Ram questions
#1
Posted 17 April 2008 - 01:45 PM
#2
Posted 17 April 2008 - 02:26 PM
I explain it very badly, but that's always been my understanding of it.. hence I'm running 2x2gb sticks instead of 4x1gb sticks
#3
Posted 17 April 2008 - 02:31 PM
Howling said:
I explain it very badly, but that's always been my understanding of it.. hence I'm running 2x2gb sticks instead of 4x1gb sticks
you can also look at it from a different angle.
Do you want to divide the workload on 4 workers or on 2 and make them work twice as hard?
but anyways I also use 2x 2 gig in my pc, never had any complaints.. honestly I dont think there is much of a difference.
#4
Posted 17 April 2008 - 02:47 PM
The general idea is that the more RAM you have, the better. Just like the more cache you have, the better.
But the bigger your RAM, the more likely you are to miss and the more misses you can have. Misses are costly and can add up. This is especially bad if your data is nowhere in where you expect it to be because you have to have a whole bunch of misses before your computer even complains. You'll also need bigger lookup tables.
But who knows. Having more RAM probably just overpowers the damage misses do to your efficiency by reducing paging requirements ( once it is in RAM anyway ). Less paging is always a good thing.
The idea about splitting work among employees is... kinda right I guess. But a lot of that heat generation now is due to how much current is running through your RAM in order to efficiently double pump data on the very large modern buses. Plus there are tons of reasons why your computer would not evenly distribute load between four sticks. For example, if it wants contiguous blocks, the best and obvious solution is to dump it on one stick if possible. Or when your RAM starts to get fragmented, your computer is going to use the first appropriate slot that is available to stick whatever you need into, and if all of those slots happen to be on the first stick, then all of the data is going to go on the first stick. This is probably most true with worst fit RAM management.
This post has been edited by Celestin: 17 April 2008 - 02:53 PM
#6
Posted 17 April 2008 - 02:56 PM
Howling said:
I answered that question too.
2 x 1GB = more misses, but probably more even usage of RAM. Big lookup table.
4 x 512MB = less misses, but probably less even usage of RAM. Small lookup table.
Lookup tables don't matter as much as everybody uses set associative or better(?) lookup tables, but still could make a difference. In any case, I have no idea because this is how it's been drilled into me in terms of theory, but at the same time, programs don't just guess where data is.
This post has been edited by Celestin: 17 April 2008 - 03:00 PM
#8
Posted 17 April 2008 - 03:39 PM
It's a complicated matter...I understand what you guys are saying in terms of "it can spread the workload quicker"...But keep in mind these are dual channel...The workload would be split to two modules, not 4...For that you'd need quad channel in order to do so...Once the memory fills up on one, it would processor over to the other. That's how I believe the memory usage works, but not 100% sure since I never really sat and thought about how two setups of dual channel memory would work...
#9
Posted 17 April 2008 - 03:57 PM
nvm i started this post before the one above was made :(
#10
Posted 17 April 2008 - 04:02 PM
1 User(s) are reading this topic
0 members, 1 guests, 0 anonymous users
Community Forum Software by IP.Board

Register Now!
Help





MultiQuote