HISTORICAL REVIEW PROGRAM
RELEASE AS SANITIZED
DIRECTORATE OF6
Total Soviet Computing Power
SUMMARY
We have estimated the total computing power of the Soviet Union for both general-purpose and scientific explications. In general-purpose computing, the USSR hat the equivalent capacity of SOupercomputers, orf the total US general-purpose computing power. If used for sciesuifie computing, the Soviet computing inventoryapacity equivalentoray-I'l.or aboutf the total US scientific computing power.
To the Soviets, the aeqidsitieningle Western supercomputer wouldncrease in total scientific computing power. Acquisitioningleuperminicomputer would give the Sovietsiny increase in total capability,or either scientific or general-purpose computing. In terms of computational power per ruble spent, however, the Soviets certaMy will find it mghfy'attractive to actpdre Western computers of all classes. t
The large gap between the US and Soviet computing powersesult ofshortages of alt classes of computers in the USSR. TheSoviet ihortfall is particularly large for scientific computing, due to the absence of true supercomputers in the USSR. Although smaller computers can be used for scientific computing, they are very inefficient for demanding tasks such as advanced weapons system design. The Soviet Union will probably not be ready toarge-scale scientific computer comparable to theefore the.
Ttos memoiaixJom was prepared
..Office of Scientific and Wcapcos Retouch, in rupomecquctlandarc welcome,and may uldxctsed io
him or (c
LASSIFIED BY; DECLASSIFY BY: OADR DERIVED FROM
GRAPHICAL SUMMARY OF RESULTS
( Effective Computer Powers
The foul computing powerationa.ndanymjal linui on ho* much computationally tniensiwj wort that tiattoncando Trade-ofb are.ofcoune, possible: acountrycan devote more competing poweruclear weapons design aadc rumple, and lea Io weather forecasting and oUeaptoraitan the ooJ ccmpisnt;vidablr knuti thus tradlng-otT.
Quantitative estimates of the toial computing power erf tk USSR are eaeful u> many areaa la texhrtology transfermportant to know the effect of impacts on comp-aatiii power. Fornow does the sale of one VAXCs affect Soviet eomputint resources? In weapons systems development, what computer power is available to eksigners of nuclear weapons, or aerodynamic vehicles, or armor peneBaton, orecrion sysicrru? How credible are Sovsa contnhurions in the Nuclear Winter climatologynowledge of the total computing power in the USSRirst step toward answering some of fliese questions
Computers, and in paroculu to'gh perfomunce scientific computers, arc essential tools for many military and civilianrucal latge-scale scientific computing problems include rnoddinj weather, nuclear bomb explosions, aerodynamic vefuefc performance, and con trolled thermonuclear fusion systems. Large-Kale Kaetnifics also central to sastnic data analysis (for oiln quantum chemistry (for designing and endem.-nd.nr.n astrophysics.
molecular biologyand in many other advanced research fields. And there are numerous other COUBputat>jrjily-uiii:raiye tpolkatiom. signal praatsj. econoenxirrk rrwufcrocni (for eaample. strategic baSaoc nussde defemeX cita base searching, <yypac4ogy. computer viuon or tfiuge processing, and artificial inteltgence research.
Total computingioad. uveful concept It contains two important tub classes: general-purpose computing power, and scientific computing power. General-purpose coniputtnE includes word processing, industrial (eeord-keepiog. process control, accounting, and SO forth. The size of the computer system (above certain minimums) is iremienuy unimportant lor general-purpose computing Many users can effectively share one large system. Of they can each have his or her own small system. In ccmuast. scientific computing tends to icouire large, powerful machines Scientific computing problems typically cannot be broken down easily to b: tun on small computer* without suffering grave penalties Li speed or accuracy.
It is convenient io calculate the iwa) conpuorir. power ancm in Cray! Units s the archetypal iupeicomr-jlei of lV pastmcetsutg data rate (FOR) ofeak computational rate offtops (miliJons of ftoanr. point anthmeut opeiaticms pernd an average enmputaag rate tor reahstk scientific probterm of aboutfiopsodem large grnerd purpose mainfia^OOtrpeter isf aiy (xaj tu perm, rut ncipwi (tuck ts thenp) from Digital Equipments1nit An IBM-POAI or an Apple Maraasfc isna. dependseg on the details of the programmatic language beesg used, the pasblcm under inaet etc
ood measure of toiij comfMrt power ii ii.Tiplj die loraof
the inJiTXkkil computer power*ation, jusi add wp Ac hardware For icicnliric comp*ing. however.ad meawre The iraaOer aad -powerful oxnpatm no not MMr.ueh ii lieu appaienivapry
A food waynoma* ihe icica/ic conpsMir. poweraaori bwuki "quadrawc nciiu *Ncn adding up proeeajon Each comreditedhr iQjMSorcouQflg poaxiaorranr. thai.nail pnxeavon are appliedupercornputer prohlem. only <herootroeeaiort arc effectively inrd it on* ttnae)estern main fume computercnfb of (hepoenl-puipoiepo^rrorI would
O R ft A
onlyundredth the pawei for soenoTic eemponiu; Oiis penalty fintficn xtyxiactual experience of seienoik r'ipcrcornputer mm. There an alsoi n
systems.
There are three reasonable approaches iootal national computerinventory, personnel, and budget The utvnfcvy approachding up the actual current stockpile of eompuLng resourceson.he most direct approach,ampered by the difficulty ol gettmg good tnveotory figures for the USSR The ptnomtl approachepresentative sample of *ell known computing facilities, and calculate! from that the compuicr power per technical woiku. Then, knowledge of the total number of technical worker, in the nation can be used to esnma te the total national computing power. The main link with Ihii personnel approach is that ofon -representative sample of compu ting centers for the bate. The budgtl approach is simplest but also crudest One takes the total figurei for investment in data processing equipment for each nation, and calculates how much computing power that can
unmioji AoartMEli
a^npt tohe processing powers of Soviet and US conpuTcrs.forad fat the siockwJc of systems in use Data for the fini (Soviet)umpoUxd
uuomuBon li the taste should be eoaairkred ifcCRfcT, ftouga many parts
ot the taele are confirmed by oneUnified pubbeauoes The second table, showing US compute! power, ts unclassified and is based on datarom the IDC Processor Data Book foi
MM " "
t
Cone hnjons from the Inventory Approach: Depending oo Itow onccounts, the US leads by about two orders of magnitude in total computer power. The lead is greater in scientific computing than in general-purpose computing. In supercomputers, die Soviet! have essentially no systems"scientific" Soviet mac bines are the closest thing- The US lead Is relatively consistent across the whole range of computer types.nclude the most significant computers for genial-purpose and scientific applications in the US and the USSR. The broad distribution of systems which contribute to the bottom line total computing power marcs the results insensitive to etrees in individual numbers.
Personnel Amiroadi
An alternative way to estimate the total computer poweration is to take aof the computational resources at several scientific institutes, and scale that figure up tocountry.
were the places cnosen as indicators (or me Soviet Union Wc have good information about the probable computer resources ins tailed at each of these two centers. Although they may he better equipped than the typical scientific or industrial computer center, they probably arcypical military center.
The iosiituic figures inere scaled up to ihc USSRhole based on the assumption that (here ateX0 potential professional computer users in the USSR. This number includes scientists and engineers with "KanrJidaT and higher degrees. It does not include low-level programmers, keypunch operators, etc.
he Personnel method of estimation gives reasonable measures of the teal Soviet computing power, consistent with the Inventory method. The Soviet have0nits of scientific computing power,Onits ofurpose power.
The Budgetary method of estimation relies on published figures for the total Soviet inrescmeot in data processing equipment, along with the published costs for the Soviets tonil of computer power. The Statistical Abstract of the USSR (translated as The USSR in Figures) gives numbers for "Data Processing Eqmproent and Spare Parts" forime period. The IDC Processor Data Book gives similar figures for the US. (See Table 4)
TahltM; SMndlnp on Computers
1
lli IIUUoo)
Unclassified pricing information on Soviet computersough cost0 million Rubles pernit of general-purpose power,illion Rubles pernit of scientific computing power. This applies to Soviet mainframes (like thes well as to the scieo'ific machines (like thehe comparable US figures for mainframes axeillion dollars pernit of general-purpose power,illion dollars pernit of scientific computing power. (Western supercomputersfar more cost-effective, but they do not accountignificant fraction of the gross sales figures inupercomputer comparableilllor. dollars currena)1
Being generous to the USSR, one could allocate one quarter of their "Data Processing Equipment- budget to big mainframes, which contribute maximally to Soviet computer power. (Actually, the figure might be somewhat lower, since much of the computing budget has to be devoted to peripherals, mini- and microcomputers, and maintenance) For the US systems, we use the "large" sales column of Table 4 Those figures produce the following results for annual addition* to total computer power.
degree of coruisicncy uiioog the three independent nwncds tfcwnputiru: (Ma) Sovietalso evidence that our analysis is robust. We thus believe that our overall estimates^^fatvomtg are accurate to within an ccder of rnigruiude -Oat is. Inc. truth isthree and one-third of our numbers. Sinceereoces between the US andarc much greater than thai unceitainry. especiaUy for scientific computing, ourobservedarc statistically w
TheSoviets arc about oneo orders of magnrtude behind the US in total effective- 5 "PfxvMyscKOtific conrpating than it is infor
for
.IhellSSRhasi-lOCray-iUnitJ. '
To the Soviets, the acquisitionr.(rl* supercomputer wouldOOfc increase in total scientific computing power. Acquisitioningle VAX class tst^iirjnieompjier would give Ok Sovietsiny mcreaK incapabihty.oreiSer scteitific or general-purpose compunng In terms of computational power per Ruble, however, the Soviets certainly will find it highly attractive to acquire Western compuien o( all clatiei.
For further commentaryoviet large-Kale seienaiie computing, and (he cortscqueneei of itariety of Sovsei ptogrami. see the SevrntifV; and Teefcrneal Tat;-locate RepcoC
concljdrd.
Scnti deielnpnien' ofcquun!lint* Kate KwniitVeSQrhu of die United Suits S, ihnuim. Thmvfh the yea 7CCO, Sovtac IASC uoi remain ii lean five tod protxfelyIJ yeas taliinj (he At presentAc 5cien Uxfneafmi class. TW tawacrarfcaaa^ Soviet cUumb*
MlapMc>re ireadV tugfuautf Rapid hnmprejiniSC um drptndflieimfei of boUl Mfiwaic ml hardv-atr from Ok Weai.we eapreiiibiuniiillr linna.nl Sovietu wdmHif upeoigc-puuculwiy effects darcird
villas ire arqunixm.
Original document.
Comment about this article, ask questions, or add new information about this topic: