[QUOTE=chalsall;527110]I have a plan of next steps in my workbook...[/QUOTE]

Hey all. So everyone knows, I've just "pushed" v0.31 of the TF'ing "payload" into production.

Deltas:

1. The version number is now printed on the starting log lines.

2. I have moved the "Itr Time" column to be grouped with ETA and GHzD/D. More logical, and it made a regex a lot cleaner.

3. The comms spider will now report on all entries in the worktodo.txt file. The server now calculates estimated completion for all assignments held by each instance.

For anyone currently running instances, it is safe to stop and then re-run.

Once everyone is running this (tomorrow) I'll be able to have the server recycle assignments much faster. Like rather than 12 hours, more like 20 minutes.

As always, please point out any SPEs or other "Hmmmm... That's strange..." kinda stuff.

[QUOTE=Lorenzo;527135]Hello! Where I can download the latest version CudaLucas compiled for Linux/colab? On [URL]https://sourceforge.net/projects/cudalucas/files/2.06/[/URL] I found only Windows version.[/QUOTE]


Hi Lorenzo,


Try this executable. I compiled this on the Colab with the following CCFlags:


[CODE]-O$(OptLevel) --generate-code arch=compute_37,code=sm_37 --generate-code arch=compute_60,code=sm_60 --generate-code arch=compute_75,code=sm_75 --compiler-options=-Wall -I$(CUINC)[/CODE]
The first one generates code for the K80 (CC 3.7), the second for the P100 (CC 6.0) and the third for the T4 (CC 7.5).

I noticed that Kaggle seems to give a different GUID every time you launch an instance. In that case, you might want to set [c]FixedHardwareUID=1[/c] in your [C]prime.txt[/C] file. Otherwise, your CPUs page will be cluttered by multiple copies of the same computer.

[QUOTE=chalsall;527183]Dooohhh!!! Now that's an example of a profoundly stupid programmer error!!!

Please give it another go...[/QUOTE]

[QUOTE=Chuck;527186]Working....[/QUOTE]I have a 65,xxx,xxx that is completed and the spider has not seen that PrimeNet has the result.

Also, I dropped a session off-line, then restarted the same worker after restarting my browser. It grabbed new assignments. This has happened before. (Same worker ID).

[CODE]65495411
65495371
no factor for M65495411 from 2^74 to 2^75 [mfaktc 0.21 barrett76_mul32_gs]
no factor for M65495371 from 2^74 to 2^75 [mfaktc 0.21 barrett76_mul32_gs]
2019-10-02 20:03:25
2019-10-02 23:29:52[/CODE]

Reported results aren't clearing from my colab report page

[QUOTE=Chuck;527213]65495411
65495371

Reported results aren't clearing from my colab report page[/QUOTE]
I think spidey is not looking for the 65,xxx,xxx numbers, therefore they are not getting cleared.

[QUOTE=Uncwilly;527217]I think spidey is not looking for the 65,xxx,xxx numbers, therefore they are not getting cleared.[/QUOTE]

That makes sense; I remember this behavior happening in the past when new ranges were brought in.

And I just noticed that it has been fixed as of a few minutes ago.

[QUOTE=Uncwilly;527217]I think spidey is not looking for the 65,xxx,xxx numbers, therefore they are not getting cleared.[/QUOTE]

Yup; yet another SPE! Humans are /so/ unreliable!!! :wink:

Sorry guys -- I pulled an all-nighter yesterday; 33 hour day. Heading out the door again in a few minutes.

I don't *think* there's anything else seriously broken, but if there are please point them out; I'll be BAC this evening.

[QUOTE=Chuck;527213][CODE]65495411
65495371
no factor for M65495411 from 2^74 to 2^75 [mfaktc 0.21 barrett76_mul32_gs]
no factor for M65495371 from 2^74 to 2^75 [mfaktc 0.21 barrett76_mul32_gs]
2019-10-02 20:03:25
2019-10-02 23:29:52[/CODE]

Reported results aren't clearing from my colab report page[/QUOTE]

I have 4 (I think my first 4) 64.3M results from Sept 11 that don't clear either.

[QUOTE=Dylan14;527191]Try this executable. I compiled this on the Colab with the following CCFlags:[/QUOTE]

If I may please put this out there... This thing has ***amazing*** possibilities for education!

I've had the privilege of being involved with a conference going on this week in Barbados, dealing with technology. The human networking is, of course, a huge part of this -- get people talking.

Over the last few days, I've talked with several different people involved with several levels of education -- grade school to post-grad. They all ***immediately*** "get it".

[QUOTE]"My students, who might at best only have an iPad at home to do their homework with, can have access to serious compute as part of their assignments. Now *that* would be interesting...[/QUOTE]

I think it would be really cool (and a useful resource) if we could start building a collection of "Package Specific" Notebooks. I'm envisioning short and sweet.

They download the source code for any particular package of interest -- from the official source -- and builds it from scratch. Every delta for a clean build is done by way of the Python shell script, and is documented. Possibly would involve "mkdir", "chmod", "ln", "sed", "apt-get" et al.

Bonus points for being able to specify exact build options, as Dylan did above.

Absolutely no disrespected to anyone who claims to not to be able to compile from source, but it really isn't that hard. Heck, if a human can tell a deterministic computer how to do it (by way of a script), it's easy!

And now you (and anyone!) can do this. Off on someone else's machine where it doesn't matter if you make a mistake... :smile:

I agree with you chalsall on these points.

Now since you say that you’ve talked to several people with varying levels of education, some may not be well versed with python or with working with Linux commands.

So before we’d ask people to build from source, maybe it would first make sense to give people a “quick and dirty” introduction to Python coding and interacting with the shell and the Google Drive before we cut them loose to do builds from source. At least, if I was teaching programming to people...