R10000
A superscalar MIPS IV microprocessor for workstations and servers.
The R10000 is a high-performance microprocessor implementing the MIPS IV instruction set. Developed by MIPS Technologies and fabricated by NEC and Toshiba, it launched in 1995. It introduced out-of-order execution to the MIPS line, powering SGI workstations and supercomputers.
- Implements MIPS IV instruction set with out-of-order execution
- Contains 6.8 million transistors on a 0.35 μm process
- Recall of 10,000 units in 1996 due to current faults
- Supports up to 1 TB of physical memory addressing
- Issued up to five instructions per cycle via superscalar design
What it is
The R10000 is a reduced instruction set computing (RISC) microprocessor that implements the MIPS IV architecture. MIPS Technologies designed the chip, while NEC and Toshiba handled manufacturing. It debuted in July 1995 with clock speeds of 175 MHz and 195 MHz. This processor marked a shift from simple in-order designs to complex out-of-order execution, aiming for higher performance in scientific and graphical tasks. It served as the central processing unit for high-end workstations and servers during the mid-1990s.
How it came about
Development began in the early 1990s, with initial plans for a 1994 release. MIPS Technologies collaborated with licensees like NEC and Toshiba, pooling resources for a reported $150 million investment. The chip was unveiled in October 1994, targeting parallel computing applications. Volume production faced delays due to fabrication issues at the foundries. In September 1996, Silicon Graphics recalled 10,000 units made by NEC because they drew excessive current, causing system shutdowns. A faster 250 MHz version arrived in 1997 using a smaller 0.25 μm process.
Inside the chip
The R10000 uses a four-way superscalar design capable of issuing up to five instructions per cycle. It fetches four instructions each cycle and decodes them into integer, floating-point, or load/store queues. The architecture supports out-of-order execution and register renaming to keep pipelines full. The integer unit has two pipelines, while the floating-point unit includes separate adder, multiplier, and divider blocks. On-chip caches hold 32 KB for instructions and 32 KB for data. The chip addresses up to 1 TB of physical memory using a 40-bit physical address space.
How it was made
NEC and Toshiba fabricated the R10000 using a 0.35 μm process. The die contains approximately 6.8 million transistors, with nearly half dedicated to the primary caches. The physical die measures roughly 16.6 mm by 17.9 mm, covering an area of 298 mm². It ships in a 599-pad ceramic land grid array package. Earlier reports suggested a pin grid array, but the final product used the land grid format. The external secondary cache connects via a 128-bit bus with error correcting code protection.
Where it went
Silicon Graphics integrated the R10000 into its Indigo2, Octane, and O2 workstations. It also powered the Challenge and Origin 2000 servers, as well as the Onyx and Onyx2 supercomputers. NEC used the chip in its Cenju-4 supercomputer. Siemens Nixdorf included it in servers running the SINIX operating system. Tandem Computers utilized the processor in its Himalaya fault-tolerant server line. These systems relied on the chip for heavy computational loads and graphics rendering.
Versions and family
The R12000 succeeded the R10000 in November 1998, adding a pipeline stage and better branch prediction. The R12000A reached 400 MHz on a 0.18 μm process. The R14000 followed in 2001, hitting 500 MHz with copper interconnects. The R16000, introduced in 2003, operated at 700 MHz and featured larger caches. The R16000A pushed speeds to 800 MHz and 900 MHz, with some custom units reaching 1.0 GHz. The R18000 was planned but cancelled before release.
Why it matters
The R10000 established out-of-order execution as a standard for high-performance MIPS processors. It enabled Silicon Graphics to maintain leadership in visual supercomputing during the late 1990s. The recall incident highlighted the risks of relying on external foundries for critical components. Despite the setback, the architecture evolved through several generations, sustaining performance growth for nearly a decade. It supported large-scale symmetric multiprocessing configurations in enterprise environments.
Problems and reception
Initial reception focused on its projected performance metrics, such as 300 SPECint92 at 200 MHz. Supply shortages kept prices high, around $3,000 for the 195 MHz model. The 1996 recall of faulty NEC-manufactured chips damaged confidence and affected Silicon Graphics earnings. Customers preferred the 175 MHz version over the slower 150 MHz variant released later. The chip was valued for its ability to scale in multiprocessor systems despite early reliability issues.
Questions people ask about R10000
Who designed and manufactured the R10000?
MIPS Technologies designed the R10000 microprocessor. The company was a division of Silicon Graphics at the time. NEC and Toshiba fabricated the chips using their semiconductor manufacturing facilities. Other previous partners like Integrated Device Technology did not produce this specific model.
What makes the R10000 different from earlier MIPS chips?
Unlike the simpler in-order R4000, the R10000 uses out-of-order execution and register renaming. It is a four-way superscalar design that can issue multiple instructions per cycle. This allows it to handle complex tasks more efficiently by keeping its execution units busy even when some instructions stall.
Why were some R10000 processors recalled?
In September 1996, Silicon Graphics recalled 10,000 R10000 chips fabricated by NEC. These specific units drew too much electrical current during operation. This excess current caused computer systems to shut down unexpectedly. The recall impacted the financial earnings of Silicon Graphics.
What systems used the R10000 processor?
The R10000 powered Silicon Graphics workstations like the Indigo2 and Octane. It was also used in SGI servers and supercomputers such as the Origin 2000 and Onyx2. Other users included NEC for its Cenju-4 supercomputer and Tandem Computers for fault-tolerant servers.
Good reading and sources
The facts on this page come from the Wikipedia article "R10000", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryChip rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.
Like R10000
All CPUs and microprocessorsMore from MIPS Technologies
All of them
WorldR4200
Low-power MIPS III microprocessor for embedded systemsCPU · MIPS · 1993