OC Report - Cellshock DDR3 1866 BLUE D9JNL

[saaya]

makaka, can you post some everest scores?
what chipset is that with? x38?

[camouflage]

Was a lot of work, really, but finally I got these sticks 32M stable at 1.92 Vdimm on Maximus Extreme:

[makaka]

makaka, can you post some everest scores?
what chipset is that with? x38?

The chipset is the good old p965 . motherboard asus p5b-e plus .
memory timing 5-5-5-10 (1000 mhz)

Was a lot of work, really, but finally I got these sticks 32M stable at 1.92 Vdimm on Maximus Extreme:

As reference here is a screen shot of my current super pi @ 4 ghz 500(5-5-5-15 , 1:1 ) x 8 .
11.29 vs 12.33 (nice DDR 3 boost)

[saaya]

The chipset is the good old p965 . motherboard asus p5b-e plus .
memory timing 5-5-5-10 (1000 mhz)
As reference here is a screen shot of my current super pi @ 4 ghz 500(5-5-5-15 , 1:1 ) x 8 .
11.29 vs 12.33 (nice DDR 3 boost)

hmmm weird, i thought pi gets a bigger boost from extra mem performance...
you make it sound like your ddr2 1000 555 is as fast as ddr3 at 2000, well, its definately not
what this really shows what we all knew already, pi has become a cpu cache benchmark

compare the everest results, mike has a 50% higher read bandwith and 50% lower latency...

[makaka]

hmmm weird, i thought pi gets a bigger boost from extra mem performance...
you make it sound like your ddr2 1000 555 is as fast as ddr3 at 2000, well, its definately not
what this really shows what we all knew already, pi has become a cpu cache benchmark

compare the everest results, mike has a 50% higher read bandwith and 50% lower latency...

well i am trying to find now the exact equation for DDR2 vs DDR3 in super pi .
Currently 4100 mhz with 512 mhz memory (5-5-5-15) was slower , i am re benchmarking now with 4150 mhz (518 mhz same latency) . let see how much boost and cpu frequency give this nice DDR3 .
Now 4100 mhz cpu + 512mhz ram (5-5-5-15) < 4000 cpu + 1000 mhz ram (8-8-8-20)

[makaka]

4150 mhz + 518 mhz memory should result in slightly better result . But my memory crapped out in the loop number 17.
This geil memory do not like high voltage with 2.3v this memory was instable @ 518 mhz and even @ 500 mhz while it is completely stable @ 500 with 2.1 v.
So test proof that :
DDR3 1000 (8-8-8-18)+ 500x 8 = DDR2 518 (5-5-5-15) + 518x 8 cpu .
about (+140 mhz) cpu + DDR2 (500) is equal to 1000 mhz DDR3 memory in super pi .

[mike]

Here's a 16M - sadly I need tRFC 82 for full stability:

unfortunately I am not able to view the image of your run atm - but don't worry to much about trfc - you won't loose too much as long as your mobo is all nice and tight

hmmm weird, i thought pi gets a bigger boost from extra mem performance...
you make it sound like your ddr2 1000 555 is as fast as ddr3 at 2000, well, its definately not
what this really shows what we all knew already, pi has become a cpu cache benchmark

compare the everest results, mike has a 50% higher read bandwith and 50% lower latency...

965 is like pre y2k technology -waaay slow in mem perfromance - forget about comparing everest.

Malaka - it's hard to compare 1ms; or in fact any SuperPi without knowing what OS tweaks you and the other person are using. It's easy with a couple of OS tweaks to smoke someone who is running a "much faster" base system, keep that in mind comparing X with Y.
1m is all about CPU cache...



But you're right Saaya -seems like SuperPI is turning more and more into a CPU cache benchmark - it's hard measure real world performance improvement from memory bandwidth with CPU cache taking over a lot of the work; espcially when you are trying to tighten your secondary timings, for tests like SUperPI. Maybe we need 64m test

never used DI/LN2 myself heh

nordichardware LN2 pipeline LOL

These 1866 are looking pretty good if you can match it with right motherboard it seems from these results

Good that you don't - you'd prolly smoke most of us if you were to go cold!

Yes the right motherboard is really key for DDR3 performance - I just received 2 more sets of D9JNL and are hitting pretty much the identical memory clocks with those new sticks. It makes me really wonder how well this memory would perform with an even better board!

CAS 8, CAS9 or even CAS10 results could potentially blow our minds - if there was a board/NB which could cope with it.

[camouflage]

Now I got it 32M stable tRFC 72 @ 1.92 Vdimm:

[makaka]

unfortunately I am not able to view the image of your run atm - but don't worry to much about trfc - you won't loose too much as long as your mobo is all nice and tight


965 is like pre y2k technology -waaay slow in mem perfromance - forget about comparing everest.

Malaka - it's hard to compare 1ms; or in fact any SuperPi without knowing what OS tweaks you and the other person are using. It's easy with a couple of OS tweaks to smoke someone who is running a "much faster" base system, keep that in mind comparing X with Y.
1m is all about CPU cache...



.

For me there was no tweak just winxp 32 (daily use) with a tone of crap in the background .
What are the most common OS tweak for SuperPi .
I am really interested in DDR3 and if it worth it i will update

[KanGaXx]

someone can explain me what voltage for default setting mean ?
i am explaing better: in the cellshock box i found: 1866mhz at 1,7 / 1,9V.
It mean that i can find modules that can stay at 1866 with 1,7 stable or modules stables at 1866 at 1,9v ?
or is only a range for warranty ?

[mike]

someone can explain me what voltage for default setting mean ?
i am explaing better: in the cellshock box i found: 1866mhz at 1,7 / 1,9V.
It mean that i can find modules that can stay at 1866 with 1,7 stable or modules stables at 1866 at 1,9v ?
or is only a range for warranty ?

1.7 nor 1.9v is NOT touching potential warranty issues. Would you mind to show us some screenies of your runs??? Maybe we can get more suppportive as a community after we see what TF you are talking about

[saaya]

its actually 1.7v-2.0v
and yes, of course some need more volts than others, but yeah, the 1.7-2.0v is mostly about warranty
All kits should run 1866 888 with 1.8v, some even with 1.7v

[mike]

well
i wanna get ddr3, i use liquid nitrogen , phase change ecc.
whats better for me to use with e8xxx and qx9650, this new blue 1866 or old 1800 ??

best to wear protective gear!!!!

[maxxor]

Why is my bandwidth so low ? bios p04 and p1 p2 enabel

[Kasparz]

best to wear protective gear!!!!

Thats why you always get better results

[DMH]

Why is my bandwidth so low ? bios p04 and p1 p2 enabel

For those speed I think they're good

[maxxor]

Thanks a lot Mike - will keep on trying. My Vfsb is 1.44 - the board needs that much.

NB is under chilled water - no temp probs at all, ~30°C @1.66 Vmch.

Your last comment is funny - it just happened to me. OCed Dominators removed and Cellshocks plugged in = no boot. Then I did CMOS clear, = no boot. tried this 5 times, = no boot. Has to install back the Corsairs again and set them down to default settings, = boot now !

Then I installed the Cellshocks again and all was fine....... - took me about one hour....

BTW, here's my screen:



12.5k read at 4GHz isn't bad - though it is only Cas 8.

compare whith this

[saaya]

wow, thats weird... your score sucks big time man! no offense!
i have no idea why... i DONT think p1 and p2 are really enabled...
This looks more like a score without p1 and p2... actually even then it looks very low

790 is such a weird chipset...
Ive seen some awesome bandwidth results at 1600 777 that were better than 1800 777...
Something is very weird about clocking mem and fsbs high on 790... it seems the performance actally gets worse the higher you go.
Dinos insists its not the case if you enable p1 and p2, but from what ive seen even with those enabled and tight timings, the higher you go, the lower the scores seem to drop..
Can yo run 1866 777 please and post that everst score?

[camouflage]

Mike - You still owe us the 7-7-7-14 pics on ME promised in Your first post......

[maxxor]

What is this now i am real confused

[saaya]

What is this now i am real confused

hmmm almost the same score as mike had on the x48 board at 2000 777...
well, what can i say... 790 is... weird...
what about 1866 888?
Id say go for 1866 888 and then up step by step and check at what point performance degrades. then use baredit and make a register dump at both speeds, before and after the perf drop.
use examdif or another tool to compare the dumps to see what register change.
then change one register after the other and check what happens... and with some luck you find what settings change that degrae the mem perf at higher speeds, then you can tweak the mem perf to scale even at high speeds.

is there a clockgen or something?
if yes, then use that tool to increase the fsb and memspeed from within windows. the perf should scale the same if you do this as register should remain at the same setting.

[maxxor]

8.8.8 18 no clockgen straight from bios

[saaya]

k... hmmm well idk man... id say... use clockgen and try to up with clockgen bit by bit.
That way perf should scale and not degrade the higher you get.
In case you have the time and are curious you can boot in steps and check how the performance degrades or changes, and at what point it changes.

[camouflage]

I never use clockgen or getFSB - all my results are bootable....

[mike]

Mike - You still owe us the 7-7-7-14 pics on ME promised in Your first post......

Give me a few more days - I am currently in Costa Rica, enjoying the beach


790i results...argh what a weird board. I think it very very important to have the right memory divider - prolly only linked really works well. I had lost lot of mem bandwidth going from 1600 to 2000, leaving everything else unchanged ( CPU , timings etc. ). Can't say much about 790i - just that it has been a pain so far - but I'll give it another shot soon