goafrit wondered: Still a mystery that Intel has not bought ARM.
This is JMO: I think the main problem is that it would be admitting defeat, and admitting that RISC is superior to CISC at least in some applications. Ironically, Intel bought the StrongARM architecture from DEC. But they couldn't figure out how to make it successful and sold it to Marvell. Intel couldn't do anything with the 64-bit DEC Alpha either. x86 is the "cash cow", and as such I would assume the x86 people call the shots at Intel.
Intel is trying to get power consumption down with Atom, and seems to be making progress. However, it's hard to change your mind-set to low power after decades of producing highly-profitable hot chips that require increasingly sophisticated cooling schemes. ARM has had a mind-set of low power and low transistor count from the beginning.
I suppose Intel could just buy ARM and kill it, but there's a huge ecosystem out there. Plus, many ARM vendors have architectural licenses so if Intel were to buy and kill ARM, the architectural licensees could simply forge ahead with their own versions or collaborate on common extensions. At least, that's how I understand it.
JMO/YMMV -- feel free to correct any misconceptions
>> Power has been losing ground to the x86 in servers and embedded and to ARM in embedded.
Still a mystery that Intel has not bought ARM. I know they can absorb them easily if the regulators will approve. There seems to be no reason why they alllow ARM to continue to expand with all the disruptive innovations they have brought in the industry.
tb100 wrote: I thought IBM sold the rights to the embedded versions of the Power processors (401, 405, 440) to AMCC.
This is correct for the 405 and 440. I think the 401 and 403 were orphaned. However, AMCC (now Applied Micro) seems to have lost interest in PowerPC and are convinced 64-bit ARM is the way to go. They have "low cost" development kits for those products -- $1495 and $2495 MSRP.
I tried searching for the 405 and 440 at apm.com and got bupkis. I think you have to register even to find out if they still offer the products. A great way to expand PowerPC business I think not.
I have fond memories of the PPC401GF. It came in an 80-pin TQFP with multiplexed address/data bus. Worked great with 32-bit synchronous SRAM. Fun little chip -- so much easier to deal with than enormous BGAs.
I really like the PowerPC 32-bit instruction set. It's so much cleaner than x86 or ARM. I did a lot of designs with IBM PPC403, 405, and even the 401. Very clean interfacing and excellent documentation.
IMO PowerPC needs something like BeagleBoard/Bone or Raspberry Pi to give it a restart.
I'm sorry but this shocks me. Is the US Government being a little reckless with our tax dollars? IBM Power? Really?
It's going to be a huge amount of work to get all these new technologies working well together. A new GPU interconnect from NVIDIA. A new processor interconnect, CAPI. On top of that you are using the most bug infested interconnect on the planet, Mellanox.
Intel is coming out with Omni Scale fabric. It's going to obsolete Mellanox. I think it will be much better than needing that x16 interface and another chip.
Even if you double the speed of PCIe G4, you can only reach 200G. The era of connecting the network via a PCIe x16 interface will be coming to an end when Omni Scale comes out.
The IBM POWER 8 is a nice server processor, but when I compare the POWER (or more precisely PowerPC) ISA based SoCs for embedded with the ARM V7 ISA based SoCs, PowerPC based SoCs lag behind. And by now the first ARM V8 ISA (Cortex A53, Cortex A57) based SoCs are available.