>teaching MIPS as the most elegant example of a RISC processor
elegant as in simplified to the bone, but also a market failure compared to not so reduced ARM
>You’ve been here for two months, and now you want to give our IP away!
haha, nice try Imagination Technologies, you arent giving anything for free. Later on in the blog post is a mere mention of all the proprietary IP, patents and licenses they are NOT granting/licensing. This is a huge win for Imagination, somehow they managed to convince a ton of universities to base their curriculum on Img proprietary design, this is as big as Nvidia "helping out" with heterogenous computing courses resulting in universities universally (hah) ignoring OpenCL, and exclusively teaching proprietary CUDA.
This is not simplified, we include the full RTL and all the building blocks required to run Linux (cache controllers, MMU, debug interface, load/store unit etc.).
We are granting an open license, which means you can see the code. CPU IP usually includes many options (i.e. parameters) that need to be set by design teams before doing any FPGA integration work (synthesis, functional simulation and verification, place and route, etc.). Setting those parameters up requires a lot of expertise which is usually accumulated over years of work in SoC design.
The blog article mentions a case where a university struggled to configure the MIPSfpga CPU because (naturally) IP can be quite complex, especially if you haven’t used it before. So we’ve actually done a lot of work to create a pre-configured, pre-validated version that is easy to use by academics (lecturers, students, etc.) for FPGAs. Of course, we also state in the press release that users can change it but they need to talk to us first to ensure that the changes they make are valid and sensible. The last thing you want is to break a CPU architecture; this will affect not just the behavior of the hardware, but also the toolchain.
I know it's become a bit of a tradition to abuse the word "open" in technology circles, but this is really pushing it a bit far. Even Microsoft didn't have the gall to call its "shared source" initiative "open".
but also a market failure compared to not so reduced ARM
There are tons of MIPS SoCs out there, mostly in various Chinese devices (the ones which are too cheap for ARM?) so I wouldn't say they failed completely, but in terms of performance or power efficiency MIPS is definitely behind compared to ARM or x86:
What would the "real CPU from MIPS" be, and how much difference would that make...? Loongson is a licensed design.
As for compilers, neither ARM's nor Intel's compiler were used for the other CPUs, so I think that's a fair comparison. ...and I thought one of the reasons to choose a very simple RISC like MIPS was that its simplicity made compiler optimisations easier?
AFAIK Loongson just licensed the patent necessary to implement the MIPS ISA but the implementation was done by chineese academia without any intervention from MIPS. You cannont judge the quality of MIPS processors in general by looking only at one chineese clone...
"IMHO that's a far more interesting architecture to study"
By interesting do you mean awful? The only reason for clinging on to x86 is maintaining compatibility with old proprietary (almost entirely windows) software.
In addition to all of the other examples given, countless SoHo routers (just about everything from the old WRT54G workhorse on up) run MIPS processors.
This matters because you get access to the full Verilog code for a modern CPU and a complete toolchain to boot Linux and write software. This also matters because these two vital elements (hardware and software) are coming straight from the source.
Open cores might get close to an implementation, but usually they break code compatibility or implement older versions of the architecture, which limits your ability to run (or update to) a modern Linux kernel or use a standard toolchain from established providers.
But the reason they break code compatibility or implement older versions of the architecture is that parts of the ISA are patented by MIPS. Other ISAs like OpenRISC, SPARC and RISC-V don't have these limitations.
Yes, that is true. MIPS Technologies held patents on the MIPS ISA and most of those were transferred to Imagination when we completed the acquisition. However, many of these patents are also owned by an internationally-backed consortium which includes ARM:
http://www.arm.com/about/newsroom/arm-announces-participatio...
I think this announcement is particularly important because it gives students the chance to study a piece of RISC history. When I was in University studying for my Electrical Engineering degree, we were asked to clone an 8-bit MCU. While it was a useful exercise, everyone was dreaming of seeing a fully open MIPS CPU. We were taught it yes, in the Computer Architecture course, but never got a chance to see any RTL code.
RTL code + slides beats slides in my book so that's why I feel this announcement is so important for me, personally - but also for universities worldwide.
You are right, I would haved loved this as a student too. But today there are real open source alternatives. Especially in an academic environment it is essential to be able to modify and share source code without restrictions so you can experiment and collaborate as you wish. E.g. students should be able to add custom instructions, experiment with cache architectures, build multicores, etc. All with the architecture they learned about in class.
So back to your point, a lot of universities teach computer architecture based on MIPS since it was defined by Patterson and Hennessy (Hennessy also co-founded MIPS Computer Systems Inc. and went into production with MIPS R2000 based on the MIPS I ISA).
MIPS is a very popular architecture, having shipped in over 800 million devices last year.
Now these universities that teach MIPS in their coursework have a chance to enhance the course by giving students access to the RTL too. When those students leave university, they will have the skills required to understand and work with a commercially-available CPU architecture.
Hopefully, this will get more traction than OpenSPARC did. I was really surprised that one didn't get adopted heavily. It runs Linux (and Solaris and OpenBSD), is multicore, CMT, low power.
There is also the RISC-V ISA (http://riscv.org/) being developed at Berkeley and the lowRISC SoC incorporating a 64-bit RISC-V CPU (http://www.lowrisc.org/) by a team in Cambridge.
The package includes a microAptiv CPU (MIPS32 Release 3) written in Verilog, a getting started guide, and various files for debugging, simulation and synthesis (free SDK, tools from Modelsim and Xilinx, etc.)
elegant as in simplified to the bone, but also a market failure compared to not so reduced ARM
>You’ve been here for two months, and now you want to give our IP away!
haha, nice try Imagination Technologies, you arent giving anything for free. Later on in the blog post is a mere mention of all the proprietary IP, patents and licenses they are NOT granting/licensing. This is a huge win for Imagination, somehow they managed to convince a ton of universities to base their curriculum on Img proprietary design, this is as big as Nvidia "helping out" with heterogenous computing courses resulting in universities universally (hah) ignoring OpenCL, and exclusively teaching proprietary CUDA.