Legit Reviews

Just thought I'd take a look at Phenom memory bandwidth at CL4 and CL5 timings with four 1GB modules at 800MHz on the 790FX chipset. I used Sandra and Everest to check the memory bandwidths that I got on the MSI K9A2 Platinum V1.0 motherboard.
With the TLB patch enabled the memory bandwidth was 8105MB/Sec according to Sandra and the read score was 3813MB/Sec according to Everest v4.50.1330.
With AMD OverDrive 2.0.17 fired up we disabled the TLB patch and memory bandwidth was 9690MB/Sec according to Sandra and the read score was 5802MB/Sec according to Everest v4.50.1330. That is a 19.5% perforamnce increase on Sandra and a 52% increase on Everest.

Trying to get better scores, I dropped the timings from 5-5-5-15 to 4-4-4-12 and redid the tests...
With the TLB patch enabled the memory bandwidth was 8213MB/Sec according to Sandra and the read score was 3972MB/Sec according to Everest v4.50.1330. By going from CL5 to CL4 timings the performance in Sandra went up by 1% and Everest performance increased by 4%.
We again used AMD OverDrive 2.0.17 to disabled the TLB patch and memory bandwidth was 9849MB/Sec according to Sandra and the read score was 5911MB/Sec according to Everest v4.50.1330. A modest 1.6% gain on Sandra and 1.9% on Everest by using tighter timings.

Bottom Line: Our testng shows that with an AMD Phenom 9600 Black Edition Processor running DDR2 at 800MHz CL4 memory timings does improve performance over CL5 memory timings, but not by a significant amount. This testing was done on a 790FX chipset, but it would be fair to assume that other AMD 7 series chipsets would perform similar to this.

wow, i always thought cas latency was the deciding factor in memory speed. so this means we should see no difference in real world aplications right? and does this hold true for intel as well?

Yep thats why I always go for speed rather than tighter timings when I set my ram. I get better results by loosening the timing and increasing the speed.

Either way, overclocking ram still doesn't hold a candle to overclocking cpu and video card.

martini161 wrote:wow, i always thought cas latency was the deciding factor in memory speed. so this means we should see no difference in real world aplications right? and does this hold true for intel as well?

It plays a bigger role for Intel as the memory controller isn't on the CPU like it is on AMD. I was just showing this to those that are running a Phenom platform. Tighter timings is better, but only ever so slightly.

Not that it is something I will need, but thanks for putting the effort in and sharing your results

Dan