Granted credit 0???
23 Apr 2007 13:37:59 UTC
Topic 192620
(moderation:
Hi can anyone explain why I was not granted any credit for the following work unit?
Result id 83433960
work unit id 33365901
outcome Success
Client state Done
claimed credit 173.88
granted credit 0.00
Any ideas??
Many thanks
Matt
Granted credit 0???
)
You result (withresult-id 83155831, workunit 33233633) was invalid!
Udo
any ideas why this was
)
any ideas why this was reported as an invalid result???
the difference between your
)
the difference between your answer and the second one returned were wider than the maximum allowed by the validator. A 3rd result was commisioned and it was closer to the other persons result than yours. The reason could be an error occured on your machine while running the app, or that the validator is set to restrictively.
Checking my own results I found:
9@ good
2@ fail at WU start.
1@ WU aborted at 60%.
1@ 2 intel, 1 amd: intel lost - loosing machine did have a successfuly prior result
1@ 1 intel, 1 amd: 3rd result pending.
I suspect that the validator needs to be tweaked to allow slightly larger rounding errors before reporting a problem, but without access to serverside data it's impossible to say one way or the other.
thanks for the
)
thanks for the explanation..... hopefully all will be well with my future results...
RE: Hi can anyone explain
)
I too have just received 2 null (zero) results for otherwise successful 187.92 point Work units:
33352765
33232689
Is your processor a PowerPC? Mine is a 2000 vintage PowerMac 500 MHz single G4 upgraded with a PowerLogix Dual G4 1.3 MHz (Motorola PowerPC 7457 v1.1) that has never had a client error or null result before during its 181,862.86 pts worth of work. PowerPCs are certified for Space & Flight duty due to their reliability.
Is there a discrepancy between Intel's & PowerPCs' Floating Point Units (FPUs)? Is there an outright error as there was about 10 years ago when the Pentium's FPUs were found to have a 'bug'?
This is too much work to be getting thrown out, and with the heat of summer coming, why run Einstein when such a high percentage of work will be thrown out?
RE: Is there a discrepancy
)
There are differences. There are also differences between amd and intel cpus even though both are x86. There are also (potential) differences between subsequent generations within the same family. All modern CPUs implement the IEE754 floating point standard for 32 and 64bit floats. It defines minimum levels of accuracy but leaves it up to the chip developers to determine the best way to meet them. For extremely lengthy calculations differences in the values within the allowed error tollerances can ultimately be promoted into significant bits. The result is that in really long calculations two different hardware implementations will give slightly different answers, and that hte validator must check not for identical values but for ones that are within a certain tollerance of each other. The current validation problems could be a result of actual bugs in the app, or a validator that's being too picky with the longer WUs that s5r2 is issuing. If the latter based on prior history, bruce will probably rerun the validator on failed results to give credit once it's updated. A post in the s5r2 thread in crunchers corner says that the next app release will only be a bugfix not a speedup so it's possible that the problem is a less than perfectly working app, in which case I don't know what will be done about busted WUs.
I don't know how the PPC does it's floating point math but I can describe the basic strategy of the x87 (intel/amd). It's architecture uses 80bit floats internally having decided that a slightly larger physical chip was an acceptable tradeoff instead of a more complex design that used slower math implementations that had a lower percentage error. The 80bit design allows the use of designs that are simpler and require fewer steps to run but result in errors in the last few bits.
The issue with any finite length decimal representation is that ultimately it can't represent all numbers exactly and ultimately must round them. The article below goes into alot more detail about floating point caculations and the rounding problems that occur but without a technical background (ideally compsci, but math or egineering will probably also suffice) is likely to be over your head.
http://docs.sun.com/source/806-3568/ncg_goldberg.html
Discussion on floating point
)
Discussion on floating point math
That is the same place that Dan referenced, I just made it fit BBCode...
BTW, thanks Dan for reminding me of the reasons why I decided not to pursue much math beyond Calculus w/ Analytical Geometry... I'm sure if I kept with it, all that stuff would make sense, but now, some 15 years removed from that area of study......nope, not a chance...
Hello, I´ve the same
)
Hello,
I´ve the same problem with no granted credits.......
Now i let my PC the last time to try get credits.....if not.......I´ll leave this Project......and that was it......hehehehe.....
Regards
Assasine
RE: RE: Is there a
)
I am familiar with floating point math as I have a 2nd BS in CS & minor in Math & learned long ago never to do an equality test on FP results (even though from the same FPU). Thank you for the refresher though.
The question remains if the problem is due to the very different FPU (Floating Point Unit) architectures where this problem is cropping up. In Mattcatterall's situation (WU #33365901) his AMD FPU result was compared against a new & older Intel Xeon FPU.
In my situation, I had 2 rejected WUs in as many days, which was unprecedented for that long running PowerPC machine. Furthermore, both of my rejected work units (WU #33352765 & WU #33232689) were my PowerPC's FPU being compared against various new & old Intel FPUs.
I made public this possibility as there seems 'to be smoke' when a machine that has never had a bad result gets 2 in 2 days. Whether or not there 'is fire', is not my job nor do I have time to pursue it; I'm just giving my CPU time and if it's going to be thrown away then there are other projects to subsidize.
For the record, I have 2 PowerPC & 2 Intel based Macs. I am also aware of the extreme lengths Intel has gone through to maintain benchmark supremecy, which became public when they got caught with their FPU 'bug'. Thus, 'once burned twice shy'.
My Macintosh G3 has been
)
My Macintosh G3 has been doing just fine until the work units quadrupled in run time see 33019401 vs. 33216553 and 33376147. The latter two resulted in zero credit. How can I get credit, or get work units small enough so I don't end up getting no credit for lots of cycles?
Rick