Alpha - Moving Toward Its Omega?
R.A. Krause - August, 1999
The Alpha RISC processor was developed in the early 1990's
to provide a generational leap in CPU (and system) performance. In addition
to the higher performance of the chip itself, Alpha is based on a 64-bit architecture,
providing (according to Digital/Compaq) far superior performance to the 16-bit
and 32-bit architectures in use by the Intel x86 architectures. The other strength
is Alpha's ability to run multiple OSes: Windows NT, Unix (Compaq's "Tru64 Unix"),
Linux, and VMS. Although Alpha-based products exist in multiple market spaces,
their greatest success so far has been in two areas: high performance workstations
and high-end servers. Alpha's cost structure has not allowed it to compete effectively
in the low-end (PCs, etc.) marketplace. The present competition has been any
Intel-based server and workstation, as well as workstations from Sun Microsystems
(Unix), and Unix-based servers.
Workstations: Sun (52% of Unix mkt./0% of Win mkt),HP (16%U/23%W),
Dell (0%U, 23%W)
Intel-based servers: Compaq ProLiant (30+% of Intel market),
Dell(~15%), HP, IBM(~10% ea.)
Unix server competition: Sun, HP, IBM
Sales for Alpha are increasing slightly (year-over-year). Alpha-based
products are about 7% of the market (NT server and workstation volume) and about
7% of the Unix market [Source: Compaq]. Alpha is unlikely to overtake any other
CPU (that it hasn't already passed) in market share.
Power: Alpha-based products are strongest in compute-intense
applications, such as computer-aided design (CAD) and very large database
applications. Its SPECfp scores have been consistently superior (50%-150%
higher) to the nearest Intel-architecture processor. [CAD applications often
require this power, and the difference is evident to users.]
Customer Satisfaction: Surveys of the "mid-range"
market indicate customers are very happy with Alpha running Unix, its main
shortcoming being lack of application software (relative to Sun and HP).
Flexibility: Alpha products are able to run any
of four OSes - Windows NT, Unix ("Tru64 Unix"), Linux, and VMS. This flexibility
theoretically allows Compaq to sell Alpha to companies with a multiple-OS
environment. However, since Compaq VMS system sales are essentially flat,
and Alpha/NT is not presently a significant player, this flexibility is
losing its value.
Reliability: In its Unix and VMS implementations,
high-end Alpha products are often used in 24x7 environments. Alpha's power
combined with system reliability provides an advantage over Windows NT products
for critical applications. (Note: Windows products are not noted for their
reliability, and are not generally used in critical-application situations.)
can strengthen its revenue stream in a few ways:
Build on its 64-bit Unix performance advantage
Make a concerted effort to get Win64 apps available ASAP
Reduce product cost where possible/feasible
alliance partners include Intel for chip manufacturing and Microsoft for Windows
NT development. It should be noted that the Microsoft relationship has yielded
little result for Alpha NT.
Cost: Alpha products have been expensive, and their
entry-level cost has been high. This has made Alpha unsuitable for low-end
products. In addition, Intel is catching up on "raw performance", so Compaq
may no longer be able to charge a premium for having the highest numbers
on the TPC-C scale. Additionally, the $/tpmC figures for Alpha, although
improving in recent test results, still are only competitive in the Unix
market, not Windows NT - and only in certain performance bands (~25K tpmC).
Few true 64-bit applications: Although Compaq/Digital
has a fair amount of 64-bit Unix applications, there are few/none for Windows
NT, due to Win64 apps not being generally available. Alpha's performance
on 32-bit apps is often no better than Intel, especially when the applications
are not "Alpha-native". Software Development Kits for Win64 were released
around December '98, so robust applications should not be expected until
around December '99 at the earliest. Microsoft also does not appear to be
making a significant commitment to NT/Alpha - releasing Alpha apps simultaneously
with Intel apps provides small advantage to Alpha.
Merced/McKinley: When Intel finally ships Merced,
Alpha's primary architectural advantage over Intel will be significantly
reduced, if not disappear altogether. Even though it is likely that Merced
(and its follow-on, McKinley) will be expensive - perhaps as costly as Alpha
- and that Merced's performance will be lower than Alpha in 2000, its presence
will draw customers away from Alpha.
Reduce Cost: Alpha's current price/performance ratio
($/tpmC) is typically higher than Intel and HP/Sun. Compaq should make a
concerted effort to cost-reduce the Alpha product set, thereby improving
this figure. Although some of the needed cost-reduction can be accomplished
by improving manufacturing efficiencies, there may be "structural" redesign
required to make significant improvements. Cost reduction will also allow
better market penetration in the mid-range segment. In addition, if Compaq
is unwilling to take the necessary steps to make Alpha/NT cost-competitive,
then Alpha should exit the NT space. Its new DS10 is an attempt to be low-cost,
but it appears geared toward Unix.
Take advantage of Merced's delay(s): To build market
share before Merced becomes a reality, Compaq should make a concerted effort
to get Win64 applications up-and-running on Alpha. If significant Win64
apps are not available and running until just before Merced starts shipping,
the less likely it is that non-legacy customers will choose Alpha. Granted,
optimized compilers/apps will take awhile, but having Merced there is a
powerful deterrent to Alpha. In addition, if Unix on Merced becomes reality
(vs. vaporware), it will be another nail in Alpha's coffin.
Play into Alpha's strengths - performance and Unix:
Build up the number of robust 64-bit Unix apps - pay ISVs to develop apps
- and sell it aggressively. Improve the performance numbers, Sun and HP
are catching up. Compaq has recently announced a $100 million Unix-on-Alpha
campaign. If this is successful, it may make Alpha a more serious contender.
If it fails, Alpha probably will, too.
Alpha servers - Unix only: Compaq has not truly
committed to having Alpha as a market presence for NT environments. Until
it decides to do so, and until a 64-bit version of NT is available, customers
are better served with Intel-based systems for NT.
Alpha servers for mission-critical applications:
Because of Alpha's solid system design and reliability, Alpha Unix servers
should be always be considered for critical applications, such as 24x7 operation
of a factory floor, or powering eBay (or a similar E-commerce company).
In addition, Unix shipments went up in 1998, so its demise at the hands
of NT does not appear to be imminent.
Alpha for high-performance workstations and servers:
Alpha workstation performance is excellent, consistently winning AIM benchmark
awards. Users should consider this for scientific/technical/engineering
applications. Please note that this recommendation is only for "top tier"
applications requiring lots of compute power. (In other words, running AutoCAD
on an Alpha workstation is not cost-effective.)
(and now Compaq) squandered whatever lead/advantage Alpha had over the Intel
architecture. Alpha is now, and should continue to be, at most a niche player.
Compaq must now expend considerable effort if it wants Alpha to be more than
that. Unless significant inroads are made before Merced and McKinley ship, Alpha
risks becoming another flashy-but-dying technology.
the exception of legacy VMS systems, and a relatively small number of NT systems,
Alpha seems destined to become a Unix-only system. Although Unix is not dead/dying,
NT has passed it in quantity of licenses. Alpha will survive at least another
three years (90% survival likelihood), but its prospects diminish after that
(30% survival likelihood after five years), unless Compaq can turn things around
in the next 12 months. Users should take these factors into consideration when
deciding on a technology platform.
Instruction Set Computing
VMS: DEC/Compaq proprietary OS
SPEC: Performance benchmark rating system
TPC, TPC-C and tpmC: Transaction Processing Performance Council and
its performance rating system
24x7: 24 hours a day, 7 days a week
Win64: 64-bit Windows NT
SDK: Software Development Kit
ISV: Independent Software Vendor
eBay: Online auction house, recently experiencing uptime problems
AIM: Yet another performance rater
PowerPC: CPU formerly produced by Motorola