FX-6200 memory scaling

[wutske]

So, I was trying to figure out which of my DDR3 kits performed best (lowest latency, highest frequency) and I noticed that pretty much nothing I did improved my scores.

I used my FX-6200 for this, mainly because it was what was sitting next to me. I overclocked the CPU tot 4,4Ghz, nothing fancy but I needed something stable enough so that crashes could be nothing but RAM.

 

I tested 7-zip, SuperPi 1M and WPrime32 since those benefit more from faster RAM compared to e.g. Cinebench (and they are rather quick to finish). Running DDR3-1833 CL8-8-8, DDR3-1833 CL10-10-10 and DDR3-2133 CL10-11-11 all yielded pretty much the same results, with less than 1% difference. I've read that Bulldozer wasn't exactly the best thing in the world, but ... is it normal to have such little speed difference ?

 

FYI, I was lazy, so for 7-zip green is the best result and for the other red is the best result.

 

 

[Leeghoofd]

None optimal selection of benchmarks, Wprime doesn't scale with memory timings, SuperPi 1M is a slot machine

How do the secondarys look ?

I need to  provide you some better sticks than these man

[wutske]

To be honest, SuperPi was surprisingly consistent, all results are within 0,11% ... though that's the problem I'm having and I have seen wprime scale with memory frequency, but that might have also been more of a lucky shot.

 

These are my full timings:

 

I know I need better RAM, the best stick I have is a single stick of 2GB OCZ Platinum which does DDR3-1600 at CL7-7-7 (at 1,9V that is). The Kingston sticks I'm using in this test can do DDR3-1866 at CL10-10-10 at 1,8V...

[Leeghoofd]

I'll send you some Hypers and PSC, SuperpI 1M is too short for seriously evaluating mem timings. Its a quick test, that's the only benefit...

[wutske]

That would be awesome.

I'll see if I can do some better benches this weekend to get a better understanding of what impacts what.