mersenneforum.org Not about Mersenne prime coin thread
 User Name Remember Me? Password
 Register FAQ Search Today's Posts Mark Forums Read

2021-09-09, 14:12   #23
unk

Sep 2021

17 Posts

It turns out that there is already a coin that implements useful calculations. https://primecoin.io / A coin with a very low price.
Quote:
 Originally Posted by LaurV Indeed, you really fail to understand how a currency (and especially a decentralized one) works.
I agree with your arguments. I knew about it, I just didn't take it into account. The server can be hacked. It was possible to steal the same Ethereum from The DAO. The server can deceive network participants.
Quote:
 Originally Posted by chalsall There is no panacea. Deal with it.
A panacea? The smallpox vaccination is too good to be true. The knife is too simple to be useful. There are no absolutely unbreakable fonts (in fact, there are cipherblocks). There are a lot of panaceas.
Quote:
 Originally Posted by kriesel Yes, and far too many of them are in positions of power or taking to the streets violently. And their emotion-driven wrongheaded votes count equally as much as those of the clueful.
https://slatestarcodex.com/2015/06/0...bit-hole-goes/
Quote:
 You once read about something called Gell-Mann Amnesia, where physicists notice that everything the mainstream says about physics is laughably wrong but think the rest is okay, doctors notice that everything the mainstream says about medicine is laughably wrong but think the rest is okay, et cetera. You do not have Gell-Mann Amnesia. Everyone is terrible at everything all the time, and it pisses you off.
The arguments of LaurV and kriesel convinced me. The server, the centralization, are really too vulnerable. Developers really won't update numerous types of applications. And if the second problem can be solved by" simply " writing a new project from scratch, then the first one cannot be solved. We are required to trust the server, but there is no reason for such trust.
Quote:
 Originally Posted by kriesel Only prime95/mprime developed and maintained by George Woltman has the secret security code software for secured interaction with the PrimeNet server.
The results of calculations can not be faked? I'm glad.
This is the end of the list of questions. Thank you all for the conversation, for the consultations.

Quote:
 Originally Posted by kriesel I suggest you learn more about the project & personnel, try to get a current understanding.
There are a lot of them. I just participated in the project, why do I need detailed information about it? Everything is clear to me about Mersenne prime coin. We can't trust a server that will collect statistics and send out rewards.

(machine translation)

Last fiddled with by unk on 2021-09-09 at 14:21

2021-09-10, 12:32   #24
unk

Sep 2021

17 Posts

Quote:
 Originally Posted by LaurV The main idea of a criptocoin is decentralization, getting rid of the bank. Banks can be robbed and servers can be hacked.
You can't trust cryptocurrencies either. They say about Bitcoin that its creator Satoshi Nakamoto has mined the cryptocurrency for himself. From January 3, 2009 to January 25, 2010, only one person was engaged in mining, and the absolute majority of coins mined at that time are still stored with him. This is about a million coins. At the Bitcoin exchange rate of $46209, this will amount to$ 46 billion 209 million. Just for the fact that a person was mining first. In the branches of Bitcoin, these coins are saved, that is, the person who first mined receives several more types of coins, because they are based on the Bitcoin block chain.
The Ethereum currency issued 72,009,995 coins at the time of launch. DASH was also dishonest. In the first hour of the launch, about 500,000 DASH were mined. The next 1,000,000 DASH were mined in 7 hours and finally another 400,000 in the next 36 hours. The authors of Chia just drew 21 million coins. To sum up — everything in Mersenne prime coin depends on the honesty of the developers. But the same problem with developers exists in Bitcoin and Ethereum. So there is nothing new here, and you just need to trust the developers.
Another thing, I think that the development of the coin will not benefit the developers. In addition, in addition to the coin, it will be necessary to create a client for calculations, encrypt its code so that the client is not hacked and fake statistics are not sent to the server. And this requires qualified programmers. That is, everything depends on money.
Quote:
 Originally Posted by kriesel Financial gain is not what drives GIMPS participation though.
Yes. Most of them participate because of the desire to find a Mersenne prime number or because of the competitive effect.
I want to ask you something. Can the LL and D tasks be calculated on video cards?
If someone wants to mine, then you can go and mine. GIMPS does not imply mining yet. Perhaps in 40 years they will combine mining and GIMPS. But there remains a problem of trust in the organizers. In real cryptocurrencies, this is a very serious problem.

(machine translation)

2021-09-10, 13:04   #25
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

22·1,447 Posts

Quote:
 Originally Posted by unk Can the LL and D tasks be calculated on video cards?
Yes, (and P-1, and TF) that has been possible for several years. But it is terribly inefficient to do LL & LLDC anywhere for production testing, compared to performing PRP with GEC and proof generation, almost eliminating chance of error and work of verification. See gpuowl, cllucas, cudalucas in the available software summary attachment in the reference info. Fastest is gpuowl 6.11-380 or near it, on Radeon VII or perhaps RX6900XT, or A100 on Colab Pro. Read https://www.mersenneforum.org/showpo...35&postcount=1
Rough rule of thumb, most NVIDIA GPUs (especially RTX20xx, GTX16xx or newer) are more effectively employed at TF due to low DP performance, while AMD GPUs do quite well at PRP or P-1and less well at TF due to higher DP performance.

Last fiddled with by kriesel on 2021-09-10 at 13:14

 2021-09-10, 15:54 #26 unk   Sep 2021 17 Posts In mining, video cards are much more powerful than processors. It turns out that processors are useless in the project? Do I have to buy a video card? On my processor, one task counts for more than two months. And how long does it take to run PRP with GEC and proof generation on a video card? Three days? (machine translation)
2021-09-10, 16:17   #27
Zhangrc

"University student"
May 2021
Beijing, China

22·33 Posts

Quote:
 Originally Posted by unk Do I have to buy a video card? And how long does it take to run PRP with GEC and proof generation on a video card? Three days?
You can if you want. RX 6900 XT or Radeon 7 or other recent AMD Radeon cards can complete a PRP in less than two days, while a 3090 needs 4 to 5 days.
You need to run a software on it, called GPUowl.

Last fiddled with by Zhangrc on 2021-09-10 at 16:18

2021-09-10, 16:57   #28
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

22×1,447 Posts

Quote:
 Originally Posted by unk how long does it take to run PRP with GEC and proof generation on a video card?
PRP etc times depends greatly on the video card model (or CPU for mprime) and exponent. See run time scaling reference posts. Following is a recent example that just completed, ~24.22 hours for 100M, on a Radeon VII running at 189W reduced power for economy:

Code:
2021-09-09 11:13:18 gpuowl v6.11-380-g79ea0cc
2021-09-09 11:13:18 config: -user kriesel -cpu asr2/radeonvii0-w2 -d 0 -use NO_ASM -maxAlloc 13000 -cleanup -proof 10
2021-09-09 11:13:18 device 0, unique id ''
2021-09-09 11:13:18 asr2/radeonvii0-w2 100006129 FFT: 5.50M 1K:11:256 (17.34 bpw)
2021-09-09 11:13:18 asr2/radeonvii0-w2 Expected maximum carry32: 2C960000
2021-09-09 11:13:19 asr2/radeonvii0-w2 OpenCL args "-DEXP=100006129u -DWIDTH=1024u -DSMALL_HEIGHT=256u -DMIDDLE=11u -DPM1=0 -DAMDGPU=1 -DWEIGHT_STEP_MINUS_1=0x9.4558a9978254p-4 -DIWEIGHT_STEP_MINUS_1=-0xb.bd51cfcb92d2p-5 -DNO_ASM=1  -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2021-09-09 11:13:24 asr2/radeonvii0-w2 OpenCL compilation in 4.36 s
2021-09-09 11:13:25 asr2/radeonvii0-w2 100006129 OK        0 loaded: blockSize 400, 0000000000000003
2021-09-09 11:13:25 asr2/radeonvii0-w2 validating proof residues for power 10
2021-09-09 11:13:25 asr2/radeonvii0-w2 Proof using power 10
2021-09-09 11:13:26 asr2/radeonvii0-w2 100006129 OK      800   0.00%;  876 us/it; ETA 1d 00:20; dbd4f470588ccb47 (check 0.61s)
2021-09-09 11:16:18 asr2/radeonvii0-w2 100006129 OK   200000   0.20%;  863 us/it; ETA 0d 23:56; d0fd564e891c752a (check 0.53s)
...
2021-09-10 11:20:24 asr2/radeonvii0-w2 100006129 OK 100000000  99.99%;  876 us/it; ETA 0d 00:00; (redacted) (check 0.83s)
2021-09-10 11:20:34 asr2/radeonvii0-w2 CC 100006129 / 100006129, (redacted)
2021-09-10 11:20:36 asr2/radeonvii0-w2 100006129 OK 100007200 100.00%;  928 us/it; ETA 0d 00:00; (redacted) (check 0.54s)
2021-09-10 11:20:36 asr2/radeonvii0-w2 proof: building level 1, hash (redacted)
2021-09-10 11:20:37 asr2/radeonvii0-w2 proof: building level 2, hash (redacted)
2021-09-10 11:20:38 asr2/radeonvii0-w2 proof: building level 3, hash (redacted)
2021-09-10 11:20:40 asr2/radeonvii0-w2 proof: building level 4, hash (redacted)
2021-09-10 11:20:43 asr2/radeonvii0-w2 proof: building level 5, hash (redacted)
2021-09-10 11:20:48 asr2/radeonvii0-w2 proof: building level 6, hash (redacted)
2021-09-10 11:21:00 asr2/radeonvii0-w2 proof: building level 7, hash (redacted)
2021-09-10 11:21:21 asr2/radeonvii0-w2 proof: building level 8, hash (redacted)
2021-09-10 11:22:02 asr2/radeonvii0-w2 proof: building level 9, hash (redacted)
2021-09-10 11:23:25 asr2/radeonvii0-w2 proof: building level 10, hash (redacted)
2021-09-10 11:26:10 asr2/radeonvii0-w2 PRP-Proof 'proof\100006129-10.proof' generated
2021-09-10 11:26:10 asr2/radeonvii0-w2 Proof: cleaning up temporary storage
2021-09-10 11:26:12 asr2/radeonvii0-w2 {"status":"C", (redacted)}
The cpu on this system is a mere Celeron G1840, making proof generation, GCDs etc unusually long.

Last fiddled with by kriesel on 2021-09-10 at 16:57

2021-09-10, 17:47   #29
unk

Sep 2021

1116 Posts

Quote:
 Originally Posted by Zhangrc You can if you want.
I can't, I don't have any money.
Quote:
 Originally Posted by kriesel The cpu on this system is a mere Celeron G1840, making proof generation, GCDs etc unusually long.
This means that a processor is also needed.
It is curious that the link https://www.mersenne.org/primes/ there are no PRP tests, only L-L. PRPs will prove themselves in the future, video cards in this test are much more efficient than processors in the L-L test.

(machine translation)

2021-09-10, 18:34   #30
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

22·1,447 Posts

Quote:
 Originally Posted by unk I don't have any money.
substitute your time for money. Colab (and some other cloud computing) offer some free use.

Quote:
 This means that a processor is also needed.
Of course. A CPU is needed to boot the system and for many other things too. Nearly all the PRP, P-1, or LLDC work is done on the GPU in gpuowl. Saving interim restart files and logs to disk is done by the CPU. Developers chose to use library (cpu-based) functions for GCD in P-1 rather than attempt to write a new GPU-based GCD.

Quote:
 It is curious that the link https://www.mersenne.org/primes/ there are no PRP tests, only L-L. PRPs will prove themselves in the future, video cards in this test are much more efficient than processors in the L-L test.
LL and PRP are both performable in CPU-only or GPU applications. PRP with proof is the better algorithm, independent of hardware and implementation, for production first tests. This exponent's LL TC was by gpuowl v6.11-380.
Robert Gerbicz announced his excellent PRP error check in August 2017, too late for Mersenne primes through Mp#49* (found 2016-01-07) or 50* really (found 2017-12-26). It took a while to get software to use it created, tested, announced, and widely deployed. In small sampling in January 2021, I saw a 20-40% rate of LL still lingering; in July it was down to ~1%. As of 2021-04-08, the server no longer issued LL first time test assignments, but some slow LL first test assignments may still be completing and getting reported.

Last fiddled with by kriesel on 2021-09-10 at 18:50

2021-09-11, 14:41   #31
unk

Sep 2021

17 Posts

Intel Core i3-10100F OEM costs 6550 rubles, 90 dollars. PowerColor AMD Radeon RX 6900 XT Red Devil costs 149999 rubles, 2061 dollars. It is not yet available, but it is available for 159999 rubles, 2199 dollars.
A video card is 22.9 times more expensive than a processor. And 45 times faster.
Quote:
 Originally Posted by kriesel substitute your time for money. Colab (and some other cloud computing) offer some free use.

I wanted to. I thought my computer would mine Mersenne prime coin. Now it is clear that a video card is needed for decent indicators. Even in mining, even in GIMPS. The fact that Colab is free is a good thing. What should I do to save money for the RX 6900 XT?

(machine translation)

2021-09-11, 15:06   #32
M344587487

"Composite as Heck"
Oct 2017

3×7×41 Posts

Quote:
 Originally Posted by unk Intel Core i3-10100F OEM costs 6550 rubles, 90 dollars. PowerColor AMD Radeon RX 6900 XT Red Devil costs 149999 rubles, 2061 dollars. It is not yet available, but it is available for 159999 rubles, 2199 dollars. A video card is 22.9 times more expensive than a processor. And 45 times faster.
It's a bit more complicated when you consider that every consumer CPU needs decent RAM, motherboard, PSU. A GPU needs to be plugged into something but many GPU's can share that same something, and that something can also be relatively junk (PSU excepted). Worst case (one machine per GPU) including the host the numbers are more like ~7 times more expensive not 22.9.

2021-09-11, 15:27   #33
unk

Sep 2021

1116 Posts

My computer cost me 25060 rubles. The video card is built into the processor. If I took a processor without built-in graphics, but would have bought a video card, I would have paid 171059 rubles. 171059/25060=6.83 times. You are absolutely right.
I took the prices from https://www.dns-shop.ru/, the processor costs $90.02, the video card costs$ 2061.56. The initial figures were inaccurate. These are accurate, I rounded them up according to the rules.

(machine translation)

Quote:
 Originally Posted by unk It is not yet available, but it is available for 159999 rubles, 2199 dollars.
But this figure was accurate. I rounded it up correctly.

(machine translation)

Quote:
 Originally Posted by kriesel As of 2021-04-08, the server no longer issued LL first time test assignments, but some slow LL first test assignments may still be completing and getting reported.
The project got rid of a lot of unnecessary work.

(machine translation)

Quote:
 Originally Posted by unk Intel Core i3-10100F OEM costs 6550 rubles, 90 dollars.
The price has changed right now. Already 6799 rubles, 93.44 dollars.

(machine translation)

 Thread Tools

 Similar Threads Thread Thread Starter Forum Replies Last Post thorken Software 66 2019-01-13 21:08 Uncwilly Miscellaneous Math 85 2017-12-10 16:03 petrw1 Puzzles 1 2015-02-06 23:17 davieddy Puzzles 62 2010-05-29 01:14 mfgoode Puzzles 65 2007-01-26 00:25

All times are UTC. The time now is 07:05.

Fri Oct 22 07:05:38 UTC 2021 up 91 days, 1:34, 1 user, load averages: 1.19, 1.31, 1.30

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.