mersenneforum.org Getting others to do the work on exponents I like (was: Trial Factoring Progress)
 Register FAQ Search Today's Posts Mark Forums Read

 2021-08-14, 21:50 #221 tuckerkao   "Tucker Kao" Jan 2020 Head Base M168202123 22·3·47 Posts I finished the P-1 factoring of M120202123 and will run the TF from 2^76 to 2^77 some time next week. It seems to be floating on the lower edge of the Cat 4 exponent group. Try to test and see whether curtisc will take it and run the PRP test as soon as it switches to Cat 3. At least curtisc acquired some newer school computers faster than the ones he had back a year ago. Z343094Ɛ7 <- M120202123 Z483Ɛ7237 <- M168202123 Z52437Ӿ37 <- M186202123 Z3773054Ɛ <- M130202123 and Z4Ӿ81214Ɛ <- M175202123 seems to be another pair of the series. I got curtisc on M108377323 when I was testing M168377323 around half a year ago. Both M168175123 and M168934123 have reached the PRP Ready status, NF-PM1 and NF to 2^78. Last fiddled with by tuckerkao on 2021-08-14 at 22:49
2021-08-14, 22:49   #222
drkirkby

"David Kirkby"
Jan 2021
Althorne, Essex, UK

44810 Posts

Quote:
 Originally Posted by tuckerkao I still need some time to trial factor up more exponents to the PRP Ready status if to ask another user for an entire month. Ben Delo finished 3 M168M exponents within only 4 days and 6 hours, thus at least 28 PRP ready exponents will be required to fulfill that time consider a full month with every components run up to the maximum efficiency.
I personally don't feel you are being fair to Ben to ask him to run your exponents. You could easily drive him away from GIMPS if you pester him too much. I expect if you got lucky (and it would be just luck), that one of your exponents was prime, he would probably give you the money. But you will know that he found the prime - not you.

Running AWS instances is not exactly easy. I almost gave up due to the difficulty I had with their blinking captcha. You can get a free account for a year. It's painfully slow, but you will get to learn how to use it. Alibaba also offer cloud computing, but they wanted too much personal information for my liking, so I did not complete the sign-up, once I realised they wanted a copy of my passport. Google is another.

There's a big thread somewhere about using Amazon AWS. Find the thread and read about it. The cheapest way is to use "spot instances" where you pay for computer resources that nobody else wants to use at that time. But if another user comes along, willing to pay more, your job stops. You do this on an "elastic file system", so the data is available to another instance you start.
i've only ever paid by the hour, so the process keeps running. That costs a lot more, but I got free credits from Amazon, so it did not cost me a penny.

I would give you my thoughts on buying computers on a limited budget. You are better off not buying the fastest available, as you pay a hefty premium for a little extra speed. Over the course of the years, you will get more work done by keeping away from the very latest chips, but upgrading more frequently. If you have $4000 to spend, and intend upgrading in 4 years, that has cost you$1000/year. If instead you spend $2000 today, and another$2000 in 2 years time, it will have cost you the same over 4 years, but you will get better throughput. A $2000 computer in 2 years time is likely to have much better performance than a$4000 computer today.

2021-08-14, 22:54   #223
tuckerkao

"Tucker Kao"
Jan 2020

22·3·47 Posts

Quote:
 Originally Posted by drkirkby A $2000 computer in 2 years time is likely to have much better performance than a$4000 computer today.
A top end PC would be like AMD Threadripper 6970X(Zen 4), AMD Radeon RX 7900 XT(RDNA 3) by Aug 2023.

I'm sure that the M168M range will still be stay solidly deep in Cat 4 during that time.

I want a PC that runs as fast as Ben Delo's AW Instance, the good question is when should I buy this PC for the best price per efficiency output. I don't have plans to go after exponents that are much larger than M168M. My current guess will be it's definitely not the end of this year or the beginning of next year.

Which will generate more Prime95 output? 2 AMD Ryzen 5900X or 1 AMD Threadripper 5970X? The total prices are about the same between the 2 choices.

Quote:
 Originally Posted by chalsall That is sound counsel.
chalsall seems to be happier today.

Last fiddled with by tuckerkao on 2021-08-14 at 23:08

2021-08-14, 22:56   #224
chalsall
If I May

"Chris Halsall"
Sep 2002

2×5,023 Posts

Quote:
 Originally Posted by drkirkby I would give you my thoughts on buying computers on a limited budget. You are better off not buying the fastest available, as you pay a hefty premium for a little extra speed.
That is sound counsel.

2021-08-14, 23:27   #225
chalsall
If I May

"Chris Halsall"
Sep 2002

273E16 Posts

Quote:
 Originally Posted by tuckerkao chalsall seems to be happier today.
Please take this in the namaste context within which it is intended...

Babylon 5 -- Humor is such a subjective thing...

IMHO, reading (and watching) both history and "hard sci-fi" is worth the time to be ready for "now".

2021-08-15, 10:18   #226
drkirkby

"David Kirkby"
Jan 2021
Althorne, Essex, UK

26·7 Posts

Quote:
 Originally Posted by tuckerkao A top end PC would be like AMD Threadripper 6970X(Zen 4), AMD Radeon RX 7900 XT(RDNA 3) by Aug 2023.
A lot can happen in the computer world in 2 years. You have two competing manufacturers (AMD and Intel) both with big R&D and marketing departments.
Quote:
 Originally Posted by tuckerkao I'm sure that the M168M range will still be stay solidly deep in Cat 4 during that time.
Yes, because most people can see it's better to test smaller exponents first.
Quote:
 Originally Posted by tuckerkao I want a PC that runs as fast as Ben Delo's AW Instance, the good question is when should I buy this PC
It will probably be affordable 4 years from now. I doubt you have Ben's money, so I think it is unrealistic for you to buy hardware in 2021 that Ben is using in 2021.

Since I received $500 of free resources on Amazon, I had some discussions with Ben about what Amazon AWS instances he used. He said he had not benchmarked it for a couple of years, but suggested a few types. One of those was c5.metal https://aws.amazon.com/ec2/instance-types/c5/ in North Virginia. A bit of research showed me that North Virginia was the cheapest location for almost any instance I could find. So I decided to try an Amazon c5.metal instance. Those instances used 3 GHz Intel Xeon Platinum 8275CL CPUs. They are not a common CPU, and almost certainty supplied only supplied to Amazon. But they are the same 82xx series as used in my Dell 7920. I can tell you the new cost of CPUs of comparable performance will be around$10,000 each in 2021, and since two were used, that is \$20,000 in CPU cost alone. Also bear in mind that CPUs designed for servers need ECC RAM, and that is expensive.

Using a GPU seems to be more cost-effective for GIMPS than using a CPU. The Radeon VII is well known to be good.

By far the most sensible thing for you to do is move away from the 168 million exponents.
For any given hardware, they are going to take you more than 4x as long to run a PRP test on as a category 0 or 1 exponent. You also need to do the trial-factoring and P-1 factoring too, adding further to the time to test. So maybe it is taking you 5~6 times longer to test than a 105 million exponent.

2021-08-15, 13:25   #227
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

22×1,481 Posts

Quote:
 Originally Posted by drkirkby For any given hardware, they are going to take you more than 4x as long to run a PRP test on as a category 0 or 1 exponent. You also need to do the trial-factoring and P-1 factoring too, adding further to the time to test. So maybe it is taking you 5~6 times longer to test than a 105 million exponent.
For PRP or LL or the P-1 optimized for them, (168/105.5)2.1 ~2.66 times. Even ignoring the additional TF effort, the underlying point is still valid.

2021-08-15, 14:29   #228
drkirkby

"David Kirkby"
Jan 2021
Althorne, Essex, UK

26·7 Posts

Quote:
 Originally Posted by kriesel For PRP or LL or the P-1 optimized for them, (168/105.5)2.1 ~2.66 times. Even ignoring the additional TF effort, the underlying point is still valid.
Oops, I'm not sure how I miscalculated that, to get more than 4x.

2021-08-16, 03:40   #229
tuckerkao

"Tucker Kao"
Jan 2020

22×3×47 Posts

Quote:
 Originally Posted by kriesel For PRP or LL or the P-1 optimized for them, (168/105.5)2.1 ~2.66 times. Even ignoring the additional TF effort, the underlying point is still valid.
Too many users are throwing the dices in the M105M region now, thus it's very hard to acquire large amount of the consecutive properties. I prefer to have a monopoly of an entire row of TFs and P-1s written in my name.

I have the desire to put a skyscraper on M168202123, then expand out from there. My special numbers include M168,***,*23 and it's year 2021 now, so an ideal location to establish my headquarter and fits my favorite theme. I want to own the entire avenue of M168,202,*** the same way as how Roland Clark owns the street of M160,374,***.

Drkirkby built a skyscraper on M105211111, but many of the surrounding properties were already owned by Ben Delo, thus Drkirkby will have to find another area in the upcoming months. He picked this exponent probably because of the repunit digits pattern from the right-end, so it's easier to remember and he can come back to visit this place often.

Last fiddled with by tuckerkao on 2021-08-16 at 04:39

2021-08-16, 06:48   #230
tuckerkao

"Tucker Kao"
Jan 2020

22·3·47 Posts

Quote:
 Originally Posted by Viliam Furik Please, report your TF results. No credit doesn't mean it's completely worthless...
If there was a factor found, I'll submit the result for sure. If no factor, I sometimes skip the report if not done in my own place.

I tried to submit "no factor for M168202123 from 2^78 to 2^79 [mfaktc 0.21 barrett79_mul32_gs]" to the server, got suspicious results 3 times, it went through the 4th time. For the copy and the paste, if I by chance entered 1 letter or 1 number wrong, it didn't work. I believed I put in barrett76 at first, then corrected it to barrett79.

There seem to be some issues when I want to submit something that hasn't been done by my own computer but on a friend's gaming PC. Nvidia Geforce 3070 Ti is of course much faster than my current Geforce 780 GTX.

Yes, I knew there were no factor found within a certain range of bits, but was that really my work?

Last fiddled with by tuckerkao on 2021-08-16 at 07:44

2021-08-16, 13:32   #231
Viliam Furik

"Viliam Furík"
Jul 2018
Martin, Slovakia

2×353 Posts

Quote:
 Originally Posted by tuckerkao If there was a factor found, I'll submit the result for sure. If no factor, I sometimes skip the report if not done in my own place. I tried to submit "no factor for M168202123 from 2^78 to 2^79 [mfaktc 0.21 barrett79_mul32_gs]" to the server, got suspicious results 3 times, it went through the 4th time. For the copy and the paste, if I by chance entered 1 letter or 1 number wrong, it didn't work. I believed I put in barrett76 at first, then corrected it to barrett79. There seem to be some issues when I want to submit something that hasn't been done by my own computer but on a friend's gaming PC. Nvidia Geforce 3070 Ti is of course much faster than my current Geforce 780 GTX. Yes, I knew there were no factor found within a certain range of bits, but was that really my work?
There is an output from the mfaktc program, you know? You just copy-paste it...

 Similar Threads Thread Thread Starter Forum Replies Last Post LaurV GPU to 72 81 2020-12-02 05:17 jfamestad PrimeNet 3 2016-11-06 20:32 lidocorc PrimeNet 4 2008-11-06 18:48 jocelynl 15k Search 0 2003-07-11 14:23 ThomRuley Software 5 2003-05-30 20:34

All times are UTC. The time now is 13:15.

Fri Dec 3 13:15:37 UTC 2021 up 133 days, 7:44, 0 users, load averages: 1.21, 1.22, 1.23