I have a few words to say about IBM's proprietary technology: IBM PC Technical Reference Manual.
Regarding PowerPC, I've used a lot of IBM PowerPC 4xx series embedded processors over the years, and when IBM owned the technology the documentation was excellent and almost completely open. A while back they sold the technology to AMCC (now Applied Micro), and now you have to be a registered customer even to look at the list of products. I wonder how many sales Applied Micro misses out on because prospects can't even see what's offered?
I'm delighted to see IBM still pursuing PowerPC technology. IMO it's a great architecture, much cleaner instruction set than ARM. Go Big Blue!
I've used the IBM PowerPC 4xx embedded processors too. They were very power efficient for the amount of processing they could do, so much so that IBM built a supercomputer filled with them that was the fastest in the world at the time (the Blue Gene/L).
But IBM stopped building low power chips, which is why I think Apple dropped the IBM Power architecture for Intel when they realized that portable PCs where becoming more and more important.
ARM, which I think started all the way back with the Acorn processor, is very low end and therefore very low power, so it was a natural with smart phones, and has grown tremendously since then. Intel is working hard to come up with something that can compete in low-power tablets.
But low power isn't as important in servers, which is where Intel shines, and where the latest PowerPCs and Sparcs compete. Whether or not ARM can scale up enough to compete with them remains to be seen.
Note that Fujitsu has a Sparc license and still makes processors and servers.
I had to add this since it mentions all these processors:
The present ranking of the fastest supercomputers in the world (TP500):
1. Tianhe-2, in China, uses Intel Xeon
2. Titan/Cray XK7 uses AMD Opterons with NVIDIA graphics/processing chips
I am glad someone pointed out ARM's history. It came way back in the early 80s. When I was an undergrad, my school had a BBC lab with the first ARM boxes. But never go to do assembly level hacking on those boxes. For some reason the lab was not in the EE dept., it was in a special room next to the Principal's office, not the best of places to have a lab !
For those of you who were in diapers in the 80s or were just being contemplated by your parents in the 80s, ARM was essentially funded by BBC's computer literacy project. Govt. mandated technology does not seem too bad , does it ! Somehow cannot imagine NPR or the big broadcasters in the US sponsoring something like it. Can you imagine an NBC funded CPU and what kind of wierd copyright enforcing and DRM functionality it would have ? Probably a contant scanning DFA engine like the micron processor to scan all items in the data cache for copyright violations. Which will probably be hacked by someone to be an ultra fast virus scanner !
On a more srious note, ARM scales up pretty well with the new v8 inst set. Actually it wouldnot be too difficult to take a Power8 and change the front end to decode ARMv8. Exceptions and status registers would be a pain.
Gentlemen this consortium is about HPC and scale out clusters. And while I support ARM server as an analyst, this announcement is all about Power and heterogeneous compute riding Nvidia Pascal GPU to Processor interconnect.
No doubt Nvidia ARM Tegra will also benefit from Pascal in the Hadoop batch processing appliance space. But what this Power Consortium is all about is real time analytics and competing with Intel in the HPC space. On this front IBM and Nvidia can definitely wrestle share from Xeon + Phi acceleration. Which leaves AMD where?
The consortium is about surfing the performance wave for Nvidia and Tyan. And for IBM and Mellanox protecting what they do best, knowing Intel has their Infiniband block under development.
The only question then is what of Nvidia x86? That is where the lion's share of gold from Intel monopoly mine remains for the taking.
Nvidia's PCI bus license from Intel, assured by FTC v Intel Docket 9341, is essentially over as time left on the consent agreement is down to a single board product design cycle. Because of the bad blood between Nvidia and AMD, I seriously doubt any merger soon.
My bet is that Nvidia buys VIA, retires Glenn, and instills some Intel kick ass in the Centaur design team.
When Sun was struggling to stay relevent in servers it mades Sparc open and it went nowhere. I see no reason why IBM would not face a similar fate with Power8. I see vultures circling this architecture.
Open is the trend and is likely the way to go in the future. However, open hardware like open sources is a tricky "business". You let it open but you don't stop supporting. Nowadays, the successful open sources are grown by a couple engineers close to a mature point and then, a big company comes in to become a backer of the open sources. Ideally speaking, if the open source start out with a backer, the chance of success will become higher.
I see Rick point of the failure of Sparc. It is a great architecture too. I'm too afraid Power8 will go down the same path. Yet, who knows! Time is different now. Back then, x86 was the market favorite. Nowaday, market doesn't seem to be fond of x86 as much. In addition, if Power8 can deliver better performance with lower power consumption, it might actually dilute ARM in the mobile market.
1. Power ISA actually has an embedded variant that can be very power efficient, has VLE too like ARM Thumb. . Freescale had an app processor, the MPC823 that was power efficient. Technology politics I believe made Freescale decide on ARM, it had nothing to do with the Power arch.
2. If you look at the Freescale T4240 at 28 nm, it is about 10% faster than a sandy Bridge Core i7 (6 core). Has twice the no of cores but is much more power efficient. TDP is 60-80 W and typical power is 30-40 W. It does all this with a 7 stage pipeline, very short compared even to a Cortex A9. Bottom line efficiency is a function of the micro-arch, ISA really has nothing much to do with it. The PPC compnat that Apple bought (forget the name) also had an efficient Power implementation.
3. PowerISA is still licensed. if the ISA was made free, then this effort really has a chance. It was one of the reasons why we dumped Power for RISC-V in our open cpu effort. Even here we have kind of borrowed the POWER hypervisor model from the server profile simply becuase it is well thought out.
4. Sparc is alive in small efforts like LEON apart from Oracle. A lot of research uses the OpenSparc RTL. Their threading model is still top notch.
BTW, there is an intesting chip using the LEON3: the Skytraq Venus GPS chip, features in a successful Inidegogo campaign for an Arduino compatible GPS board (I thought about backing, but decided I wasn't likely to use GPS anytime soon).
Yes, sounds familiar to the move to make the Sparc T1 open source. But there was no group of companies around the Sparc CPU then like the OpenPower initiative, this could make a difference.
But my point here: Yes, nowadays you see seldom a Sparc server, but they are still around. Oracle has modified the Sparc architecture so that it is now a "CPU optimized for running a Oracle Database optimized for Sparc". It's not really a "general purpose CPU" anymore more like a very big and fast ASIC :-)
Main difference between Sun's attempt to go open vs IBM is that Sun always had an angle, a catch or a gotcha such as using the CDDL vs the GPL license for Solaris. When they tried Linux it was Sun Linux and was heavily restricted making it worthless except for the specific Sun servers which Sun was peddling. They were even willing to kill Solaris x86 instead of promoting it on x86 for fear of it undermining the SPARC business. IBM looks to be laying it all out there. They have embraced open source products like OpenStack and KVM for use on Power servers, added more Linux distro's plus the ability to run in LE or BE mode. OpenPower Foundation members seem to have the keys to the kingdom to develop to the platform whereas the platform they have supported for years is locking them out as they either product it themselves or put it on the chip. Plus, IBM is diverse and not as dependent on a single line of business like Sun was with most of its business in one way or another relying on SPARC or Solaris whether it was software, storage or services. Customers are realizing they are being preyed on by software companies with x86 technology as it is a never ending game of more core x86 servers with each refresh requiring more software licenses. The global economy is helping many CFO, CTO, IT Directors and others who care about budgets to evaluate technology that controls costs reliably and not perpetually keeping their hand in the proverbial cookie jar.