IBM today jointly announced with their partners, Nvidia, Tyan, Mellanox and Google, that they would be establishing the Open Power Consortium. For those unaware, this Open Power Consortium actually has very little do with power, but rather IBM’s own Power processors for enterprise computing which utilizes IBM’s own Power ISA. Considering the fact that IBM has slowly moved towards a heavier and heavier IP-based business model, it seems reasonable that they would want to keep the Power ISA alive through something like today’s announcement of the Open Power Consortium. With pressure from ARM and x86 in the server and HPC markets, IBM has refocused their efforts to include themselves and optimize greater for GPU based parallel computing.
IBM’s own Power CPUs actually have little to nothing to do with GPUs or parallel computing, however, they do have Nvidia onboard as their partner for the Open Power Consortium, meaning that they have a strong parallel HPC partner. With Mellanox, IBM gains the interconnect capabilities to connect their IBM Power cores with Nvidia’s GPUs and to connect the GPUs together in the future when a CPU may no longer be necessary. Tyan will obviously be the partner that will help them develop the right motherboards and power supplies to optimize for the consortium’s ultimate goal. The Consortium ultimately wants to to establish and build the necessary networking, server and GPU-acceleration technology aimed at the next-generation of hyperscale and cloud data centers.
Basically, right now, there is very little streamlining of parallel computing and it looks like IBM wants to keep their Power cores alive by working with all of the other HPC companies to keep them in the loop. By doing this, IBM could potentially keep themselves in the HPC game for longer as companies like Nvidia are constantly looking for ways to push their primary competitors, AMD and Intel, out of their server deals. Afterall, Nvidia would ultimately like to be able to integrate ARM cores into their Teslas and sell more Teslas if they don’t have to bundle with Xeons or Opterons.
So, we could theoretically see Google and Nvidia integrating IBM’s own Power cores into their own custom computing architectures. IBM clearly recognizes that companies like Nvidia and Google would like the ability to expand themselves beyond ARM or x86 and by offering an alternative, IBM could potentially give ARM and x86 competition in the server market. Something that they would like to do considering the new found growth of ARM in servers and the need for lower power in the data center.