Bitcoin Forum
November 17, 2024, 11:28:38 AM *
News: Check out the artwork 1Dq created to commemorate this forum's 15th anniversary
 
   Home   Help Search Login Register More  
Pages: [1]
  Print  
Author Topic: The real deal with designing a mining rig  (Read 2276 times)
rocksalt (OP)
Newbie
*
Offline Offline

Activity: 51
Merit: 0



View Profile
June 23, 2011, 08:33:36 AM
 #1

hi guys,

As a rule i tend to follow what you lot specify with hardware which is brilliant.
I've been looking and comparing and searching and im not finding some hard info on why certain items are better than others....

Can someone enlighten me please...


GRAPHICS CARD

1.Is it the number of GPU /CORE?
2.speed of said GPU /CORE?
3.Memory speed on card?
4.How about shaders, is this related? i see lots of post about shaders and unlocking them.. what's the general rule of thumb here?
5.Bus speed access, pretty much is PCIe x16 ( no crossfire needed as apparently it works against you )

RAM & CPU

The other related aspects of a mining rig, under-clocking RAM and CPU are pretty much solid, and also the motherboard types as well. I think it's a given that CPU isn't a real issue neither is RAM, but i was browsing youtube last night and watching some vid on bitmining and one guys had what looked like a solid set-up, a quad core amd cpu but each core was totally maxed out and he was only GPU mining... might be his set-up but it was food for thought. He was running windows 7 as far as i could tell.

PSU

How about PSU? say a 1000w PSU, the 12v line, multiple lines, or mutiple 650w PSU's ? is there a point of diminishing return here? i know some of the top flight PSU's have great energy efficiency but is there a way to maximise this?



My basic idea is to construct a rig that is efficient as possible, not just in terms of H/s but in terms of heat generation ( gonna love this in the winter ) but also the amount of power being drawn to the PSU and then in turn used by the various components. Is there any sort of guide apart from the general hardware list that dictates what is and what combination is used? like Linux with a type of motherboard, 4x 5830 cards, a 1000w psu and a quad intel cpu performs worse in terms of efficiency than the same set-up with windows and AMD cpu ?

I think now that the difficulty is approaching the point where most average miners are going to expend more on power than they can get back in bitcoins, i think we need to start delving into this area and research it more think nuclear against wind/tidal energy lol

As you lot are uber experienced in this and pretty much most of us have experimented a lot with set-ups and rigs for different purposes i think it would be good too pool knowledge and try to bring the cost of mining down a bit, at least until new technology like AMD's new GIGAFLOP CPU lines become available next year... CPU mining might become more attractive or at least a close run thing.
detroit
Member
**
Offline Offline

Activity: 69
Merit: 10


View Profile
June 23, 2011, 12:03:01 PM
 #2

Number of GPUs is certainly the biggest factor!  Nothing can touch the two-chips-on-one-card cards!
Number of shaders is how many simultaneous operations you can perform
Speed of core is how fast each one goes.

Clearly the last two work together to determine a cards potential hash rate.

Tradehill.com referral code: TH-R1494
Please consider using it if I've said something useful!
Waltibaba
Newbie
*
Offline Offline

Activity: 38
Merit: 0



View Profile
June 23, 2011, 12:44:24 PM
 #3

Number of GPUs is certainly the biggest factor!  Nothing can touch the two-chips-on-one-card cards!
Number of shaders is how many simultaneous operations you can perform
Speed of core is how fast each one goes.

Clearly the last two work together to determine a cards potential hash rate.

What do you mean? My two-chips-on-two-cards (5830+6950) cost 1/2 of a 6990 and give more MH/s, less heat, less noise, less power use. ok they use 2 pcie slots, but for the difference in price I can get an entire new mobo+cpu+ram setup...

OP: in short, there's no such comparison for complete systems. You dont need a good cpu for mining, nor ram/video ram. You even downclock video ram to give less heat... It looks like the # of shader units determines the hash rates (and then the core speed of course).

Try to add up costs, and find whats best for you based on hardware prices/electricity cost. I don't care about electricity costs, so I run 3x5830s and a shader unlocked 6950. If you do, consider getting 5770s; they're the best in MH/joule. PSU is scaled to your PC needs. I have found that the best system for me is a small 2 slot mobo for 55 euros, sempron 140, 2gb ram, and 500W antec psu with 2 5830s.

If you wanna run a bigger system, check the prices of psus vs. watts they deliver. I really recommend using multiple psus (if you're running without a case), its no hassle to set up only takes 1 minute, and can be scaled better. Plus, most of the time, its cheaper at >900W.
rocksalt (OP)
Newbie
*
Offline Offline

Activity: 51
Merit: 0



View Profile
June 23, 2011, 01:04:40 PM
 #4

Yup i was considering the dual psu approach in an open case design. i would have thought splitting the load across multiple psu's would give better stability and lower power draw. it's much the same with a server, you split your os across two disks and you get a total fast read/write ( the down side is if you loose one disk, you lose the OS as its stripped across both, but thats the same as a single disk anyway )

for example, single motherboard with 4 PCIe slots with 4 5830's. Each one powered from a fairly good 250/300w psu that draws only the absolute required amount of watts for each card.
I think I might have to put a power meter on the pus to work it out for each card... kind like scientific investigation Tongue
rocksalt (OP)
Newbie
*
Offline Offline

Activity: 51
Merit: 0



View Profile
June 23, 2011, 02:47:19 PM
 #5

Ok, so i've been doing a lot of digging and asking questions and i've come up with the following....

The basis of this is asking why certain cards are favoured over others...

READ: http://en.wikipedia.org/wiki/Stream_Processing

With this in mind and a brain frazzled from cross checking certain technical data then comparing to the specs of the cards most favoured.. i find the following:

5670
- Core Clock: 775MHz
- Memory: 512MB GDDR5
- Memory Clock: 4000MHz (Effective)
- Processing Cores: 400
- Bus Type: PCI-Express 2.1

5770
- Core Clock: 850MHz
- Memory: 1024MB GDDR5
- Memory Clock: 4800MHz (Effective)
- Processing Cores: 800
- Bus Type: PCI-Express 2.1

5850
- Core Clock: 725MHz
- Memory: 1024MB GDDR5
- Memory Clock: 4000MHz (Effective)
- Processing Cores: 1440
- Bus Type: PCI-Express 2.1

6770
-Memory: 1024 MB GDDR5
- Core Clock: 800 MHz
- Memory Clock: 4400 MHz
- Processor Streams: 800
- Bus Type: PCI-Express 2.1

6790
- Core Clock: 840MHz
- Memory: 1024MB GDDR5
- Memory Clock: 4200MHz (Effective)
- Processing Cores: 800
- Bus Type: PCI-Express 2.1

6950
 Core Clock: 800MHz
- Memory: 1024MB GDDR5
- Memory Clock: 5000MHz (Effective)
- Processing Cores: 1408
- Bus Type: PCI-Express 2.1

The choice of the two top cards is based upon performance and in this Processing streams are the key element. Now i think i understand the technical reasoning behind such a choice, i compared to Nvidia and realised why they suck donkey balls.

Both the 5850 and the 6950 have extremely fast streams and for hash calculation, stream win hands down as the data is constant and only changing by a small variance, as opposed to graphics rendering which changes constantly.
I know this might be old news to some of you, but its pretty interesting when you discover the reason why behind it Smiley
SlipperySlope
Hero Member
*****
Offline Offline

Activity: 686
Merit: 501

Stephen Reed


View Profile
June 23, 2011, 03:28:29 PM
 #6

Quote
CPU mining might become more attractive or at least a close run thing.

The x86 and ARM CPU ALU are way too powerful, and way too few in number, to be efficient bitcoin miners.  Their architecture is designed to execute general purpose threads and thus are not sufficiently optimized when compared to the Stream Processing Units on the AMD Radeon graphics cards.

For example the HD 5770 cards that I use each have *800* Stream Processing Units overclocked to 960 MHz.  Each Stream Processing Unit takes a tiny fraction of the chip die space occupied by an x86 CPU ALU.  So for each lithography generation, there will be many more Stream Processing Units per chip than x86 ALUs - like 100 times more!

Its likely that in the future that chips will be manufactured expressly for bitcoin mining - see the ASIC discussions on this forum.  These are supposed to easily surpass existing AMD graphics card performance and power-efficiency.

Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!