Feeds

back to article Look out ARM, Intel, here comes MIPS – again

Imagination Technologies, best known for its PowerVR graphics-core designs used by a host of system-on-chip (SoC) purveyors from Apple to Intel, wants world+dog to know that it is dedicated to grabbing a hefty chunk of the compute-core market with its MIPS-based designs resulting from its acquisition of that venerable company, …

COMMENTS

This topic is closed for new posts.
Gold badge

It'll be amusing...

It'll be amusing to me if they come up with some "tons of MIPS cores on a board" type solutions, and SGI (as a vendor now of mainly massively parallel computers) started shipping a MIPS-based system again.

Anyway... MIPS is a bit odd, it's *REALLY* RISC, it makes ARM look laden down with excess instructions. I've used a MIPS-based DECStation, as well as several SGIs. My WRT54G access point is also MIPS, and I know some of the wifi chips have embedded MIPS so I've probably got a few non-user-accessible ones ticking over in the machine I'm using right now.

2
0
Silver badge

Re: It'll be amusing...

> it's *REALLY* RISC

that doesn't fit my recollection of it. If you want big and ugly then IBM's power is worth looking at.

(disclaimer: not a low level guy, I just read up on this stuff)

0
2
Silver badge

Modern MIPS isn't as RISC as it used to be...

The original MIPS design rationale was to exceluded anything from the CPU that could be done in the development tools.

For example, the original MIPS cores had no pipe interlocks which mean you had to be careful about how you accessed registers. You had to wait enough pipeline steps before the register values would be valid. That time was to be filled with other instructions and it was the compiler's job to make this all work.

This made MIPS assembly very hard to write, in fact it made it a real bastard.

More modern MIPS cores step away from that minimalism and now include interlocks making it easier to write assembly. There are still some odd-balls though: the instruction after a branch gets executed, making MIPS code pretty hard to undersntand when you first pick it up.

4
0
Anonymous Coward

Re: Modern MIPS isn't as RISC as it used to be...

"the instruction after a branch gets executed,"

That's called a "delay slot" (as in "branch delay"). Several RISCs had them (e.g. SPARC), some didn't (e.g. Alpha) [others, I can't remember]. Delay slots are also found in some DSP architectures.

If you think it's bad looking at assembler listings for chips with delay slots, you should try looking at a state trace of instruction fetch addresses on a logic analyser (instructions that are fetched may not actually complete their journey through a multi-stage pipeline, and sorting out this maze can be quite interesting, even without the added benefit of on-chip cache making some instruction fetches invisible).

Come back Motorola 68000. You could trust them (usually).

4
0

Re: Modern MIPS isn't as RISC as it used to be...

ARM doesn't have a delay slot, but if you perform any computations using the Program Counter register (e.g. retrieving local pool data, immediate data that's too big/complex to go in the immediate part of a MOV instruction) you find that it's actually pointing two instructions (8 bytes) beyond the instruction that does the computation. That's a bit mind-bending.for anyone who grew up on a CISC processor.

0
0
Bronze badge

Re: Modern MIPS isn't as RISC as it used to be...

Regarding delay slots - I've done a bit of SPARC assembler, and I got used to the delayed instructions being executed "out or order" quite quickly. Much easier than trying to remember the horrendous number of addressing options for the M68K!

0
0
Bronze badge
Boffin

Re: Modern MIPS isn't as RISC as it used to be...

That's a compatibility throwback to the ARM1 & ARM 2 which had a 3 stage fetch/decode/execute pipeline, so the program counter set by the fetch stage would be two instructions ahead when read in the execute state. Modern ARMs have far longer pipelines, but the 8 byte offset is always maintained (except for the one place it's PC+12) to make it architecturally independent.

0
0

Really great, but! ..

There needs to be a ground swell of development for MIPS based devices. The only way this will happen is if they get two things out there:

1) Something low-end similar to the Rasberry Pi. Runs BSD or Linux (or both). Cheap and cheerful and like the Pi with a decent amount of power and better developer documentation.

2) Something mid-range. A standard motherboard that if you want a bit more power you can get a 2 (or more) core system with PCIe expansion slots and it goes in a standard PC case. Eg: a modern version of the Malta boards. You want to get people USING MIPS gear for tasks. Give them a ready made example solution they can use daily and point to that remarks "this is a MIPS system".

Both these things need to be affordable. Profits need to be spread over a large inventory rather than a small one. MIPS *NEEDS* hackers and enthusiasts that when they need an embedded or high-end processor will turn to MIPS gear first rather than ARM or x86 because they're familiar with it.

MIPS needs new generation hearts and minds. Not just slightly plumper existing markets. They need decent affordable access to the devices.

1
0
Silver badge

re: why?

Why does there need to be a groundswell of development for mips?

I don't care what CPU is in my phone so long as the apps work, and they are written in Java

I have a Pi but I really don't ae if it has an ARM or Atom CPU - I couldn't even tell you if it was 32 or 64bit without checking /proc/cpu - I just care that is is a little Linux machine and has python.

What will make this a success is a customer using it rather than an ARM in sufficiently large numbers.

Still it will be nice if ARM had some competition - look at what a monopoly did to Intel

2
1
Silver badge

@Uridium

I couldn't disagree more.

There are already a few projects using MIPS. Many OpenWRT devices use MIPS because it is very common in routers.

ARM was a smash hit long before RPi. 99.9% or RPi hackers are programming in C and the ARM connection is irrelevant.

ARM succeeds because they make it really easy to design ARM into chips, the debugging is common (one debugger covers you from $1 micros to top end multi-cores), cheap per unit cost, and they have a very good story for power etc. On top of that ARM are healthy and are a safe bet.

MIPS needs to beat ARM convincingly on most of those fronts or they won't get a look in. It does not help if they are 10% better on power if the chip makers don't find it easy to integrate or people worry they will be out of business in 5 years.

ARM has the lead and chip makers will need a damn good reason to change.

5
0
GM

A usb stick format "mini-pc" would be good too: cheap, plays to the current TV strength, and would reinforce its importance in the Android market.

1
0
h3
Bronze badge

Nothing wrong with MIPS it just needs some of the stuff like DMA that SGI had.

ARM has started to become a total mess.

(The docs / devboard and Netbsd mips and you have a chance of getting it working.)

0
3
Anonymous Coward

Re: "ARM has started to become a total mess."

Really?

Sounds interesting. Care to share?

1
0
h3
Bronze badge

The best reasonably priced router is mips. (Ubiquiti EdgeRouter Lite). Got stuff that you don't normally get under £1000

0
0
Anonymous Coward

"The best reasonably priced router is mips. (Ubiquiti EdgeRouter Lite). Got stuff that you don't normally get under £1000"

Kudos to them for building a good router. But that's the bit that counts.

How much of that could (not) have been done with ARM Inside, and does it matter whether it's ARM, MIPS, or whatever?

I've not looked at network-centric SoCs for quite a while so what follows may be dated (I didn't even know Cavium had bought MontaVista, presumably for their embedded + realtime MontaVista Linux product family which I have used in the past).

This Ubiquiti box looks like a Cavium http://www.cavium.com/OCTEON-Plus_CN50XX.html) plus a relatively widely available open source software package, right? The SoC looks like a fairly generic router-centric SoC, with one exception: it's a 64bit (MIPS64, specifically) SoC rather than a 32bit SoC.

ARM (and partners) can't really do 64bit yet. Does a router need 64bit? I can't say. But apparently this one works well. Does it work well because it's MIPS? Because it's 64bit? Because it's based on a nice SoC which just happens to be 64bit?

Meanwhile, back to the main news: there are already various MIPS SoC designers out there (some better known than Cavium, e.g. who hasn't heard of Broadcom?). How does the Imagination Technologies deal affect them now the dust has had time to settle? The article made little (no?) mention of SoC design partners. Did something fall off the press release?

ARM (and partners) make nice stuff. But there may well be room for others too. Who wants to design the first Atom-ased SoHo router? Hmmm, thought not.

1
0
Bronze badge
Paris Hilton

Synergies ?

As the PRC has licensed MIPS instruction sets for the native Dragonball CPU development (ITIRC), would this also expand developer resources and create an alternative very non-USA based source of hardware ? This would also offer the oportunity for Chinese developers to cut into coding and apps on a big scale.

I miss Paris in the springtime.

0
0
Anonymous Coward

They really shouldn't be trying to aim for Android. Sure they Java apps can run on it but the native apps will not. They should have started sooner so that Sailfish OS (Jolla), Ubuntu and Firefox OS could have started from scratch using MIPS. At least then you don't have fragmentation as the entire platform could have evolved around MIPS from the start. Trying to get people to move to MIPS later doesn't make much sense and takes development cycles that many do not want to allocate.

Where MIPS might be able to make a comeback, the embedded market. Factory GPS and other computer functions in cars would be much easier to make the transition than smartphones.

0
0
Silver badge

That's what the MagicCode product mentioned in the article deals with — executing ARM code on MIPS. So they've got the native code problem solved.

0
0

did you follow the link to the MagicCode page and see the FAQ? I wouldn't say they've got the native code problem solved.

1
0
Anonymous Coward

MIPS emulating ARM

"I wouldn't say they've got the native code problem solved."

Quite. But does that matter, as long as they've done enough to fool the decision makers with the checkbooks?

0
1

Hopefully, now that the patent that caused Lexra so much trouble (in which processor instructions dealing with misaligned memory) is resolved, MIPS64 can really take on real life.

The Chinese Loongson (Godson) processors have been used in their supercomputers for several years. MIPS is not a dead technology by anyone's opinion. A simple RISC core even more easy to scale and license than SPARC is exactly what the world needs, and China knows it. Now the rest of the world can revive this somewhat forgotten alternative to the hundreds of hyped 32-bit ARM-based derivatives flooding the market. MIPS is has been 64-bit for over a decade.

3
0

Quite a bit more than a decade

> MIPS is has been 64-bit for over a decade.

Actually, something like 22 years. MIPS was the first mainstream architecture to go 64 bit and the R4000 came out in 1991.

http://en.wikipedia.org/wiki/R4000

If anything, mature 64 bit MIPS architectures would be its key advantage over ARM, although that will be a transient window of opportunity. It's also got a much richer background in server architectures going back to the 1980s.

Also, you can already get MIPS based laptops if you want a native dev box.

4
0
Anonymous Coward

"MIPS is has been 64-bit for over a decade."

Mid 1990s was when 64bit caught on as bleeding edge technology.

DEC's Alpha was a new architecture, MIPS64 followed the previous MIPS, SPARC64 followed the previous SPARC, and Power from IBM had been around before too.

IA64 got there eventually, but by the time it did, the emergence of AMD64 had showed that Intel's claims that "x86-64 is impossible" were a little bit misleading, and Intel had to invent EM64T in a bit of a hurry.

There may be other 64bit chips, I forget, but realistically available (if not necessarily widely affordable) 64bit processors are around two decades old.

ARM are way behind the industry's broader 64bit curve. So what. It doesn't matter, in their markets. And they have a cunning plan. There's time.

[Definition of 64bit here: 64bit logical address space. ie printf("sizeof (char *) is %d\n", sizeof (char *)) says 8]

0
0
Bronze badge

Re: Quite a bit more than a decade

MIPS was the first mainstream architecture to go 64 bit and the R4000 came out in 1991.

Agreed (though it's possible I'm forgetting something), but outside the mainstream there are a handful of predecessors worth mentioning. The Intel i860 came out in 1989, and had a 64-bit address bus. National Semiconductor's Swordfish (derived from the never-released NS32732) had a 64-bit address bus and appeared around 1990.

And then there's the Cray-1, from 1975, which is generally reckoned to have been a 64-bit machine (by register size), though its addresses were only 24-bit.

0
0
Silver badge

New lamps for old

It seems all the oppressed architectures of the byegone era are showing shoots of green these days. Almost as if there were a change in the climate, the end of an ice age.

2
0
Silver badge

Re: New lamps for old

Companies finally figured out shelving them due to the coming Itanium wave was a wah wah (trombone sound for epic fail).

3
0
Silver badge
Coat

Re: New lamps for old

Intel got such a big part of the market with a rather poor architecture and now there seems to be room for more competition. The Intel/Microsoft hegemony is no longer there as strong as before.

4
0
Silver badge
Boffin

Re: New lamps for old

Ice Age? More like the end of the Dark Ages. That the x86 crap architecture has lived on as much as it has is surprising; if the car industry had gone the same way the microprocessor industry has gone, we would still be using horse-powered vehicles and just now "horseless carriages" would be making a comeback.

1
0
Silver badge

Re: New lamps for old

> if the car industry had gone the same way the microprocessor industry has gone ...

Cars would go 1000mph, would travel 500miles on a pint of fuel, and would be 2inches long.

1
0
Anonymous Coward

Re: New lamps for old

"Cars would go 1000mph, would travel 500miles on a pint of fuel, and would be 2inches long."

And everyone would have to buy a new car/bus/truck every time Microsoft put a new logo on their chain of petrol stations and associated franchises.

1
0

Bring it on!

Speaking as an ARM shareholder (it's 10% of my pension fund), I welcome the competition. And this is genuine competition, not like thingummy-inside who compete not so much with ARM as with ARM's licensees. Good luck to them!

A reinvigorated MIPS doesn't have to be taking ARM's cake. Its success can grow the total cake, to everyone's benefit.

(Can you tell I'm just back home from overindulging a bit?)

4
0
Silver badge

No publicly available PowerVR documentation

Without documentation, only the manufacturer can fix the drivers. Even if the drivers are not broken now, they will break on a kernel or Xorg upgrade. When the manufacturer releases new hardware, they stop maintaining the old drivers. Most 3D acceleration hardware suffers from this fault.

For open source, changing CPU architecture just means changing the name of the cross compiler. Qemu detects the new target architecture of your application and the self run as usual. The way to get open source developers is to release proper hardware documentation.

4
1
Silver badge

Re: No publicly available PowerVR documentation

Talk about missing the point… has the lack of docs on the Power VR chips hindered the sales to Apple? Do the lockdowns on the Qualcomm and Broadcom chips stop them finding their way into our gadgets?

Anyway open source OS on MIPS does have a long and successful history and long may it continue.

0
1
Bronze badge

MIPS can chat about the good old days with Power in the old-processors home

If IBM's is struggling to try and make Power processors look modern and relevant, and failing, it should be pretty clear that developers are quite happy with just a choice of Intel or ARM.

Finding niche applications for a processor used in embedded applications won't cut it, if it did we would have Parallax Propeller and PIC phones and tablets by now and that sure hasn't happened.

Let's just say that the N64 was as near to mainstream as MIPS will ever get, and even with the might of IBM behind it, Power architecture ultimately couldn't compete even when it was seemingly entrenched in Apple products.

1
0
Silver badge

Re: MIPS can chat about the good old days with Power in the old-processors home

Well the GameCube/Wii/WiiU is PowerPC so by that definition Power is still competing (after a fashion)...

0
0
Go

I miss my IRIS Indigo

I loved that little purpley box - just the right size to rest your feet on :-)

The Indigo2, Indy and O2 just weren't as practical (although there was fun to be had building a 'mousetrap' style course in front of the I2 and seeing how far you could get the CD caddy to travel after ejecting it... ah, happy days as a graduate plonked in front of lots of expensive SGI kit :-)

2
0
Bronze badge

New and Old Markets

>For MIPS cores to achieve a quarter of all CPU sales, however, Imagination will have to do more than move them into new markets. It will have to crack existing markets,

No MIPS needs to create new mass markets and dominate them. ARM didn't crack existing markets, they made the low power market their's and went on to dominate it. It has taken a years for Intel - the dominate player in existing markets, to deliver products suitable for the markets ARM dominates.

0
0
Anonymous Coward

Re: New and Old Markets

" It has taken a years for Intel - the dominate player in existing markets, to deliver products suitable for the markets ARM dominates."

Intel aren't there yet in the real SoC market, whatever the press releases and churnalists may say. Intel today in integration terms are probably not much further on than DEC were in the 1990s with their high-integration 21066 Alpha [1] SoC (except they weren't called SoCs back then, and they certainly weren't the affordable SoCs seen today).

MIPS-based SoCs have been there or thereabouts for ages with affordable SoCs (just ask Broadcom and other multi-architecture SoC builders).

[1] 21066: you won't have heard of it. Nearly twenty years ago, an Alpha 21064 64bit RISC core plus all the SoC basics including PCI controller and basic VGA frame buffer (just add RAMDAC). Just add DRAM and a SuperIO chip and a hard drive and you've got an NT/Alpha PC. Add SCSI and a network and you could even have an Alpha VMS or UNIX box. Sadly it ran like treacle, unlike the real Alphas, allegedly due to issues with the bus interface (pre BGA, not enough pins, too much multiplexing?). Used in the Tadpole Alphabook and a few others. Look for "21066 Data Sheet" for a 70-page chip data sheet in PDF, or "21066 Product Brief" for a short glossy - in gzipped PostScript (!).

1
0
Bronze badge

I almost bought a MIPS machine

It was one of those Chinese sub-netbooks, a bit too sub to be useful, too tight on RAM. It's assembly cost that strangles the low end of GP computing. A new chip won't change that.

0
0
Anonymous Coward

Re: "assembly cost that strangles the low end of GP computing. "

"It's assembly cost that strangles the low end of GP computing."

May have been the case once. May no longer be as true as it once was.

Motorola have started to onshore handset assembly. The cost impact: $4 per handset.

http://www.theregister.co.uk/2013/08/28/motox_teardown_onshoring_handset_costs/

Raspberry Pi cards are assembled in Wales

http://www.theregister.co.uk/2012/09/06/raspberry_pi_production_moves_to_uk/

1
0

The success of a MIPS core in the IP market for android devices, or any other devices for that matter, will be mostly predicated upon the ease of integration and library of functional units that can be mixed and matched with it to build the SoC. Assuming a similar license cost to ARM of course....

Speed to market is everything with these small devices, and if a licensee can "bish bash bosh" together his SoC faster with a MIPS core then he will do so.

AG

0
0
Silver badge
Linux

Hurts Intel's attempt

As MIPS is already more established in Domestic Routers/Modems than x86, this likely hurts Intel's attempt at getting into the phone market see Intel: Our new mobile chip SoCs it to its predecessor. I can't see it taking more than a few percent of ARM's juggernaut.

It's likely to be more attractive for Tablets than Intel's "Clovertrail" too.

If Intel want to succeed in phones and tablets then they need to get MS to stop calling all versions of their OS "Windows" and only call the one compatible with 2000, XP, Vista and Server applications "Windows". Maybe MS also need to market XP with a new service pack as Legacy Windows and examine why the XP tablet failed outside of specialist markets. I'd say it wasn't the UI as it was a large screen, but cost and battery life.

The Zune like TIFKAM is a good idea for phones and small tablets (The classic desktop style CE UI on a 320x 240 PDA was daft). Larger tablets where the thumbs can't address all of the screen need a different GUI. 10" approx and larger tablets need a different approach. A screen on a desk used with a keyboard needs the traditional WIMP GUI.

So MIPS has some future. Hard to see one for Intel Cloverdale/Medfield unless they sell them cheap for cheap windows x86 compatible notebooks. Phones don't need x86 compatibility as the screen is too small for traditional Windows applications like Sage, Excel, Powerpoint, Visio, MS Project, or even traditional Outlook with Exchange server for meetings etc. Applications for handheld devices need not just a different GUI but need to be written differently too.

There is no "one size fits all".

3
0
This topic is closed for new posts.