GPU72 Weird stats

[Gordon]

Hi,

IF I go to the page www.gpu72.com/account then I have apparently found

43 factors in 2,412 DC checks and
77 factors in 5,631 LL checks.

On the other hand I go to the Overall Individual Statistics page from the link in the menu in the left that changes to

97 factors in 6,242 DC checks and
23 factors in 1,791 LL checks.

They don't even come to the same totals!

Which one is correct? I have certainly done a load more DC than LL checks, but is either set of figures correct??

[Mark Rose]

Originally Posted by Gordon

Which one is correct? I have certainly done a load more DC than LL checks, but is either set of figures correct??

The graph of work done matches the Individual Overall Statistics numbers for me.

The numbers at /account are incoherent. It shows me having LLTF assignments when I only have DCTF assignments, etc. I suspect it's counting lots of DCTF work done as LLTF work done as well.

[petrw1]

Originally Posted by Gordon

Hi,

IF I go to the page www.gpu72.com/account then I have apparently found

43 factors in 2,412 DC checks and
77 factors in 5,631 LL checks.

On the other hand I go to the Overall Individual Statistics page from the link in the menu in the left that changes to

97 factors in 6,242 DC checks and
23 factors in 1,791 LL checks.

They don't even come to the same totals!

Which one is correct? I have certainly done a load more DC than LL checks, but is either set of figures correct??

Same total # of factors....I get the same discrepancy there but same totals

[chalsall]

Originally Posted by Gordon

They don't even come to the same totals!

Which one is correct? I have certainly done a load more DC than LL checks, but is either set of figures correct??

Thanks for pointing that out. Clearly a SPE on my part.

[Gordon]

Originally Posted by chalsall

Thanks for pointing that out. Clearly a SPE on my part.

It's even more broke now, showing me as having done a P-1 and the other totals are still trash

[chalsall]

Originally Posted by Gordon

It's even more broke now, showing me as having done a P-1 and the other totals are still trash

Please forgive me for this, but I don't know what GPU72 account you are referring to.

Rather than providing a link to a page which is dependent on log-in credentials (thus, gives no information), could you please provide a public page of your stats, or PM me your GPU72 username (but, importantly, not your password).

The only Gordon I was able to find has done no work. But, as usual, I'm probably wrong....

[Gordon]

Originally Posted by chalsall

Please forgive me for this, but I don't know what GPU72 account you are referring to.

Rather than providing a link to a page which is dependent on log-in credentials (thus, gives no information), could you please provide a public page of your stats, or PM me your GPU72 username (but, importantly, not your password).

The only Gordon I was able to find has done no work. But, as usual, I'm probably wrong....

Okay, let's see if this works

screenshot from this page http://www.gpu72.com/account/ is nitro-overview-01

screenshot from this page http://www.gpu72.com/reports/worker/blahblahblah is nitro-stats-01

[Gordon]

Originally Posted by Gordon

Okay, let's see if this works

screenshot from this page http://www.gpu72.com/account/ is nitro-overview-01

screenshot from this page http://www.gpu72.com/reports/worker/blahblahblah is nitro-stats-01

The stats are STILL wrong, the last two DC factors I found appear in the "account" report as LL factors, the total candidates tested have different totals as does the number of days of work. differs by 11 candidates and 1,346 days of work

If I was creating the report it would simply sum by type, clearly this isn't happening.

So which figures are correct, the "by account", the "overall statistics" or neither?

[chalsall]

Originally Posted by Gordon

So which figures are correct, the "by account", the "overall statistics" or neither?

OK... Thanks for the prodding; you have indeed identified bugs in the GPU72 code.

The fundamental issue is I never imagined that GPU72 would last this long, and so stupidly assumed in the SQL queries that DCTF'ing would always be below 40M (this was after DCTF'ing was bolted on as an afterthought, by request). I thought I have fixed all of this a year or so ago, but clearly I missed a bit.

I've updated the code which renders the "real-time" data at https://www.gpu72.com/account/.

Please review this, and let me know if you see any issues. So you (and everyone) knows, the stats for DCTF and LLTF may change (read: be more accurate), but the totals shouldn't chance (beyond what you've completed in the last few hours).

Please further know that a batching script seems to suffer the same bug. This is why some "Rank / Of" cells are wrong. I'll review this this weekend.

[Mark Rose]

I still see a discrepancy between my Overall Statistics and Account Overview. The Completed and GHz Days columns differ.

Overall Statistics

Work Type  Rank   Of       Candidates            Factors        GHz Days
                      Assigned  Completed  Predicted  Found         
DC TF         2  138       258     50,839    573.651    532   252,408.59
LL TF        23  206               11,933    165.963    164   157,551.39
P-1          64   89                   13   Not sure               54.01
Overall      16  224       258     62,785     739.61    696   410,014.01


Account Overview

Work Type  Rank   Of  Assigned  Completed             Facts     GHz Days
DC TF         0    0       258     61,624               532  301,031.831
LL TF         0    0               12,622               164  178,693.801
P-1           0    0                   14                         57.461
Total                      258     74,260               696  479,783.094

I assume the Rank on the account overview page will update when that job runs. :)

[Gordon]

Originally Posted by chalsall

OK... Thanks for the prodding; you have indeed identified bugs in the GPU72 code.

The fundamental issue is I never imagined that GPU72 would last this long, and so stupidly assumed in the SQL queries that DCTF'ing would always be below 40M (this was after DCTF'ing was bolted on as an afterthought, by request). I thought I have fixed all of this a year or so ago, but clearly I missed a bit.

I've updated the code which renders the "real-time" data at https://www.gpu72.com/account/.

Please review this, and let me know if you see any issues. So you (and everyone) knows, the stats for DCTF and LLTF may change (read: be more accurate), but the totals shouldn't chance (beyond what you've completed in the last few hours).

Please further know that a batching script seems to suffer the same bug. This is why some "Rank / Of" cells are wrong. I'll review this this weekend.

The DC totals are correct, but the LL ones are out by 10 tests and the 1346 days still. The "account" version still has me down for a P-1...all my work of that type is on the CPU.