CL3 vs CL2

[Thorsson]

For arguments sake (real world difference):
Dual Channel vs Single Channel = 5%
CL2 vs CL3 = 5%
200 v 166 (memory divider) = 5%
1T v 2T = 5%

It begins to add up...

[Dark Horse]

For arguments sake (real world difference):
Dual Channel vs Single Channel = 5%
CL2 vs CL3 = 5%
200 v 166 (memory divider) = 5%
1T v 2T = 5%

It begins to add up...

You can ignore dual channel because as long as you have 2 sticks of ram you can do it.

Cl yes small drop, a few fps at most in a game

Memory divider gives no performance decrease at all on a64

No really sure what to say about 1t vs 2t as its only really relevant when overclocking and you shouldn't need to change to 2t unless you really have crap ram.

[Zad]

Of course, another catch is that whenever you use more than 2 sticks, it drops to 166, and in this case, single channel for the extra stick. So the speed drops even further.

When it comes to memory, there are so many opinions, so few facts

Memory must be so unsexy that none of the major review sites have done any reviews of them, they all do the bling motherboards, glamourous CPUs, and supermodel graphics cards but nobody seems to want to touch memory. Sure theres the occasional A vs B comparison review, but for every review that says that it makes no difference what memory you use, there is another that swears blind that CL2 or PC4400 is essential and will give your system a huge lift. Or in my case, ones that say Value Select is fine with DFI LanParty, and others that say it's a nightmare.

*Shrug*

Seems to me that it's a lucky dip!

[kalniel]

The thing is, it doesn't all add up usually.

Real world improvement is mostly about reducing bottlenecks for whatever it is you are trying to do. If you're playing a game and your gfx card just can't display things faster than 42fps, increasing the speed of the ram etc isn't magically going to make the gfx card faster.

In those kind of situations you might gain 1 or 2fps extra with faster memory timings etc. .. but then that's all, as you've just eased things around to shift the bottleneck slightly, and it will take a much bigger theoretical increase elsewhere to see any improvment at all.

I went from 3500mb/s on sandra to 4500mb/s by going from 166 to 200mhz. I got to 5000ish by going 1T on top of that. In all of that, 3dmark 2001se went from 12080ish to 12800ish at most. So from the worst case senario (not including single channel here, sorry) to best (and not stable on my board) was only 5% improvement. Even assuming I'm playing a game that isn't utterly gfx card limiting, say half-life 2 (3D2001se is known to test more than just gfx card) and getting 50fps, that then goes up to 52.5. Wow. That whopping 1500mb/s theoretical increase in memory speed only translates to a few fps.

[chinny]

OK - did some benchmarking with BF2 on dual channel and single channel here's the results from an 800 x 600 test:

Running in single channel mode with 1Gb (2 x 512Mb)
------------------------------------------------------------------------------
T I M E D E M O - R E P O R T
------------------------------------------------------------------------------
Demo: mods/bf2/Demos/sample.bf2demo
Camera: mods/bf2/Demos/sample.bf2cam
Target Frame Rate: 30
Resolution: 800x600
Field of View: 90
-
Total Frame Time: 27.9191 seconds
Number of Frames: 1839
Average Frame Rate: 65.8688
Average Frame Time: 0.0151817
Shortest Frame Time: 0.00720215 (frame #1)
Longest Frame Time: 0.210381 (frame #336)
------------------------------------------------------------------------------

and....

Running in dual channel mode with 1Gb (2 x 512Mb)
------------------------------------------------------------------------------
T I M E D E M O - R E P O R T
------------------------------------------------------------------------------
Demo: mods/bf2/Demos/sample.bf2demo
Camera: mods/bf2/Demos/sample.bf2cam
Target Frame Rate: 30
Resolution: 800x600
Field of View: 90
-
Total Frame Time: 30.3049 seconds
Number of Frames: 2150
Average Frame Rate: 70.9456
Average Frame Time: 0.0140953
Shortest Frame Time: 0.00686264 (frame #651)
Longest Frame Time: 0.20631 (frame #647)
------------------------------------------------------------------------------

[Thorsson]

That's a 7.7% increase. Not too shabby - people buy one graphics card over another for less...

I think there's a mythology on Hexus that memory makes no difference. It may not be particularly good bang per buck in games, but it does make a difference.

[Elessar_VPR]

OK - did some benchmarking with BF2 on dual channel and single channel here's the results from an 800 x 600 test

Thanks for posting these, nice to see some actual numbers instead of hearsay and conjecture.

[chinny]

Ok just out of interest I've borrowed some RAM and ran the bench in BF again but with 2Gb. I know it won't make much odds in a small map but thought I'd post the results up to complete the results.

------------------------------------------------------------------------------
T I M E D E M O - R E P O R T
------------------------------------------------------------------------------
Demo: mods/bf2/Demos/sample.bf2demo
Camera: mods/bf2/Demos/sample.bf2cam
Target Frame Rate: 30
Resolution: 800x600
Field of View: 90
-
Total Frame Time: 31.2566 seconds
Number of Frames: 2273
Average Frame Rate: 72.7205
Average Frame Time: 0.0137513
Shortest Frame Time: 0.0072937 (frame #5)
Longest Frame Time: 0.206619 (frame #770)
------------------------------------------------------------------------------

thanks for all the replies.

[Thorsson]

I believe the big difference on BF2 with 2Gb is in the map loading times. not in the framerate.

[chinny]

I believe the big difference on BF2 with 2Gb is in the map loading times. not in the framerate.

Yeah - definately shouldn't have to resort to paging file so much. Just thought the result might be interesting given I'd done the others

[kidzer]

Quick hijack (sorry!)

I've read stuff about RAM @ 1T and 2T...what's that all about?

Cheers,
David

[kalniel]

the 'T' is the command rate - how long memory waits for commands to be carried out before taking the results back to the main system (roughly ) 1T means it's only allowed 1 cycle, 2T means it's given 2. 1T is faster but more risky as if you try and get things back before they're done your system will crash.

Most memory can handle 1T quite easily, but instead the performance of the motherboard or the memory controller are usually what limit things. On AMD chips for example it's usually not possible to run 1T when you have 4 banks of DIMMs filled. On my ASrock motherboard I also can't run 1T at 200mhz even with just 2 banks.

The gain in speed from 1T is more than you get from tightening up other timings in RAM, but still only translates to a few fps in real world terms.

[kidzer]

Excellent explanation, thanks

[Thorsson]

http://www.xbitlabs.com/articles/cpu...64-e3-mem.html

Y'all might want to take a look at this.

Conclusions (for the lazy):
1T vs 2T worth only 2-3% !!
DDR480 vs DDR400 worth around 3% (i.e. about the same as CL2 v CL3).

Therefore the conclusion must be:
1. Dual Channel gives most benefit
2. 2Gb is going to become inceasingly important (I suggest even more so for dual core chips)
3. The rest are worth little in real world terms, and the sensible approach has to be buy based on price and warranty.