ean be used for technical calculations, 
some, such as the 702 and 705, were 
primarily intended for business applica- 
tion. Likewise, the majority of File 
Computers and IB\I 650’s undoubtedly 
will not be used for technical computing. 
Future machines will have faster and 
larger memories and, in the immediate 
future at least, will cost more. (See 
p. 60 for more detail.) 
Staffing a Computer 
The size of the computing group 
associated with a ligital computer 
seems to be closely proportional to the 
size of the machine. For example, 
computing groups associated with the 
650 and Datatron usually contain be- 
tween 3 and 20 people, while groups 
associated with the 704 and 1103 con- 
tain between 12 and 60 people. 
Much debate is centered about 
whether the computing should be car- 
ried out in a ‘‘closed shop” or an ‘‘open 
shop.’”’ In a closed shop, all the pro- 
gramming or coding is carried out by a 
group directly associated with a digital 
computer. Problems are presented to 
this group by the originator, and the 
solution is returned to him. 
In an open shop, the originator of a 
problem does the coding himself or 
supervises someone. in his group who 
does it. In the open shop, the person- 
nel associated directly with the ma- 
chines merely operate them, prepare 
utility and mathematical routines and 
perhaps act as consultants on the use 
of the machine. 
An appropriate compromise for 
groups involved in nuclear calculation 
is the ‘“‘service shop,’ in which the 
computing group assists the engineer 
or scientist in whatever fashion is de- 
sired, and also carries on certain longer- 
range problems and programs. 
As in most technical professions to- 
day, there exists a great shortage of 
good technical computing personnel. 
Even more than in most technical 
fields, including nuclear energy, the ex- 
pansion in computing during the last 
five years is astounding. An estimate 
of 10,000 mathematicians has been 
made for computing needs in 1958. 
Longer-range predictions become as 
high as 150,000 by the late 1960's. 
Inasmuch as the 1956 combined 
membership list of the American 
Mathematical Society, the Mathemati- 
cal Association of America, and the 
Society for Industrial and Applied 
Mathematics contains slightly less than 
34 
10,000 names, it becomes shockingly 
apparent that there will not be enough 
trained mathematicians to go around. 
Even today the majority of com- 
puting groups, including those con- 
cerned with reactor calculations, make 
poor use of their machines because they 
are unable to obtain enough well- 
trained personnel. It is not uncom- 
mon for an individual to waste an 
amount equal to his monthly salary in 
a few hours of machine time owing to 
lack of technical knowledge of either 
the computer or the problem. 
Computer Economics 
Use of a digital computer can now be 
obtained in four ways: purchase, rental, 
rental with the option to purchase, and 
costs per year are between $100,000 and 
$300,000, and for large-machine in- 
stallations, between $500,000 and 
$1,300,000. 
Cost per Problem 
The cost of performing a given nu- 
clear calculation depends on both the 
character of the problem and the kind 
of machine used. Since the greatest 
effort has gone into multigroup caleula- 
tions, the costs of making these kinds 
of calculations will be considere:| in 
some detail. 
One method of estimating the cost of 
using a high-speed digital computer to 
solve a multigroup problem is first to 
estimate the number of ‘‘equivalent”’ 
multiplicative operations for the prob- 
TABLE 2—Computer Costs 
Operations /sec Computer costs 
Floating Rental Purchase Per operation 
Machine Fixed point point ($/mo) ($) ($) 
Desk 2.5 X 10? 5 X 10? 3X WOR 
calculator 
650, Datatron 10? 15 4 xX 108 2X 105 6 X 10-5 
UNIVAC 5 xX 10? 10? 2 X 104 10° 6X 10-5 
701, 1103, 2 X 103 10? 2 X 104 108 15 xX 10ss 
IAS type 
704, 1103A, 6 X 103 Gixe03 4 xX 105 2 X 10° 107° 
709, X-301G 
LARC 105 10 105 3.5 X 108 10-5 
STRETCH 106 106 2105 6 X 105 2X 10m 
the hourly rental of time on a machine 
at a service bureau. Medium ma- 
chines rent for $2,000 to $12,000 per 
month, can be purchased for $40,000 to 
$300,000 and can be used hourly for 
$40 to $200. Large machines now rent 
for about $15,000 to $55,000 per month, 
can be purchased for $500,000 to $3,- 
000,000 and have an hourly rate of 
$150 to $750. The monthly rental 
cost generally refers to a one-shift oper- 
ation, with each additional shift costing 
50% more. 
Besides the cost of the machine itself 
personnel costs represent a major con- 
tribution to the total cost of operating 
a digital computer. A rough rule that 
applies fairly well to installations rent- 
ing machines on a one-shift basis is 
that the personnel costs should be 
equal to the machine costs. For me- 
dium-machine installations, the total 
lem and then multiply by the dollar 
cost per multiplicative operation. A 
rule of thumb for the number of equiv- 
alent multiplicative operations is to 
count the number of multiplications 
and divisions and multiply by two to 
account for all other arithmetical and 
logical operations. The final column 
in Table 2 gives some representative 
costs per operation. The following 
word formula can be used to estimate 
the number of operations for a multi- 
group problem: 
(source iterations) X (groups) X 
(mesh points) X (operations per point) 
x (iterations per group) = operations 
Before we give numerical examples, 
some general comments on the use of 
this estimate are appropriate: 
1. In one dimension, the number of 
regions and the operations needed for 
