SGI: Discussion

As part of my 2014 Winter Warmup RetroChallenge project I thought it would be fun to compile cpuminer for IRIX and see how my trusty old SGI O2 would work at Litecoin mining.



Download
If you are curious to benchmark your own system you can grab an IRIX 6.5 copy of cpuminer here:
irix-cpuminer.zip

Benchmarks
Here is a graph with the results from my systems:


For comparison I tested the same cpuminer code on a few computers in my house:

• 2008 Apple Mac Pro Xeon - 8 core 3.2GHz System (25.6 GHz) = 8 threads x 4.78 khash/s
• 2006 HP DC5750 Desktop AMD Athlon 64 X2 - 2 core 2.1 Ghz (4.2Ghz) = 2 threads x 1.6 khash/s
• 2013 Raspberry Pi ARM 700 MHz = 0.33 khash/s
• 1996 SGI O2 195 MHz MIPS R10K = 0.08 khash/s

Installation
Step 1. Install the required Nekoware library: neko_curl-7.19.7.tardist

You can get the nekoware tardist files from the primary mirror:
http://nekoware.dustytech.net/index.php?path=current/

I ended up adding the following Nekoware compiled packages to my system so I could develop and run cpuminer:

• neko_curl-7.19.7.tardist
• neko_cyrus_sasl-2.1.22.tardist
• neko_db4-4.4.20.tardist
• neko_gawk-3.1.5.tardist
• neko_gmp-4.2.1.tardist
• neko_krb5-1.8.tardist
• neko_libidn-0.5.4.tardist
• neko_openldap-2.3.26.tardist
• neko_openssl-0.9.8x.tardist
• neko_readline-6.1-4.tardist
• neko_unixodbc-2.2.10.tardist

Step 2. Give the minerd executable permissions


Step 3. The minerd program should be copied to the /usr/local/bin folder.


Step 4. Launch minerd with the stratum server URL and login info.


Notes
For some reason the IRIX compiled version of minerd only accepts the single dash version of arguments. This means you can use the -o and -O options to provide the server path and login details.

ah that's a nice one
you're not gonna get rich there tho i guess

as for the options, irix doesn't support long options (double dash) in general but brief options do the job just as well.

Octane2 2xR12000 400MHz does 2x 0.17 khash/s .......

0.34 khash/s on the "mighty SGI Workstation" vs. 0.33 khash/s on the raspberry LOL

I'm half-tempted to see what the quad G5 makes of it.

I'm interested to see, what your Fuel is making of it

Saw this on Twitter and got it running on my Onyx2; launched with 4 separate threads (one per CPU) with each thread pulling around 0.23 khash/sec. Hardware info here: viewtopic.php?f=14&t=13448

So one R14000 550 MHz is 35 % faster than one R12000 400 MHz in Octane2 when 'mining' Litecoin

Geoman wrote: So one R14000 550 MHz is 35 % faster than one R12000 400 MHz in Octane2 when 'mining' Litecoin

So clock for clock the R12k & R14k are basically the same it seems, at least for this test.

Not really a surprise as from my reading the R10k was the last major upgrade to the MIPS based CPUs and future versions were just minor tweaks and die shrinks. Shame really.

They're actually *500MHz* CPUs, not 550MHz

But yes, the big reason I decided to get rid of my 4x700MHz Tezro was because it wasn't much faster CPU-wise than the Onyx2 (and IR3 blew the doors off of V12).

I want to see an R16.000 in 22nm with on-die-caches @3 GHz --that would be fun!

Besides, it would be so great to upgrade the R1X.000-based SGIs to the newest imaganation/MIPS incarnations like it has been done with O2s R5K and PMC-Sierra 700 MHz.

i wonder what how much a Fuel would produce ?

my guess is: fastest Fuel: 0.5 per CPU

@Neko: a 25% increase in clockrate (and refined design) yields a 35% increase in computing-power (at least in litecoin-mining) --actually this is very good imho

nekonoko wrote: (and IR3 blew the doors off of V12).

Ha ! Should have kept your mouth shut, now I'm going to bother you !

In what way ? This is not just an intellectual exercise : I can go with 4 V12's or try to get IR shovelled into 4 dvi connectors. Both methods have advantages / disadvantages. But the thing you can't tell without real-world experience is, how do IR and VPro graphics compare in use ?

hamei wrote:

nekonoko wrote: (and IR3 blew the doors off of V12).

Ha ! Should have kept your mouth shut, now I'm going to bother you !

In what way ? This is not just an intellectual exercise : I can go with 4 V12's or try to get IR shovelled into 4 dvi connectors. Both methods have advantages / disadvantages. But the thing you can't tell without real-world experience is, how do IR and VPro graphics compare in use ?

I am very interested in this product and/or service as well. (Though in my case, it is purely intellectual.)

I wonder how much faster it would run if compiled with MIPSpro

Octane with 2x R14k 600MHz does 2x 0.28 khash/s.

At least that's faster than the Raspberry...

So I let the Octane mining for around 3 hours: earned one share

I'm seeing .11 Kh/s out of my Octane (250MHz R10K).

pip wrote: I wonder how much faster it would run if compiled with MIPSpro

Doesn't want to compile for me with MIPSpro (-pthreads not supported or something like that)

I've only got GCC installed on my SGI O2. If someone wants to provide ssh access to an IRIX box with MIPSpro installed I'm willing to compile an optimized version of cpuminer.