Impact of ram speed

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,334
Credit: 1,849,892,931
RAC: 1,489,620
Topic 190552

I'm fixing up my old athlon 1400 as a crunch box, and was wondering what level of performance gain I'd see between ddr266 and 333. I've got 512mb of the former and only 256 of the latter and was wondering what level of gain (if any) I'd see from buying a 2nd stick of 333 and using it instead.

Sharky T
Sharky T
Joined: 19 Feb 05
Posts: 159
Credit: 1,184,389
RAC: 0

Impact of ram speed

If you're only building a dedicated crunchbox I think you only need 256MB of RAM in the box anyway.
All my dedicated crunchboxes have 256MB and have at least 130MB free physical memory while running EAH and they don't use the swapfile while crunching.(Win XP).
So I'd go for 256MB DDR333 in your case.
But of course you could test if you see any performance difference by testing 256MB DDR266 vs. 512MB DDR266.


Winterknight
Winterknight
Joined: 4 Jun 05
Posts: 482
Credit: 143,883,784
RAC: 165,029

RE: I'm fixing up my old

Quote:
I'm fixing up my old athlon 1400 as a crunch box, and was wondering what level of performance gain I'd see between ddr266 and 333. I've got 512mb of the former and only 256 of the latter and was wondering what level of gain (if any) I'd see from buying a 2nd stick of 333 and using it instead.


What's the fsb of the mobo?
What are the RAS, CAS etc timings of the sticks?
If the fsb is low can the timings be lowered?

If you need guidance on memory timings, and your BIOS allows you to alter them, some mobo's of that period only allowed default (SPD) or turbo, (whatever that meant), see Corsair memory basics

Andy

NikolZ
NikolZ
Joined: 31 Dec 05
Posts: 6
Credit: 4,741
RAC: 0

I use win98se and this os

I use win98se and this os reliese needet power like winxp.
Xp use lots of cpu power to run(lots if theres primary work)

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,334
Credit: 1,849,892,931
RAC: 1,489,620
gravywavy
gravywavy
Joined: 22 Jan 05
Posts: 392
Credit: 68,962
RAC: 0

Linux distros come with a

Linux distros come with a useful program call memtest
(or memtest+ or suchlike). It runs a series of tests
on (almost) every memory address in sequence, and repeats
the cycle over and over again. An unintended use of
it is to run those tests overnight, let it run to the
end of whichever cycle its on, and use the number of
cycles / hour to compare actual memory performance.
This is especially useful as it is designed to be a
very memory intensive program so in theory will
exaggerate the effect of running with diffeent
RAM speeds.

A while back I did some tests comparing PC100 and
PC133 memory in a 833MHz box. What was interesting
is that the 133 memory only ran 8% faster.

OK, whenever the box goes into wait-on-RAM that
wait will be over in 3/4 of the time - but clearly
a lot of the time the time critical component is not
the RAM at all.

At first I was surprised, but after some thought
decided it is what I should have expected. A well
designed system will have components that are all
pretty near their critical point, to have any one
compnent well over is a waste of money. Like, if
you got a 33% boost when going from PC100 to PC133
it would mean that the processor was faster than
needed for the rest of the components.

Since I ran those tests I've never gone out and
bought memory just to upgrade the speed. My
suggestion is to see whether the larger slow memory
or the small fast memory does better on your
favourite project; don't buy extra RAM but put
the funds towards your next box (whether new or
secondhand)

R~~

edit: broke lines because of stretching.

PS: Dan

Your long url has made this thread stretch
sideways.

Please check out how to post a url as a link
rather than as text, then it won't stretch the
thread, which makes it harder to read as your
readers need to scroll left and right to see
everthing. It's not hard to do, just click the
nearby BBtags link to see examples. :-)

~~gravywavy

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,334
Credit: 1,849,892,931
RAC: 1,489,620

RE: Linux distros come with

Message 23778 in response to message 23777

Quote:
Linux distros come with a useful program call memtest
(or memtest+ or suchlike). It runs a series of tests
on (almost) every memory address in sequence, and repeats
the cycle over and over again.

THere're standalone tools to do the same. Bootable CD's so probably the same origins

Quote:

A while back I did some tests comparing PC100 and
PC133 memory in a 833MHz box. What was interesting
is that the 133 memory only ran 8% faster.

that's relatively inline with my expectations as well.

Quote:

Since I ran those tests I've never gone out and
bought memory just to upgrade the speed. My
suggestion is to see whether the larger slow memory
or the small fast memory does better on your
favourite project; don't buy extra RAM but put
the funds towards your next box (whether new or
secondhand)

I wasn't planning on spending extra towards more ram, and with people saying their crunchboxes are getting by while only using half I don't think I'd need to.

Quote:


PS: Dan

Your long url has made this thread stretch
sideways.

Please check out how to post a url as a link
rather than as text, then it won't stretch the
thread, which makes it harder to read as your
readers need to scroll left and right to see
everthing. It's not hard to do, just click the
nearby BBtags link to see examples. :-)

Too late to fix before I realized it was a problem.

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 769
Credit: 201,984,256
RAC: 395,884

I guess the question is not

I guess the question is not urgent any more, but here's my take: E@H doesn't need much memory bandwidth. Your can clearly see this when comparing AXP and A64 at the same clockspeed, they're almost the same speed while the A64 has a vastly superior memory interface. So good timings don't help einstein either. And size doesn't, of course ;) (win taskmanager shows 7MB per einstein client if I remember correctly)

MrS

Scanning for our furry friends since Jan 2002

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.