PDA

View Full Version : How much RAM to disable swap file?


SquireSCA
12-24-05, 10:29 AM
[Specs in sig]

Obviously this rig is blistering fast. But right now, I am copying 145GB of data from my HD to an external drive over the Firewire port.

While it is not having any issues so far, I am noticing that when an app like Thunderbird(any app really) is minimized, and I go to bring it up to the foreground, it takes a few seconds, like 4 or 5 and you can hear the HD chugging away, telling me that it is having to load it from swap.

Granted, one does not copy such massive amounts of data on a daily basis, but it does indicate that in some cases, 2GB is not enough and your system has to tap the swap file, thus slowing things down.

I am looking on the net to see if people are able to disable the swap file in XP with no ill effects when they have 4GB of RAM installed.

Theoretically, it should work, as I have 2GB of RAM and a swap file set to a static 2GB. So I only have 4GB available now, it is just that 2GB of it is on the HD which would slow certain things down.

If I had all 4GB in fast RAM at 250fsb, the system should speed right up, yes?

And would having 4 sticks lower my abilty to overclock? I am running 250fsb, but the memory is divided down to 208FSB CAS2 1T. I would be getting high end stuff and would love to not only get rid of the swap file, but have the 4GB running in dual-channel mode at 250fsb.

Thoughts?

rev.
12-24-05, 12:14 PM
(rudolf) Someone correct me if I am wrong, but 4 GBs of memory would mean using all four RAM slots...meaning running at 333mhz (not full DDR 3200). You would be forced to run at a divider if you fill all the slots.

SquireSCA
12-24-05, 02:46 PM
OK, then what RAM in a 1GBx2 config is the most likely to run at 250fsb without having to mess around with the timings?

Cost is not a factor.

TierMann
12-24-05, 03:01 PM
DDR 500 (pc4000).
With your Opteron 175 though it shouldn't have the same problem with running at 333 with 4 dimms filled as older A64's do. Anything newer than Venice core should have that fixed.

Anyone know for sure on that?

SquireSCA
12-24-05, 03:26 PM
I know the speed I need(DDR500), but what brand and series is the best?

It *must* run at 250fsb with the timings that the chip's SPD wants it to. I am not gonna fiddle with settings to get it to run stable.

I need the best memory around, and cost is not a factor.

Suggestions?

Roliath
12-24-05, 03:47 PM
read up on xtremesystems.org
I recall someone using 4x1GB Crucial Ballistix Tracers and ran them at PC4000, albeit with 2T.

and IMO you NEVER want to disable the page file completely.

EDIT: here's the LINK (http://www.xtremesystems.org/forums/showthread.php?t=73675&highlight=crucial) to the thread on XS

Rakeesh
12-24-05, 04:34 PM
In most systems, the pagefile size should be identical to how much memory you have. E.g. 3GB of ram should have a 3GB pagefile.

retsam
12-24-05, 05:09 PM
here (http://www.microsoft.com/resources/documentation/Windows/2000/server/reskit/en-us/Default.asp?url=/resources/documentation/Windows/2000/server/reskit/en-us/core/fnec_evl_wafw.asp)

The default size is equal to 1.5 times the amount of physical memory.

if you have like 4 gigs of ram, its better to run a ramdrive then to turn paging off.you would put the page file on the ram drive. some programs and dont like the absence of a page file.


here check this out
http://cyberwizardpit.net/article3.htm
(xmassign)

superklye
12-24-05, 05:46 PM
(rudolf) Someone correct me if I am wrong, but 4 GBs of memory would mean using all four RAM slots...meaning running at 333mhz (not full DDR 3200). You would be forced to run at a divider if you fill all the slots.
For pre-San Diego core Athlons, that is true, but the San Diego (and now the Venice) have fixed memory controllers that no longer force 4 DIMMs of DDR400 to DDR333.

Four DIMMs of of single-sided DDR400 will stay at DDR400 1T. Four DIMMs of double-sided DDR400 will stay at DDR400 but will be at 2T.

I am almost positive this applies to all of AMDs products post-San Diego.

Capt. Picard
12-24-05, 06:29 PM
I need the best memory around, and cost is not a factor.

Sheesh! I wish I could say that.

SquireSCA
12-24-05, 10:19 PM
Well, sales and consulting pays pretty well these days...

In any case, I want to find the bestest RAM there is. I want to run at 250fsb with tight timings, as I am worried that 250fsb CAS3 might be slower performing than 200fsb CAS2.

Thoughts?

superklye
12-24-05, 10:41 PM
Well, sales and consulting pays pretty well these days...

In any case, I want to find the bestest RAM there is. I want to run at 250fsb with tight timings, as I am worried that 250fsb CAS3 might be slower performing than 200fsb CAS2.

Thoughts?
The difference between CAS2 and CAS3 is practically nothing, especially compared to the fact that you're gaining 50MHz (100 effectively).

1T > 2T is also a MUCH bigger hit than CAS2 to CAS3.

EDIT: And a quick search of DDR500 PC4000 RAM on NewEgg came up with not a single CAS2 set of 2x1GB. If you want DDR500, you're getting CAS3

SquireSCA
12-24-05, 10:51 PM
Reading some reviews, it looks like having the lower latencies often makes more difference than the higher fsb... Hmmm... I wonder if I can just add more of the RAM that I have and keep the existing timings?

superklye
12-24-05, 10:59 PM
Reading some reviews, it looks like having the lower latencies often makes more difference than the higher fsb... Hmmm... I wonder if I can just add more of the RAM that I have and keep the existing timings?
That's really only relevant to synthetic benchmarks though. The performance in game and general tasks will be, for the most part, unchanged.

SquireSCA
12-24-05, 11:06 PM
Well, that is the idea. Have enough RAM on tap that I will rarely have to swap to disc.

I could shrink the page file to like 100MB for the few apps like Photoshop that require SOMETHING no matter how much RAM you have.

superklye
12-24-05, 11:09 PM
Or stop using crappy Photoshop and use a man's program...like Fireworks 8. :p

uOpt
12-25-05, 10:18 AM
It is never a good idea to disable paging to disk. Doing that messes with the OSes decision-making. In particular, with no paging area you now force the OS to drop readonly pages (code) from more programs, instead of dropping readonly pages and paging out read-write pages from few programs.

To makes this actually work, you don't only have to have enough RAM to hold you programs, but also enough RAM to hold every piece of file youever touch. This is not possible on a socket 939 board and a decent workload.

SquireSCA
12-25-05, 10:28 AM
OK, then I will look into getting 4GB of RAM, and then get one of those PCI cards that takes DDR memory and make a 4GB RAMDRIVE and put the swap file on that.

uOpt
12-25-05, 11:22 AM
OK, then I will look into getting 4GB of RAM, and then get one of those PCI cards that takes DDR memory and make a 4GB RAMDRIVE and put the swap file on that.

You wouldn't need 4 GB on the PCI memory card.

If you just take 4 times of whatever max usage you see for a disk-based page file, that'll be fine. It will tell the OS that it is free to make its normal decisions so that it doesn't screw things up trying to obey to your "only page out readonly pages policy".

However, don't expect a speed increase from the PCI memory card over a harddrive. If the OS works correctly disk accesses will be so few it shouldn't be noticable.

SquireSCA
12-25-05, 08:44 PM
So what is your final advice? 4GB of RAM and let Windows make up it's own mind on the swap file sizing?

superklye
12-25-05, 08:59 PM
Why is it such a big deal to disable the swap file? I'm just curious.

SquireSCA
12-25-05, 09:11 PM
Well, I thought that by elinating the need to swap data to a HD, the overall system performance would go up.

If that is not the case, then I will not spend the money.

uOpt
12-25-05, 09:14 PM
Why is it such a big deal to disable the swap file? I'm just curious.

People assume that no pageout will happen when you have no paging space.

These people don't understand that swapspace is used only for read-write pages, but that a major part of a program's pages are readonly or copy-on-write not yet triggered. This majority of pages is just dropped from memory, not written to paging space.

So not only do you do nothing against kicking the majority of pages out of memory, you now increase the OSes tendency to kick out pages from possible valuable programs - because by removing paging space you force the OS to touch valuable programs instead of first kicking all the known junk out.