How would your computer rank among the fastest computers in the world?

Computational power since 1993

The Top500 list ranks the 500 most powerful computers in the world and has been released twice a year, since 1993.

This list allows us to keep track of computer evolution, for instance, computing power.

flops <- read.table("flops.txt")
fitsum <- lm(log(sum, base = 2) ~ adate, data = flops)
ratesum <- fitsum$coefficients[2]
2^ratesum

## adate 
##  1.84

The computing power is multiplied by approximately 1.8 every year.

Then, your computer could be as powerful as a supercomputer of the nineties!

Computational power of the fastest machine and the 500th at every ranking, as well as the sum of the power of all 500 machines in the list.

plot of chunk unnamed-chunk-3

The $y$ scale is logarithmic and the units are:

GFLOPs	TFLOPs	PFLOPs	EFLOPs
$10^9$ FLOPs	$10^{12}$ FLOPs	$10^{15}$ FLOPs	$10^{18}$ FLOPs

Computing power: definitions

Computing power is measured in FLOPs (Floating Point Operations Per Second). The theoretical (or peak) computing power $R_{peak}$ can be computed as follows:

$R_{peak}$ = Number of cores $\times$ Frequency $\times$ Operations per clock cycle

The measured comuting power $R_{max}$ is the computing power as measured by a specific benchmark application, the High Performance Linpack.

The efficiency is defined as $R_{max}/R_{peak}$. For desktop and laptop computers with standard Intel processors, this value is between 80 and 95%.

A quick comparison

429th fastest machine
in June 1994
20 MHz
512 cores
$R_{max}$ : 1.0 GFLOPs

MacBook Air from 2010
Frequency: 1.4 GHz
2 cores
Efficiency: 90%
$R_{max}$ : 10.08 GFLOPs

fastest machine
today
GPUs
3'120'000 cores
$R_{max}$ : 33'862'700 GFLOPs

Now you play

Other facts about supercomputers

Machines by operating systems

plot of chunk unnamed-chunk-4

Historically, high performance computing was dominated by proprietary UNIX operating systems. Today, 96% of the 500 most powerful computers in the world run Linux.

Fastest computers, today

Rank	Country	RMax [PFLOPs]	Electricity [MW]	Accelerator
1	China	33.9	17.8	Phi
2	USA	17.6	8.2	NVIDIA
3	USA	17.2	7.9	--
4	Japan	10.5	12.7	--
5	USA	8.6	3.9	--
6	Switzerland	6.3	2.3	NVIDIA
7	USA	5.2	3.5	Phi
8	Germany	5.0	2.3	--
9	USA	4.3	2.0	--
10	USA	3.1	N/A	--

With great computing power comes great electricity consumption. For comparison, a town of 5'000 inhabitants uses approximately 5 MW.