The Senate may consider a bill today that would limit export of next year's desktop-class computers and business servers.
The 1998 Defense Department authorization bill, which requires the Commerce Department to license and monitor for compliance the export of certain classes of machines, passed the House by a vote of 286 to 123 on Tuesday. The bill was taken up by the Senate yesterday, but it did not come to a vote and is now caught up amid other end-of-term business. It could be taken up at any time over the next few days.
According to a provision originally attached as an amendment by Reps. Floyd Spense (R-South Carolina) and Ron Dellums (D-California), computer makers shipping machines that can perform 2,000 million theoretical operations per second (MTOPS) to some 50 countries, including Russia, China, and India, must apply to the Commerce Department for an export license.
So far, however, the export provision's potential effect on trade has largely escaped the notice of most lawmakers, who have been more concerned with revelations that supercomputers shipped to Russia and China were appropriated for non-civilian use. "Unfortunately, what really got their attention was some of those sales. That was their initiation to the issue," commented a House staffer familiar with the so-called Spense amendment.
Meanwhile, the bill has been held up because senators from California and Texas object to unrelated clauses regarding the provision of services to defense bases in those states, an issue that's part of the post-Cold War closing of bases. If passed, only a presidential veto would stop the export controls from going into effect, but President Clinton has indeed threatened a veto because of concern with the base services issue.
The standard may include machines that will reach the general market in 1998. Current 300-MHz Pentium II computers can perform at 1,000 MTOPS, estimated Nathan Brookwood, semiconductor analyst at Dataquest. Next year, a machine with the 450-MHz version of the Intel processor will in theory exceed 2,000 MTOPS, he said. Chips from other manufacturers, including Digital Equipment, will also top the mark.
Leading chipmaker Intel disagrees, however, saying a 300-MHz Pentium II runs at 350 MTOPS. According to a spokesman, PCs equipped with current Intel processors and likely next year's chips would not be affected by the provision.
A Commerce Department spokesperson also suggested that desktop-class machines were unlikely to top the 2,000 mark.
Even if single-processor machiness fall below the 2,000-MTOP mark, many manufacturers plan to come out with Intel-based multiprocessor machines that use two, four, or more chips, a compounding effect that will certainly push many computers over the limit. Four-processor Pentium II machines are scheduled for release next year.
Next-generation NT servers from IBM will pass the hurdle and Unix machines from Sun, SGI, and Digital already exceed it, said Paul Freedenberg, a Commerce Department undersecretary in the Reagan administration.
One reason for the disagreement surrounding the issue lay in the fact that MTOP calculations vary depending on the data is used, said Dean McCarron, an analyst with Mercury Research.
Roughly, MTOP performance equals the number of instructions a chip issues times the speed of the chip. Realistically, a Pentium II issues 1.5 instructions per cycle, giving a 300-MHz chip a MTOP rating of 450. Theoretically, however, the Pentium II can issue three instructions, two integer execution functions and a floating point instuctrion, he said. Others put the number of instructions higher than three to accomodate MMX. Brookwood used four instructions per cycle for his estimates.
An Intel Web page on the microarchitecture of the Pentium II in fact states that one of the chip's advantages is that it moves beyond the two instruction speed of a Pentium, a statement which contradicts Intel's current math on the subject.
Theoretically, therefore, next year's 400-MHz Pentium II would have an MTOP rating of at least 1200, while a two-processor workstation would theoretically have an MTOP rating of 2400, triggering the export license requirement.
In any event, McCarron termed the regulation ineffectual because these countries could string together a number of Pentium II desktops and achieve massive MTOP performance.
The 2,000-MTOP standard was adopted in October 1995, following to a government-wide study that included Pentagon input on military use of high-performance computers. A study looking into revising the standard, one of several export control parameters, is currently under way at Stanford, according to various sources. It's expected to wind up by the end of the year.
The defense bill changes the U.S. export control regime by rolling back trade liberalization effected by the Clinton administration in 1995. Under the current regime, PC manufacturers and resellers can export without a license to "Tier 3" countries machines capable of more than 2,000 MTOPs but less than 7,000 MTOPs, provided the machines are intended for non-military, non-weapons research use. The exporters are not required to investigate the machines' end uses but cannot ignore red flags that suggest the machines would be misappropriated for military ends.
In the future, vendors would submit export notifications to the Commerce Department identifying the country of destination and ultimate end user. The Department would then circulate the petition to a number of agencies, including the Defense Department. If the federal government fails to act within ten days, the vendor can export the computers without restrictions.
If an objection arises, the Commerce Department can then demand a license for export, which can take up to 90 days to obtain, according to Jason Mahler, legislative assistant for Rep. Zoe Lofgren (D-California). (Lofgren brought concerns over the 2,000-MTOP standard to light earlier this week.) The license can also be denied. Further, the bill requires the Commerce Department to perform compliance verification.
"In many ways the [Spense amendment] is a vote of no confidence in the export control regime," said Jay Kraemer, who practices trade law at Fried, Frank, Harris, Shriver & Jacobson in Washington, D.C., and teaches at the Georgetown University Law School. "It's an unfortunate example of micromanaging the export control process."