...............................................................................................
..............................
Date....................Runtime...................Daily Points.......Daily Results.........Place
08/01/2005..... 1:056:09:26:13.......... 254,135............. 1,280................. #2
07/31/2005..... 1:012:08:46:10.......... 236,163............. 1,297................. #2
07/30/2005..... 1:050:08:55:58.......... 260,039............. 1,351................. #2
....................................................................................................
.........................
(Stats-Image Generator by WCG Member Börni)

So you're saying the composite scores of the client are ignored and it uses the device profile - this seems strange. It was interesting that the client correctly would say that the composite was lower when I was running of the small hard drive. I was also under the notion that hourly points were approx a factor of ~5.2-5.3 of the composite which means I would expect that it is really using the 62 and not the 70 (which would be like 5.8 divisor). This sort of begs the question on which one it is using since the 70 should reflect what the system is using if it is using the device settings and not the local client score. That is a bit confusing but it smells like it is ignoring both the local client score and the device settings and it is using the original local hard drive score.

My real concern on posting this was not the 1.56 points an hour I thought I might be losing but the issue of it not recognizing the amount of hard drive available. I just purchased another motherboard to upgrade a relative and I was planning on playing with a network boot so I would not need to run with a hard drive before I did the upgrade. Playing with the concept of a computing farm with minimum components to reduce power requirements and cost of creating. If the network host was a problem than this plan would be an issue.

The other thing I have been trying to understand is if it is better to buy a few screamers or buy really low end machines. From a cost perspective I can buy a Pentium III / 866 mHz machines as complete systems for $45 each and I'm getting 400 points a day on one of these - so 2.5 of these machines would equal a top end machine given the current limit on 2X score. So for $115 dollars I get the equivalent of a top end machine points wise. It seems like this would be cheaper unless the power costs are a big delta. The motherboard / CPU I just bought was a Sempron 2200 with 0.5 GB of memory and it came in at $170 and that doesn't include OS, a case, power supply or hard drive. So without a hard drive or OS I'm near $200 and this machine might not score a 200 without more memory or playing with overclocking. Therefore, it seems like low end machines will win if you are trying to build point capability at the lowest cost. That might change if the 2X limit was lifted of course. Just some thought I'd share some ramblings on my hobby

.___ Andrew R ___

________ for Joining our Top 4 Team
__ All Our New Members - Please - Get Your Team Banners
___For all the Members who would like a Signature Banner in the Messages
_ Click Here and please follow the Instructions: My Online Signature Banners
__ Click Here for instructions to get the: Stats Generator in your Signature
_____Please check your Member Profile at: My Online Team Website
-Click to Email David Autumns our Captain with any Info you would like added

Good morning Robert
I hope you are well
I see your enthusiasm for posting proceeds you
With regards to the Member Intro, I try to cover all the bases

Hi DAJ,



You're buying machines dedicated to WCG? Cool

Honestly, I think cheap machines would win hands down. Why? Because the client is low footprint and so cheapo mainboard architecture is not punished very hard. This is a mere crunching job so all that counts is CPU cycles. If you need a fast high end server or video workstation then bus speeds, I/O performance and chipset qualities are your main target -- you can even use a slower CPU with more cache, it might easily outperform the one with more speed and less cache when you're looking at overall performance.

Crunching, cracking, benchmarks and other specialized tasks can be done faster on a couple of cheap machines than on a large monolithic power machine. That's why crunching supercomputers (chess machines, climate prediction etc.) for computational tasks nowadays are more and more built from clusters of cheap standard machines than built as a traditional supercomputers.

But then do it right -- don't buy complete PCs -- build a ventilated rack and install bare cheapo mainboards (no cards, video and HD controller onboard) that boot from a single network drive. Build a crunching farm like CompuDude (only cheaper as you don't need the fast I/O of rendering farms). Also reduces your heating costs (but your power consumption might be a bit higher )


Best, Stefan.