High Performance Computing Products
High Performance Computing (HPC) hardware is designed to handle complex computational tasks that would be impractical or impossible on traditional computers. IBM Power servers are built to accelerate scientific research, simulate complex systems, solve large-scale problems, support artificial Intelligence and machine learning, and drive innovation in various industries. Tachos delivers HPC systems based on IBM Power servers. All IBM Power servers share the following features:
- Industry-leading reliability
- Cutting-edge, zero-overhead security technologies
- 32 channels of DDR5 memory per socket
- Support for popular Linux distributions such as RHEL, Rocky Linux, and SLES
- Support for both IBM i and AIX operating systems
- Built-in matrix accelerators for accelerating AI/ML inferencing
- Optional SMT-4 and SMT-8 configurations (2 SMT-4 cores = 1 SMT-8 core)
- Low-latency networking via ConnectX6 network cards
IBM POWER S1022
The Power S1022 server is a slim 2U server with up to 80 SMT-4 cores running at up to 4.0 GHz and fed by memory running at 1.6 TB/s of bandwidth, depending on the configuration. When rack space is at a premium, the S1022 is the perfect choice.
IBM POWER S1024
The Power S1024 server is a slim 4U server with up to 96 SMT-4 cores running at up to 4.0 GHz, depending on the configuration. Like the 1022, the 1024 is capable of up to 1.6 TB/s of memory bandwidth. The larger chassis gives plenty of space to dissipate heat, enabling more cores and higher clock speeds than the S1022, making the S1024 an excellent choice for HPC clusters in larger datacenters.
IBM Power E1050 "Hot Rod"
Our special E1050 “hot rod” isn’t your standard enterprise server. This powerful machine is configured especially for HPC. 240 SMT-4 cores deliver up to 15 TFLOPs of SIMD-accelerated floating-point power, and up to 3.2 TB/s of memory bandwidth to ensure those hungry cores stay fed. This server is leaner than the standard enterprise product in order to be cost-effective for HPC applications
Why Choose IBM Power: The Quality Factor
Few computers are pushed to their limits the way HPC systems are. Whether you’re simulating the aerodynamics of a next-generation aircraft or molecular dynamics of drug antibodies, your servers will be running at maximum power draw for long stretches of time.
That kind of heat stresses components, and when a stressed component causes a breakage, that means lost time, delaying your time to market, and possibly even losing business. And no matter how generous your warranty is, your IT department has better things to do than track down and troubleshoot burnt-out server nodes.
IBM has decades of history delivering HPC systems to the most demanding users in government, aerospace, pharmaceuticals, and more, which is why IBM servers are first and foremost engineered for reliability. We measure downtime in seconds per decade, and that’s not an exaggeration. IBM Power’s reliability begins delivering benefits to your business when your hardware first arrives. No matter whether your order is large or small, you can rely on every IBM Power server working out of the box, ensuring that your cluster is commissioned right on schedule.
Once you’re up and running, the engineered cooling solutions in IBM Power servers maintain operation inside their thermal envelope, ensuring your batch jobs get done on schedule. After all, for large-scale HPC computations, if one server is unable to perform due to thermal throttling, the entire job is hamstrung. A server that stays cool is a server that delivers results when you expect them – just ask the satisfied users of the Department of Energy’s Summit supercomputer. From the time you stand up your HPC system, you can rely on your compute resources being ready at hand to get your work done.
All good things must come to an end, and eventually, it will be time to decommission your reliable workhorse. IBM’s industry-leading reliability will come through for you one last time. Since IBM servers last decades, we have a robust secondary market. We’re happy to take your old servers off your hands and resell them. There’s always a buyer who needs the robustness of an IBM Power server, and this means more money in your budget when it’s time to upgrade.
From the time your brand-new hardware arrives to the time you decommission, the industry-leading reliability of IBM Power will continually provide value to your business. When you choose IBM Power, you’re choosing a fast, flexible, reliable solution that will always be there when you need it.
Why Choose IBM Power: The Performance Factor
IBM Power has a long legacy in supercomputing, beginning in the 1990s with the RS/6000 series of servers. Over the years, one thing has remained consistent with IBM Power processors, and that is that they are designed for maximum data throughput and performance at the extreme edge of computing Power10 is the latest in a long line of high-performance computer products and the best offering yet.
Power10’s value for HPC begins with its novel memory architecture. Unlike other CPUs, which interface directly to standard memory DIMMs and thus have highly constrained bandwidth, Power10’s unique OMI memory fabric enables up 32 channels of DDR5 per socket, delivering twice the data throughput of similar CPU solutions.
The fully modular memory architecture means that when IBM releases new memory technology, you can upgrade your server for even more bandwidth. Our largest single-box server, the E1050, is capable of a total of 4.8 TB/s of memory bandwidth.
That memory bandwidth isn’t just a theoretical figure, either. It’s proven to be achievable with STREAM Triad benchmarks, making IBM Power10 the best CPU on the market for memory-bound applications.
All that memory bandwidth doesn’t do you much good without the compute power to make use of it. A Power10 server can host up to 960 hardware threads, each with its own 128-bit SIMD unit, clocked at up to 4.0 GHz.
And with up to 16 TB of memory in a server, this means a single Power10 server offers a combination of flexibility, performance, and capacity unlike any other computing platform on the market. That’s why across a wide variety of benchmarks, Power10 servers consistently show performance at the top of their class.
Modern HPC is inseparable from machine learning, which is why for AI/ML applications, each Power10 socket has up to 120 512-bit matrix engines. These engines accelerate everything from inferencing on the edge to matrix-heavy physics calculations due to their support of both low-precision ML-oriented operations and high-precision operations used in scientific computing.
With the high performance and reliability of IBM Power, you’ll get your results when you need them every time you need them.