中文版 | News | Archives | Reviews | Forum | $ DealsLinks | History | Contact | Privacy

tRAS setting on NF3, can it affect performance (1)
Gortok 22 June 2004

Many tales have been told of how setting memory timings can lead to better performance. This can lead to some serious confusion since there are so many different things to "play" with...

It is not my goal here to describe what these settings are or how they affect system performance, however, I HAVE done some testing using my A64 3000+ and the Kingston Value 3200. This was the setup used for the latest review of the Chaintech VNF3-250, which by the way, is still working beautifully. I have heard a lot about NForce2 and NForce3 boards that like the higher values of Tras...the largest value in the settings string...ie. 2.0-11-2-3 where 11 is the Tras value. It seems the NVidia chipsets perform better at a higher latency value than other chipsets. Most boards would perform better at something like 2.0-6-2-2.

What I've done today is check a large assortment of timings by using SiSoft Sandra Membench. I've used the Max3! version for this test. You may ask, "Why not a newer version like 2004?". The fact here is that this is what I have available, and since we're looking for differences on the same system...the version won't matter...only the differences in the results.

So, here we go....First, I checked the throughput at the lowest (quickest) latencies possible 2.0-5-2-2. Since this is only possible on "my" RAM at 166FSB, that's what I set it to. Again, this is just to check for differences in the results...

Setting                                Result Int Buff/Float Buff

2.0-5-2-2                           2576/2576

2.0-6-2-2                           2574/2576

2.0-7-2-2                           2576/2576

2.0-8-2-2                           2575/2575

2.0-9-2-2                           2576/2576

2.0-10-2-2                         2575/2576

2.0-11-2-2                         2575/2576  

I stopped at this point. It appears there's no difference at all using one over another... On to 200...

Next >>>

Copyright 1998-2004
http://www.ocworkbench.com