Higher memory timing/speed or CPU FSB/Mhz better?

[BoboKatt]

hey all..

still trying to figure out if I should be more concerned with running my 1 GIG of RAM at 1/1 ratio or as close to as possible with lowest timings on my AMD64/3200 939 on a Neo2Plat board or go for as much MHz/FSB from the CPU and just keep the ram at stock (i.e? 200mhz)? I have heard from folks here and other places that the increase in FSB on the AMD/64 is not that great of a difference like it once was say on an older Celeron 300 running 66mhz and overclocking it to 100mhz to get 450. Or I would assume the benefits you would get from oc?ing your FSB on a Intel p4 Prescott. Also as the AMD64 has it's memory controller on chip I think that again some benefits are not as evident by cranking up the memory speed? Am I totally off? In that case heck to get the best performance in games is all that matters total MHz at the end of the day?

I spent some good money on buying quality memory that should allow good low latency timings at insane RAM speeds and just wondering if that was really worth it. Someone had posted that I could have saved some coin and just kept my original crap no name PC3200 ram and just clock my divider to a number that would have kept the final speed of the ram at no higher than 200mhz. Currently I am running my system at 2400 MHz... and keep the 1/1 ratio by just keeping it at 200 in the bios.. rather than say 166 that would effectively bring my ram down to 200 MHz. Is it really doing much in performance? I benchmarked and noticed little change but that could be from my "older" ATI 9800 pro holding everything back. Sandra is crap and I don?t really trust that. Any which way playing games I cannot notice ANY discernible difference in frame rate? so?. LOL maybe I just answered my question.

thanks for your suggestions..

 

[Devlyn]

Someone on OC Forums ran tests last night btwn 2-2-2, 2.5-2-2, and 3-2-2, and while benchmarks showed 1% gains btwn each, gaming showed very little FPS gains such that you will never know. I'm thinking of returning my Kingston Hyper-x 2-3-2 ($170 AR) for something a little less costly and picking up something a little bit cheaper.

 

[BoboKatt]

Good point and glad there are some test like that out. I will search for some more on my own. I still think though that if the better RAM allows you to reach higher clocks and stay at a higher frequency such as 200 MHz or 240mhz or even 250... that there should be a noticeable difference in performance. I am going to test more of that now... gona see just how much difference returning my ram to 200 MHz and keeping FSB at 250 or 240 and then seeing what I notice or get with ram running 240 or 250. I don't think I worked that test scientifically enough...

Thanks

 

[DAPUNISHER]

With A64 CPU clockspeed >* If you can obtain max stable clockspeed and bring the ram along for the ride that is great, but if the ram is limiting your max stable CPU speed then use a divider because there is more performance to be had from CPU speed and the fact the on-die memory controller performance improves with it.

Timings mean very little to gaming performance on A64 with the difference between 2-2-2 and 3-4-4 being less than 5% in every game I've seen tested, with 2-3% the average. LL ram is perhaps the worst price to performance ratio hardware for A64 IMO. The performance gains from high ram frequency and tight timings are easily overcome just by obtaining higher CPU clockspeed on this platform, if a choice must be made between them. IOW It is great if you can have it all, but when forced to make a choice, give up timings first, ram speed second, but don't sacrafice max stable CPU clockspeed for either of the others.

 

[sdack]

You are asking the wrong question (sorry if that disappoints you). Both values are theoretically independent. So you should try to get the highest core _and_ highest memory speed. In paractice however you usually start with overclocking the FSB to find a first limit. You will then have two frequencies and the question that remains is if it is the core's frequency which cannot run higher or the memory. You then increase the mutlipier and thereby increase the core clock further. If you seem to be going nowhere, it is likely that the core cannot run faster. Otherwise its the memory which is limited. Try following the three rules:

- Aim for a high core frequency (multiplier * FSB)
- Aim for a low mutliplier and a high FSB
- Aim for a 1:1 or 2:3 (1:1.5) ratio

The first rule is obvious... The second should make sense, too: The multiplier is the amount of clock cycles a CPU has to wait for its I/O (from/to the chipset). Naturally, you want to keep this low and increase the FSB over the multiplier. The third however is difficult since it depends on the chipset and how well it was designed.
Early nForce2 boards were bad at everything besides 1:1. The nForce2 ultra atleast does 2:3 as good as 1:1. Think about it: With a ratio of i.e. 3:5 (aka 1:1.2) and an FSB at 200MHz it means for a chipset to have a timing as precise as to a fifth of the FSBs frequency. In other words its logics would effectively need to have the capability to be running at 1GHz. It is likely that not all chipsets can match this and simply pretend to do 3:5.
Best is to use a memory test like "memtest" which shows you the speed of memory accesses. It is good enough to display the difference between the ratios and you need to run benchmarks anyways to see if what you did really has the expected result.

Sven

 

[BoboKatt]

Hey all

thank you much for all the great advice and help. I am totally gonna start trying some of these out tonight.