Comprehensive Memory Overclocking Guide

[MageTank]

5 hours ago, steffeeh said:

Slightly off topic, but how do I know if I run single rank and 1DPC?

 

This is my exact RAM kit:

http://www.corsair.com/en-eu/vengeance-lpx-16gb-2x8gb-ddr4-dram-2400mhz-c14-memory-kit-black-cmk16gx4m2a2400c14

 

EDIT: Finally found some basic info on DPC, and I assume you mean by 1DPC that my two RAM sticks are properly running in dual channel mode, instead of single channel per stick?

By DPC, I mean "Dimms Per Channel" as in, you have 1 DIMM in each channel (4 sticks, 1 in each channel, running quad channel). If you had 2DPC, You would be using all 8 sticks of memory.

 

As for how to determine if you are running single or dual rank, use Aida64 or thaiphoon burner to check your ram's SPD information.

[steffeeh]

So I have 1DPC which is nice given that I have Haswell-E, but I also have dual rank memory, which I've understood is not the best solution for my system.

 

I'll probably start by setting voltages to 1.35V vDIMM, 1.15 VCCIO & VCCSA/SystemAgent, and do the usual half DRAM VTTDDR.

I'll then bump up the RAM frequency, probably hitting 2666MHz with proper timings before IMC issues starts to occur.

Then I'll try to hit 2800MHz and up by using generous primary timings just as a temporary failsafe, and then aim to tweak the tertiary _DR-timings (I suppose you loosen them?) to see if the IMC allows me to post with both RAM sticks and hope I don't "lose" one of the sticks while in Windows.

Then if I make it happen I tighten the primary timings to reasonable latency and wait a period of time to check if it's long term stable.

Then after that I've gotten compability out of the way, I can start to experiment with the rest; tightening primary timings, testing secondary and tertiary timings, etc.

 

I don't think I'll fiddle with the tRDWR unless I have to in order to ensure compability, since I don't have much temperature headroom left... I even feel anxious just to bump up the VCCIO and the VCCSA/SystemAgent voltages as it will add more heat from the CPU.

Then of course, if loosening tRDWR timings both prevents the IMC from hating you, and lessens the heat from the CPU (hopefully without giving a negative night and day performance difference), then it could be a good thing to adjust.

 

Sorry for the number of posts and the length of them, but RAM overclocking feels like a pain when you have poor IMC's

[MageTank]

So, a slight update. AMD has released a new AGESA revision, that has unlocked access to secondary and tertiary timings. This may take time for your specific board to receive the update, but it's exciting news nonetheless. https://community.amd.com/community/gaming/blog/2017/05/25/community-update-4-lets-talk-dram

 

With access to these timings, it means we can make ANY IC's work with Ryzen, and should be able to fine-tune the performance to achieve much better latency and bandwidth results. I will be working on the Ryzen part of the guide while I have time off work, and I'll update that as soon as I can. For now, I'll post this AGESA information in the Ryzen half of the thread, and guide people with information regarding specific timings should they decide to tinker with ram manually. 

[DrMikeNZ]

17 minutes ago, MageTank said:

So, a slight update. AMD has released a new AGESA revision, that has unlocked access to secondary and tertiary timings. This may take time for your specific board to receive the update, but it's exciting news nonetheless. https://community.amd.com/community/gaming/blog/2017/05/25/community-update-4-lets-talk-dram

 

With access to these timings, it means we can make ANY IC's work with Ryzen, and should be able to fine-tune the performance to achieve much better latency and bandwidth results. I will be working on the Ryzen part of the guide while I have time off work, and I'll update that as soon as I can. For now, I'll post this AGESA information in the Ryzen half of the thread, and guide people with information regarding specific timings should they decide to tinker with ram manually. 

A beta BIOS with the AGESA update has been available for my Ryzen motherboard for a few days, and I have some 3200CL14 RAM lying around which I plan to put into it. Just need a window of availability in my workload to test and play.

[LooneyJuice]

Just chiming in here as well to say thanks for the guide, very much appreciated! 

[Jade]

Looking forward to the Ryzen section when you get the information you need. Thanks for the post!

 

Also, would it still be valid to test stability with the same steps in the Intel post?

[MageTank]

3 minutes ago, Jade said:

Looking forward to the Ryzen section when you get the information you need. Thanks for the post!

 

Also, would it still be valid to test stability with the same steps in the Intel post?

Yes, I've been currently stress testing a 3466 C14 setup on Ryzen with my previous stress test methodology and it seems to be working exactly the same. I will say this. For as much of a headache Ryzen is when it comes to general memory overclocking, once you find something that posts, it's pretty difficult to make it crash outright. The hard part is finding something the board/IMC tolerates in order to post in the first place, lol. 

[Jade]

5 hours ago, MageTank said:

Yes, I've been currently stress testing a 3466 C14 setup on Ryzen with my previous stress test methodology and it seems to be working exactly the same. I will say this. For as much of a headache Ryzen is when it comes to general memory overclocking, once you find something that posts, it's pretty difficult to make it crash outright. The hard part is finding something the board/IMC tolerates in order to post in the first place, lol. 

Awesome, I'll put my set through the tests over the next few days, probably.... after what's down below. 

 

I've been having unstable behavior when my RAM (G.Skill FlareX 2x8GB DDR4-3200 CL14) is at 3200MHz (however, it does consistently post & boot into Windows, with & without boot training) -- applications will crash -- but it passes memtest86 on default settings (4 pass, all tests) with no errors and validates on CPU-Z happily. Seems to play nice at 2933MHz, though... any tips on how I might fix those problems? Exact build's below, but if you need more information feel free to ask. 

 

Exact specs:

PCPartPicker part list / Price breakdown by merchant

CPU: AMD - Ryzen 7 1700 3.0GHz 8-Core Processor  (Purchased For $310.00) 
CPU Cooler: NZXT - Kraken X61 106.1 CFM Liquid CPU Cooler  (Purchased For $120.00) 
Thermal Compound: Arctic Silver - 5 High-Density Polysynthetic Silver 3.5g Thermal Paste  (Purchased For $7.79) 
Motherboard: ASRock - X370 KILLER SLI/ac ATX AM4 Motherboard  (Purchased For $145.00) 
Memory: G.Skill - Flare X Series 16GB (2 x 8GB) DDR4-3200 Memory  (Purchased For $184.00) 
Storage: Kingston - Predator 240GB PCI-E Solid State Drive  (Purchased For $128.00) 
Storage: Crucial - MX100 256GB 2.5" Solid State Drive  (Purchased For $103.00) 
Storage: Western Digital - Caviar Blue 1TB 3.5" 7200RPM Internal Hard Drive  (Purchased For $54.00) 
Storage: Western Digital - BLACK SERIES 3TB 3.5" 7200RPM Internal Hard Drive  (Purchased For $156.00) 
Video Card: Asus - GeForce GTX 1070 8GB Video Card  (Purchased For $425.00) 
Case: NZXT - S340 (Black) ATX Mid Tower Case  (Purchased For $62.00) 
Power Supply: SeaSonic - 760W 80+ Platinum Certified Fully-Modular ATX Power Supply  (Purchased For $110.00) 
Operating System: Microsoft - Windows 10 Pro OEM 64-bit  (Purchased For $15.00) 
Case Fan: Noctua - NF-P14s redux-1500 PWM 78.7 CFM  140mm Fan  (Purchased For $14.00) 
Case Fan: Noctua - NF-P14s redux-1500 PWM 78.7 CFM  140mm Fan  (Purchased For $14.00) 
Case Fan: Noctua - NF-P14s redux-1500 PWM 78.7 CFM  140mm Fan  (Purchased For $14.00) 
Case Fan: Noctua - NF-F12 industrialPPC-2000 PWM 71.7 CFM  120mm Fan  (Purchased For $24.00) 
Monitor: Acer - H236HLbid 23.0" 1920x1080 60Hz Monitor  (Purchased For $100.00) 
Monitor: Acer - H236HLbid 23.0" 1920x1080 60Hz Monitor  (Purchased For $168.00) 
Monitor: BenQ - XL2411Z 24.0" 1920x1080 144Hz Monitor  (Purchased For $125.00) 
Total: $2278.79
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2017-06-28 01:06 EDT-0400

[MageTank]

19 minutes ago, Jade said:

Awesome, I'll put my set through the tests over the next few days, probably.... after what's down below. 

 

I've been having unstable behavior when my RAM (G.Skill FlareX 2x8GB DDR4-3200 CL14) is at 3200MHz (however, it does consistently post & boot into Windows, with & without boot training) -- applications will crash -- but it passes memtest86 on default settings (4 pass, all tests) with no errors and validates on CPU-Z happily. Seems to play nice at 2933MHz, though... any tips on how I might fix those problems? Exact build's below, but if you need more information feel free to ask. 

 

Exact specs:

PCPartPicker part list / Price breakdown by merchant

CPU: AMD - Ryzen 7 1700 3.0GHz 8-Core Processor  (Purchased For $310.00) 
CPU Cooler: NZXT - Kraken X61 106.1 CFM Liquid CPU Cooler  (Purchased For $120.00) 
Thermal Compound: Arctic Silver - 5 High-Density Polysynthetic Silver 3.5g Thermal Paste  (Purchased For $7.79) 
Motherboard: ASRock - X370 KILLER SLI/ac ATX AM4 Motherboard  (Purchased For $145.00) 
Memory: G.Skill - Flare X Series 16GB (2 x 8GB) DDR4-3200 Memory  (Purchased For $184.00) 
Storage: Kingston - Predator 240GB PCI-E Solid State Drive  (Purchased For $128.00) 
Storage: Crucial - MX100 256GB 2.5" Solid State Drive  (Purchased For $103.00) 
Storage: Western Digital - Caviar Blue 1TB 3.5" 7200RPM Internal Hard Drive  (Purchased For $54.00) 
Storage: Western Digital - BLACK SERIES 3TB 3.5" 7200RPM Internal Hard Drive  (Purchased For $156.00) 
Video Card: Asus - GeForce GTX 1070 8GB Video Card  (Purchased For $425.00) 
Case: NZXT - S340 (Black) ATX Mid Tower Case  (Purchased For $62.00) 
Power Supply: SeaSonic - 760W 80+ Platinum Certified Fully-Modular ATX Power Supply  (Purchased For $110.00) 
Operating System: Microsoft - Windows 10 Pro OEM 64-bit  (Purchased For $15.00) 
Case Fan: Noctua - NF-P14s redux-1500 PWM 78.7 CFM  140mm Fan  (Purchased For $14.00) 
Case Fan: Noctua - NF-P14s redux-1500 PWM 78.7 CFM  140mm Fan  (Purchased For $14.00) 
Case Fan: Noctua - NF-P14s redux-1500 PWM 78.7 CFM  140mm Fan  (Purchased For $14.00) 
Case Fan: Noctua - NF-F12 industrialPPC-2000 PWM 71.7 CFM  120mm Fan  (Purchased For $24.00) 
Monitor: Acer - H236HLbid 23.0" 1920x1080 60Hz Monitor  (Purchased For $100.00) 
Monitor: Acer - H236HLbid 23.0" 1920x1080 60Hz Monitor  (Purchased For $168.00) 
Monitor: BenQ - XL2411Z 24.0" 1920x1080 144Hz Monitor  (Purchased For $125.00) 
Total: $2278.79
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2017-06-28 01:06 EDT-0400

Well, memtest86 isn't exactly a stress test. It's mostly used to validate whether or not the IMC itself is okay with the timings. If you want to actually stress the ram and I/O lanes, you need to use Prime95 512k-4096k @ 75% ram capacity. As for random bouts of instability, please be sure to check tFAW. By default, Ryzen boards have been incorrectly setting it's value. At minimum, it must be 4x tRRD_S. Meaning, if tRRD_S is 6, tFAW cannot go lower than 24 or else you will risk instability. I've noticed some boards breaking this rule, and it scares me that they even allow it. For extra stability, you can loosen tFAW up beyond that rule, to x8 the value of tRRD_S, but I don't really think it's necessary. Just make sure it's adhering to the tRRD_S x 4 rule and you should be fine.

 

You can also check your ODT settings. I personally use a value of 96, but your mileage may vary. 

[Jade]

14 hours ago, MageTank said:

snip

I appreciate the effort you put into your posts, can't thank you enough for them. I ran the Prime setup you provided in the OP at 3.8GHz & 1.36v / 2933MHz CL14 1.350v over night and encountered no errors (woohoo!) Looks like I might have to just stick with that though...

 

Upping the RAM frequency to 3200MHz fails consistently (usually within about 30 minutes) even with SoC voltage at 1.2v (runs at 1.1v at 2933MHz and below) and CPU at stock speeds & voltage. 

 

As for tRRD_S & tFAW, they were set to 6 and 39, respectively, before I even looked at them... Just over the middle of the 4-8x you suggested. Haven't tinkered with ODT as I haven't really found any information on it except in a tomshardware post which only suggests 40-60, but has no explanation as to what logic to follow for using a lower or higher value.

Edited June 28, 2017 by Jade
Grammar

[MageTank]

27 minutes ago, Jade said:

I appreciate the effort you put into your posts, can't thank you enough for them. I ran the Prime setup you provided in the OP at 3.8GHz & 1.36v / 2933MHz CL14 1.350v over night and encountered no errors (woohoo!) Looks like I might have to just stick with that though...

 

Upping the RAM frequency to 3200MHz fails consistently (usually within about 30 minutes) even with SoC voltage at 1.2v (runs at 1.1v at 2933MHz and below) and CPU at stock speeds & voltage. 

 

As for tRRD_S & tFAW, they were set to 6 and 39, respectively, before I even looked at them... Just over the middle of the 4-8x you suggested. Haven't tinkered with ODT as I haven't really found any information on it except in a tomshardware post which only suggests 40-60, but has no explanation as to what logic to follow for using a lower or higher value.

https://community.amd.com/community/gaming/blog/2017/05/25/community-update-4-lets-talk-dram

 

Quote

ProcODT(CPU On-Die Termination):     A resistance value, in ohms, that determines how a completed memory signal is terminated. Higher values can help stabilize higher data rates. Values in the range of 60-96 can prove helpful.

It's highly dependent on your ram and CPU as to which value works best, but on your higher-clocked kits, 96omhs seems to be better all around. There are also threads on other forums that seem to agree with this, and my kit certainly proved to enjoy 96 better than whatever was set at auto. I have not tinkered enough with the other settings (aside from 60, which wasn't exactly stable) but I'll get around to it once I have the time. Impact on performance seems negligible.

[WereCat]

Hi @MageTank I need a bit of advice.

 

I had Crucial Ballistix Sport 2x8GB DDR3 kit 1600MHz CL9-9-9-24-CR1 which I was able to successfully OC to 2200MHz CL10-11-11-30-CR2.

I was also able to make it boot at 2400MHz but not stable and unable to make it boot at 2600MHz+.

 

Now I got Patriot Viper 3 2x8GB DDR3 kit 2400MHz CL10-12-12-30-CR2 and I am not at all able to make it boot at 2600MHz+ even if I loosen the timing by a lot.

I increased all the IMC voltages to not make it a factor and also increased the DRAM voltages from 1.65V to 1.75V but I still don't know why it doesn't boot.

I am using 4770k with the Gigabyte Z87X-UD4H motherboard and I tried two different BIOS versions. I was using the F10betra which has issues with BCLK, any change to BCLK would crash the system, now I am using F9 version which lacks some of the memory OC settings like Boot Mode for example (not timings).

 

If I increase BCLK to +1MHz the RAM goes higher, +2MHz still works but I am starting to have some SATA issues at that point.

I tried 1.25 divider but I have not even booted with it at 2400MHz for some reason.

 

What happens is that PC looks like it is going to retrain the the RAM and after it is about to boot I get BIOS screen saying there was boot issue and it resets all the settings. It is very annoying since my BIOS is unable to load any saved settings for some reason...

 

The sub timings on BIOS version F10beta were way more tight than on the F9 version where if I enter the ones from F10beta manually it won't boot.

 

It looks to me that I am fighting more with my motherboard than with the actual RAM overclocking.

 

There I have written everything down from the XMP profile at 2400MHz.

Everything behind the slash "/" is the settings on the F9 BIOS, everything else is from the F10b BIOS.

Channel B has only RTL differences from the Channel A.

 

[MageTank]

5 hours ago, WereCat said:

Hi @MageTank I need a bit of advice.

 

I had Crucial Ballistix Sport 2x8GB DDR3 kit 1600MHz CL9-9-9-24-CR1 which I was able to successfully OC to 2200MHz CL10-11-11-30-CR2.

I was also able to make it boot at 2400MHz but not stable and unable to make it boot at 2600MHz+.

 

Now I got Patriot Viper 3 2x8GB DDR3 kit 2400MHz CL10-12-12-30-CR2 and I am not at all able to make it boot at 2600MHz+ even if I loosen the timing by a lot.

I increased all the IMC voltages to not make it a factor and also increased the DRAM voltages from 1.65V to 1.75V but I still don't know why it doesn't boot.

I am using 4770k with the Gigabyte Z87X-UD4H motherboard and I tried two different BIOS versions. I was using the F10betra which has issues with BCLK, any change to BCLK would crash the system, now I am using F9 version which lacks some of the memory OC settings like Boot Mode for example (not timings).

 

If I increase BCLK to +1MHz the RAM goes higher, +2MHz still works but I am starting to have some SATA issues at that point.

I tried 1.25 divider but I have not even booted with it at 2400MHz for some reason.

 

What happens is that PC looks like it is going to retrain the the RAM and after it is about to boot I get BIOS screen saying there was boot issue and it resets all the settings. It is very annoying since my BIOS is unable to load any saved settings for some reason...

 

The sub timings on BIOS version F10beta were way more tight than on the F9 version where if I enter the ones from F10beta manually it won't boot.

 

It looks to me that I am fighting more with my motherboard than with the actual RAM overclocking.

 

There I have written everything down from the XMP profile at 2400MHz.

Everything behind the slash "/" is the settings on the F9 BIOS, everything else is from the F10b BIOS.

Channel B has only RTL differences from the Channel A.

 

Interesting, you seem to be having the exact same problem I had on Haswell. For me, 2600-2800 was a complete dead zone, and nothing would post within that zone. I was able to get 2933 to post, but the tertiary timings were so loose that it completely defeated the purpose to use it,as both it's raw bandwidth and latency were weaker. Judging by your tertiary timings, I can say they look quite tight. Did you do this by hand, or did the board train them this tight? Running tight tertiary timings while also trying to increase frequency means you are likely going to get nowhere, as the IMC is already tapped out on the tertiary timings alone.

 

If it were me, I would settle with 2133 or 2400, and focus on finishing the rest of the timings. You have copious amounts of bandwidth, but you can always use lower latency. Tighten tRFC up a little more, and experiment with higher tREFI (assuming your case has decent airflow) for lower latency. If you still wish to try for 2600+, then loosen your tWRRD timings a little more, and hope for the best. Also, your tFAW needs to be 28 in that configuration. With tRRD at 7, tFAW must be at minimum 4x that. The window will not stay active long enough in your current configuration, and may result in instability or corruption. 

[WereCat]

4 minutes ago, MageTank said:

Interesting, you seem to be having the exact same problem I had on Haswell. For me, 2600-2800 was a complete dead zone, and nothing would post within that zone. I was able to get 2933 to post, but the tertiary timings were so loose that it completely defeated the purpose to use it,as both it's raw bandwidth and latency were weaker. Judging by your tertiary timings, I can say they look quite tight. Did you do this by hand, or did the board train them this tight? Running tight tertiary timings while also trying to increase frequency means you are likely going to get nowhere, as the IMC is already tapped out on the tertiary timings alone.

 

If it were me, I would settle with 2133 or 2400, and focus on finishing the rest of the timings. You have copious amounts of bandwidth, but you can always use lower latency. Tighten tRFC up a little more, and experiment with higher tREFI (assuming your case has decent airflow) for lower latency. If you still wish to try for 2600+, then loosen your tWRRD timings a little more, and hope for the best. Also, your tFAW needs to be 28 in that configuration. With tRRD at 7, tFAW must be at minimum 4x that. The window will not stay active long enough in your current configuration, and may result in instability or corruption. 

Thanks.

This is what the board set automatically after I enabled XMP, that is not manual configuration. I was just trying to work my way up from there which did not work for me and I haven't yet got to try and tighten the timings on 2400MHz which I will definitely do later next week as I am expecting to receive my delid kit and two other fans for my case because my airflow right now is not great.

I manually tried only 2600-2800, I may give a shot to what you said about the 2933MHz and see if that would work. I also read somewhere that maybe disabling DUAL BIOS may help but I have not tested that yet.

[MageTank]

Just now, WereCat said:

Thanks.

This is what the board set automatically after I enabled XMP, that is not manual configuration. I was just trying to work my way up from there which did not work for me and I haven't yet got to try and tighten the timings on 2400MHz which I will definitely do later next week as I am expecting to receive my delid kit and two other fans for my case because my airflow right now is not great.

I manually tried only 2600-2800, I may give a shot to what you said about the 2933MHz and see if that would work. I also read somewhere that maybe disabling DUAL BIOS may help but I have not tested that yet.

If the board is aggressively training the tertiary timings that tight, that may be what's preventing your higher clock speeds. You can go through and slightly loosen each one by 1-2T and see if that gives you enough headroom to push for higher clocks. Either way, I couldn't hit 2600-2800 on my ASRock board, and my friend never could on his EVGA classified either. Some combination of memory kits and motherboards (and i guess to some extent, even the IMC) simply refuse to post in that zone. 

[WereCat]

2 hours ago, MageTank said:

If the board is aggressively training the tertiary timings that tight, that may be what's preventing your higher clock speeds. You can go through and slightly loosen each one by 1-2T and see if that gives you enough headroom to push for higher clocks. Either way, I couldn't hit 2600-2800 on my ASRock board, and my friend never could on his EVGA classified either. Some combination of memory kits and motherboards (and i guess to some extent, even the IMC) simply refuse to post in that zone. 

So I tried this:

 

I copied the XMP settings manually into the BIOS and made sure it boots (increased the tFAW to 28 as you suggested).

Then I was playing and loosening RTL and IO as much as I could until I couldn't loosen it anymore.

 

I got to 53/54 RTL and 7/8 IO. I couldn't loosen RTL anymore because after every step I had to loosen IO as well and 8 is max. (also I made sure it boots)

 

Then I tried to boot 2600MHz+ but with no success.

After, I went and loosened up the timings and tried 2600+MHz again... nope.

 

So I manually set the BCLK to 100.00MHz and set the divider to x1.25 and I tried 2500MHz... it booted, yay!

 

After, I worked down and tightened the RTL down to 49/50 IO 4/4.

The AIDA 64 benchmark results were quite bad though. Latency improved by 2ns but Read/Copy suffered 5000MB/s loss and Write improved by 2000MB/s.

Not really worth it. Besides it was not passing Memtest86, I still had to loosen the CL to 11 which made things even worse.

 

Still, I need more time to play around with this but I guess that trying to tighten the 2400MHz is my best option. (would be sweet if I manage to tighten the RTL even more).

 

[WereCat]

@MageTank

 

So after further testing...

 

XMP profile is a total garbage.

It is not passing the Memtest86 at all, it starts to spit errors 5 minutes into testing.

 

Changing tFAW from 26 to 28 as you suggested makes it complete PASS1 without errors but it starts spitting out errors in TEST 4,5,7 in PASS2.

 

So I was playing around (still will be for some time) and found out that I can run CR1 instead of CR2 at 2400MHz and complete Memtest86 without errors by increasing DRAM voltage to 1.7V from 1.65V (I assume that this is still safe voltage for 24/7 usage, right?).

 

However, I don't understand this:

 

CAS Latency - 10

tRCD -12

tRP -12

tRAS -30

_____________________________________________________________________________

tRRD - 7 / 8                                                                                                 

tFAW - 28 / 32

CR - 1 boots with no errors in Memtest86 / 1 doesen't boot at all  with loosened tRRD and tFAW, why is that?

 

Everything else set to AUTO so far (trying to go one timing at a time systematically and figure out where and what could cause issues).

 

EDIT:

Now I cannot boot with the same settings at all, whaaat?

EDIT2:

Looks like the retraining fails sometimes after I change other timings and I have to set CR2, boot, go back to BIOS and set CR1 again for it to boot.

 

 

[TheRandomness]

I'm quite happy, following this guide once again I managed to get 8-10-10-28 1T stable, and get these results:

So thanks, I guess  

[MageTank]

4 hours ago, TheRandomness said:

I'm quite happy, following this guide once again I managed to get 8-10-10-28 1T stable, and get these results:

So thanks, I guess  

I know people are going to see this, and scoff at the low bandwidth of 16,559MB/s, but allow me to put this into perspective for those that do not understand. At 2133, you have a peak theoretical memory bandwidth of 17,064MB/s in single channel. You are achieving 97% of that bandwidth on your read speed. If that alone does not impress you, allow me to point out that read speed is ALWAYS slower than write, in regards to how Intel's IMC works, so your write efficiency is automatically higher. Meaning it can be 98/99% efficient in writes. This is absolutely impressive. The only thing that's left, is if you have access to tRFC and tREFI, to tweak those to bring that latency down to about 45ns. Also, another stick of ram to activate dual channel would be pretty neat as well, but that's totally optional.

 

Overall, a very impressive overclock. 2133 at C8 has an absolute latency of 7.5ns, which is just as low as my 3733 at C14. 

[TheRandomness]

43 minutes ago, MageTank said:

I know people are going to see this, and scoff at the low bandwidth of 16,559MB/s, but allow me to put this into perspective for those that do not understand. At 2133, you have a peak theoretical memory bandwidth of 17,064MB/s in single channel. You are achieving 97% of that bandwidth on your read speed. If that alone does not impress you, allow me to point out that read speed is ALWAYS slower than write, in regards to how Intel's IMC works, so your write efficiency is automatically higher. Meaning it can be 98/99% efficient in writes. This is absolutely impressive. The only thing that's left, is if you have access to tRFC and tREFI, to tweak those to bring that latency down to about 45ns. Also, another stick of ram to activate dual channel would be pretty neat as well, but that's totally optional.

 

Overall, a very impressive overclock. 2133 at C8 has an absolute latency of 7.5ns, which is just as low as my 3733 at C14. 

I mean, following on from the PMs we had a while ago, I'm sure I'd be able to get that latency even lower, as I have access to every timing available. Also, I do think the slot closest to my CPU might be dead, so dual channel may not be an option. Though, I might attempt 8-9-9-28, or 7-10-10-28. Hey, it's the only overclocking in the PC world where lower numbers are better, right?

[MageTank]

Just now, TheRandomness said:

I mean, following on from the PMs we had a while ago, I'm sure I'd be able to get that latency even lower, as I have access to every timing available. Also, I do think the slot closest to my CPU might be dead, so dual channel may not be an option. Though, I might attempt 8-9-9-28, or 7-10-10-28. Hey, it's the only overclocking in the PC world where lower numbers are better, right?

For most timings, that is absolutely correct. I am thinking about investing in the Z270 Apex, so I can get 4000 C15-15-15-30-2 running 24/7. Would put me at exactly 7.5ns flat (no remainders) absolute NS, and my bandwidth would be about 60GB/s across the board, with 35ns latency. Sadly, talking myself into investing $270 on a motherboard while still paying for my own education is rather difficult at the moment. A matter of wants vs needs, lol. 

[Carclis]

@MageTank

Thank you so much for making a guide that gives beginners something to actually work with. On that note this guide has helped me to tweak all my primary and secondary timings to be as tight as possible, but being a beginner I have absolutely no idea what I'm looking at when it comes to the IOL/RTL timings and as such have left them at their default XMP values. My kit is a Corsair 3200Mhz 16-18-18-36 one which I've managed to bump up to 3333Mhz 14-18-18-32. Would you have any idea where to go from here?

 

-Sorry for the craptastic photo quality 

[MageTank]

32 minutes ago, Carclis said:

@MageTank

Thank you so much for making a guide that gives beginners something to actually work with. On that note this guide has helped me to tweak all my primary and secondary timings to be as tight as possible, but being a beginner I have absolutely no idea what I'm looking at when it comes to the IOL/RTL timings and as such have left them at their default XMP values. My kit is a Corsair 3200Mhz 16-18-18-36 one which I've managed to bump up to 3333Mhz 14-18-18-32. Would you have any idea where to go from here?

 

 

 

-Sorry for the craptastic photo quality 

That is some very impressive latency. Your tertiary timings need work, as there is still about 5GB/s bandwidth left for you to obtain on your write bandwidth, and maybe another 3-4GB/s on read. As for RTL/IO-L, they are tricky. It's impossible for me to give you a value to input, as it's completely random based off your IMC, motherboard, and the specific IC's you are using. It's seriously the most brutal trial and error one can attempt when memory overclocking, and all I can say is, wire your reset button to clear CMOS if you intend to take that kind of challenge on.

 

If it were me, I would focus entirely on tertiary timings, and leave RTL/IO-L alone. You can still get a pretty significant difference in bandwidth from tertiary timings alone, while also improving latency a little more. As it sits now, your latency is already lower than 95% of all XMP kits on the market (even the 4266mhz kits) so congratulations on that. You also have plenty of memory bandwidth to go around. Anything beyond this point, is simply for bragging rights. 

[Carclis]

12 hours ago, MageTank said:

That is some very impressive latency. Your tertiary timings need work, as there is still about 5GB/s bandwidth left for you to obtain on your write bandwidth, and maybe another 3-4GB/s on read. As for RTL/IO-L, they are tricky. It's impossible for me to give you a value to input, as it's completely random based off your IMC, motherboard, and the specific IC's you are using. It's seriously the most brutal trial and error one can attempt when memory overclocking, and all I can say is, wire your reset button to clear CMOS if you intend to take that kind of challenge on.

 

If it were me, I would focus entirely on tertiary timings, and leave RTL/IO-L alone. You can still get a pretty significant difference in bandwidth from tertiary timings alone, while also improving latency a little more. As it sits now, your latency is already lower than 95% of all XMP kits on the market (even the 4266mhz kits) so congratulations on that. You also have plenty of memory bandwidth to go around. Anything beyond this point, is simply for bragging rights. 

I was afraid that might be the case. I guess I'll have to play around with the other tertiary timings for now until I actually get some time to dedicate to tweaking the RTL/IOL.

[Carclis]

22 hours ago, MageTank said:

If it were me, I would focus entirely on tertiary timings, and leave RTL/IO-L alone. You can still get a pretty significant difference in bandwidth from tertiary timings alone, while also improving latency a little more.

Did you ever experience any strange behaviour when playing around with the tRDRD timings? I can't for the life of me get them to settle on the values I input. For instance, I have set:

tRDRDSG: 6
tRDRDDG: 4
tRDRDDR: 6 (7)
tRDRDDD: 8
 

Despite these settings being input my tRDRDDR seems to always be equal to the value I input +1 unless I set it to 9 or greater, in which case it actually does what I tell it to. I also had trouble with tRDRDDD which only seems to like even numbers, but that at least makes sense and is easy enough to work with.

Or could this perhaps just be a terrible bios from MSI?