I need someone to confirm my theoretical findings

[flexy]

I posted this over at xtreme already, but noone replied yet what my theoretical calculations showed - and i want to know whether this reflect to real world results also. I cant test myself since my build is on the way and not finished yet.

I did some preliminary calculations which i based on that *excellent* excel table at http://www.thetechrepository.com/showthread.php?t=195


Given:

Q6600 might do 3600Ghz
memory might to 500+ mhz @ CAS5
+ memory could do 400-450 mhz at CAS4
Q6600 max. realistical FSB 450

My findings are now that there are the following options if goal is hitting 3600 on a Q6600:

[514x7 CAS5 9.7ns lat 8,227MB/s bandwidth (CAS6 11.7ns)] <- not possible!!

500x9 DIV4:5!!! CAS5 10.0ns lat theor. 8,000MB/s bandwidth (FSB@400) !!!!
450x8 DIV1:1 CAS5 11.1ns lat 7,200MB/s bandwith (CAS4 8.9ns)
400x9 DIV1:1 CAS4 10.0ns lat 6,500MB/s bandwidth

According to this (with results from the excel table) the BEST way to overclock with a goal of 3.6 Ghz would be using divider 4:5 with 9x500, FSB at 400 and memory at 500/CAS5

This would result in pretty ok latency of 10.0ns and pretty good theoretical mem bandwidth of 8000MB/s.
I need someone to confirm whether those findings show in (cough) real life performance, eg. what everest says....


Eg. keeping it "conventional" with eg. 9x400 1:1 at CAS4 would be not enough bandwidth...514 1:1 the CPU would never do, and 8x450 CAS5 would result in a bit higher latency and "only" 7,200 bandwidth.

So..agree with my findings? Whats the deal now with "1:1" always being better if in fact 5:4 here gives the best results?

G.

 

[sutahz]

I may be beating a dead horse here, but do you not care that having fast memory w/ tight timings offers negligible performance gains? Or is this just a personal project of yours to have fun with and give you a warm and fuzzy inside knowing you have lots of bandwidth?

"be using divider 4:5 with 9x500, FSB at 400 and memory at 500/CAS5" should be
"be using divider 4:5 with 9x400, FSB at 400 and memory at 500/CAS5" if you care (9x4=36, 9x5=45).

 

[flexy]

what are "negligible performance gains".

Do you know that there are MANY people who run at 9x4 having a THEORETICAL bandwidth of 6,500 because its said left and right that 1:1 divider is best. Just using a divider, keeping FSB at 400 and use memory at 500 gives you 8000 (now i actually typed (=== .)...but this has nothing to do with an e-penor.

It was just calculations i did based on that excel table and mainly for the reason to decide what memory to get, whether DDR800 CAS4 or DDR1000 or whatever.

I also do NOT consider a bandwidth increase from 6.500 --> 8000 "negligible"..respective would be a retard if i wouldnt use the setting which gives me max performance....thats what we OCers usually do

 

[sutahz]

http://www.bit-tech.net/hardwa...2_duo_memory_timings/1

http://forums.guru3d.com/showthread.php?t=199461 (post #3)

lets not forget about our host,

http://www.anandtech.com/printarticle.aspx?i=2813

maybe you'll get something outta all this too:
http://www.diy-street.com/forum/showthread.php?t=70031

But yeah, you'll see 1-2% performance increase in the real world seems to be a pretty accurate consensus.

I'll reitterate though, if you can run tighter timings at the freq. you end up at, why not, its free performance. And if you want to run your ram as fast as you can to get that bandwidth thats great. Its what you find fun, its your computer. I was simply beating the dead horse of "memory speed/timings has negligible (1-2% increase) effects on your computers performance.
I've shared my view, sorry if you don't agree w/ it, I wont post again in this thread.

 

[Conjugal Visit]

Originally posted by: sutahz
http://www.bit-tech.net/hardwa...2_duo_memory_timings/1

http://forums.guru3d.com/showthread.php?t=199461 (post #3)

lets not forget about our host,

http://www.anandtech.com/printarticle.aspx?i=2813

maybe you'll get something outta all this too:
http://www.diy-street.com/forum/showthread.php?t=70031

But yeah, you'll see 1-2% performance increase in the real world seems to be a pretty accurate consensus.

I'll reitterate though, if you can run tighter timings at the freq. you end up at, why not, its free performance. And if you want to run your ram as fast as you can to get that bandwidth thats great. Its what you find fun, its your computer. I was simply beating the dead horse of "memory speed/timings has negligible (1-2% increase) effects on your computers performance.
I've shared my view, sorry if you don't agree w/ it, I wont post again in this thread.

Nice links. Well thought out and well done. Since I carry a sig around, you'll note the 4-4-4-12 timings @ 1:1 ratios. Why? Because I can and still maintain a moderate O/C at stock voltages (1.3500 vcore). I completely agree with the links you've shared and others like it; I'm just not going to "see" 1-2 fps bumps, particularly if my rig is already providing 60+ fps in whatever game I run. Ratios - running synchronous - continue to be important to me, primarily because I haven't seen a wealth of articles blowing that belief out of the water - yet. So in my case, my memory decision was based upon the CPU, it's multiplier, and the O/C that I wanted to achieve, balancing performance and heat and noise. Running good memory, with a real functioning heatspreader, and bumping it from 333 to 440 @ 2.1V (SPD value at 400MHz) has certainly "worked" for me. I'll take the 1-2% performance boost running tight timings if it's available. If it isn't available - as you've said - it's 1-2% real world.

Stating the obvious, though, these discussions (and the pursuit of every little performance increment) is what sustains the hobbyist, and of course, forums like these. I enjoy them.

Regards,