAN##]

10 minutes ago, Ebony Falcon said:

where do I start ? 

Set 1.45V (1.5-1.6V if you have active cooling) for the DRAM voltage, set 1.4V VCCSA, make sure to set Gear 1, and loosen out the primaries a bit. Start increasing the frequency until it stops being able to POST, then start lowering the frequency until you're able to quick memory stress tests (I like Y Cruncher 2.5B since it only takes a minute and spits out a performance result, but you can pick something else). Also, disable the E cores and setup a static overclock, it doesn't need to be anything crazy high or anything but you will be checking to make sure there aren't any performance regressions so having a locked core clock helps make sure that it's changing (locking it at max turbo isn't a bad idea). Disabling the E cores is because they can cause issues with some memory stress tests. 

 

Once you have the frequency dialed in as likely to be stable, start lowering the primary timings one by one until you fail to POST, then start walking them back up till you can pass memory stress tests. Primary timings don't really make that much of a performance difference, but on Intel they do have some relation to other, more important timings so you want to get them dialed in first. tRAS you can just punch in 28 (the lowest that timing goes IIRC on the DDR4 Raptor Lake IMC, though it might be different from previous Intel memory controllers or the DDR5 Raptor Lake controller), 2T command rate will be a must, and tRCDWR can just be set to 8, meaning the only three primary timings you need to worry about configuring are tCL, tRCDRD, and tRP. Try to get those dialed in, lowering them till they won't boot, then raising them till they pass stress tests. Once those are dialed in, move on to tRFC and tREFI. Set tREFI to something reasonably high, 66666 is almost guaranteed to not have issues, 111111 is slightly more likely (my kit of DDR5 will error instantly with a tREFI this high) but still incredibly rare to cause issues, and 262000 (maxed out) will give the best performance and probably be OK, it can sometimes run into hard to diagnose errors so it's not necessarily a good idea to run this, even if you can boot and run stress tests short term with it. tRFC doesn't really matter so much once tREFI is as high as it is, so I'd just set it to 300 and call it a day. It can probably go lower but it's not worth the hassle of tuning it lower since the performance difference is so low. After those are set, set the tRRD timings and tFAW. tFAW should go to 16, tRRD_S should go to 4, and tRRD_L will likely go to 4, but if there's stress test instability set this to 7 (4 rarely causes issues, but it technically can, while 7 almost certainly won't). If you want to tune no more timings, have those tuned, those are what make the biggest performance difference. No matter what though, I'd do an overnight stress test on those just so you don't have to think about them when tuning the tertiary timings. 

[Ebony Falcon]

9 hours ago, RONOTHAN## said:

Set 1.45V (1.5-1.6V if you have active cooling) for the DRAM voltage, set 1.4V VCCSA, make sure to set Gear 1, and loosen out the primaries a bit. Start increasing the frequency until it stops being able to POST, then start lowering the frequency until you're able to quick memory stress tests (I like Y Cruncher 2.5B since it only takes a minute and spits out a performance result, but you can pick something else). Also, disable the E cores and setup a static overclock, it doesn't need to be anything crazy high or anything but you will be checking to make sure there aren't any performance regressions so having a locked core clock helps make sure that it's changing (locking it at max turbo isn't a bad idea). Disabling the E cores is because they can cause issues with some memory stress tests. 

 

Once you have the frequency dialed in as likely to be stable, start lowering the primary timings one by one until you fail to POST, then start walking them back up till you can pass memory stress tests. Primary timings don't really make that much of a performance difference, but on Intel they do have some relation to other, more important timings so you want to get them dialed in first. tRAS you can just punch in 28 (the lowest that timing goes IIRC on the DDR4 Raptor Lake IMC, though it might be different from previous Intel memory controllers or the DDR5 Raptor Lake controller), 2T command rate will be a must, and tRCDWR can just be set to 8, meaning the only three primary timings you need to worry about configuring are tCL, tRCDRD, and tRP. Try to get those dialed in, lowering them till they won't boot, then raising them till they pass stress tests. Once those are dialed in, move on to tRFC and tREFI. Set tREFI to something reasonably high, 66666 is almost guaranteed to not have issues, 111111 is slightly more likely (my kit of DDR5 will error instantly with a tREFI this high) but still incredibly rare to cause issues, and 262000 (maxed out) will give the best performance and probably be OK, it can sometimes run into hard to diagnose errors so it's not necessarily a good idea to run this, even if you can boot and run stress tests short term with it. tRFC doesn't really matter so much once tREFI is as high as it is, so I'd just set it to 300 and call it a day. It can probably go lower but it's not worth the hassle of tuning it lower since the performance difference is so low. After those are set, set the tRRD timings and tFAW. tFAW should go to 16, tRRD_S should go to 4, and tRRD_L will likely go to 4, but if there's stress test instability set this to 7 (4 rarely causes issues, but it technically can, while 7 almost certainly won't). If you want to tune no more timings, have those tuned, those are what make the biggest performance difference. No matter what though, I'd do an overnight stress test on those just so you don't have to think about them when tuning the tertiary timings. 

Ok thanks bro, il start going through this process tonight 

[Ebony Falcon]

On 11/30/2022 at 10:47 AM, RONOTHAN## said:

Set 1.45V (1.5-1.6V if you have active cooling) for the DRAM voltage, set 1.4V VCCSA, make sure to set Gear 1, and loosen out the primaries a bit. Start increasing the frequency until it stops being able to POST, then start lowering the frequency until you're able to quick memory stress tests (I like Y Cruncher 2.5B since it only takes a minute and spits out a performance result, but you can pick something else). Also, disable the E cores and setup a static overclock, it doesn't need to be anything crazy high or anything but you will be checking to make sure there aren't any performance regressions so having a locked core clock helps make sure that it's changing (locking it at max turbo isn't a bad idea). Disabling the E cores is because they can cause issues with some memory stress tests. 

 

Once you have the frequency dialed in as likely to be stable, start lowering the primary timings one by one until you fail to POST, then start walking them back up till you can pass memory stress tests. Primary timings don't really make that much of a performance difference, but on Intel they do have some relation to other, more important timings so you want to get them dialed in first. tRAS you can just punch in 28 (the lowest that timing goes IIRC on the DDR4 Raptor Lake IMC, though it might be different from previous Intel memory controllers or the DDR5 Raptor Lake controller), 2T command rate will be a must, and tRCDWR can just be set to 8, meaning the only three primary timings you need to worry about configuring are tCL, tRCDRD, and tRP. Try to get those dialed in, lowering them till they won't boot, then raising them till they pass stress tests. Once those are dialed in, move on to tRFC and tREFI. Set tREFI to something reasonably high, 66666 is almost guaranteed to not have issues, 111111 is slightly more likely (my kit of DDR5 will error instantly with a tREFI this high) but still incredibly rare to cause issues, and 262000 (maxed out) will give the best performance and probably be OK, it can sometimes run into hard to diagnose errors so it's not necessarily a good idea to run this, even if you can boot and run stress tests short term with it. tRFC doesn't really matter so much once tREFI is as high as it is, so I'd just set it to 300 and call it a day. It can probably go lower but it's not worth the hassle of tuning it lower since the performance difference is so low. After those are set, set the tRRD timings and tFAW. tFAW should go to 16, tRRD_S should go to 4, and tRRD_L will likely go to 4, but if there's stress test instability set this to 7 (4 rarely causes issues, but it technically can, while 7 almost certainly won't). If you want to tune no more timings, have those tuned, those are what make the biggest performance difference. No matter what though, I'd do an overnight stress test on those just so you don't have to think about them when tuning the tertiary timings. 

Just did a xmp test just to get base line results and the read bandwidth was 48,000 ish isnt that very low even at xmp for this kit ?

[Ebony Falcon]

@RONOTHAN##

I tried to boot 3773 at 1.45v and wouldn’t boot, I have to option in bios dram voltage and eventual dram voltage which one should I be changing to 1.4/1.45

 

primarys loosened to all 18’s 1.4v sa 

[RONOTHAN##]

32 minutes ago, Ebony Falcon said:

@RONOTHAN##

I tried to boot 3773 at 1.45v and wouldn’t boot, I have to option in bios dram voltage and eventual dram voltage which one should I be changing to 1.4/1.45

 

primarys loosened to all 18’s 1.4v sa 

DRAM voltage, eventual DRAM voltage when set to auto will run at the same speed as DRAM voltage. 

 

Just to make sure nothing is weird, make sure the settings for 3733 you're trying also work at 3600.

 

4 hours ago, Ebony Falcon said:

Just did a xmp test just to get base line results and the read bandwidth was 48,000 ish isnt that very low even at xmp for this kit ?

It seems a bit on the low side, though it could be a subtimings that's messed up on the XMP screwing I've performance. That said, it's been a while since I've ran memory at stock, so I don't really know the numbers off the top of my head. 

[Ebony Falcon]

3 minutes ago, RONOTHAN## said:

DRAM voltage, eventual DRAM voltage when set to auto will run at the same speed as DRAM voltage. 

 

Just to make sure nothing is weird, make sure the settings for 3733 you're trying also work at 3600.

 

It seems a bit on the low side, though it could be a subtimings that's messed up on the XMP screwing I've performance. That said, it's been a while since I've ran memory at stock, so I don't really know the numbers off the top of my head. 

It’s with xmp enabled and I just changed the dram frequency to 3733 is that the problem should I have xmp off ? 
 

but also I tried tightening 3600 and it’s booting 14-14-14-14-28 not tried stress test yet just seeing how far I can drop them till it won’t boot 

[RONOTHAN##]

7 minutes ago, Ebony Falcon said:

It’s with xmp enabled and I just changed the dram frequency to 3733 is that the problem should I have xmp off ? 
 

but also I tried tightening 3600 and it’s booting 14-14-14-14-28 not tried stress test yet just seeing how far I can drop them till it won’t boot 

That seems about right, nothing that should be too bad. 

 

You might've just gotten a kit that doesn't do 3733. It's not unheard of. Maybe try bumping voltage to 1.5V to see if that works, but if it doesn't just give up on it and go for 3600C14

[Ebony Falcon]

1 minute ago, RONOTHAN## said:

That seems about right, nothing that should be too bad. 

 

You might've just gotten a kit that doesn't do 3733. It's not unheard of. Maybe try bumping voltage to 1.5V to see if that works, but if it doesn't just give up on it and go for 3600C14

Yeah I tried 1.5 with no joy  

 

latency so far just from primary’s has dropped to 63ns 

 

trefi is at 14055 when I went 66666 my latency got worse and went to 68ns

[Ebony Falcon]

These are current sub timings in bios with xmp

[RONOTHAN##]

3 minutes ago, Ebony Falcon said:

Yeah I tried 1.5 with no joy  

 

latency so far just from primary’s has dropped to 63ns 

 

trefi is at 14055 when I went 66666 my latency got worse and went to 68ns

That shouldn't have happened. I'd probably blame it on Aida64 being a terrible memory benchmark (it's super inconsistent, I have seen a 5ns delta from one run to another). It sounds like you should be tuning it for 3600 though, 3733 is just out of the cards. 

 

5 minutes ago, Ebony Falcon said:

These are current sub timings in bios with xmp

They aren't amazing, but they're not bad either. Do note that a lot of them aren't real though, only tRRD, tFAW, tCWL, tCKE, and tREFI from that list actually exist on the Intel memory controller, everything else is controlled by timings lower down in the BIOS. tWTR, for example, is controlled by the tWRRD timings instead, and they're calculated off the some other timings

[Ebony Falcon]

5 minutes ago, RONOTHAN## said:

That shouldn't have happened. I'd probably blame it on Aida64 being a terrible memory benchmark (it's super inconsistent, I have seen a 5ns delta from one run to another). It sounds like you should be tuning it for 3600 though, 3733 is just out of the cards. 

 

They aren't amazing, but they're not bad either. Do note that a lot of them aren't real though, only tRRD, tFAW, tCWL, tCKE, and tREFI from that list actually exist on the Intel memory controller, everything else is controlled by timings lower down in the BIOS. tWTR, for example, is controlled by the tWRRD timings instead, and they're calculated off the some other timings

Whats a better memory bench mark ? 
And sorry to be a pain do u have a link to the version of y cruncher u we’re on about 

[RONOTHAN##]

3 minutes ago, Ebony Falcon said:

Whats a better memory bench mark ? 

Intel MLC is relatively consistent, though it's numbers are incomparable to Aida's. PYPrime is really consistent for performance though doesn't give a ns rating that Aida does. 

[Ebony Falcon]

12 minutes ago, RONOTHAN## said:

Intel MLC is relatively consistent, though it's numbers are incomparable to Aida's. PYPrime is really consistent for performance though doesn't give a ns rating that Aida does. 

Ok that’s where where at atm 

[RONOTHAN##]

6 minutes ago, Ebony Falcon said:

Ok that’s where where at atm 

That's actually not that bad. Good kits should get 60GB/s and about 40ns, but since you're at 3600 43ns and 58GB/s should be your goals. 

[Ebony Falcon]

20 minutes ago, RONOTHAN## said:

Intel MLC is relatively consistent, though it's numbers are incomparable to Aida's. PYPrime is really consistent for performance though doesn't give a ns rating that Aida does. 

Is that the one u said to set to 8 ? 

[RONOTHAN##]

4 minutes ago, Ebony Falcon said:

Is that the one u said to set to 8 ? 

Yes, that one should be set to 8. 

[Ebony Falcon]

12 minutes ago, RONOTHAN## said:

Yes, that one should be set to 8. 

Ok I have all that punched in and booted 

should I run occt 80 percent sse overnight now ? 

[RONOTHAN##]

Just now, Ebony Falcon said:

Ok I have all that punched in and booted 

should I run occt 80 percent sse overnight now ? 

Not a bad idea. No idea how good OCCT's memory stress test is though, usually I do TestMem5 Anta777 Extreme1. 

[Ebony Falcon]

1 minute ago, RONOTHAN## said:

Not a bad idea. No idea how good OCCT's memory stress test is though, usually I do TestMem5 Anta777 Extreme1. 

Ok well this is where we’re at now which I think is decent with all the settings we discussed 

 

if I play a few games now before stressing over night ( already passed 10 min test) 

it’s not gona break the pc or anything is it ? Worse case blue screen?

[RONOTHAN##]

Just now, Ebony Falcon said:

it’s not gona break the pc or anything is it ? Worse case blue screen?

 

Pretty much, that's the worst thing that could happen. 

[Ebony Falcon]

1 minute ago, RONOTHAN## said:

Pretty much, that's the worst thing that could happen. 

80 percent memory usage ok for test or should I do 100 I don’t really understand 

 

and will this have me competing in gaming perf with something like 5600 ddr5 xmp kits ? 

[RONOTHAN##]

3 minutes ago, Ebony Falcon said:

80 percent memory usage ok for test or should I do 100 I don’t really understand 

 

ideally you'd want 100% though that can cause some weird issues with Windows. 80% should be OK though. 

 

4 minutes ago, Ebony Falcon said:

and will this have me competing in gaming perf with something like 5600 ddr5 xmp kits ? 

Yeah, it should be somewhere in that neighborhood. You'll want to tune some of the subtimings though, since they help a lot with performance. 

[Ebony Falcon]

1 minute ago, RONOTHAN## said:

ideally you'd want 100% though that can cause some weird issues with Windows. 80% should be OK though. 

 

Yeah, it should be somewhere in that neighborhood. You'll want to tune some of the subtimings though, since they help a lot with performance. 

Ok well il test this tonight and make sure it’s all ok and if u don’t mind helping me tomorrow to tune others ? 

[RONOTHAN##]

11 minutes ago, Ebony Falcon said:

Ok well il test this tonight and make sure it’s all ok and if u don’t mind helping me tomorrow to tune others ? 

yeah thats fine