Eclipse's Memory Bandwidth Contest!

Heligrin said:
Whats with all you people and your opterons
Click to expand...
i have an E6 venice on the way. opterons are overrated ;)


justin82 said:
figured it would be nice to get on the list for a little while b4 all the heavy hitters start bringin the bh5 and opterons out. :p
Click to expand...
we're not doing a latency test here, bandwidth. tccd and some micron will own the voltage loving ram in the face


Budwise said:
how about we seperate these into a 2GB, vs Other category. Us 2GB guys cannot compete...
Click to expand...
i'll show you something in a while that proves otherwise.. 2x1gb definitly can compete. what really needs to be done is seperate things into single channel, dual channel and NUMA capable :p
 
(cf)Eclipse said:
i have an E6 venice on the way. opterons are overrated ;)



we're not doing a latency test here, bandwidth. tccd and some micron will own the voltage loving ram in the face



i'll show you something in a while that proves otherwise.. 2x1gb definitly can compete. what really needs to be done is seperate things into single channel, dual channel and NUMA capable :p
Click to expand...

Opterons are overrated, stick with teh 3700 san dieg0z0rz :p

Where/HOW do these people get 300 mhz frequency ram modules :eek:
 
Heligrin said:
Where/HOW do these people get 300 mhz frequency ram modules :eek:
Click to expand...
easy, just know where to look. i've got 4gb of ram sitting within 3 feet of me that can easily break 300mhz on an opteron that doesn't like going over 305 ;)
 
(cf)Eclipse said:
easy, just know where to look. i've got 4gb of ram sitting within 3 feet of me that can easily break 300mhz on an opteron that doesn't like going over 305 ;)
Click to expand...

IN THE FACE!111
 
(cf)Eclipse said:
easy, just know where to look. i've got 4gb of ram sitting within 3 feet of me that can easily break 300mhz on an opteron that doesn't like going over 305 ;)
Click to expand...
I like, but when you mean look, what exaclty do you mean? :D
 
just do some research, and have a rig capable of doing 300mhz.. there are some chips that are known to do pretty high clocks, samsung TCCD and TCC5 as well as most well binned micron (mainly ballistix)
 
Best i could muster...

Sandra%20Bandwidth.jpg
 
Well here is my somewhat dated machine but hey I made the top 10 as they are listed right now.. and that is with 2x1gb of not so fast ram!

This is where I started:
Sunin%20Memory%20Tweak%20Run%201.JPG

This is my final after all tweaks and settings were played with... a bit of an improvement
Sunin%20Memory%20Tweak%20Run%203.JPG

Now to rerun 3dmark05 and see if I got any performance boost from the tweaks!
 
easy, just know where to look. i've got 4gb of ram sitting within 3 feet of me that can easily break 300mhz on an opteron that doesn't like going over 305
Click to expand...

Have a weird feeling they might be tracers... :p
 
A few hints for people trying to eek out some speed... this is what worked for me.

1. Play with the Dram Latency parameter... mine was initially set to 01, but 11 added about 20-30mb/s

2. Refresh Mode Select (RMS) parameter move this around and retest each time... I gained about 50-60mb/s by moving this from the default of 7.8u to 64u

3. Tweak the FSB, find out where your sweet spot is... get this... at 250x11 was slower then 230x12. This gained me about another 50mb/s

I started at about 5318, and ended up at 5481 in about an hour or two. Ths also helped when I ran superpi 1M run dropping me down to 26.797s

Just a few hints to help people...

P.S. what had not effect to much of my surprise was changing from 3,4,4,8 to 2,3,3,5. I changed 1 parameter at a time one step down at a time and saw no significant change in bandwidth!!! Which kinda made me look at my screen crosseyed!
 
MC FLMJIG said:
Have a weird feeling they might be tracers... :p
Click to expand...
nope, 2x512mb g.skill LE, 2x512mb crucial ballistix, and 2x1gb teamgroup xtreem :cool:


i'm outta time for now though, i'll update when i get back from class
 
Once I remount my loop with it's hella pink solution, I'll show you some high voltage memory owning you in the face! But first I have to go get braces. :( :mad: No pretzels for 2 years! So lame.

Full Folding Ahead!

 
Sunin said:
P.S. what had not effect to much of my surprise was changing from 3,4,4,8 to 2,3,3,5. I changed 1 parameter at a time one step down at a time and saw no significant change in bandwidth!!! Which kinda made me look at my screen crosseyed!
Click to expand...

This should not surprise you. Timings are a function of latency, not bandwidth.
 
Stellar said:
This should not surprise you. Timings are a function of latency, not bandwidth.
Click to expand...

so what then produces higher bandwith if not timings?

i think eclipse started this as a reverse psych way of gettin us to get our systems clocked right. kinda like how the best teacher you had in school would give you outlandish but fun tasks to learn some boring bit of physics or chem or math or englsh even. that eclipse... such a tricky guy
 
actually timings do effect bandwidth, just not a huge amount in sandra. a64's bandwidth is more sensitive to timings than p4's, because it changes the latency to a higher percentage.

it's mostly depenant on memory mhz though.. faster ram = higher theoretical bandwidth
lower latency = higher efficiency, thus more bandwidth is utilized properly :cool:
 
yea cuz i got my bandwith up to 7200/71xx by fuking around in tweaker last night man. did the stuff like tighten up max async, and the seecondary timings. got my 1M times to low 28's with it.

get on AIM eclipse ;)
 
(cf)Eclipse said:
actually timings do effect bandwidth, just not a huge amount in sandra. a64's bandwidth is more sensitive to timings than p4's, because it changes the latency to a higher percentage.

it's mostly depenant on memory mhz though.. faster ram = higher theoretical bandwidth
lower latency = higher efficiency, thus more bandwidth is utilized properly :cool:
Click to expand...

No longer running a P4, I'm onto PM's Dothan 760 to be exact. I have DDR4000, but the mobo is not letting me get a stable clock much above 240... :( Infact the 240 times are slower then the 230 times... shakes head...

But yes I did expect as you say eclipse to see a little more movement when changing the latency... no dice though... I'm literally talking less then 5mb/s from going 3448 to 2335
 
(cf)Eclipse said:
actually timings do effect bandwidth, just not a huge amount in sandra. a64's bandwidth is more sensitive to timings than p4's, because it changes the latency to a higher percentage.

it's mostly depenant on memory mhz though.. faster ram = higher theoretical bandwidth
lower latency = higher efficiency, thus more bandwidth is utilized properly :cool:
Click to expand...
I wouldn't use Sandra to measure bandwidth. I mean, it's purely synthetic. I had a P3, that Sandra thought ran at 10GHz for some reason, and I had sky-high bandwidth scores to go along with that. That orange bar actually went out of the window all the way accross the desktop...
 
el rolio said:
so what then produces higher bandwith if not timings?

i think eclipse started this as a reverse psych way of gettin us to get our systems clocked right. kinda like how the best teacher you had in school would give you outlandish but fun tasks to learn some boring bit of physics or chem or math or englsh even. that eclipse... such a tricky guy
Click to expand...

Speed (Mhz) produces bandwidth. As eclipse stated, improved timings use available bandwidth more efficiently, which is effected in synthetic memory bandwidth scores, but do not increase available memory bandwidth.
 
New picture as I didn't read the rules thuroughly enough... Gained some points while I was at it :)

 
iddqd said:
I wouldn't use Sandra to measure bandwidth. I mean, it's purely synthetic. I had a P3, that Sandra thought ran at 10GHz for some reason, and I had sky-high bandwidth scores to go along with that. That orange bar actually went out of the window all the way accross the desktop...
Click to expand...
ah, i normally don't either, but it's a fun activity. don't spoil it for the others ;) :cool:

BLiTzKRiEG, i guess we can't follow the rules so everyone knows the settings you used to get there? the reason i wanted you guys to post your settings is so that other can work off what we have set to try to help them figure out how to tweak their own rig ;)
 
Yea, I didn't read enough... lol

I repasted the pic in my first post,... here it is again though.

 
Ok. Just received my G.Skill with black spreaders, DDR 500.
These actually did:
2.5 3 2 @ 2.65 200 FSB (almost what the baby blue ones do ;) )
3 3 3 @ 2.6 250 FSB
The top is what you see here.

 
nice mc, though can i ask you to edit that pic with one that shows memory timings too? i'm presuming 3-4-4, considering it's UCCC

it's updated :D
 
Anyone want to help me find out why im scoring so low? I just scored the following

int: 4518
float: 4543

This is with gskill hz's running at 250mhz 3-4-4-8-1t and an opteron 170 at 275x10. The mobo is a dfi rdx200. I wouldnt have known that it was so slow until i saw this thread and ran the tests :(.
 
hmmm, a bit too high to be in single channel.. are you running the unbuffered test by chance?
 
You must log in or register to reply here.
Back
Top