... Here is another interesting one to test your wits on :). It's a totally different group of machines but once again there are a total of five results. My Linux result was #4 in the full list. My question is why were #2 and #3 both given no credit? :).
Hi Gary,
your 'riddle' from '6:39:19 UTC':
#2 was given no credits because the new validator was still not happy with the windows 4.17 result (it seems that uninitalized variables lead to totally different results...)
#3 was given no credits because: "Result was reported too late to validate"
...there also is a deadline where no more credits are granted...
I fear that #5 in your example from '5:41:33 UTC' will also get no more credits due to returning its result too late...
#2 was given no credits because the new validator was still not happy with the windows 4.17 result (it seems that uninitalized variables lead to totally different results...)
#3 was given no credits because: "Result was reported too late to validate"
...there also is a deadline where no more credits are granted...
Hi Udo,
Yes, you are correct. The first validation attempt was done on June 27 and #2 & #4 became CNBC. The next validation attempt was on July 4 where #2 (Win), #4 (Linux) & #5 (Linux) were involved. That's when #2 was given zero credit and the Linux results prevailed. #3 came along later on July 8 and outside the deadline - so zero credit.
Quote:
I fear that #5 in your example from '5:41:33 UTC' will also get no more credits due to returning its result too late...
I don't share your fear :). Whilst the box is a PIII Katmai and therefore almost as slow as Tullio's famous PII, how many deadlines has Tullio missed? :) I work on the theory that anyone who is running Linux on one of these old dinosaurs knows exactly what they are doing and will be fairly skilled in not missing deadlines :).
Of course he might look at the full quorum, see the action is all over and decide to abort so as not to waste any more of his time :).
Digressing for a moment, have you ever considered replacing your veteran PII with a slightly less veteran PIII or P4? I'm sure that all moderately sized cities around the planet would have auction houses which sell off surplus government and business computers for just a few Euros. You could have a lot of fun with something like a 1Gig PIII or a 1.6Gig P4 which would probably double or triple your production rate for about the same electricity cost. in fact if you were to select one of the quite efficient Tualatin PIIIs, you might be able to reduce your electricity consumption at the same time :).
The problem is, my CPU is a Slot 1 CPU, with a 512k L2 cache on a chip connected to the CPU by a fast bus. My Mobo is an Intel 440BX. Should I buy a PIII or a P4 they would not fit into my slot, and I should have to buy another Mobo too. But in that case I would surely go to a Core Duo chip, which outperforms Athlons and also consumes less electricity. The problem with modern hardware is that almost as soon as you have bought a new chip it is already obsolescent. So I am sticking to my (already famous) PII until it dies a natural death. Thanks anyway for your suggestions.
Tullio
Yes, I know, and I'm sorry I didn't make my comments fully clear to you. I wasn't talking about upgrading your machine. I was suggesting that you completely replace it with a fully working government or business surplus unit. You should be able to get a PIII 1Gig with 256MB RAM and a 20Gig HDD for maybe 20-30 Euros or so if the auction market in Italy is anything like it is here in Australia. If you know what you are looking for you can get some real bargains.
For example, just over one week ago I bought a HP e-vectra small form factor PC which runs from a laptop style 60W power pack and uses a laptop style slim CDROM. It had a PIII 1Gig CPU, 128MB RAM and a 10Gig HDD. It cost me about $USD10. It fired straight up and even had the OS left on it. This is not normal as they usually wipe the HDD. This one must have been overlooked. So I installed EAH and got it ready for crunching. I also jacked up the FSB from 133MHz to 153MHz and tested it for stability. It is running quite stable at 1150MHz. It has now been crunching for just over a week with no issues and here is its results list. It should be able to complete around three results every week whereas your PII might only do about one per week.
Here is another example. It is a mini tower PC with a Tualatin Celeron 1.1GHz CPU, 256MB RAM and a 20GB HDD. Once again I was easily able to overclock the FSB from 100MHz to 120MHz so it is running at 1320MHz. It is doing the monster WUs of the type that Bikeman was talking about and it has just about finished its third one already. Once again it is a fully functioning machine and it cost me around $USD20. This is exactly the type of machine you should replace your PII with as it has several spare PCI slots for any add-in cards you might already have.
Thanks Gary. But the computers I see on ebay start with prices from 150/200 euros up if they are at least decent machines. So the difference between used machines and brand new ones is getting smaller and smaller. I frequent a PC shop where I buy peripherals, modems, scanners, printers etc. So one day maybe a see a spanking new PC and buy it. Cheers.
Tullio
Oh no, not ebay. You need to go to where the sellers on ebay do their buying - the traditional auction houses which sell off hundreds of computer lots every week. Stick the italian equivalent of "traditional computer auctions milano" into a google search and see what you get. You might be surprised :).
EDIT: If you wish to continue, let's do so in Bikeman's new thread :).
OK Gary. I have a technical question on cross validation. Windows uses double precision (64 bit), while Linux uses extended precision (80 bit) in floating point arithmetics. How do you compare a Windows result against a Linux result? Do you pad the 64 bit number with zeros until it reaches 80 bit or do you truncate the 80 bit number bringing it down to 64 bit? I think this may be the heart of the (validation) matter, with apologies to Graham Greene.
Tullio
How do you compare a Windows result against a Linux result? Do you pad the 64 bit number ...
I'm absolutely no expert on this so someone else can fill in the details or make corrections.
I don't think it's a matter of padding or truncating an answer. It's to do with zillions of intermediate calculations in which the incredibly small differences can compound to give a different answer. The difficulty is to find routines or to design calculation procedures which don't allow answers to diverge over large numbers of iterations.
If it were a straight forward business to do this, I'm sure we wouldn't have had this issue in the first place.
Please realise that as a mathematician, I'd make a pretty good floor sweeper :).
Edit: If you trawl through Bernd's posts, you might find that he has expanded on this somewhere ....
OK Gary. I have a technical question on cross validation. Windows uses double precision (64 bit), while Linux uses extended precision (80 bit) in floating point arithmetics. How do you compare a Windows result against a Linux result? Do you pad the 64 bit number with zeros until it reaches 80 bit or do you truncate the 80 bit number bringing it down to 64 bit? I think this may be the heart of the (validation) matter, with apologies to Graham Greene.
Tullio
As Gary said, the difference is in intermediate results during the rather complex computations, the final results that the validator will receive are all cast to much lower precision on all platforms. The validator will then compare the results and will allow them to vary by a certain relative value.
RE: ...Here is another
)
Hi Gary,
your 'riddle' from '6:39:19 UTC':
#2 was given no credits because the new validator was still not happy with the windows 4.17 result (it seems that uninitalized variables lead to totally different results...)
#3 was given no credits because: "Result was reported too late to validate"
...there also is a deadline where no more credits are granted...
I fear that #5 in your example from '5:41:33 UTC' will also get no more credits due to returning its result too late...
Udo
RE: #2 was given no
)
Hi Udo,
Yes, you are correct. The first validation attempt was done on June 27 and #2 & #4 became CNBC. The next validation attempt was on July 4 where #2 (Win), #4 (Linux) & #5 (Linux) were involved. That's when #2 was given zero credit and the Linux results prevailed. #3 came along later on July 8 and outside the deadline - so zero credit.
I don't share your fear :). Whilst the box is a PIII Katmai and therefore almost as slow as Tullio's famous PII, how many deadlines has Tullio missed? :) I work on the theory that anyone who is running Linux on one of these old dinosaurs knows exactly what they are doing and will be fairly skilled in not missing deadlines :).
Of course he might look at the full quorum, see the action is all over and decide to abort so as not to waste any more of his time :).
Cheers,
Gary.
RE: Digressing for a
)
The problem is, my CPU is a Slot 1 CPU, with a 512k L2 cache on a chip connected to the CPU by a fast bus. My Mobo is an Intel 440BX. Should I buy a PIII or a P4 they would not fit into my slot, and I should have to buy another Mobo too. But in that case I would surely go to a Core Duo chip, which outperforms Athlons and also consumes less electricity. The problem with modern hardware is that almost as soon as you have bought a new chip it is already obsolescent. So I am sticking to my (already famous) PII until it dies a natural death. Thanks anyway for your suggestions.
Tullio
RE: The problem is, my CPU
)
Yes, I know, and I'm sorry I didn't make my comments fully clear to you. I wasn't talking about upgrading your machine. I was suggesting that you completely replace it with a fully working government or business surplus unit. You should be able to get a PIII 1Gig with 256MB RAM and a 20Gig HDD for maybe 20-30 Euros or so if the auction market in Italy is anything like it is here in Australia. If you know what you are looking for you can get some real bargains.
For example, just over one week ago I bought a HP e-vectra small form factor PC which runs from a laptop style 60W power pack and uses a laptop style slim CDROM. It had a PIII 1Gig CPU, 128MB RAM and a 10Gig HDD. It cost me about $USD10. It fired straight up and even had the OS left on it. This is not normal as they usually wipe the HDD. This one must have been overlooked. So I installed EAH and got it ready for crunching. I also jacked up the FSB from 133MHz to 153MHz and tested it for stability. It is running quite stable at 1150MHz. It has now been crunching for just over a week with no issues and here is its results list. It should be able to complete around three results every week whereas your PII might only do about one per week.
Here is another example. It is a mini tower PC with a Tualatin Celeron 1.1GHz CPU, 256MB RAM and a 20GB HDD. Once again I was easily able to overclock the FSB from 100MHz to 120MHz so it is running at 1320MHz. It is doing the monster WUs of the type that Bikeman was talking about and it has just about finished its third one already. Once again it is a fully functioning machine and it cost me around $USD20. This is exactly the type of machine you should replace your PII with as it has several spare PCI slots for any add-in cards you might already have.
Cheers,
Gary.
I have an ancient dual P III
)
I have an ancient dual P III crunching for E@H ...
You know what.... let's discuss this in a separate thread here we are kind of off-topic here but it's so interesting.
CU
BRM
Thanks Gary. But the
)
Thanks Gary. But the computers I see on ebay start with prices from 150/200 euros up if they are at least decent machines. So the difference between used machines and brand new ones is getting smaller and smaller. I frequent a PC shop where I buy peripherals, modems, scanners, printers etc. So one day maybe a see a spanking new PC and buy it. Cheers.
Tullio
RE: ... the computers I see
)
Oh no, not ebay. You need to go to where the sellers on ebay do their buying - the traditional auction houses which sell off hundreds of computer lots every week. Stick the italian equivalent of "traditional computer auctions milano" into a google search and see what you get. You might be surprised :).
EDIT: If you wish to continue, let's do so in Bikeman's new thread :).
Cheers,
Gary.
OK Gary. I have a technical
)
OK Gary. I have a technical question on cross validation. Windows uses double precision (64 bit), while Linux uses extended precision (80 bit) in floating point arithmetics. How do you compare a Windows result against a Linux result? Do you pad the 64 bit number with zeros until it reaches 80 bit or do you truncate the 80 bit number bringing it down to 64 bit? I think this may be the heart of the (validation) matter, with apologies to Graham Greene.
Tullio
RE: How do you compare a
)
I'm absolutely no expert on this so someone else can fill in the details or make corrections.
I don't think it's a matter of padding or truncating an answer. It's to do with zillions of intermediate calculations in which the incredibly small differences can compound to give a different answer. The difficulty is to find routines or to design calculation procedures which don't allow answers to diverge over large numbers of iterations.
If it were a straight forward business to do this, I'm sure we wouldn't have had this issue in the first place.
Please realise that as a mathematician, I'd make a pretty good floor sweeper :).
Edit: If you trawl through Bernd's posts, you might find that he has expanded on this somewhere ....
Cheers,
Gary.
RE: OK Gary. I have a
)
As Gary said, the difference is in intermediate results during the rather complex computations, the final results that the validator will receive are all cast to much lower precision on all platforms. The validator will then compare the results and will allow them to vary by a certain relative value.
EDIT:
I think Gary referred to this post of Bernd
CU
BRM