This site may earn chapter commissions from the links on this page. Terms of use.

The story of how Intel rose to dominance in the server and workstation markets was written more a decade ago. One time upon a time (the late 1980s and early 1990s), various RISC, CISC, and SPARC architectures dominated the high-end PC manufacture, even every bit consumer PCs overwhelmingly opted for x86 designs. Intel leveraged both economies of calibration and its alliance with Microsoft to push x86 processors into both information centers and workstations. By 2000, Intel had seized the majority of the market, forcing companies like IBM and Sun into smaller and smaller niches.

Today, IBM wants to reverse that trend — and it thinks Ability 9 is the chip to do it. The new cores will come in two flavors; a "scale-out" design meant for large HPC applications and supercomputers with back up for iv or more than CPU sockets and a "scale-up" pattern with i-2 sockets. Side by side Platform reports that at that place will be iv flavors of Power nine in total, varying in terms of total core count, direct-attached memory, the corporeality of simultaneous multi-threading (SMT) per-cadre, and support for various co-processors and other types of compute hardware (ASICS, FPGAs, and GPUs). Power 9 processors can exist fastened directly to various accelerators rather than communicating via the PCI Express coach. Some of the variants will also support IBM's Centaur memory buffer technology to meliorate performance and implement an L4 cache.

The Power 9 die.

The Power nine die.

Ability 9 CPUs volition be built by GlobalFoundries on its 14nm process node and volition exist the get-go chips to implement Power ISA iii.0. This is the offset version of the ISA to be released since IBM launched its OpenPOWER initiative and the offset to back up AltiVec 3 instructions. Like SSE or AVX, AltiVec is a single instruction multiple information (SIMD) instruction set, just how much additional performance it'll evangelize over its predecessor isn't currently articulate.

Challenging Intel, ARM… and AMD?

ARM, AMD, and IBM accept all announced fresh challenges to Intel's most-total dominance of the data center market in the past few weeks. Data centers aren't but a lucrative market — they're one of the simply areas of computing that'south really growing. The tablet market is shrinking, smartphone markets in the developing globe are maturing rapidly, and the PC industry has been in gratis-fall for nearly v years. The growth of deject computing and services similar Google and Facebook means that selling to these customers has become essential to superlative-end server providers.

Each visitor is making different moves to address these markets. AMD, of course, has Zen, its new architecture that it hopes volition offer potent contest for its long-fourth dimension rival. ARM added a flexible vector instruction set to boost performance in HPC workloads, and IBM is launching an all-new iteration of the Power architecture.

"Nosotros started with a 64 flake compute building block and we coupled that with a 64 bit load store building cake," Brian Thompto, senior technical staff at IBM said in his Hot Chips presentation. "Each compute building block has a heterogeneous mix of compute, fixed and float, supporting scalar and vector. This allows us to obtain high utilization of our compute resources while also providing seamless exchange of information and shared information paths. Information technology also serves as an efficient building cake for managing education menses through the machine. Nosotros couple two of these 64 scrap slices together to make a 128 scrap superslice, and this is our concrete blueprint building block."

ibm-hot-chips-power9-execution-slice

IBM's new cores and their peripheral accelerators will be tied together past its custom on-fleck fabric. L3 enshroud bandwidth is said to be 256GB/s, with support for PCI Express iv.0 (when that standard is available), multiple 25Gb/southward "Bluelink" ports (these are used for the fastened accelerators and for Nvidia's NVLink ii.0 protocol). Aggregate bandwidth from all these sources is over 7TB/southward, though the scrap would rarely leverage all of these devices simultaneously.

IBM claims that Power ix volition offering up to 2x the functioning of Power 8 per socket — and it's already courting at least one significant customer. Google has reportedly field-tested Ability 8 and has said information technology would move to Power 9 if doing then would salve it 20% power compared with using Intel hardware.

Betting against Intel in this space is risky — the CPU giant is well enlightened of how critical data centers are to its bottom line, and its more than than willing to spend big bucks to defend its marketplace share. At the same fourth dimension, Intel's near-total potency of the market gives its competitors nowhere to go simply up. If ARM, AMD, and IBM collectively took 15% of the market away from Intel over the next few years it would be a meaning achievement and acquirement generator for each — and a meaning loss for Intel.