Why was MIPS used in workstations and servers during the 80s and 90s while modern day MIPS CPUs are mainly used in embedded systems?
In: Technology
When MIPS was first created it was in the early days of the PC. MIPS was superior to other available microprocessors from Intel and Motorola in that it was more focused on things like floating point calculations and intensive graphics. This made it very good for the workstation market and for video game consoles, but not really optimized for the PC market. At the time this was probably smart, since the workstation market was very lucrative and PCs were just becoming a big market. Game consoles were probably mostly a fortunate side benefit (not sure about that, though).
As PCs caught on, Intel locked up the DOS/Windows market into the x86 instruction set, and Motorola got the Apple Macintosh market with their 68000 family. It wasn’t easy to efficiently change an operating system to a different architecture, so those tended to be locked up. Which meant MIPS couldn’t get into that market once those operating systems became completely dominant.
MIPS didn’t even have all of the workstation market. I don’t remember what Daisy used. Apollo used Motorola processors, and Sun eventually used their own SPARC. IBM had their own Power architecture (later PowerPC in conjunction with Motorola).
Eventually Intel and Motorola came out with co-processors and then integrated processors with better floating point and graphics performance, and Sun’s SPARC took away a significant part of the workstation market from Silicon Graphics (the creator of MIPS).
Meanwhile, Motorola and particularly Intel were rolling in cash due to the strong success of the PC market. They could outspend most of their competitors on R&D. It became harder and harder for other general purpose microprocessors to compete with them, although Sun’s SPARC still did rather well in the dwindling workstation marketplace.
Silicon Graphics decided to spin MIPS off as as separate company and MIPS decided to target the embedded processing market rather than hitch their wagon totally to SGI’s workstations, which was probably wise. Sun was a strong competitor, and dedicated workstations were becoming used less and less. High performance PCs became a more cost-effective competitor for many uses (if the software was made available), and supercomputers took over the highest performance needs. This made workstations somewhat of a niche product.
Of course, the embedded market is not free of tough competition either, but I’d say it was probably one of the best moves that MIPS could have made at the time.
**TL;DR** – *Sometimes you have to adapt to survive*.
MIPS was one of the first RISC architectures around. It was heavily pipelined which let it keep up a 1 instruction per cycle execution rate and claw back performance against its CISC competitors. Also, because of the simplicity of the architecture and the lack of complexity in the wiring of the ICs it could scale to high frequencies on similar process nodes.
For instance, the R4000 was released at the start of 1993 starting at 100MHz on the 0.6um (600nm) process node while Intel was struggling to get the 66MHz P5 out the door. By 1994 Toshiba had a 200MHz part fabbed on 0.3um (300nm) while Intel was struggling to get its 75MHz P54C parts out the door.
Because of the predictable in-order pipeline and the 1 clock instruction latency they still make great embedded CPUs. The compiler can make sure metrics for realtime programming are met far easier than with a machine that has weird instructions that can take dozens of cycles.
Latest Answers