Hardware Canucks

Hardware Canucks (http://www.hardwarecanucks.com/forum/)
-   HardwareCanucks F@H Team (http://www.hardwarecanucks.com/forum/hardwarecanucks-f-h-team/)
-   -   X79, ram, and f@h (http://www.hardwarecanucks.com/forum/hardwarecanucks-f-h-team/50920-x79-ram-f-h.html)

G777 January 25, 2012 11:35 AM

X79, ram, and f@h
 
Thought this was interesting, while fooling around last night with a little overclocking I borrowed a couple sticks from my X79 rig to run some benchmarks on P67. I left the rig to fold overnight with only the two remaining sticks, and woke up this morning to a significant drop in PPD.

The socket 2011 machine is a 3960X@4.7Ghz paired with 4x4Gb of G.Skill 2133 ram cas 9. So 16GB of memory running quad channel.
I've had nothing but P6904 units for weeks so I'm pretty familiar with it's performance. (Numbers are running Linux client via VirtualBox.)

TPF is around 51mins, PPD ~114,00

So overnight the rig was running in dual channel, 2X4GB of the same 2133 c9 ram, memory usage was under 6GB in total so it wasn't maxed out.

TFP rose to 61 mins, PPD sunk to 95,000

This was a bigger hit than I would have thought going from dual to quad channel, maybe there's something else going on that I'm missing, but thought I'd share.

Dead Things January 25, 2012 12:07 PM

Quote:

Originally Posted by G777 (Post 594578)
This was a bigger hit than I would have thought going from dual to quad channel, maybe there's something else going on that I'm missing, but thought I'd share.

We saw something similar with the 2P and 4P G34 rigs insomuch as there was a considerable amount of performance to be gained from higher frequency/lower latency RAM in quad-channel. It's interesting because that was essentially the first platform to witness a gain in FAH performance from RAM. By comparison, LGA1366 provided no discernible difference in switching between tri-channel or dual-channel.

Having said that, I share your surprise. I figured part of the reason why memory bandwidth made as much of a difference on G34 is because each processing thread needs its own address space, so a large MP system would depend more heavily on RAM performance. Your findings, though, would suggest it has more to do with the quad-channel bandwidth itself being particularly agreeable to FAH. This is indeed interesting. Thanks for sharing!

Tim_H January 25, 2012 12:11 PM

I know that it's an issue with a AMD G34 setup, regardless of the number of processors in it so it makes sense I guess.

edit: Ninja'd by DT

chrisk January 25, 2012 12:44 PM

Interesting...I would have said prior to your post it makes no real difference.
http://3.bp.blogspot.com/_b8suo6dUxR..._you_know2.jpg


All times are GMT -7. The time now is 06:47 AM.