Re:"God's Number" is 20 - Rubik's cube has been "solved"
by slig2 on Monday August 09, 2010 @08:18AM
(#1587735)
Any guesses of how much this would cost to do on EC2?
[ Parent
| on HN ]
Re:"God's Number" is 20 - Rubik's cube has been "solved"
by bd on Monday August 09, 2010 @11:35AM
(#1588287)
With some handwaving it could be about $82.5K (computation spread over 1 year) or $125K (computation done in 1 month).
total cost was 35 CPU years (Intel Nehalem, 4-core, 2.8GHz)
1 EC2 compute unit is about 1 GHz Xeon
let's assume 1 Google unit = 12 EC2 units (3x4 cores)
High-CPU Extra Large instance = 20 EC2 units
35*12 = 420 EC2 unit years = 5040 EC2 unit months
= 21 HiCPU XL instances for 1 year
= 252 HiCPU XL instances for 1 month
Putting this into AWS calculator [1] yields: $82,491.36 (21 HiCPU XL reserved instances for 1 year)
$125,435.52 (252 HiCPU XL on-demand instances for 1 month)
[1] http://calculator.s3.amazonaws.com/calc5.html------
Edit: It could be actually a bit more, Nehalems seem to be faster than my first estimate.
Plugging in EC2 computing unit comparison from Cluster Compute instance (which has 2x quadcore Nehalem), 1 Google unit can be 16.75 EC2 units.
This gives:
$114K for 1 year of 29 reserved HiCPU XL instances
$166K for 1 year of 18 reserved cluster compute instances
$173K for 1 month of 348 on-demand HiCPU XL instances
$246K for 1 month of 210 on-demand cluster compute instances
-----Edit2: Dedicated server hosting would be much cheaper. For example, from Hetzner [2] it would cost just about 26K EUR (= $34K) to rent 35 dedicated quadcore Nehalem servers for 1 year.
[2] http://www.hetzner.de/en/hosting/produktmatrix/rootserver-pr...
Re:"God's Number" is 20 - Rubik's cube has been "solved"
by fragmede on Monday August 09, 2010 @11:31AM
(#1588259)
Roughly $43000, using spot-instances.
Back of the envelope... From the link, it would take
1.1 billion seconds @ 4x 2.8ghz
= 4.4 billion @ 2.8 Ghz
= 12.32 billion seconds @ 1 Ghz = ~3430000 hours
Using ec2 unit performance as 1Ghz, assume computing is comparable, Ghz for Ghz (it isn't...), and storage and xfer is negligible.Normal instances
instance-size|ec2units | $/hr | $ total
sm | 1 | 0.085 | 291550
large | 4 | 0.34 | 291550
xl | 8 | 0.50 | 214375
himem-xl | 6.5 | 0.69 | 364108
himem-double-xl | 13 | 1.20 | 316615
himem-quad | 26 | 2.40 | 316615
hicpu-med | 5 | 0.17 | 116620
hicpu-xl | 20 | 0.68 | 116620
cluster | 33.5 | 1.60 | 163821
Using spot instances
sm | 1 | 0.031 | 106330
large | 4 | 0.14 | 120050
xl | 8 | 0.245 | 105044
himem-xl | 6.5 | 0.172 | 90763.1
himem-double-xl | 13 | 0.427 | 112662
himem-quad | 26 | 0.871 | 114905
hicpu-med | 5 | 0.06 | 41160
hicpu-xl | 20 | 0.249 | 42703.5