45nm crunchers
24 Jan 2008 14:28:44 UTC
Topic 193466
(moderation:
Got Intel 45nm stuff? Post results and considerations here please!
I just upgraded to Intel "Wolfdale" E8400 3GHz 6MB Dualcore. These guys run cold, I'm currently running it at 4.2GHz with 0.2volts above stock under watercooling and the radiator fan blows out almost cold air while crunching E@H on both cores. Very polar cap friendly processor indeed :D
Im expecting E@H times to go from ~17,000 down to or below 14,000 for similar WUs, very significant speedup at much lower power consumption, performance per watt is no less than amazing.
The Host: http://einsteinathome.org/host/1096348
nb: Im still playing with overclocking and stability testing, i usually try to not use Boinc for testing my overclocks but accidents happen, so trashed WUs could occur.
45nm crunchers
)
Sound good to me!
There might be better test to see if overclocking is still within safe margins, but E@H seems to be nice for testing thermal control. I was observing the temperature on my Pentium M @ 1.5GHz lately (using the new Linux Power User App). Under Linux, this is as easy as
cat /proc/acpi/thermal_zone/*/temperature
The Pentium M can sustain up to 99° C so I was not concerned at all about the 80 degrees result, but I was concerned when in the middle of the result temperature suddenly dropped by 10 dregrees, and the fan slowed down. What had happened????? Simple enough: BOINC had decided it was time to do a litlle benchmarking and suspended the E@H app. Obviously the BOINC benchmark doesn't stress the CPU nearly as hard as the E@H app itself.
I'd be very much interested in more results for the new 45nm Penryns. People have overclocked this CPU to really insane levels :
CU
Bikeman
The Boinc benchmark is quite
)
The Boinc benchmark is quite outdated, good example is Rosetta and the old credit system there when someone at Team XS wrote an SSE optimized mod for boinc. The original bench dont use much of a modern CPUs capabilities.
TEMP: I use icspll for on-die sensor thermal monitoring in Linux, E@H cant make it as hot as sprime2414 but its putting a decent stress on the CPU. Off topic, Icspll is quite good, Linux users who dont need the fancy stuff from lmsensors can check it out: http://sourceforge.net/project/showfiles.php?group_id=182331
Just unpack, cd to icspll-1.1/cpucoretemp, type ./run (on some distros you should do this as root), then after that check temps with cat /proc/cpucoretemp. Add the cd and ./run lines it to /etc/initrc to autoinstall it on boot.
About my Wolfdale again, i already seen E@H times just above 12,000 sec for 236.47 credit WUs (4.27 power users app), and the CPU is still at (only) 4.2GHz. 12,200 sec for 236.47cr is more than 1 credit per minute, per core.
Team Philippines
RE: I'd be very much
)
Am I correct in assumin that is a liquid nitrogen benchmark?
RE: RE: I'd be very
)
Yes!
LN2 as the big boys in
)
LN2 as the big boys in overclocking call it. Allegedly the 6GHz mark has been passed with 45nm quads under LN2 cooling.
My E8400 rig just finished 2 WUs at 4.32GHz, voltage is now up to 1.525, not worth it for a poor ~3% performance increase, the crunchtimes are still above 12,100, the increased heat is much more noticeable than the performance gain. Im going back to 4.2GHz and 1.425v to be more energy efficient.
Team Philippines
My waterpump died, currently
)
My waterpump died, currently running stock speed (3GHz) with a boxed cooler, its still pretty fast and coretemps are around 43-46c with this crappy aluminum cooler. I think this E8400 outperforms E6850 at same clock speed, and for performance per watt its a one man show. These results were run with 800MHz RAM at very loose timings (5-5-5-18):
http://einsteinathome.org/task/91673612
http://einsteinathome.org/task/91677200
I believe the low power consumption has almost nothing to do with the 45nm production process, its almost entirely because Intel basically reinvented the transistor with hafnium High-K/Metalgate technology. Should not expect a similar drop in power consumption from the 45nm AMDs coming later this year, smaller process = more leakage, AMD (and IBM) will have their first 45nm chips made with traditional gates, it could be well into 2009 before they implement hafnium into their production.
Team Philippines
RE: TEMP: I use icspll for
)
AWESOME. I'm going to give that a shot since I couldn't get lmsensors to work on my old crappy Dell systems. It'd be nice to have a good handle on how fast I'm murdering these systems.
RE: AWESOME. I'm going to
)
Old crappy Dell as in Pentium 4? Sorry, those processors dont have ondie sensors, and icspll reads only from ondie sensors.
Did you check in /proc/acpi/thermal_zone? You could be lucky and find something there.
Team Philippines
Well folks, I finally made
)
Well folks, I finally made the plunge. I've just built a fancy new 45-nm Yorkfield Extreme Edition machine, and put it into operation yesterday evening. It's impressive, even though it's still running at stock speed.
Yorkfield results
I'll try overclocking it later, after giving it a good burn-in time.
Of course, lest any of you think that I'm made of money. . .
This was an "open-box" special from Newegg. So, the savings was considerable.
RE: Well folks, I finally
)
Very nice rig, looks like its able to reach a RAC of 5000+ without overclocking.
You are using a default 32bit no PAE kernel with 4GB RAM i think, i would install a bigmem kernel + the kernel modules you need with apt, that will give you ~3.96GB RAM. To bad you will not get that option when installing from an Ubuntu CD, but the debian DVD will give you bigmem kernel option if you start the install with the expert option (just type expert at the boot prompt).
Team Philippines