Ram - clock vs timings

[Caspase]

Hi there,

I got bored today so I started to see the limits my ram could overclock to.
It's corsair vengeance artic white low profile low voltage 1600 9-9-9 2t .
Since it starts at 1600 with only 1.35V I was hoping for some great OC... The results were... Meeeh...
I got it stable at:
1600 7-8-7 1t @1.60
1800 10-11-10 1t @1.65
1866 11-11-11 1t @1.656 (Failed at the last cycle of stability testing @1.65 )

Just wanted to share and to know which one would you keep. Thanks!

 

[erixx]

Keep the one that benchmarks the best!

 

[McSteel]

Depends on the usage scenario. Much like with regular storage, there are sequential and random access patterns. Apps and games requiring loads of consolidated data will benefit most from shear throughput, meaning higher clocks. On the other hand, apps/games that make a lot of mixed random-sized read/write combinations would benefit more from reduced latency.

The catch is in the fact that both clock and latency play a role in both final response time and throughput... So as erixx nicely put it, keep the one that benches the highest

 

[Caspase]

What benchmarks? Cristal disk mark (after creating a ram disk) prefer the 1866 version. In games there really seems to be no difference. 3d Mark had a very subtle preference towards 1600 cas 7.
I use a ram disk for browser caching, I've started to fold now (ram OC actually seemed to make a difference here), and I game basically. Any real world benchmark I can use?

 

[MrSeanKon]

Try Sysoft Sandra. MemTest shows a value although it is a stability program mainly.
Beside of them I don't suggest RAM voltages >= 1.65

 

[Caspase]

Will try it out but as I remember it Sandra is mainly synthetic right? I was more interested in everyday performance!

 

[MrSeanKon]

Yeap Sysoft Sandra is a multibenchmarking program, it monitors and informs for your system etc....

 

[McSteel]

For everyday performance I'd stay at 1600 @ 7-8-7-1T. There is a nice explanation/guide to calculating the best scenario for particular benches here (Ian Cutress/AnandTech). Generally, single-word performance is more important to apps that do a lot of cache swapping and to highly-parallel simple-data calculations and lookups (like database operations or WEP/WPA cracking/phone unlocking). Multi-word performance is important to intensive GPGPU work, most games and complex-data crunching.

Using RAM as a RAMdisk is about halfway between, because you're doing both random reads/writes (browser cache) and sequential ones (when there are larger chunks of data like large pics and video streams). Though, in both cases, the RAMdisk will outperform any form of non-volatile storage by a large factor, no matter what your RAM settings are.

AIDA's memory testing is done pretty well.

 

[Caspase]

Seems like I'll be keeping 1600 then! Thanks.