overclock RAM

[Alan1969]

I have the 8 pack RAM 3200

https://www.overclockers.co.uk/team...3200mhz-dual-channel-kit-black-my-08l-tg.html

and I have used that DRAM calculator to tweak the sub timings - don't get much benefit when I used a benchmark before and after (passmark I think to test the overall system).

I would like to reduce my CAS timings however. currently the kit is CL 14-14-14-31

Do I just up the voltage in the bios to being 1.40v and then gradually change the timings to say 13-14-14-31 and then 13-13-14-31 and then 13-13-13-31 then 13-13-13-30 and then 12-13-13-30 etc? running something like AIDA64 between each adjustment?

 

[Alan1969]

update:

rather than reducing my timings and still running at 3200 MHz I keep reading articles about how running at 3600mhz is better for the infinity fabric - obviously you have a trade-off with higher latencies.

eg. the stock xmp for the 8 pack memory are:
3200mhz 14-14-14-31 for https://www.overclockers.co.uk/team...3200mhz-dual-channel-kit-black-my-08l-tg.html
and for
3600mhz 16-16-16-36 for https://www.overclockers.co.uk/team...3600mhz-dual-channel-kit-black-my-08q-tg.html

anyway I was thinking:
1) is 3600mhz 16-16-16-36 better than 3200mhz 14-14-14-31?
2) can simply change my 3200mhz 14-14-14-31 kit settings and run at 3600mhz 16-16-16-36?
3) can I overclock my kit using either of these timings?

the 3200mhz memory kit defaults to either 2133 or 2400mhz at 1.20v, but when xmp is activated it uses 1.35v. the kit has been quoted as being tested up to 1.40v (but can you run more say 1.45v?).

answers:
1 & 2) I tried just changing the timings in the bios to 3600mhz 16-16-16-36 when at 1.35V and it didn't boot! perhaps the 3200mhz kit is a bit different to the 3600mhz kit and my 3200mhz kit needs more than 1.35v to achieve the 3600mhz 16-16-16-36 timings. Hence I didn't not test the theory in question 1)
3) I overclocked my 3200mhz 14-14-14-31 kit by increasing the DRAM voltage in the bios from 1.35v to 1.40v and then gradually increasing the frequency and running the passmark test (which tests cpu and gpu & disk performance as well as memory) to get an idea of what effect it was having on my system and to make sure that the entire system was stable. it took about 9 iterative attempts (3200, 3266, 3333, 3400, 3466, 3533, 3600mhz etc) to get my frequency up from 3200 to 3600mhz.

so at the moment I am running this kit https://www.overclockers.co.uk/team...3200mhz-dual-channel-kit-black-my-08l-tg.html at 3600mhz at timings of 14-14-14-31 at 1.40v and with the improved sub-timings from the Ryzen DRAM calculator and it seems to have a small beneficial effect on my passmark benchmarks.

What I will try tonight is to reduce the voltage gradually down from 1.40v.


my questions for you guys is who has the 8 pack 3200mhz kit and how have you overclocked it? have you achieved more than what I have got up to (note that I didn't not get to a point above 3600mhz where it failed yet)?
Also what is the max voltage that I should put through the 8 pack memory kit? (can I go to 1.45v)

 

[Joxeon]

I have the 8 pack 3200/14 running at 3800 16 16 16 32 1.42v in a b450 tomahawk max + r5 3600 with fclk 1-1 @1900 though this is CPU dependant. I used the fast times from ryzen dram calc but with gear down mode enabled and some further timing trimming. I believe up to 1.45 is fine for daily use on this ram and even a bit more with some airflow over the dimms.

 

[Joxeon]

delete.

 

[Alan1969]

Delete.

great thanks. I may try pushing mine higher tonight like yours. How do you know that Fclk is 1-1 when running the memory at 3800 MHz?

 

[Joxeon]

great thanks. I may try pushing mine higher tonight like yours. How do you know that Fclk is 1-1 when running the memory at 3800 MHz?

Over 1800 you have to set the fclk manually, basically wants to be half the speed of your ram speed although stability over 1800 is not a given so your best bet is to keep ram low/stock and push the fclk as high as it will go while stress testing, I found aida 64 memory and cashe test would just crash out when it wasn't stable, then once you know your highest stable fclk you can work on ram then as if you just go and set both to 3800 + 1900 and it's not stable then you won't know if it's the ram or infinity fabric or both that needs adjusting.

Ryzen master shows if the fclk is running 1-1 and also shows ram and fclk speeds.

 

[Alan1969]

Over 1800 you have to set the fclk manually, basically wants to be half the speed of your ram speed although stability over 1800 is not a given so your best bet is to keep ram low/stock and push the fclk as high as it will go while stress testing, I found aida 64 memory and cashe test would just crash out when it wasn't stable, then once you know your highest stable fclk you can work on ram then as if you just go and set both to 3800 + 1900 and it's not stable then you won't know if it's the ram or infinity fabric or both that needs adjusting.

Ryzen master shows if the fclk is running 1-1 and also shows ram and fclk speeds.

I haven't used ryzen master much - I tried using it a month ago but was not getting anywhere with my ryzen 5 3600.

 

[Joxeon]

I haven't used ryzen master much - I tried using it a month ago but was not getting anywhere with my ryzen 5 3600.

I don't use it for any overclocking, just use it for monitoring when I'm overclocking ram as it shows,all the timings. I make all the changes in the bios.

 

[Awahwah]

I could not get mine to run 14-14-14-31 @ 3600, had to loosen to 15-15-15-31 with sub timings. I went up to 1.5v to test but they did not want to know. @ 15's I can run 1.4v

 

[Alan1969]

I could not get mine to run 14-14-14-31 @ 3600, had to loosen to 15-15-15-31 with sub timings. I went up to 1.5v to test but they did not want to know. @ 15's I can run 1.4v

thanks for this.

As mentioned I kept running the passmark benchmarks and had no problems (in any of the tests). However last night in real world gaming on world of warships I did get a blue screen of death and then later a game crash - which is unusual - I was running the 8 pack 3200mhz at 14-14-14-31 @ 3600 at 1.40v (and the sub timings had been left as the Ryzen DRAM calculator settings for 3200 at 14-14-14-31 at 1.35v). So then I upped the voltage to 1.41v.

I will see what happens tonight - but perhaps I will try your settings of 15-15-15-31 at 1.40v - when you mention sub-timings was that using the Ryzen DRAM calculator?

or perhaps I will try the 3800 16 16 16 32 1.42v as mentioned by @Joxeon

 

[Awahwah]

thanks for this.

As mentioned I kept running the passmark benchmarks and had no problems (in any of the tests). However last night in real world gaming on world of warships I did get a blue screen of death and then later a game crash - which is unusual - I was running the 8 pack 3200mhz at 14-14-14-31 @ 3600 at 1.40v (and the sub timings had been left as the Ryzen DRAM calculator settings for 3200 at 14-14-14-31 at 1.35v). So then I upped the voltage to 1.41v.

I will see what happens tonight - but perhaps I will try your settings of 15-15-15-31 at 1.40v - when you mention sub-timings was that using the Ryzen DRAM calculator?

Started with, yes. Reduced some down and tested for improvement and stability. To be honest though, apart from reduced latency and greater throughput - I've not seen much of a difference between XMP 3200 and what I have now in real world.

 

[Joxeon]

and the sub timings had been left as the Ryzen DRAM calculator settings for 3200 at 14-14-14-31 at 1.35v).

This would likely be the problem.

First you want to use typhoon burner to read the SPD and convert it to HTML then import this into the dram calc and set the speed your trying for to 3600 then make sure you use the sub timings it gives you for this speed, start with safe timings then try fast if fast doesnt boot then try fast but with gear down mode enabled.

 

[Alan1969]

This would likely be the problem.

First you want to use typhoon burner to read the SPD and convert it to HTML then import this into the dram calc and set the speed your trying for to 3600 then make sure you use the sub timings it gives you for this speed, start with safe timings then try fast if fast doesnt boot then try fast but with gear down mode enabled.

tonight I may just reset the bios and leave the sub timings alone - maybe just rather change the main timings settings to 15-15-15-31 @ 3600mhz at 1.40v and see what happens.

yes I used Thiaphoon burner previously but didn't change it over the past few nights when changing the main timings.

 

[Alan1969]

last night I reset the mobo by loading optimized settings (to remove the memory sub timings). Then applied 15-15-15-31 @ 3600mhz at 1.40v and had boot problems. So increased to 15-15-15-31 @ 3600mhz at 1.41v and it worked fine even through passmark benchmarking and during a long session of world of warships (where previously I have had issues).

Benchmark (passmark that tests CPU, GPU, memory, disk etc) results show only slight gains by overclocking from 14-14-14-31 @ 3200 at 1.35v 9even with sub timings from the Ryzen DRAM calculator) - however somehow it does feel more zippy. I don't think that I can be bothered to optimized the main timings any more from 15-15-15-31 @ 3600mhz at 1.41v or even to now apply improved sub timings from the Ryzen DRAM calculator.

Note that I did see a significant increase (in passmark score) when applying xmp and also recently when overvolting and overclocking my vega56.

next I will have another go at overclocking my ryzen 5 3600 and using ryzen master. I tried a month or so ago and for my system I could not get a stable system - even if just setting the multiplier to 41 or 42 in the bios and upping the vcore to between 1.42 and 1.45V! I did see a hardware unboxed youtube that said that they had trouble with a gigabyte mobo (I have the £259 x570 auros pro).

 

[Awahwah]

last night I reset the mobo by loading optimized settings (to remove the memory sub timings). Then applied 15-15-15-31 @ 3600mhz at 1.40v and had boot problems. So increased to 15-15-15-31 @ 3600mhz at 1.41v and it worked fine even through passmark benchmarking and during a long session of world of warships (where previously I have had issues).

Benchmark (passmark that tests CPU, GPU, memory, disk etc) results show only slight gains by overclocking from 14-14-14-31 @ 3200 at 1.35v 9even with sub timings from the Ryzen DRAM calculator) - however somehow it does feel more zippy. I don't think that I can be bothered to optimized the main timings any more from 15-15-15-31 @ 3600mhz at 1.41v or even to now apply improved sub timings from the Ryzen DRAM calculator.

Note that I did see a significant increase (in passmark score) when applying xmp and also recently when overvolting and overclocking my vega56.

You may find that your RAM is like mine. However, I was able to use the tightened 'Fast' sub timings from the tool with 15.15.15.30. This improved things a little but not a lot.