Battle royal breaks out over Russian chip claim
But it's jaw-jaw really
A bitter argument has broken out over a claim that Russian technologists were the first to develop superscalar microprocessors, later adopted by Intel. Mark Smotherman, writing from a US university, said: "I would like to correct, for the record, the statement in Intel uses Russia military technologies" by Andrei Fatkullin in which he says: "Superscalar architecture was invented in Russia." "The first superscalar design was the IBM ACS-1 supercomputer,designed in Menlo Park, California, in the mid-1960's at IBM's Advanced Computing Systems by a team that included John Cocke. In fact, the vision for a computer that decoded and issued multiple instructions per cycle was due to Cocke. "The ACS-1, which was cancelled in 1968, would have issued up to seven instructions per cycle: three index register instructions, three floating-point register instructions, and one branch. The floating-point register instructions could be issued out-of-order from an eight-entry "contender stack" that was scanned on each cycle. "Each floating-point register had a "backup register" so that a new value could be loaded into the backup register while dependencies still existed with the previous value. (This is a simple form of register renaming). "There was a condition register with 24 individually-addressed bits so that independent compare instructions could be executed in parallel. Moreover, each instruction included a skip flag so that the instruction would be executed or not based on a current skip condition, which was specified as a logical operation between two condition bits. (This is a simple form of predication). " Branches were implemented using two types of instructions: a "branch" instruction that evaluated a logical operation between two condition bits and placed that result along with a branch target address into a branch table; and, an "exit" instruction that selected the first tabled branch that evaluated to true.(This is a form of multiway branching.) "Simulations of the ACS-1 on the innermost loops of a scientific code such as Newtonian Diffusion showed a sustained instructions-per-cycle value of 1.8. On this code, which contained branches within the body of the innermost loop, a 100 MHz ACS-1 would have been 76 times faster than the CDC 6600 and 22 times faster than the IBM S/360 Model 91. " The ACS-1 was described in the "Western literature" in 1971 (see reference 3)." However, Alexei Pylkin, aprogrammer-researcher at the Supercomputer Software Department, Russian Academy of Sciences Novosibirsk, Akademgorodok, has refuted Smotherman's argument. He said: "I'd like to note that ACS-1 was experemental design while Elbrus-1 was a mainstream computer. Serial production of Elbrus-1 started in 1977 but research on superscalar architecture was done in Russia in the same time frame as in IBM. "Experimental ACS-1 had no successors. The second Western superscalar processor has appeared only in 1992. Serially produced Elbrus-1 had successors - serially produced Elbrus-2 (1984), microprocessor El-90 (1990, prototypes)." Smotherman produced references to support his argument, which follow: 1. John Cocke, "The search for performance in scientific processors," Communications of the ACM, vol. 31, no. 3, March 1988, pp. 250-253. 2. Emerson Pugh, Lyle Johnson, and John Palmer. IBM's 360 and Early 370 Systems. Cambridge, MA: MIT Press, 1991. [see section 7.5, which covers ACS] 3. Herb Schorr, "Design principles for a high-performance system," Proceedings of the Symposium on Computers and Automata, New York, New York, April 1971, pp. 165-192. 4. Mark Smotherman, "IBM Advanced Computing Systems -- A Secret 1960's Supercomputer Project," available online. ®
Sponsored: What next after Netezza?