Points Per Day: Windows SMP
Leonardo
Wake up and smell the glaciersEagle River, Alaska Icrontian
Gentleman, points per day (PPD) are looking mighty fine for SMP! I STRONGLY consider you trying the new SMP client if you run a dual CPU or dual core machine! (In case you are reading this thread and wonder what "SMP" is, it is Symmetric Multi Processing. What SMP accomplishes for Folding@Home synchronize both cores of a dual core CPU or two CPUs in a dual-CPU computer to work on the same work unit.)
This thread is the announcement for SMP (beta) and some discussion. Post No. 13 has links for the core and FAHMon monitor files.
The shot below shows my four rigs in my signature and my laptop. The laptop is not running SMP. The other four ARE running SMP.
The first work unit (top of the chart) is 2651 Gromacs SMP; the other three are 2652.
Average time per frame is 19:30.
This thread is the announcement for SMP (beta) and some discussion. Post No. 13 has links for the core and FAHMon monitor files.
The shot below shows my four rigs in my signature and my laptop. The laptop is not running SMP. The other four ARE running SMP.
The first work unit (top of the chart) is 2651 Gromacs SMP; the other three are 2652.
Average time per frame is 19:30.
0
Comments
Around 32min/frame on my Opteron @ 2475.
Question: does FahMon need time to calculate correct ppd like Electron Microscope does, or is it roughly correct as soon as it loads the log file?
19-20 minutes per frame with 2651 is very good, Leo. I think WinSMP is running better on Intel processors. I do not get this performance with an Opteron-165 running at 2.8 GHz. Something like 25 min/frame with P2610. Anyway, good to know.
I want to do this, but I am lazy. What is the easiest route?
Just make sure to follow the instructions in the FAQ and in the readme file that comes with the client before you run it. Nothing difficult there, just a couple of important details.
Mirage, as far as comparing Intel PD's with Opterons - keep in mind the wild overclocks I have on my machines - 1000MHz ea on the D930s and 1.2GHz OC on the D915.
Is it running 4 instances and eating over 200MB for you guys? Thats fine with me if that is what it's supposed to do, I just want to make sure I didn't goof something.
:bigggrin: :bigggrin: :bigggrin:
If I try to run as a service I freeze up on boot so It runs in an open command line interface, is there a way to hide that?
TrayIt should be fine, I just didn't want my wife to close the window by accident or something. Thanks Leo.
Assuming the massive PPD increase is related to a real productivity increase, this is fantastic.
Yup. I was a little concerned that it was taking up so much RAM on my gaming rig, but there's still enough left for me to play without noticeable impact
Not for everybody, but seems like it has been for most of us.
The short deadlines are what worry me.
I am testing on a 3.06HT system (533 FSB) with P2652. It seems like it will make it just 5-6 hours before the deadline (3 days).
The Stanford guys don't recommend using the SMP client unless you have two physical cores. They don't advice using HT unless you have two physical cores, and turning on HT would make it appear as four.
Personally, I would stick with that, because they're obtaining information on how the client is performing during this beta period, and they're not expecting people to use single-core HT rigs.
Sledge, turn off HT on that Xeon rig. I don't understand why that computer shouldn't be stable.
I'm thinking of running SMP on my E1405 (dual core 1.83GHz) laptop as soon as it finishes a nearly-completed work unit. Will it be powerful enough to meet the deadlines?
I'll bet so. It sounds like Intels in general and Core2's especially are performing well. I'd give it a shot and see if the frame times look reasonable.
I'm having the same problem on my FX-62 rig. The farthest I've gotten on a WU is 38 frames, and then is has an "EARLY UNIT END." I had these occasionally before but I'm on the eighth or ninth WU that has failed in a row.
Any Ideas? I would really like to do SMP to give a boost to ppd, but I don't get any credit for these early ending WUs.
FreeC - sorry, no clue to your problem. FX62 is dual core, right?
EDIT: "<i>NOTE: The account under which the client runs MUST have a password. Blank passwords will cause the FAH cores to fail.</i>" from here
I usually don't password my accounts at home so I created one and used Tweak UI to login automatically on boot.