Memory - How much difference between 3-3-3-11 & 2-2-2-11 ?

[Butcher]

Anyone got similar tests/number for A64s?

[Lee H]

10? 11? Mysterious articles? Have any of you benchmarked this yourselves?

http://www.madshrimps.be/?action=getarticle&articID=133

http://www.dvhardware.net/review/409

I actually used the 2-2-2-11 timings on my old NF7-S based system and it does seem that thats the optimum memory setup on the nforce 2 chipsets and it does seem to actually make a difference when testing with programs such as Sandra

[iranu]

the 11 at the end is actually faster than lower numbers. there's an interesting article around somewhere about it. I don't claim to know all there is to know but its quite interesting

When I switch between 166fsb to 200fsb on a xp2500 the ram (muskin 1 gig 3200 level 1) automatically goes from 2-3-2-5 to 2-3-2-6. So I assume from above that this is faster/better. Not sure why this happens but may try some other timings.

[SilentDeath]

Right, please all of you able, goto bios, set your nice shiny pc3200 2-2-2-6 mem to pc2700 (166/333ddr) @2.5-3-3-7.

Change CPU multi so cpu clockspeed doesnt change much (otherwise this whole test is pointless..)

Now run some benchmarks.3dm2k1, pifast.. etc.. stuff like that (not 3dm2k3/3dm2k5.. thatspoinltess for this..). Also real world game benchmarks like q3 would probably give reliable results.

I think you would be suprised how little differnce this makes to overall performance..
It would make much less difference in most new games as they are graphics limited more than anything and untill you have the fastest GPU - an x850XTPE (overclocked ofcourse) there are probably better things you can do for performance that cost less...

[Thorsson]

Wildmonkeys second link directs to the first...

Anyway I did tests on my ASUS SLI/Crucial Ballistix running at stock, 1T, using 2-2-2-5/6/8/10/11. With Sandra 2005 SR1 the numbers were extremely close, but 2-2-2-5 was marginally fastest and 2-2-2-11 was marginally slowest. But I'd probably need to run a hundred tests to get a statistically valid result as they are all so close

[Thorsson]

Just a thought - is the mobo actually changing these timings or is it just "pretending" to? Anyone know of a way to definitely check? Sandra tells me that they're running at 2-2-2-8, which funnily enough seems to be the default "auto" speeds.

[bouncin]

i think using amd tweaker would be a tad quicker than all the bios changes

i found the latencies make little or difference in real world on a64 systems upto 250ht but cant really comment above this as my ram only does around 250 1-1

[Thorsson]

Well if I'd heard of AMD Tweaker I no doubt would download it and use it. But it doesn't google.

[Lee H]

I think the 2-2-2-11 applies only to the nvidia nforce 2 chipset. I've not really had time to play with the Asus A8N board at home yet due to other reasons but I can have a fiddle tonight and benchmark the system in normal "2-2-2-5" and "2-2-2-11" to see if there is any differences

[Rave]

It's called A64 tweaker I think.

[Kumagoro]

This does ram latency make much diff has been going on for so long it seems like
something that will never be done and dusted. Nearly everything I have ever read
when it comes to timings has shown that knocking of an extra nano second doesnt
give that much more performance and not worth the extra money. Normally saying
that between the tightest most expensive and the normal slack cheapo ram there is
only UP TO a few % diff and not a constant improvement.

I would only buy it if i was going to buy near enough the top of the range components
in all other performance areas.

[Butcher]

I would only buy it if i was going to buy near enough the top of the range components in all other performance areas.

Does spending over £100 just on cooling for graphics cards count?

[Firelord]

I think for nForce3 the magic number is x-x-x-10

[bouncin]

Cheers for clearing that up rave. my bad i just assumed every1 knew what it was.

[Thorsson]

Well I do now...

Anyway I've confirmed that with my board changing TRAS has next to no effect. However lowering TRTW (or TRWT as it is in the BIOS) plus TRC and TRFC to a lesser extent, gives some nice gains in bandwidth - I got an 11% gain overall in Memtest86.

[Lee H]

Anyone got similar tests/number for A64s?

http://www.anandtech.com/cpuchipsets...oc.aspx?i=2330

They have memory tests and show the increases in FPS depending on the memory speed.

What makes me laugh is a A64 3200 (2.0Ghz) is better in some of the tests than a 3.6 Ghz P4