The MIPS 1 instruction set is small compared to those of the contemporary
80x86 and
680x0 architectures, encoding only more commonly used operations and supporting few
addressing modes. Combined with its fixed instruction length and only three different types of instruction formats, this simplified
instruction decoding and processing. It employed a 5-stage
instruction pipeline, enabling execution at a rate approaching one instruction per cycle, unusual for its time. The architecture makes use of a
branch delay slot. The
compilers for the R3000 available from MIPS Computer Systems were typically able to fill the delay slot some 70 to 90 percent of the time. In some military applications, the figure was 75 to 80 percent occupied. The CP works as a
coprocessor. Besides the CP, the R3000 can also support an external R3010 numeric coprocessor, along with two other external coprocessors. The R3000 CPU does not include level 1
cache. Instead, its on-chip cache controller operates external data and instruction caches of up to 256 KB each. It can access both caches during the same clock cycle. The R3000 was a further development of the
R2000 with minor improvements including larger
TLB and a faster bus to the external caches. The R3000 die contained 115,000 transistors and measured about 75,000 square
mils (48 mm2). MIPS was a
fabless semiconductor company, so the R3000 was fabricated by MIPS partners including
Integrated Device Technology (IDT),
LSI Logic,
NEC Corporation,
Performance Semiconductor, and others. It was fabricated in a 1.2 μm
complementary metal–oxide–semiconductor (CMOS) process with two levels of
aluminium interconnect. ==Use in workstations and servers==