Missing big picture
Guys, You are missing the bigger picture. Processor is just one thing. It is a good piece of HW, but can do everything by itself.
OpenPower is about several other stuff:
- Higher core performance. Less cores = less licensing, less space, less power... other comments covered it but companies would love use less licenses on smaller datacenters.
- Higher multi-threading. Each core = 8 threads (with perf) vs 2 threads on x86.
- Virtualization via HW = less licensing, higher security, better stability (both PowerVM and KVM), open standards, less overhead. One less critical piece of SW to manage.
- Little Endianess & Big Endianess = Ease to migrate old and build new multiplatform apps
- CAPI - It not about supporting GPUS. It is about having GPUS and CPUS sharing the same memory space. Like having a server with 2 flavors of CPU and allowing it to decide where it will run the code. GPUs can access server RAM natively, no need for programming it via SW = This allows for servers that can offload processing to specialized GPUs without fancy coding or specialized SW.
- Same goes for FPGAs. Having FPGAs for compression, encryption, network acceleration, you name it... all natively being accessed by the servers CPU and sharing memory space.
- Tighly NIC coupled HW - In the yesterday event, Mellanox talked about Tightly Coupled CPUs via NICs. A Bunch of small servers, working as a single entity via a high-speed low-latency network.
The overall picture is bigger and the CPU itself is just part of it... by having an ecosystem developing things around the CPU, you can have servers optimized to almost anything you can imagine. X86 can do something similar ? Yeah, they have GPUs and FPGAs, and Networking... but is is all about the SW making all these parts work together. The HW itself does not care what other HW you have on the same box.
That is all the fuss about the openPower. HW aware of other HW pieces working together to allow for for easier and more powerful SW.