COMPUTE ATLAS Supercomputer supply-chain graph
1153 systems 482 sites Sourced data

Parts/CPUs/Hitachi SR2201 processing unit

CPUs · SR2201 · 1996

Hitachi SR2201 processing unit

Hitachi implementation of HP PA-RISC at 6.7ns with pseudo-vector processing added in hardware, so long-vector code avoids the cache-miss penalty that normally ruins RISC on this workload.

Why this part

A licensed instruction set with a local answer bolted on. Hitachi implemented HP's PA-RISC, then added what it called pseudo-vector processing: hardware that prefetches long vectors into registers so that streaming code does not pay the cache-miss penalty that normally makes RISC processors disappointing on physics workloads. This is the compromise of the mid 1990s in one part. Vector machines were fast on the science but priced themselves out, and RISC was cheap but stalled constantly on exactly the access patterns lattice QCD generates. CP-PACS was designed by physicists who needed the second option to behave like the first.

Systems using this part

Systems
2
Rmax underneath
588.6 GFlop/s
Units recorded
3,072
2 of 2 edges disclose a quantity
System Role Country Units System Rmax Supplied by Rank
CP-PACS
1996-10
CPU Japan 2,048 368.2 GFlop/s Hitachi #1 →
Hitachi SR2201/1024
1996
CPU Japan 1,024 220.4 GFlop/s Hitachi #1 →

“Units recorded” sums only the edges where a public source states a quantity. It is a floor, not a total, and should never be read as installed base.

Further reading