RAM overclock benefits ?

[xerces8]

Hi!

I posted this first in the "HEXUS.help - buying advice & technical queries" > "HEXUS.hardware" / "C2D Overclocking Guide For Beginners P5K Add-On" thread, but I think it should be in its own thread. (I can't delete my post in that thread, can a moderator assist, please ?)

A question about RAM overclocking. In the guide, there is a Sandra screenshot showing about 2% improvement after tweaking, but in the text it says some real world applications have greater benefit.
Which real world applications are those ? How much is the performance improvement ?

Thanks,
David

[Bounty_Bam]

Good question

[xerces8]

Good question

... with no good answer. Or any answer. I asked this also on two other forums and no reply whatsoever ...

[Thorsson]

Programs that would benefit would be those, like say Photoshop, that make heavy use of memory. I've no idea how you could measure it apart from using an application-based benchmark like PCMark (although that's getting a bit long in the tooth).

[pmanington]

I think it depends heavily on the ram you have. You have to have really high quality ram to make it worth it as otherwise you have to raise the cas latency etc to get a decent overclock.

[GSte]

Most real world applications will see little to no benefit though.....

[kalniel]

Hi!

I posted this first in the "HEXUS.help - buying advice & technical queries" > "HEXUS.hardware" / "C2D Overclocking Guide For Beginners P5K Add-On" thread, but I think it should be in its own thread. (I can't delete my post in that thread, can a moderator assist, please ?)

A question about RAM overclocking. In the guide, there is a Sandra screenshot showing about 2% improvement after tweaking, but in the text it says some real world applications have greater benefit.
Which real world applications are those ? How much is the performance improvement ?

Thanks,
David

You'd have to ask the author of the guide as to which applications they were referring to. In general a 2% increase in the theoretical Sandra benchmark would translate to no real world performance improvement. Where you might see one is in any application that is strictly memory bandwidth limited - they're hard to find because generally to pass data through memory you have to create it or load it up from somewhere else, which is usually slower than the transferring data through memory. If you had an application that cached all of it's data and then performed relatively simple but numerous operations on it you might notice memory speed - I would guess that cached database manipulations, spreadsheet calculations etc. might fall unto that category.

[Blitzen]

Most real world applications will see little to no benefit though.....

Absolutely.
The only time its useful to OC memory is with a big overclocking CPU so it can keep up or stay as near to 1:1 as possible.

[Hunain]

Absolutely.
The only time its useful to OC memory is with a big overclocking CPU so it can keep up or stay as near to 1:1 as possible.

Yeah that is correct, I have OCd my ram from 667Mhz to 856Mhz, quite large benchmark gains though, but nothing significant in games etc..

[iranu]

Yep. I'm always surprised when people run ram above 1:1. I'm sure there are apps that can benefit but it's not required for everyday use or gaming.

I think overclocking ram is using the same mindset as overclocking a cpu. Extra performance without spending the extra £££. See My system. £60 of ram running at 450mhz (1:1) rather than default 400mhz. If I wanted that guaranteed with those timings then I'd have to spend more cash on PC2-8500 or higher quality ram.

[GSte]

I don't fully understand keeping the RAM at 1:1 myself, cas latency and many other factors seem to make it difficult to know if that ratio is useful or not..... Perhaps that's another thread though

[xerces8]

I don't fully understand keeping the RAM at 1:1 myself, cas latency and many other factors seem to make it difficult to know if that ratio is useful or not..... Perhaps that's another thread though

Maybe people do it for the same reason: higher numbers and the feeling that it improves the system ?
Now, if I had the (free) time for days of RAM settings testing

[xerces8]

Hi!

I took the time to do some testings myself.
System Asus P5K-E Wifi BIOS v0906
2x2GB DDR800 RAM (Transcend JetRAM JM800QLU-2G) 5-5-5-18
Q6600, nVidia 8800GT 512MB

I used 7zip v 4.57 and 3DMark 2006 Basic 1.1.0 as benchmarks.

Timings change:

I slowed the RAM timing to 6-6-6:

3DMark gave 1% slower CPU Marks. Everything else was below 1% change.

Frequency change:

Increased RAM clock from 800 to 852:
3DMark gave 1.5 % better results.
7zip had 4% faster compression.

Increased RAM clock from 800 to 1066 and set timings to 6-6-6 (to save the time searching for lowest working timing):

3DMark 1.5 % higher score.
CPU Score : 1.5 % higher.

7zip had 7% faster compression.

Conclusion : (mostly) waste of time.
7zip compression gains from clock frequency, but is mostly independent of timing.
3DMark gives almost negligible improvement. With lower timings maybe up to 5%.


PS: During tests I had gfx curruption for all changes to RAM settings on my LCD monitor connected over DVI. So I used my good old VGA CRT. The corruption appeared right after reset in BIOS screens and endured thru the Windows session.
I guess RAM clock changes might also affect PCI-E clock, but why would relaxed RAM timings do any harm to gfx ?