Linux powered Blue Gene/P becomes the fastest computer in the world

newsguy 0 Tallied Votes 590 Views Share

IBM has just announced the arrival of , the newest and fastest supercomputer around. How fast? Well Blue Gene/P just about triples the performance of Blue Gene/L, which currently holds the official title of the world’s fastest computer.

The IBM System Blue Gene/P Solution is designed to operate continuously at speeds exceeding one “petaflop” or one-quadrillion operations per second if you prefer. That is 100,000 times more powerful than an average home PC, and means that drug researchers could run simulated clinical trials on 27 million patients in one afternoon using just a fraction of the machine’s power. IBM claims it can process more operations in one second than the combined power of a stack of laptop computers nearly 1.5 miles high.

Not least because Blue Gene/P can be configured to reach speeds in excess of three petaflops.

With Blue Gene, the IBM design team has sidestepped two key constraints on state-of-the-art supercomputing, namely power usage and space requirements. It has been purpose-built to fit in smaller spaces and use less electricity compared to other commercially available supercomputer designs. The end result is a machine that is, according to IBM, “at least seven times more energy efficient than any other supercomputer.” The breakthrough Blue Gene design uses many small, low-power chips each connected through five specialized networks inside the system.

Four IBM (850 MHz) PowerPC® 450 processors are integrated on a single Blue Gene/P chip. Each chip is capable of 13.6 billion operations per second. A two-foot-by-two-foot board containing 32 of these chips churns out 435 billion operations every second, making it more powerful than a typical, 40-node cluster based on two-core commodity processors. Thirty-two of the compact boards comprise the 6-foot-high racks. Each rack runs at 13.9 trillion operations per second. The one-petaflop Blue Gene/P configuration is a 294,912-processor, 72-rack system harnessed to a high-speed, optical network. Blue Gene/P can be scaled to an 884,736-processor, 216-rack cluster to achieve three-petaflop performance. A standard Blue Gene/P configuration will house 4,096 processors per rack.

There are some key differences between Blue Gene/L and Blue Gene/P. In hardware, Blue Gene/P moves to more (four vs. two) and speedier (850 MHz vs. 700 MHz) processors per chip; more memory and an SMP mode to support multi-threaded applications. This new SMP mode moves Blue Gene/P to a programming environment similar to that found in commercial clusters. Blue Gene/P dramatically scales up collective network performance to minimize common bottlenecks inherent in large parallel-computing systems. Software marks the third key upgrade for Blue Gene/P -- system management, programming environment and applications support have all been refined in Blue Gene/P.

Blue Gene’s operating system is based on open-source Linux. Applications are written in common languages such as Fortran, C and C++ using standards-based MPI communications protocols.

Dani AI

Generated

Quick clarification and practical notes for the thread: the Blue Gene family isn’t “Linux on every core” in the usual sense. To get predictable, low-noise performance IBM runs a very small, single-purpose Compute Node Kernel (CNK) on the compute nodes while Linux runs on the I/O/service and front-end nodes that handle file systems, login, and system management. That split is deliberate — it reduces OS jitter on tight numerical codes and pushes I/O and system services off the fast compute fabric. (research.ibm.com)

What that means for porting and tuning: treat compute ranks as single-process, minimal-OS workers. Rely on MPI and parallel I/O (MPI-IO, HDF5, staging data on front ends), avoid frequent syscalls or spawning helper processes from the compute job, keep per-process memory small, and use topology-aware rank mapping to minimize torus hops. Pre-stage libraries/data on front-end nodes or I/O nodes and use static builds when possible; use vendor tools and the Blue Gene performance toolchain to spot communication hotspots. These are the usual levers for getting good sustained performance on this class of machine. ()

On the OS/market side: ’s Top500 point is spot on — Linux-based stacks are the norm in HPC, while Windows HPC Server has only occupied niche pockets (for example, some enterprise/financial workloads). Linux’s dominance in the Top500 and similar lists explains why most HPC tooling and middleware target Linux-first. (osnews.com)

For and anyone thinking of buying hardware: full Blue Gene installations were custom, large, and sold to national labs and big research centers — procurement is handled through vendor channels and integrators. For business uses consider whether a custom supercomputer is needed or if a smaller commodity cluster or cloud-based HPC (pay-for-use) would give better ROI once you factor in space, power, cooling, and ops staff. Contact vendor sales/partners with a clear workload profile (memory per process, I/O patterns, concurrency needs) for a proper quote; use a lab deployment (e.g., Argonne’s Intrepid) as an example of the class of system and its operational scale. (anl.gov)

pty 882 Posting Pro

Linux has been doing in that arena for a long time.

lasher511 185 Veteran Poster

Lol yeah but its not like they have many other choices what are they gonna do run windows on it?

happygeek 2,411 Most Valuable Poster Team Colleague Featured Poster

Well the Microsoft Windows Compute Cluster Server seems to be doing OK, especially within the financial sector :)

Miacol 0 Newbie Poster

It must be about time for Linux to start overtaking Windows as the most used software - and not a minute too soon ;)

njagi 29 Newbie Poster

How much does the`Blue Gene/P and Blue Gene/L cost on the at the market price i need to by a few for my business.

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.