Nutanix today announced it is adding IBM Power processors to its data center servers for AI. Last year Microsoft, Amazon and Intel announced they were developing Field Programmable Gate Array-based servers. Google announced it was developing an ASIC-based A.I servers and Nvidia launched the DGX-1, a GPU-based supercomputer in a shoebox for A.I. that costs $130,000 per box, a relative bargain.
What’s missing is the Intel X86 processor, the venerable workhorse from the last 40 years. The days of a do-it-all processor are coming to an end as more compute takes place in specialized data center environments. Cloud vendors will benefit from reduced capital costs (one FPGA server can do the work of 100 X86 servers), less power consumption, less floor space and less heat.
The challenge is porting code to the new architectures. FPGAs can instantiate the software logic in silicon, but it means removing 40 years of legacy code. Also, raise your hand if you can spare your senior software engineers to port code for three weeks. At this point its very hard to get developers to pay attention to this.
But the big vendors are taking the lead. Last year announcements were made. This year some vendors promise to deliver products. Next year we might see commercial loads running on these systems. It’s meaningful that one of those vendors is Intel. And then it will become much easier to get the developers’ attention.