![]() |
|
|
#1 |
|
Dec 2017
12 Posts |
Hi Everybody,
First of all: I am totally noob and English is not my mother tongue :) I started this whole mersenne prime search in 2009, when I studied matemathics on a higher level in secondary school but after a few weeks my PC broke down because of other reasons, and I finished searching for primes until today. I decided that I would like to build a PC just to search for mersenne primes. Can you recommend me some hardwares? Motherboard? CPU? RAM? What should you buy? Thank you for instance Gergő |
|
|
|
|
|
#2 |
|
"Victor de Hollander"
Aug 2011
the Netherlands
23×3×72 Posts |
- Buy a good power supply from a decent brand with good rating (80+ gold for instance)
- If you're only going to use the PC for LL-testing, than you don't need an expensive motherboard or CPU. Something like a quad-core i5 with DDR4-2400 memory will give you great bang-for-buck. Some more advice in this thread: http://mersenneforum.org/showthread.php?t=20795 Make sure to calculate your electricity costs (per year for instance), so you don't get unpleasant surprises when you get the energy bill. |
|
|
|
|
|
#3 |
|
Sep 2003
5×11×47 Posts |
A lot of graphics cards were bought for cryptocurrency mining, but I think that will soon reach the point where it's no longer practical. Either because dedicated hardware like ASICs is the only practical option, or because of a switch to different verification methods.
So maybe a bunch of used GPUs will go on sale in 2018 at attractive prices... ? |
|
|
|
|
|
#4 | |
|
Nov 2008
1111101012 Posts |
Quote:
|
|
|
|
|
|
|
#5 |
|
Serpentine Vermin Jar
Jul 2014
7·11·43 Posts |
I imagine it'd be something along the lines of an ASIC that did ridiculously fast FMA and had a lot of RAM caching or something. No idea of the complexity of that though and I have every reason to believe it'd probably still be more cost effective to get a CPU or GPU that already has that capability. But I could be wrong.
The ASICs for mining have a profit motive so people can and do pump a lot of research and fab into it. For anything else you'd have to rely on some really dedicated people with the right knowledge and a lot of free time. On the other hand, the new CPUs that have FPGA capabilities might be useful. I haven't read that much about how versatile those FPGAs are but if people developed some cool libraries of functions for them, we might come across something that helps speed things up. |
|
|
|
|
|
#6 | |
|
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2
22×23×103 Posts |
Quote:
A substantially similar math problem (genome alignments) is not only been run as a PoC but is successfully commercialized. While many genome centers still run conventional CPU alignments and get the results in 8-12 hours for a typical genome on a 32-core node (typical demand is tens of thousands genomes per year at a large center, so they don't spread one genome over many nodes), the FPGA solution runs in 45 minutes on a special node. These are now available at AWS, too! Usually, the algorithmic entry threshold is having a team of Verilog experts. This is far more involved than CUDA or OpenCL coding. |
|
|
|
|
|
|
#7 | ||
|
Feb 2016
UK
6608 Posts |
Quote:
Some ideas: on the higher end, i7-7800X - 6 cores, support for quad channel ram, and AVX-512. The last part might offer a performance increase once software has been updated to support it. The CPU isn't too expensive, but the motherboards will cost more than mainstream consumer level, and you'll need 4 sticks of ram, preferably higher speed around 3000 mark. A popular budget motherboard for this is the Asus X299 TUF mark 2. For a more mainstream idea, the i3-8100 or i3-8350k are interesting. These are 4 core CPUs, the latter supporting overclocking. You'll probably need to get a Z370 chipset motherboard to enable higher speed ram, so that it wont limit. Again, aim for ram speeds around 3000 mark, but this only needs two sticks for dual channel. For overall work throughput, you'd have to work out if it is more efficient to run multiple simpler systems, or fewer faster ones. If only building one system, it may depend more on your budget. Quote:
I've kinda wondered, what is it about the implementation that limits us to "only" 64-bit/80-bit operations? Wouldn't it be lovely to have orders of magnitude bigger registers to work with? Say kilobit? I don't know how multiplication functions are actually handled in a modern CPU, but if I were to do it from scratch I'd probably brute force a parallel long multiplication to minimise latency, with a quick sum at the end. The cost being circuit quantity (assuming integer math for now, assume principles could be expanded for floating point). |
||
|
|
|
|
|
#8 | |
|
Einyen
Dec 2003
Denmark
35×13 Posts |
Quote:
The 2 first options might be very similar in hardware, but the 3rd option would be a totally different setup. Do you only want to use CPU or do you want a graphic card for either trial factoring or for doing more LL tests on? |
|
|
|
|
|
|
#9 | |
|
"Composite as Heck"
Oct 2017
32D16 Posts |
Quote:
|
|
|
|
|
|
|
#10 | |
|
Banned
"Luigi"
Aug 2002
Team Italia
32×5×107 Posts |
Quote:
|
|
|
|
|
|
|
#11 |
|
"Kieren"
Jul 2011
In My Own Galaxy!
2×3×1,693 Posts |
Go for dual-rank DIMMs. There is a distinct performance boost. Kingston HyperX is serving me well.
Unfortunately, I see that fewer of their spec sheets state Rank than I remember. The only dual set I found is 16GB parts. |
|
|
|
![]() |
| Thread Tools | |
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| Search for prime Gaussian-Mersenne norms (and G-M-cofactors) | Cruelty | Proth Prime Search | 158 | 2020-07-31 22:23 |
| The prime-crunching on dedicated hardware FAQ (II) | jasonp | Hardware | 46 | 2016-07-18 16:41 |
| Low clock speeds on Mersenne Prime search | Ammonia | Hardware | 2 | 2016-01-21 17:46 |
| Can I specify the range to search the Mersenne Prime? | Unregistered | Information & Answers | 22 | 2012-03-20 11:38 |
| The prime-crunching on dedicated hardware FAQ | jasonp | Hardware | 142 | 2009-11-15 23:20 |