Work in progress - just rerunning everythig with 7/7/7/14....pics coming very soon
OC Report: Cellshock DDR3 1866 BLUE D9JNL
I have spend some quality time with a very nice set of Cellshock's newest line of DDR3 and was amazed to find out what nicely binned and matched Micron D9JNL can do. Unlike the Micron D9GTR which is scaling extremely well with Vdimm , the D9JNL chips are really shining at low Vdimm by outperforming the D9GTR chips at equal voltages. The Micron D9JNL chips, which most benchers consider to be inferior, are actually needing 0.1V - 0.2V less for equal clocks.
The Memory:
Part number : CS3222770
Capacity: 2GB-Kit (2x 1GB)
Configuration: 128Mx64 DRAM Micron Device
Config: 128Mx8 PCB 240pin
Module Info: unbuff. non-ECC, DDR3 DIMM
Speed: PC15000
CAS Latency: 8-8-8-16
VDIMM: 1,7V-2,0V
Warranty: 5years
The test setup:
OS: Windows XP SP2
CPU: Intel Xeon E3110 – air cooled
Mobo: Asus Maximus Extreme – Bios 9.05 - NB water cooled
PSU: PC Power & Cooling 750W
GPU: NVidia 9600GT / 9800GX2
HDD: Gigabyte I-Ram
The Windows XP installation is 100% bone stock without any tweaks, with the exception of having Large System Cache enabled for 32m runs.
The testing was done with the mindset for stability for Super Pi as well as 3D- and PC-Marks and NOT for max results ( for now...)
Super Pi Results:
Single Super Pi 1M CAS7 1T clocks:
cpu limited (air...) I think -
Single Super Pi 1M CAS8 2T clocks:
on P5E3 Premium
Single Super Pi 32M CAS7 1T clocks:
VDIMM: 1.80V = 1775mhz
VDIMM: 1.84V = 1810mhz
VDIMM: 1.90V = 1870mhz
VDIMM: 1.94V = 1910mhz
VDIMM: 2.00V = 1940mhz
VDIMM: 2.06V = 1998mhz
VDIMM: 2.08V = 2004mhz
Secondary timings are often not tweaked to perfection – often these are too loose. For example tRFC could probably have been tigher at lower mhz - and really needed lossening up at high mhz crappy vid of running 1992 2.06v @ http://www.youtube.com/watch?v=kVqnxPJuKh0
Single SuperPi32M CAS 8 1T clocks:
My Asus Maximus Extreme unfortunately appears to not be able to go past 2080 – I have had this problem with many different sets of memory.
Single SuperPi32M CAS 6 1T clocks:
VDIMM: 1.80V =
VDIMM: 1.84V =
VDIMM: 1.90V =
VDIMM: 1.94V = 1680mhz
VDIMM: 2.00V =
VDIMM: 2.06V = 1750mhz
VDIMM: 2.08V =
I have to revisit all my CAS6 results and post them later.[/INDENT]
Memory Bandwidth Results:
Everest Mem Read :2000mhz 7/7/7/14 @ 2.06V
Everest Mem Write: 2000mhz 7/7/7/14 @ 2.06V
Everest Copy: 2000mhz 7/7/7/14 @ 2.06V
Everest Latency: 2000mhz 7/7/7/14 @ 2.06V
3DMark Results:
3Dmark05
VDimm = 2.1V 7/7/7/14 1T @ 2000mhz
Asus P5E3 Prem. / 9800GTX all default settings
PC Mark Results:
PCMark05 memory suite
VDimm = 2.1V 7/7/7/14 1T @ 2000mhz
Asus P5E3 Prem. / 9800GTX all default settings
NOTES:
-Actual Vdimm was about 0.01V higher than set in BIOS – all above Vdimms are BIOS set ( so +0.01V for real value )
-This set of Cellshock DDR3 scales very well until 2.07V but at 2.11V it errors out hard to and refuses to boot had high clockspeeds.
-The max CAS7 clocks were done with 2.09V , but in memtest already a lot more errors were showing than when running 2.07V.
-This set does not like different CAS and tRCD timings, which I believe is common for this type of micron chips – hence the 7/7, 6/6, 8/8 timings used.
-Unfortunately CAS 6 results suffer a bit compared to D9GTR equipped memory, due to D9JNL's problem with handling higher voltages, which are needed for great CAS6 results.
-The Microns D9JNL allow running very tight Tras - 7/7/7/12 timings are possible to be run at almost the max clocks ( with between 4-16mhz hit in max frequencies compared to Tras 21), Tras 14 appear to be quickest for SuperPi 32m, Tras 12 is about 10 seconds slower compared to Tras14; Tras21 is about even with Tras 14
Cellshock DDR3 1866 compatibilty:
I have tested the Cellshocks on three different boards – the Asus P5E3 Premium ( X48), the Asus Maximus Extreme (X38) and on the EVGA 790I board. On all boards the memory ran without a hitch – but the X48 board really made this memory shine. Unfortunately I lost the board at the beginning of the testing due to extreme NB degradation and am not able to run with this board past 450FSB at this point.
I had pushed too much VNB; 1.89-1.91V on the P5E3 Prem. (with a very brief stunt of 1.97V to see if helps when things started going bad...to make things really really worse ...lol), while air cooling... ouch !
The Maximus now is on water cooling – which helps out greatly!
Memory Clocks were about 5-20mhz higher on the P5E3 Prem. vs. the Asus Maximus Extreme.
The X48 also allowed me to run 7/7/6 with tighter secondary timings, whereas the X38 board did not appear to like these primary timings at all.
The 790i is a decent performer, but the Intel chipsets are a bit better in terms of memory performance. Max mem clocks were also bit lesser for me on the EVGA 790i compared to the X38/X48 boards. I did not spend too much time on the 790i since massive OS corruption problems on this board ( Rev C0...hint) took all the joy out of memory clocking. With more tweaking I assume similar max memory clocks might be achievable.
Final thoughts
Compared to what Vdimm is necessary to match such clocks with older D9GTR Micron based memory, this trend is certainly a step forward for the real world user. Only for benching FSB limited Quad Core CPUs it could be preferable to have D9GTR based memory, where good sets can allow you to bench in the 1800+ region with CAS 6 timings. Something I still would love to also have in my arsenal.
All in all I really love this Cellshock set, it is a match made in heaven for a 24/7 high performance rig. The speeds that are obtainable with low voltages make them truly perfect for a insane gaming rig and/or bench systems which are running at the max all the time. Being able to run 1900mhz CAS 7 for my daily settings is more then I hoped for, especially without having to worry about potentially degrading the memory.
Mike
Bookmarks