Why bother with huge bandwidth...

[RafaDel]

After reading the article below I'm wondering if we are not just waisting time and money chasing for high HTTs... Perhaps cheap DDR400 at 2-2-2-5 is just enough for A64s...

http://www.anandtech.com/cpuchipsets...oc.aspx?i=2469

[thephenom]

from DDR400 to DDR488 yield 5% performance difference.

VX at ~265 @ 2-2-2-6 is about TCCD/5 at 300 @ 2.5-3-3-8

So 44Mhz OC (DDR488) yield 5% increase, so 130Mhz OC (DDR530) would yield ~14.77% if it's a linear relations.

So really, if you OC to 300HTT with TCCD, you would get close to 15% performance gain. 15% is not much, but certainly can't be overlooked. (15% here, more % from CPU/GPU will give you more.)

[thephenom]

Lastly, good enough is never good enough.

[Trevor]

5% performance difference would be significant if you were in the top 10 3Dmark, for the rest its just fun to get the most out of your hardware.

[jumanji969]

By that reasoning most people here shouldn't bother with cascades, phase change units, watercooling, and about 90% of the other things we do to make our computers faster. It's not like I need my computer running at 2.5ghz, I just want it there.

[[XC] moddolicous]

Eery little bit counts. I like to know that my comp is running at its fullest.

[freecableguy]

As you can see from my quick graphical illustration, the difference in performance of the TOP system and the "performance" systems is actually very little. The cost is where the real difference lies...Law of Diminishing Returns. You will note that the biggest and baddest is always the lowest "bang for the buck." But that's why this is XtremeSystem.org and not "Cost-effectivesystem.com" eheheheh.

-FCG

[omga14]

being a s754 guy i would have liked to have seen the impact of more bandwith in single channel setups. i'm waiting for the M2 before i build new again so maybe things will change at that time. for now though being single channel bound makes me wonder. even though this article reasures me that i can run comfortably at DDR400 w/tight timings that really is never enough. or else that would mean all these rma's i've been doing have been for naught.

[pRS317]

i may just be stupid, but i think their results are due to testing the ram on dividers. from what i've seen, running ram on anything but 1:1 lowers the efficiency. most of us want the ram that's capable of running high htt so that we can keep our ram running 1:1 with our oc'ed cpu which would allow the higher bandwith to shine a tad brighter.

[Ugly n Grey]

i may just be stupid, but i think their results are due to testing the ram on dividers. from what i've seen, running ram on anything but 1:1 lowers the efficiency. most of us want the ram that's capable of running high htt so that we can keep our ram running 1:1 with our oc'ed cpu which would allow the higher bandwith to shine a tad brighter.

Running 1:1 means nothing on an S939. Run a search here on the board for details.

[charlie]

As you can see from my quick graphical illustration, the difference in performance of the TOP system and the "performance" systems is actually very little. The cost is where the real difference lies...Law of Diminishing Returns. You will note that the biggest and baddest is always the lowest "bang for the buck." But that's why this is XtremeSystem.org and not "Cost-effectivesystem.com" eheheheh.

-FCG

Yeah, well done. It's true!

C

[ozzimark]

Running 1:1 means nothing on an S939. Run a search here on the board for details.

or on s754, because it never really is "1:1"

for a quick explaination:
the memory controller runs at cpu speed, with a divider setting the ram speed. the lowest divider is 5x, so the HIGHEST ratio you can use is 1:5.. typically it'll be around 1:9-1:12 in normal overclocks

[SlackeR]

And by the way, normally we increase the CPU speed, too, which increases the performance gain from higher bandwidth.. Not by much, but as they said

[i found nemo]

i'll agree with all of the pro statements, what is the point of having capable components and not releasing the potential. it's all for fun and tweaking, consider the name of the forum for further referance.

[fareastgq]

seeing the bling in my case and then seeing the performance to back it up.. well,... it's .... cool and refreshing... tastes great and less filling. Who cares about diminishing returns :P

[harpyboy]

i'll agree with all of the pro statements, what is the point of having capable components and not releasing the potential. it's all for fun and tweaking, consider the name of the forum for further referance.

and considering the price u paid for the premium components

[Truckchase!]

I've been debating this as well given my recenly aquired addiction to BF2. 1GB just doesn't cut it anymore. :-(

[red_dragon]

haha the "why do we do this?" question always seems to come up. As others say, it's for the fun usually. And if you're a real hardcore comp user, it feels good to be able to say that your computer is faster than most other systems in the world. For the XS world record holders, they can say faster than all other systems in the world .

[HARDCORECLOCKER]

No need for discussion - we need DDR600 2-2-2-5..........

[SoF]

well i had 2 GB of bh5 but everytime i tried dividers the limit was ~265MHZ.
A Superpi 1m time with 280/250HTT/RAM at 2-2-2-6 1T was ~31,5s and my yesterday 280mhz run with my new ocz 3-4-3-8 2T was 280MHZ 1:1 34s ^^
So i supose i will need at least 300mhz ram clock to get my 31s again
I have a weak mem controller i guess but in 2 weeks i will have another 3200+ for testing here to see if it's really the limiting point.
But nevertheless we need strong memcontrollers first before hitting 2-2-2-5 @300mhz...
i tested my khx on 3 boards and 3 procs. the best one was hitting 255 2-2-2-5 1T on my k8nxp-9 and 3500+ so i know that my memcontroller on the 3200+ isn't that good .

Greetz SoF

[brandinb]

No need for discussion - we need DDR600 2-2-2-5..........

ok ill take a 2x1gb kit of that please oh ill take the 2.6v 1t model as well
________
E CIGARETTE SHOP

[RafaDel]

OK so afterall should we look for latency more than bandwidth?