Lots of tech noise about IBM's recent 64 bit licensing announcements such as this generally excellent Faultline analysis at the Register.
ARM Competition
One comment made by an IBMer however is quoted without serious comment by the Faultline article. It is worth picking up this because it really shows how people can uncritically accept totaly stupid claims that don't stand up to more than about 2 seconds thought. The quote is:
"The ARM chip and the Intel chips were fine for what they did and they were ok for the mobile and PC world, but for a world which requires lots of floating point calculations and handling of video from its chips, you are going to need a multiprocessor with a 64-bit architecture," Nigel Beck, IBM
IBM is making a cheap (and generally failing) shot at ARM. It doesn't fly for a number of reasons.
Number one is that the vast majority of embedded applicatins are still 8 bit. It is certainly true that people are migrating to 32 bit ARMs because the cost etc of a 32 bit ARM7 core plus bits is about the same as the 8 bit solution and it can do a few more things. ARM can still 8 bit wide memory busses and all sorts of other tricks that reduce power and size and increase code density (e.g. Thumb instructions). People are moving to 32 bit because memory has got so cheap that it is irritating having to fit things in to 64kb or deal with ugly hacks in 8 bit CPUs to procss the extra - little things like that. Very few of them have anything to do with graphics, music and the like. In fact the average UI is probably a couple of buttons and a handful of LEDs. I do believe that we will begin to see slightly more advanced UIs with the drop in small LCD prices but even so we are looking at simple text and pictures not HDTV and Dolby Surround Sound.
Second unless you really think that IBM is willing to offer a 64 bit power PC chip for a royalty rate of 10c or so in bulk then you have a pricing problem. Recall that ARM is on course for being paid for about 1.3 Billion ARM cores this year at an average royalty rate somewhat less that US$ 0.10. Sure some multi CPU or ARM 11 cores may pay higher royalties but the average is <$0.10. The obvious market for higher power ARM CPUs is in higher end cell-phones and PDAs. Higher end Cellphones/PDAs are no more than about 10% of the half billion or so total Cell Phone market. Add maybe another 50 million game players (That was a SWAG) and you still have a mere 100 million devices - less than 10% of ARM's volume.
Related (point 2a if you like), if IBM supplants Intel as the supplier to PCs, workstations and home game players (ignoring portables) it has a 200+ million market that it can serve at a considerably higher rate than $0.1/CPU. IBM doesn't really want to compete at ARM's game - it isn't in IBM's interest to drive the cost of a Power PC core down to ARMs level. It may be in its interest to drive it down to comparable Intel/AMD levels, which are something like 10-100 times that of ARMed SoCs.
Thirdly ARM already has a significant number of DSP, SIMD/MIMD and floating point options. These come with a significant advantage over a hypothetical 64 bit Power PC - no need to recompile existing applications - while providing the incremental migration that allows users of ARM's current offerings to offer high end graphics etc where required. Oh and by the way Intel has of course also equipped some of its ARM licensed (and royalty paying) Xscale chips with various Multimedia extensions if people don't like the home grown ARM version.
The Sincerest Form of Flattery
It is true however that IBM does seem to be trying to copy part of the ARM business model - something that I have written about extensively - but that does not mean that IBM is seeking to replace ARM. Some of the Faultline article sounds remarkably like ARM's licensing terms and the way it allows extensions in some areas but not others:
Power.Org will take on the standards around the Power processors range, operating very much in the same vein as the Java Community Process. IBM wants to keep control of the core instruction set, but then again it hopes to put some silicon items outside this such as accelerators, so that the chips can become highly customizable without losing their shape.
"For instance one of our licensees has added Chinese language processing at the chip level," said Beck.
...
The first two standards that Power.org members will be asked to define are bus architecture and a reference specification for high volume servers.
The bus architecture will enable different components on the same "system on a chip" to work together. Standardizing the bus architecture eases the integration of technology from multiple vendors.
After this, and once the first meeting of Power.Org happens, the organization will set its own agenda, with IBM only able to rule on the central choice of the chip instruction set.
I believe it is far more likely that IBM is seeking to end the Wintel domination that it unwittingly ushered into the world nearly 25 years ago. This is not no more a threat to ARM than a clash of dinosaurs impacted nearby insects. Sure a couple insects may get crushed but the carrion left over at the end is a perfect breeding ground for more. ARM cores are like Beetles, of which a famous naturalist stated that "God seems to have an inordinate fondess for", Intel and IBM are large vertibrates by comparison and they perform different tasks and rarely harm each other. Indeed it would not at all surprise me if higher end Power PC devices would in fact have a few ARMed USB, Bluetooth, disk or other controller chips inside them.
Echo...
However there is an interesting echo to the IBM story in the recent history of ARM - if you look back at the discussion about the ARM/Artisan merger when it was first announced you come across this excellent EET article and in it there is this:
ARM's strength has traditionally been in meeting the needs of architectural designers, providing them with MPU cores to differentiate their designs. In contrast, Artisan's forte has been in understanding the physical layer, building libraries that ensure RTL correlates well to silicon.
As process geometries shrink there is a greater need for that kind of linkage. The merger will allow ARM to ensure that its MPU cores — and now, likely, other cores as well — will be more closely correlated with actual silicon. It will also open up the customer-owned-tooling market, or at least assure customers using foundries that ARM cores are built with Artisan libraries.
Thus, ARM may have negotiated with one eye to a future in which competitive CPU core design will be impossible without an intimate relationship among IP developer, library developer and foundry.
"IP without physical understanding is decaying in value," Lanza said. "Bringing in the expertise of the physical layer to ARM is very important for the future of ARM but also for the future of IP in general."
which has an interesting echo to this in the Faultline/Register article:
"The traditional ways of making chips are reaching natural barriers and now integration is the problem to solve," said Nigel Beck of IBM
I speculated in this thread at TMF about the resons for the ARM/Artisan merger and said the following: Base 1) We know that as we get to 90nm and smaller processes costs go up and yields seem to go down at least on the first spin. 2) Contrariwise ARM's big selling point is "reduce TTM and risk".
Hence the ever shrinking process tech is both an opportunity and a threat for ARM. The opportunity is to really nail all the competition be being indisputably quicker in TTM. The threat is that ARM designs will be either no better or worse in adapting to these smaller geometries. If everyone needs a second spin all the time then the cost of that is going to overshadow the cost (and time) saved by using ARM vs using something more customized.
However Artisan brings to ARM expertise in some of these areas. Of course I can't be certain of ARM's future sales strategy but I'm going to guess it would be making marketing messages like "ARMed Chips never need a respin". Given that respins cost $millions and take 3-6months that turns the threat into the opportunity, but to do so requires ARM to have considerable knowledge of the physical layers and the peripherals. Artisan brings this knowledge to ARM.
In other words it is possible that ARM is in fact in a situation where not getting Artisan's (or competitor's) knowhow in house would in fact be eventually fatal. And then later I don't know if this is the reason but certainly it seems that 90nm and below seems to be causing trouble for all chipmakers and fabs. The problems are clearly solvable but they are also clearly harder to solve than people thought. In that case it is possible, indeed likely, that the same issues will hit the users of ARM's cores. If so then clearly ARM needs to come up with a way to ensure that a next gen chip with an ARM core has as little a chance as possible of hitting these problems. If so then (according to the EET article) Artisan looks like a repository of the necessary knowhow and IP. I would consider the IBM quote above to be partial justification for that opinion.