Socket A ram divider - poor performance \ VOGONS

Socket A ram divider - poor performance

Topic actions

First post, by dulu

Posted on 2021-02-08, 12:06

dulu Offline

Rank Member

Rank: Member
Posts: 161
Joined: 2020-09-28, 12:05
Location: Cracow, Poland

I heard an advice that on socket A you should not use memory dividers because it creates additional latency. In fact, when I set for example 125% memory speed (166fsb, 208MHz ram speed in result), superpi performance is slightly worse than 166fsb and 166 ram speed (1:1). Therefore, I wanted to ask what the ratio of the fsb frequency to the memory frequency must be so that no latency is generated. Assuming it's even possible.

Reply 1 of 2, by bofh.fromhell

Posted on 2021-02-08, 12:40

bofh.fromhell Offline

Rank Oldbie

Rank: Oldbie
Posts: 524
Joined: 2019-02-09, 16:11

Async memory rarely made any sense back in the day.
As you already noticed, 1:1 is most likely the way to go for overall performance.
Tho typically computational stuff likes low latency while games may benefit from more bandwidth.

There's a little trap tho, usually latency is measured in cycles, which only really works when comparing at the same speeds.
Aida64 (and probably others) measures in time instead, giving a better view of whats actually happening.

Reply 2 of 2, by The Serpent Rider

Posted on 2021-02-08, 13:09

The Serpent Rider Offline

Rank l33t++

Rank: l33t++
Posts: 6278
Joined: 2017-03-25, 19:07
Location: The Hole

Async problem affects only Nforce chipsets. Other will perform fine.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Go to top of page Go to top of page

Back to General Old Hardware