Our logic verification strategy depended o n build
ing a breadboard early in t l1e design cycle. 'fhis 500 CPU modu le. The breadboard , wl1ich contai ned the holes for the NCA, was then attached to the VAX
4000 Model ';00 CPU mod u le by soldering it to the extended socket pins. CP bus clocks were not directly routed to the breadboard logic. To control clock skew. a phase lock loop (PLL) was used on the breadboard to regenerate the CP bus cl ocks. With this configuration, the breadboard system was able to run at ful l speed.
Once the breadboard system was assemblecl, we were able to execute console com mands after
Debugging was quick because a completely func
tional console and 1/0 system al really existed . Simple functions, such as register reads and writes, were debugged using the console examine and deposit com mands. More com plex functions, such as reading ami writing to an SCSI disk, were tested by writ i ng test programs in VAX :-'LACRO, download port were debugged. In addition to software debug
ging, t h is effort provided the software to perform
Vol. 4 No .. > Slllllll/er 1')')2 Digital Tecbnical journal
ROMS
CEAC C H I P
SQWF CHIP
The VAXstation 4000 Model 90
NVAX 1/0
ADAPTER (NCA)
MEMORY SIMM NVAX
CONNECTORS
Figure 2 VAX station 4000 lHodel <JO -�)'stem fl!Jodule
extended verification. The hardware group was able to use graphics test packages running under DECwinclows software. d isk exercisers. system exercisers. and other tools su pported by the VMS operating system. This prov ided a verification envi
ron ment we cou l d never achieve with traditional si m u lation met hods.
At t h is point, we were sti l l using the VA)\ 4000 Mod el 500 console. The breadboard was then used to debug t h e Model 90 console code. We d isabled the system support ch i p , which contro l s m u ch of the conso le support ha rdw:�re in the VAX 4000 Model 500, and began using the Model 90 consol e support hardware. A base console that i ncluded minimal power-up self- test, bas ic command sup
port, and SCSI hoot suppo rt was debugged by the Model 90 console tea m . Once the console was func
t iona l , the VMS group re turned and debugged boot support for the Model 90 usi ng t he bread board . When t h is was finished, the software was com
plete ly debugged and ready to be l oaded onto the first Model 90 prototype .
Digital Tech11ical journal Vol. -1 .Vo. 3 Sunnner 1')<)2
As soon as we assembled the first Model 90 pro
totype system , we rea l ized the benefits of a.l l tile work performed using the bread board system.
Dur ing the first day of debugging, we ran the console program and booted the VMS system with m i n i m a l effort. We also ran DECwi ndows software using the l.CSPX and the SI'Xg and SP:Xgt graphics opti ons. Th is quick deb ugging a l l owed additional protOtype systems to be b u i l t i m mediately a n d sh ipped to various develop
ment and verificat i o n groups th roughout the company.
Performance
The VA.Xstation 4000 Model 90 represents the fastest VAX workstation ever produced . I ts Ci'll per
formance surpasses previous VAX worksta tions and is com parable to Digita l 's RJ SC-based workstations.
By u t i l izing the NVA)( CPU chip, the Model 90 achieves 2.7 times the perform ance of the .Model 60 when measu red aga inst the SPECmark bench
marks.' Table 2 gives the CPU performance of the
89
NV AX-microprocessor VAX Systems
Table 2 CPU Performance Compa rison Workstation
VAXstation 4000 Model 90 VAXstation 4000 Model 60 VAXstation 31 00 Model 76 DECstation 5000 Model 240
Note:
SPECmark* Rating
32.7 1 2.0 6.8 32.4
•sPEC mark is a quantitative measurement of performance, determined by run n i n g a suite of ten benchmark programs.
VAXstation Model 90 compared to other Digital workstations.
LCSPX is the entry- leveL two-dimensional graph
ics option offere(l on the Model 90. The perfor
m ance of this option is better than the LCG option offered on the lVIodel 60 for most graphics opera
tions. Table :) compares the LCSPX graph ics perfor
mance to Digit<ll workstations using standard two-dimensional metric;.
SPXg and SPXgt are high-pe rformance, th ree
dim ensional graph ics accelerators offered on both
the Model 60 and the Model 90. Table 4 compares the three-dimensional graphics performance of sev
eral of Digi tal's wor ksta tions using standard three
d imensional metrics. In ad dition, Ta ble '5 gives three-d imensional performance using the picture
level bench mark (PLB) suite.
Summary
The NVAX CPU chip prov ides the high per formance that makes the \'A,'\station 4000 iVlodel 90 competi
t ive i n today's market. The design methodology used du ring the project a l lowed us to develop and ship the Model 90 quickly and to prov ide a simple upgrade path for existing VAXstation customers.
Acknowledgments
The au thors woul d l i ke to ackn owledge the fo l l ow
ing people for tneir contributions to the VAXstation Model 90 development: Gregg Bouchard , Paul Campos, .Jon Crowell, Terry Furman, Dave lves, Bi l l Laprade, .J im Lu ndberg, Curt ivlil ler. Jan Nordh. Jim
Rei l ley, Brian Rost, M i ke Su l l ivan, P a t Su l l ivan, M i ke Wa rren, ami Tom Wen ners.
Ta ble 3 Two-dimensional Graph ics Performance Compa rison
Workstation
VAXstation 4000 Model 90 LCSPX VAXstation 4000 Model 60 LCG VAXstation 31 00 Model 76 SPX DECstation 5000 Model 240 PXG
Two-d imensional Area F i l l
(Mpixels per second}
1 8.2 1 4.6 1 4.2 1 3.9
Ta ble 4 Three-d i mensional Graphics Performance Comparison Three-d imensional Polygons
Workstation (Kpolygons per second)
VAXstation 4000 Model 90 SPXgt 33 VAXstation 4000 Model 60 SPXgt 33
VAXstation 4000 Model 90 SPXg 30
VAXstation 4000 Model 60 SPXg 30
VAXstation 31 00 Model 76 SPX 6
DECstation 5000 Model 240 PXG 52
Two-dimensional Vectors
(Kvectors per second}
266.0 21 6.0 1 83.0 263.0
Three-d imensional Vectors
(Kvectors per second}
295 300 295 295 57 302
90 l'ol. ·-i No ..i S11111111�r I?'JJ Di_q,itnl Technical jounwl
Table 5 PLB Graph ics Performance Comparison
Workstation
VAXstation 4000 Model 90 SPXgt VAXstation 4000 Model 60 SPXgt DECstation 5000 Model 240 PXG
Note:
Printed Circuit Board
1 3.2 1 2.3 1 0.0
The VAXstation 4000 Model 90
GPCmark PLBiit Resu lts*
System Cyl inder
Chassis Head Head Shuttle
1 1 .8 8.5 8.3 1 3.5
11 .1 8.4 8.5 1 2.5
1 1 .7 1 4.9 1 9.2 1 8.3
"GPCmark is a quantitative measurement of performance, determined by dividing a norma lizing constant by the elapsed time, in seconds, required to perform the test.
References
1 . (;. Uhler e t a l . . '"The f\fVAX and NVAX+ High
performance VA,'\ Microprocessors:· Digital Technical Jo urnal, vol . 4, no. 3 (Summer
1992, this issue): 1 1 -2�.
2 . J. Crowell e t al . , "'Design o f the VAX 4000 Model 400, 500, a ncl 600 System s," Digital Technical jo urnal, vol. 4, no. 3 (Sum mer
1992, this issue): 60-72.
Digital Techuical journal Vol. 4 No ..'i Summer 1992
3. TURBOchannel Hardware Specification (Palo Alto, CA: Digital Equipment Corporation, TfU/ADD Program, 1990).
4. Small Computer -�:ystem. Interface (SCSI) (New York: American National Standards Institute, A.J"JSI X.) 131 - 19R6. l9R6).
5. Digital 's VAXstation 4000 Familv Pei/Or
mance Summary Version 3. 0 (Maynard:
Digital Equipment Corporation, 1992).
9 1
Brian Porter I