• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Trying to learn memory testing/overclocking

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
Hi Folks!

Well...after a long time procrastinating am finally trying to learn about testing and overclocking memory.

What I did so far is load a pair of sticks and boot once with all ram settings on auto. After that, I manually set the speed to their default values, boot to windows and run aida once or twice. Then reboot, increase the speed, boot, run aida, etc until the setup fails to post or can't make it into windows. Then, using the max speed that passed, manually set the memory timings to whatever showed in aida previously at that speed, check that it works then start dropping the timings 1 at a time until it won't boot/pass.

The current results are below. The tridentx look the best so far. Odd that the pi's with tighter timings didn't score as well at 2400. Some weird stuff...I wasn't able to get the evo2's to work at all and I could only get the gkill pi 2400's to run at 2200. After about 5 hours the setup started acting funky (I've got an insulation issue somewhere/somehow) so had to shut down. I didn't have time to try to tighten timings on the 2400 tridentx at any speeds. Figure that next time, I should see if the timings can be tightened at 2400 and up.

So...what do you think? Am I doing this right? 2400 seems to be my sweet spot atm. Any ideas on the evo2's or 2400 pis? Attaching pics of my bios settings so please let me know if there's anything that should be changed. Also didn't play with any of the memory presets so thoughts on those would (of course) also be welcomed.

Thanks mucho mucho!
 

Attachments

  • ram.jpg
    ram.jpg
    60.9 KB · Views: 210
  • xpower-1108-bios1.jpg
    xpower-1108-bios1.jpg
    188.9 KB · Views: 211
  • xpower-1108-bios2.jpg
    xpower-1108-bios2.jpg
    243.7 KB · Views: 212
  • 2200pis-2400-32377.jpg
    2200pis-2400-32377.jpg
    81.2 KB · Views: 204
  • presets.jpg
    presets.jpg
    48.1 KB · Views: 203
Last edited:
I am no expert on memory overclocking.

Takes time and patience, two things I have little of .... LOL

Not sure how MSI memory presets work ?

Historically one would argue MSI being the right choice for memory overclocking and performance.

I have had my struggles in the past with my GD 65 , but they may very well have it worked out with these new z97 boards ?

Good luck bro
 
You're better than you admit Witchyman ;P

Figure for presents...will use/try the samsung 2x4 2666 or 2800 @ 1.85v? For the pi's...guessing the psc 4x2 2400 @ 1.71v or 2x2 psc 2600 @ 1.9v?

Hoping someone that knows what they're doing with regard to settings, testing, overclocking, etc will chime in.
 
Going to just say I've ALWAYS had ****-poor luck with overclocking ram on MSI boards.

With that being said, I'd go for the 4x2 profile as a starting point since it should have a bit looser subs
 
Easiest way on PSC is to run any tighter profile @2400-2600 and write down sub timings. Then set something like 2600 9-12-9 1.85V and check if subs from that profile are working. If yes then try to set 2666 or tighten timings some more ( you should see the pattern comparing stock SPD to profile subs ).
When you reach max clock at which it's passing Spi32 1.85V then try to tighten main timings like 8-12-9, 8-12-8, 7-12-8, 7-12-7. If you make it work then check something like 8-11-8 or 7-11-7 up to 1.95V.

For Samsung the same way just try with stock settings at 2600, 2666 and 2800 10-12-12 ~1.8-1.9V. If you make it pass Spi32 then check what subs you have in the tight Samsung ~2400-2600 profile and check if it's working with previously tested 10-12-12. At the end try 9-12-12.

To make it right be patient and make it with small steps checking Spi32 stability. Check all on low clock like 4GHz or something near just to be sure it's stable. Higher CPU clock shouldn't affect memory stability on haswell but testing memory with CPU on cold is not best idea as it takes lot of time.
 
Thanks so much guys! Running on air for testing. Scores are a bit lower since (i think) the ring is so much slower. Some of the better and/or more interesting ones from today (posting the tridentx2400 at default for comparison):

edit: saw I double-posted the tridentx2400 @ 2800 screenie...will load the 2400@2400 shot when I get home tonight.
 

Attachments

  • hyp2133-2200-776t1-30155.jpg
    hyp2133-2200-776t1-30155.jpg
    80.7 KB · Views: 171
  • pi2200-2600profile812727-32247.jpg
    pi2200-2600profile812727-32247.jpg
    80.6 KB · Views: 168
  • trid2400-2800profile-31191.jpg
    trid2400-2800profile-31191.jpg
    80.8 KB · Views: 173
Last edited:
Thanks so much guys! Running on air for testing. Scores are a bit lower since (i think) the ring is so much slower. Some of the better and/or more interesting ones from today (posting the tridentx2400 at default for comparison):

Run Samsungs at CR1. Shouldn't be an issue on haswell.
On PSC try to set 2666 at the same timings. Not much higher clock but usually is much harder to set than 2600.
 
Thanks Woomack

Couldn't get the sammys to run CR1 :( They ran all the way up to 2933 without issue. The screen I posted (31191 result) was the best I could manage. That was using the 2800 2x4 preset. Using the presets definitely gives better results.

Wasn't able to get the 2300pi's to run at 2666 but was able to tighten them up a bit using the 2600 psc preset (32247 result). Again...significantly better results with the profile than without it.

The other result (30155) was with a pair of kingston hyperx 2133. Even though I wasn't able to run them at 2400, they sure did tighten up pretty nicely (stock is 8-8-8-24). Don't have a preset for them so all sub-timings were left on default/auto. Have a feeling they'd do much better with some good secondary/tertiary settings. Thinking these sticks would be great for 775 and Sandy? For 775, I've been using the cellshocks (7-7-7-20 but so far they struggle to run faster than ~1750).

Feedback of course is very much welcomed! Have a few more sticks to check and will post any results. Still cannot get the evo2's to work in this setup and can only get the 2400pi's running at 2200 so far. :/
 
Try to tighten subs manually. Profiles are never perfect. It's good if you have board which is showing you all subs that you are actually using. Then it's easier to switch from auto.

I had no 775 DDR3 board so I won't tell you what will be best. Hypers or Microns seem good for these boards looking at other people's results.
 
Thanks again Woomack

What I've been doing:
- load a set of sticks and boot them at default
- increase speed to stock speed, boot, run aida
- increase speed to next step above stock, boot, run aida...continue until I find max speed
- select preset, boot, run aida...repeat until I find the best speed/present/aida result
- manually tighten timings one at a time until I find the tightest possible and best aida result
- move to next set of sticks and repeat process

Once I get through all the sticks will start seeing what each of the best sets can do speed/timings on spi32. Right now, looking like the 2300pi's or 2400tridentx (in that order) will go to spi32 testing on the z97 setup. That said, if I can get some good secondary/tertiary settings on the 2133hyperx, those may be worth a look for z97, too? If my testing conclusions are correct...the hyperx might be the best sticks for wprime on the current setup just because the timings are so tight.
 
Playing with an a8-6600k on a giga f2a85x-up4. Tested a few sets of sticks...the tridentx 2400's seem to be the ones that run happiest on this setup. The aida read speed with these was 1k-3k higher than with anything else I tried.

Just curious...is 26.6k ram read speed on this chip/platform considered at all decent?
 

Attachments

  • fm-tx2400-aida-26632.jpg
    fm-tx2400-aida-26632.jpg
    432.3 KB · Views: 146
Sure does take time a patience, which is why I no longer even bother with memory. Plus its hard to test for stability, I have seen memory OCs that will pass memtest but fail OCCT with large FFTs. THe performance gain in negligible, focus on CPU and video
 
@ Zakarro

This is good stuff here ..

funsoul is a bencher, not a gamer

every hundredth of gain makes a difference in 2D when competing against some the the best guys on the planet

The bar is not stability, it is 32m :)
 
If it's 32M he's going for or anything in general, the higher you can get that NB speed the better usually.
 
You likely could if you take it cold and pour the volts to it. That's a typical speed for H20 and ambient but it'll go much higher. When I was messing with my disappointing 8320 I had it up to 3500 @ 1.5 ish volts when it was frozen.

aida.JPG
 
Ahhh....got it. Not going to be taking this chip cold for a while but that's sure to come in handy.
Thanks again Johan45!
 
@ Zakarro

This is good stuff here ..

funsoul is a bencher, not a gamer

every hundredth of gain makes a difference in 2D when competing against some the the best guys on the planet

The bar is not stability, it is 32m :)
I gottcha, I see you guys are HWbot here. hehe

Nah yeah Im a gamer I just do 24/7 OCs for gaming
 
Figured this would be the right thread ;)
Hope you don't mind funsoul.

tRAS should always equal CAS + tRCD + tRP right?

So 9-9-9 = 27
Is there any limit to this, like 9-9-9 32?
 
Back