Memory Timings

[Johnn722]

Hello so I just recently brought my ram timings down from 16-16-16-37, to 13-13-13-31 I'm wondering if this is good timing set up? This is all I was able to get out of my ram kit. The command rate is 2T and the trfc is 299. The fsb:dram is 1:23

I haven't applied any voltage either its running at stock 1.2, Also if anyone knows of any other good memory stability tests other than aida 64 please let me know.

Thanks!

[Imakuni]

Aren't you forgetting something? Like, you know.. the RAM speed? Full hardware specs would also help.

 

I hope you get those posted by the time @MageTank comes to aid you.

[Johnn722]

The ram is running at 2300mhz and 32gb of crucial ballistics ddr4

[MageTank]

1 hour ago, Johnn722 said:

The ram is running at 2300mhz and 32gb of crucial ballistics ddr4

I am curious. You are going for tighter timings over raw bandwidth, may I ask why? Are you on a platform that does not allow the clock speed to be changed? Or are you simply trying to see how low you can get the timings?

 

Either way, you test memory stability by using Prime95. Now before you freak out about it killing your CPU, it won't. Just do a custom test, 512k FFT small, 4096k FFT large. This range will stress your memory, the IO lanes, and the IMC itself. Set the intervals to 15 minutes. Stability is a hard thing to prove, as you are only stable until you are unstable. No specific amount of time will say "Hey, I am 100% stable". I generally run this test for 8 hours, and I run memtest86 in my sleep (Specifically test 6 for Block Move) as I mostly deal with tertiary timings (IMC stress determines stability in this area).

 

If you can overclock the ram the other direction (higher bandwidth) then I advise you go that route. Higher clock speeds will lower your latency as a result, far better than just tighter primary timings. Let me know if you have any questions.

[Johnn722]

Hello thanks for replying I want to tighten them up for performance. The 32gb of ram I have won't run at full 2400mhz so I wanted to see if I could squeeze any more performance out of the ram.

[MageTank]

@Johnn722

 

You can, but you will see far more performance gains from the tertiary timings. When you say they won't run at 2400, have you tried loosening  the primary timings? I would manually dial in the voltage at 1.35v (this voltage is perfectly safe by JEDEC standards, and kits are even sold at this speed) and then leave your primary timings at default. Slowly change the ram stepping one increment at a time until you fail to post. Go back to the last viable option, then do a quick memtest86 run. If it fails any of the tests, determine why it failed. For example: failing test 6 likely means something is up with the IMC not liking the configuration. Solution is to loosen tertiary timings to relieve stress, or add more VCCSA/VCCIO voltage (do not exceed 1.3v on these two values if using Skylake)

 

Keep going until you find a speed you like. Once you find a speed you like, tweak it even more by adjusting tertiary timings. This, I cannot help with, as your CPU IMC, quality of ram, and even the trace topology on your board will impact what can be done with these timings. Just take note: lower is not always better with these timings. Some ram IC's actually perform better with specific tertiary timings loose, while others prefer to have them tight. Simply test using Aida64's cache and memory test after adjusting each timing to see if it had a positive or negative impact on bandwidth/latency. I know, this sounds very time consuming, but it's the best proven method to do it right thus far. 

[done12many2]

4 hours ago, Johnn722 said:

Also if anyone knows of any other good memory stability tests other than aida 64 please let me know.

 

HCI MemTest for Windows.  Open an instance for each thread that your CPU has.  Determine what 93-95% of your total memory would be and divide that by the total number of threads.  Input that number into each instance of MemTest and let it run until around 400-500%.  If you don't get any errors, your new memory timings are solid.  AIDA64 memory stress test doesn't really expose memory error very well, if at all.

[Johnn722]

Currently I am running a skylake 6700k at 4.4ghz the ram is crucial ddr4 and running at 2300mhz. The timings started at 16 16 16 37 I'm at 13 13 13 31. Are you suggesting that I make the timings above 16 16 16 37 and up the voltage no more than 1.3?

[MageTank]

1 hour ago, Johnn722 said:

Currently I am running a skylake 6700k at 4.4ghz the ram is crucial ddr4 and running at 2300mhz. The timings started at 16 16 16 37 I'm at 13 13 13 31. Are you suggesting that I make the timings above 16 16 16 37 and up the voltage no more than 1.3?

Change VDIMM to 1.35v. Do not adjust VCCIO and VCCSA until you run into instability. Change your timings to 16-16-16-37, but leave them there. Once the timings have been manually entered at 16-16-16-37, start adjusting the clock speed of your memory. If it started off at 2133, bring it to 2400. Then 2666, then 2800, then 3000. If it fails to post at any of these values, revert to the last working configuration, and run memtest86 for a full pass. If you get no errors, then you can move on to Prime95 with the settings i listed above.

 

Once you reach that moment, THEN you optimize primary and tertiary timings for max speeds. Let me know if you have any questions.

[Johnn722]

So I again could only get the ram to work at 2300mhz instead of 2400mhz, anything past that it won't post. I did apply the 1.35v to the ram and it didnt help. Passed the test just fine for 2300. Is anything else I could try to get it too 2400mhz? 

 

[MageTank]

2 hours ago, Johnn722 said:

So I again could only get the ram to work at 2300mhz instead of 2400mhz, anything past that it won't post. I did apply the 1.35v to the ram and it didnt help. Passed the test just fine for 2300. Is anything else I could try to get it too 2400mhz? 

 

You changed the VDIMM to 1.35v, right? What motherboard are you using? I have never encountered a kit that could not do 2400mhz on 1.35v. In fact, every DDR4 kit i have used, made it to 3000mhz (Even Micron IC's, which is the weakest overclockers, which is what you have in your Crucial Ballistix kit). Something is certainly wrong with this picture. 

[Johnn722]

So I was able to get it to 2371mhz with a with a 4.4 oc. This oc is with stock voltage, I tried the 1.35 again for the 2400 and nothing. Could it be the cpu overclock? The motherboard I'm using is an asus z170 ar.

[Johnn722]

Somehow Its working at 2666mhz instead of 2400mhz I'm not complaining just confused. Using 1.35v

[MageTank]

16 minutes ago, Johnn722 said:

Somehow Its working at 2666mhz instead of 2400mhz I'm not complaining just confused. Using 1.35v

As long as its the VDIMM at 1.35, and not the VCCIO/VCCSA, you are fine. Keep going until it no longer posts, then go back to the previous working setting. If 2666 is as high as you can get, start working on primary timings, then tertiary timings. 

[Johnn722]

So currently I have the ram running at 2700mhz with 15-15-15-35. Running well no problems so far. 

[done12many2]

52 minutes ago, MageTank said:

As long as its the VDIMM at 1.35, and not the VCCIO/VCCSA, you are fine. Keep going until it no longer posts, then go back to the previous working setting. If 2666 is as high as you can get, start working on primary timings, then tertiary timings. 

 

Nice work with the help bud!

[MageTank]

16 minutes ago, Johnn722 said:

So currently I have the ram running at 2700mhz with 15-15-15-35. Running well no problems so far. 

Nicely done. JEDEC standard timings, with a free 567mhz overclock. Not bad at all, especially when you factor in that you are using Micron IC's, the hardest IC's to overclock. Would be pretty fantastic if you can get 2800 C15, as that is a pretty nice stopping point for most consumers to see the best gains out of their hardware (depending on CPU overhead). Anything beyond that is icing on the cake. Just don't push yourself too hard if you cant obtain it. You will see far more gains in bandwidth from the tertiary timings than what you will trying to force raw bandwidth at the same primary timings. It's all a balancing act.