1T vs. 2T

[STEvil]

Good thread on this going at anandtech, any of you guys up for some comparrative numbers?

http://forums.anandtech.com/messagev...VIEWTMP=Linear

As per rules/request in the other thread: Please list full system specs and do not change any values other than CMD rate.

Questions to answer:
L2 size effect on 1T vs. 2T performance?
CAS/TRCD/TRP/TRAS timings effect on 1T/2T performance?
1T vs. 2T max CPU and Mem clock difference?
2x1GB DS vs 4x512 SS/DS vs 2x512 SS/DS

Good benches to run:
Half-Life 2
Quake 3
Doom 3
Far-Cry
UT2K3/4
Cinibench
Kribibench
Aquamark
3D01 (exempt car low)
3D03 (min res)
3D05 (min res)
Novodex Rocket 1.1
Super-Pi
Hexus Pifast
etc..

Lets see some numbers

edit

added stuff

[13oost]

*nods* my RMA DFI UT 250gb should be back tomorrow from RMA, ill post some numbers as soon as i can, though my memory still needs a proper burning...

i can understand where 1T vs 2T can make a big difference in dual channel on the NF4, but single channel, that is interesting


OT::

might just be me, but every 3rd post is about timmings...

could a mod spend an hour or two and migrate all the timming related stuff to the Memory Tweaking guide section, since that is where it belongs, and maybe stick it there.... there is alot of valuable info and discussions getting pushed down and lost

[Dumo]

Not much different, exept a small gain (with 1T) on 3D benching,

[13oost]

can you give up some numbers? just to show hard facts and compare

[Dumo]

Heres....
With 2x512/1T...

[13oost]

thx ill post results as soon as i get my board in and get windows up and running...

[Dumo]

Man, I got cold boot again with this pain in the neck 'redline' Well, heres sandra cpu and cpumark results. I will post that 2T sandra mem. later...

4x512/2T...


2x512/1T...

[situman]

wow exactly the same numbers between 1T and 2T. The repeatability is almost impossible even with the exact configs. Something might be wrong with these pics.

[13oost]

its not exact, one is 4x512 at 2T the other is 2x512 at 1T, lol.... the extra gig makes up for the lost bandwidth on the 939

the 1T to 2T comparason seems to be best for 754 boards that do not have the dual channel option, most ram cannot make up for the bandwidth lost going to 2T, some can though...

but if 1T and 2T at the same frequencies have close to the same results in single channel, then us with socket 754 would run 2T since it usually allows for the extra mhz on the ram and affect them most...

[HousERaT]

I wonder if anyone is having that much success using 2T and 4 sticks of ram on a plain old non-FX San Diego?

[uscfan]

I guess that running 1T over 2T is only for record breakers, it doesnt seem that running at 1T is that much better. For bandwidth sure, but Sandra seems to be a pain in the as$. Sandra yeilds better scores with tighter timings and faster command rate, could it possibly be innacurate?

[13oost]

highly doubt it, 1T is better for Dual channel because it can handle it easier, but with single channel it has a hard time handling things at 1T, thats why the 1 gig of ram was almost exactly the same as the 2gigs at 2T...

imho this will affect the NF3 754 boards because the lack of dual channel...

my board came back in from RMA, this board smells aweful new to be a refurb... i guess ive been smiled on.... but hopefully it works properly... NONE of the cable packages has been opened, no crease on the manuel, the only thing is the cd has been removed because the flap is over the paper tag, but no smudge or scratch on it... the antistatic bag has never been opened as far as i can tell... just looks like someone got the board and decided they either wanted the new 939 NF3 or a NF4 board and returned it

ill post results within the next few days

just got everything hooked up and running, currently formatting

[STEvil]

couple figures

Winchester 3000+ @ 9x285
2x512 OCZ PC4800 Platinum 2.5-3-3-5
MSI K8N Neo2 Platinum

1T
Pifast 10M: 45.95 seconds
Diffmark 2.3: 106.59 seconds

2T
Pifast 10M: 46.73 seconds
Diffmark 2.3: 106.45 seconds.

If you're wondering why diffmark got faster, its because the CPU speed varied slightly (diffmark says it did by 25mhz, but cpu-z says it didnt). Might be related to the same reason (dunno the reason!) that DBench for D2OL does better with slightly slacker timings than tight ones also.. not really sure.

[Dumo]

Heres....
With 2x512/1T...

Sandra mem. result with 2T-4X512 .....Totally different story though...

[13oost]

yea, true

[STEvil]

bleah, sandra doesnt mean much

[dippyskoodlez]

bleah, sandra doesnt mean much

but what does something using that bandwidth mean?

a sudden loss of 2gb is a lot...

[Dumo]

Heres..
1T.....


2T....

[13oost]

currently giving my memory a good hard burnin, after i update to SP2 tomorrow and get all the patches and the necessary tweaks done, ill bost my DFI UT NF3 scores tomorrow

[EMC2]

Nice info Dumo, thanks bro

Strange that the L2 bandwidth went down like that though... TLBs apparently being affected by the 2GB vs 1GB apparently... would you mind running the 2T with the 1G? Thanks

Peace

[STEvil]

what bench program is that?

A linpack bench might prove usefull.. some in the anandtech thread made mention of 1t vs 2t effect on performance due to L2 cache size... maybe its worth looking into deeper?

[fatfreepork]

I guess that running 1T over 2T is only for record breakers, it doesnt seem that running at 1T is that much better. For bandwidth sure, but Sandra seems to be a pain in the as$. Sandra yeilds better scores with tighter timings and faster command rate, could it possibly be innacurate?

Its like cpc on or off on nf2. 230Mhz with cpc on is like 250Mhz with cpc off. and of course tighter timings and 1T will give better bandwidth.

[funkflix]

[pumbertot]

ive got 4x512mb PC4000 OCZ VX running atm.only runs in 2T, doesnt boot in 1T regardless of timings as we all know.will post up some screenies etc later but so far getting pretty much the same Superpi for 2x512MB 1T as 4x512MB 2T at same speeds and timings so looks like this is pretty much the norm.