Google Diet Colab Notebook - Page 60

Dylan14 · 2019-12-05, 22:22

Introducing tf1G.py v0.09:

In running the script, I noticed that the script counted all the lines in the results file when it ran the count_exponents_tested function. This is confirmed when I ran v0.08 on a Colab session: I got 703 tested, which is really 695 exponents plus 8 extra lines where mfaktc has printed: found 1 factor for Mxxxxxxxxxx. This has now been rectified: the function will only add lines which has the strings "no factor" or "found".
The script is attached below:

kriesel · 2019-12-05, 22:43

Quote:

Originally Posted by Dylan14

Introducing tf1G.py v0.09:

In running the script, I noticed that the script counted all the lines in the results file when it ran the count_exponents_tested function. This is confirmed when I ran v0.08 on a Colab session: I got 703 tested, which is really 695 exponents plus 8 extra lines where mfaktc has printed: found 1 factor for Mxxxxxxxxxx. This has now been rectified: the function will only add lines which has the strings "no factor" or "found".
The script is attached below:

This should also allow for users modifying the mfaktx.ini file so that additional lines providing a date and time stamp are excluded from the exponent count. Or manual edits.

Dylan14 · 2019-12-05, 23:04

Quote:

Originally Posted by kriesel

This should also allow for users modifying the mfaktx.ini file so that additional lines providing a date and time stamp are excluded from the exponent count. Or manual edits.

As for the time stamp: the script only counts lines that have those strings that I mentioned. Since the time stamp looks something like this:

Code:

[Thu Dec  5 17:56:55 2019]

the script ignores it since those strings do not appear.
As for manual edits: would you mind providing some examples as to what you mean by this?

kriesel · 2019-12-05, 23:08

Quote:

Originally Posted by Dylan14

As for the time stamp: the script only counts lines that have those strings that I mentioned. Since the time stamp looks something like this:

Code:

[Thu Dec  5 17:56:55 2019]

the script ignores it since those strings do not appear.

That's what I said. You did the right thing, matching to count a line (not matching to exclude a line, which leads to needing to identify all cases of lines to exclude).

Quote:

As for manual edits: would you mind providing some examples as to what you mean by this?

Someone manually edits the results file and puts in a text comment or just an extra line break. These are very common here. Also somewhat already handled by your approach, and also not an issue here since I'm using other means to submit results. You could have trouble if someone adds a comment that happens to fit your "no factor" or "found" string matches. "#retest for whether no factor found" would be an unfortunate choice of comment.

Dylan14 · 2019-12-06, 00:19

Introducing tf1G.py v0.10:

Now the count_exponents_tested function will look at each line and see if the line starts with # or ; (# denotes a comment line in most things, ; is used in mprime/Prime95). If it does, the program will move on to the next line (thanks kriesel for the feedback!).

Chuck · 2019-12-06, 01:10

Quote:

Originally Posted by kracker

Just got a one question survey while running colab saying "Would you be interested in a paid, premium version of colab?""

I haven't seen that survey but my answer would be YES if I could have a premium GPU 24 hours per day.

chalsall · 2019-12-06, 01:37

Quote:

Originally Posted by Chuck

I haven't seen that survey but my answer would be YES if I could have a premium GPU 24 hours per day.

+1! I'd have to compare it to the costs of GCE, EC2 etc, but I would be very interested in knowing the terms of the offer.

LaurV · 2019-12-06, 02:06

T4 and P4 seem to be two different fish. They may be two different chips, or one chip with different chops. Like two copies of the same hand which once had the index finger amputated, and once had the middle finger amputated, it seems that T4 is about 4 times as fast as a K80 for TF, and only quarter speed at LL, while P4 is about 1.5 times TF and 1/2 times LL compared with k80. The P100 seems to be another fish too, being about 2.5xTF and 2xLL. YMMV...

kracker · 2019-12-06, 02:33

Quote:

Originally Posted by LaurV

T4 and P4 seem to be two different fish. They may be two different chips, or one chip with different chops. Like two copies of the same hand which once had the index finger amputated, and once had the middle finger amputated, it seems that T4 is about 4 times as fast as a K80 for TF, and only quarter speed at LL, while P4 is about 1.5 times TF and 1/2 times LL compared with k80. The P100 seems to be another fish too, being about 2.5xTF and 2xLL. YMMV...

Some specs on all the GPU's we've gotten... no idea if it's 100% accurate.

NVIDIA Tesla K80
NVIDIA Tesla P4
NVIDIA Tesla T4
NVIDIA Tesla P100

axn · 2019-12-06, 02:35

https://en.wikipedia.org/wiki/List_o...ng_units#Tesla

P4 is Pascal (10xx), T4 is Turing (RTX 20xx, GTX 16xx). Neither is good at DP. T4 >> P4 at TF.

kracker · 2019-12-06, 04:18

I notice we've gotten to try every one of Google Cloud's GPU's, except the V100...

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Alternatives to Google Colab	kriesel	Cloud Computing	11	2020-01-14 18:45
Notebook	enzocreti	enzocreti	0	2019-02-15 08:20
Computer Diet causes Machine Check Exception -- need heuristics help	Christenson	Hardware	32	2011-12-25 08:17
Computer diet - Need help	garo	Hardware	41	2011-10-06 04:06
Workunit diet ?	dsouza123	NFSNET Discussion	5	2004-02-27 00:42

2019-12-06, 02:06	#657
LaurV Romulan Interpreter Jun 2011 Thailand 10010110111111₂ Posts	T4 and P4 seem to be two different fish. They may be two different chips, or one chip with different chops. Like two copies of the same hand which once had the index finger amputated, and once had the middle finger amputated, it seems that T4 is about 4 times as fast as a K80 for TF, and only quarter speed at LL, while P4 is about 1.5 times TF and 1/2 times LL compared with k80. The P100 seems to be another fish too, being about 2.5xTF and 2xLL. YMMV...

2019-12-06, 02:35	#659
axn Jun 2003 11735₈ Posts	https://en.wikipedia.org/wiki/List_o...ng_units#Tesla P4 is Pascal (10xx), T4 is Turing (RTX 20xx, GTX 16xx). Neither is good at DP. T4 >> P4 at TF.

2019-12-06, 04:18	#660
kracker "Mr. Meeseeks" Jan 2012 California, USA 2³×271 Posts	I notice we've gotten to try every one of Google Cloud's GPU's, except the V100...