Next plans

Author	Message
rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7445 Credit: 42,730,867 RAC: 0	Message 34 - Posted: 29 Nov 2014, 9:05:57 UTC
	After the rest of the WUs are finished I will add some tests for linux32/64 and windows 32bit with double checking (quorum 2). If I find a way with help from primegrid I could change the validation process. -Reb
	ID: 34 · Rating: 0 · rate: / Reply Quote

fzs600 Send message Joined: 28 Nov 14 Posts: 17 Credit: 1,144,473,134 RAC: 823,251	Message 35 - Posted: 29 Nov 2014, 11:30:56 UTC - in response to Message 34.
	good news
	ID: 35 · Rating: 0 · rate: / Reply Quote

rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7445 Credit: 42,730,867 RAC: 0	Message 38 - Posted: 29 Nov 2014, 13:29:31 UTC
	I got an answer from primegrid about the validation. Its much more complicated than I thought and need all the time testing to catch up the correct new values so credit new is the only way I can go here, unfortunately.
	ID: 38 · Rating: 0 · rate: / Reply Quote

frankhagen Send message Joined: 7 Jun 14 Posts: 36 Credit: 241,030 RAC: 0	Message 39 - Posted: 29 Nov 2014, 13:45:46 UTC - in response to Message 38.
	I got an answer from primegrid about the validation. Its much more complicated than I thought and need all the time testing to catch up the correct new values so credit new is the only way I can go here, unfortunately. OMFG!
	ID: 39 · Rating: 0 · rate: / Reply Quote

rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7445 Credit: 42,730,867 RAC: 0	Message 40 - Posted: 29 Nov 2014, 15:16:11 UTC - in response to Message 39. Last modified: 29 Nov 2014, 16:10:14 UTC
	I got an answer from primegrid about the validation. Its much more complicated than I thought and need all the time testing to catch up the correct new values so credit new is the only way I can go here, unfortunately. OMFG! What I only could do is to split the complete work again into some parts to award fixed credits for them. Specific on time basing you will get then less credits in a higher n value but not so dramatic. The worst thing with credit new is the missing calculation with AVX, its faster against non AVX CPUs but get sometimes less credits or equal. Edit: Another idea, I will processing the first n-value and the last one, get the average time and setup fixed credits for this.
	ID: 40 · Rating: 0 · rate: / Reply Quote

rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7445 Credit: 42,730,867 RAC: 0	Message 41 - Posted: 29 Nov 2014, 15:53:08 UTC
	I have prepared win32 and compiled the wrapper. A standalone test was ok. For linux I will go to quorum 1 too and compare the results from windows in test.
	ID: 41 · Rating: 0 · rate: / Reply Quote

rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7445 Credit: 42,730,867 RAC: 0	Message 47 - Posted: 29 Nov 2014, 18:24:25 UTC
	If you have some 1000 results calculated credit new has a good value per hour so it needs some time to catch up the best value for your CPU. The other idea with fixed credits would waste too much time and I want to load all the lines at once from the sievefile.
	ID: 47 · Rating: 0 · rate: / Reply Quote

frankhagen Send message Joined: 7 Jun 14 Posts: 36 Credit: 241,030 RAC: 0	Message 63 - Posted: 30 Nov 2014, 11:38:34 UTC - in response to Message 47.
	since this is LLR, why not adopt the credit formula they use over at PG?
	ID: 63 · Rating: 0 · rate: / Reply Quote

rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7445 Credit: 42,730,867 RAC: 0	Message 65 - Posted: 30 Nov 2014, 16:42:19 UTC - in response to Message 63. Last modified: 30 Nov 2014, 16:42:41 UTC
	since this is LLR, why not adopt the credit formula they use over at PG? Its too complicated to explain, they do this after every FFT length and other things. I will split the files again in n-length and take a good fixed credit value for it.
	ID: 65 · Rating: 0 · rate: / Reply Quote

Post to thread

Message boards : News : Next plans