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

Interesting Article on Ram

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

mrgreenjeans

Member
Joined
May 3, 2003
Location
Cleveland, GA
Informative and not prejudiced. Now I've got to go home and try a 2-2-3-5 setting. Hadn't thought of trying that one! when I've run benchies, I get much better scores w/ a 2-2-3-6 than with 2-2-2-5, so maybe I'll try the 2-2-3-5. Who knows, might be the majic bullet?
Thanks for the referral.
 

tyson-chris

Member
Joined
Jun 13, 2003
Location
[email protected] in san francisco
Just as an addendum: when they say 2-2-3-5, the 2 refers to the RAS Delay and the 3 refers to the RAS->CAS delay. (I don't know why they used this unusual order) Usually, timings are referred to in CAS, RAS->CAS, RAS, TRAS order, so 2-2-3-5 is what is more commonly shown as 2-3-2-5.

edit: also, the fact that their "fastest" timings were 2-3-2-5, it's well-known that raising RAS-CAS delay from 2 to 3 has a rather large performance hit. If they had done 2-2-2-6 has their "fastest" timings, there would likely have been a more appreciable performance increase with their "fastest timings" vs their "worst timings"...

I'm sure you've heard this before too, but personally, I think 2-3-2-5 would be worse than 2-3-2-7. This seems to be coming up a lot recently :p but yeah, optimal tRAS should be CAS + RAS->CAS + 2. MrGreenJeans, it's possible you got lower performance with 2-2-2-5 than 2-2-3-6 because your tRAS wasn't high enough. I bet if you ran 2-2-2-6, you'd get better performance than 2-2-3-6.

As for the article, I found it a bit presumptuous, in that they were making rather substantial claims using only 2 synthetic benchmarks and 2 real-world benchmarks. Seems like an awfully limited number of benches to draw conclusions from, particularly when different apps respond in differing degrees to lower-latency operation.

Also, I'm not entirely sure, but I believe Intel memory systems are less latency-dependent than amd systems. I don't have an Intel system to compare, but, I believe if they had tested 2-2-2-x and 3-4-4-x on an nf2 system, there would have been greater performance deltas. They did touch on the fact that "overclockers" modules oc higher, and that mhz increases provided more performance gains. Additionally, good "overclockers" modules can oc higher (mhz-wise) all the while keeping their low latencies as well (ie. my Mushkin does 466mhz at 2-2-2 stably), which extend the performance gains further.