SGI Colors New Shared Memory Machines Ultraviolet

By Michael Feldman

November 16, 2009

After what may be the longest development cycle ever for a supercomputer, SGI has unveiled the first commercial implementation of its Ultraviolet architecture. The company first announced “Project Ultraviolet” at SC03. Now six years later, it has launched Altix UV, the company’s first scale-up HPC system based on x86 technology. The Altix UV’s connection to the 2003 design is tenuous at best, but the new architecture does fulfill Ultraviolet’s original promise of delivering a shared memory architecture able to scale from a few sockets all the way up to a petascale supercomputer.

SGI Altix UV

Besides being simpler to program than distributed memory clusters, shared memory systems are especially well suited to I/O bound and memory-bound applications; codes that depend upon a lot of inter-processor communication; and any type of application that uses large — as in terabyte-sized — in-memory databases. These shared memory systems can also be used in conjunction with large clusters to provide an “analysis supernode.”

The two initial products, the Altix UV 1000 and Altix UV 100, are both based on Intel Nehalem-equipped blades, which are hooked together with SGI’s 5th generation NUMAlink fabric. The software stack includes everything from the OS on up, including the SGI Foundation Software, data management packages (XFS, CXFS, DMF), SGI’s ProPack and System Management tools, job schedulers (Altair PBSP and Moab) and developer tools and libraries. The machines come with either SUSE Linux Enterprise Server or Red Hat Enterprise Linux.

The blades themselves contain two eight-core Nehalem EX chips, each with a bank of four DDR3 memory channels. If a larger memory to core ratio is desired, there are 6- and 4-core options, as well as a single-socket configuration. An optional I/O riser allows for a choice of expansion slots or external I/O ports. Up to two PCIe slots are available on each blade and these can be used to plug in external storage (SGI or otherwise) or GPGPUs.

SGI’s secret sauce is the UV hub, which sits on each blade and acts as the node controller. The hub, along with the NUMAlink 5 interconnect, is the technology that makes the supersized shared memory possible. The new interconnect delivers sub-microsecond latencies and 15.0 GB/sec of aggregate bandwidth per blade. The hub itself manages data traffic between the local CPU resources and the rest of the system, arbitrating between the local QuickPath Interconnect (QPI) links and the NUMAlink fabric.

According to Jill Matzke, Altix product manager, the SGI engineers decided to limit themselves to two sockets per blade in order to avoid overtaxing the QPI bandwidth, which needs to feed the NUMAlink fabric and I/O. Since Nehalem EX is designed to support up to 8 sockets per board, one might wonder why SGI didn’t opt for the dual-socket-capable Nehalem EP chips. Apparently, EX was chosen because it offered more QPI and memory bandwidth, both of which were essential to the UV design. In any case, the Nehalem EP design does not lend itself to external node controllers, such as the UV hub.

The Altix UV 100 is aimed at the mid-range market, scaling from a single 3U rackmount unit containing two dual-socket blades, up to a 7 teraflop, 96-socket machine that fits into a couple of racks. The upper limit on memory capacity on this product is 6 TB. The UV 100 is aimed at users who need a moderate to large SMP environment for their x86 applications. At the maximum 96-socket configuration, 768 cores are available, which doubles to 1,536 threads thanks to Nehalem-style multithreading support.

The Altix UV 1000 is a cabinet solution that scales all the way to the top, that is, 256 sockets (yielding 2,048 cores or 4,096 threads) and 16 TB of memory. At the max configuration, this model delivers 18.6 peak teraflops in a 42U space. The 16 TB limit on the UV 1000 corresponds to the maximum memory reach of the Intel Nehalem processor. However, the UV 1000 design can actually scale beyond this limit by connecting multiple 256-socket systems in a 2-D torus topology. In this case, the system would be partitioned with multiple OS images but support a much larger shared global address space — up into petabytes. The upper limit supported by the UV hub is 32,768 sockets, which would equate to about 2 petaflops. SGI is certainly willing to help interested parties develop such systems, but the vast majority of customers will be able to fit their applications within the 256-socket, single system image machine.

Note the current Itanium-based Altix 4700 reaches to 128 GB because that CPU’s memory address is wider, although core count on those systems tops out at 1024. That said, just getting a handful of terabytes of global memory on an x86 platform is likely to be a big attraction for HPC users. “We are seeing people ordering many more terabytes of memory on UV than they ever did on Altix with Itanium, simply because of the overall capability and the price-performance,” says Matzke.

Although UV supports highly-scaled applications in a global memory model, today the majority of global memory applications scale to just 32 or maybe 64 threads. However, UV, like most shared memory machines, can also deliver great performance for MPI applications by properly exploiting the unified memory and the speed of the interconnect fabric. Moreover, an MPI offload engine has been incorporated into the UV hub to further accelerate this class of applications. SGI has demonstrated a 3X improvement in the HPCC GUPS benchmark with the offload engine enabled. According to Matzke, “70 percent of the people that buy these systems are running MPI, but have other application demands that make it really shine on this kind of an architecture.”

According to Geoffrey Noer, SGI’s senior director of product marketing, the company is currently taking orders for the new UV machines, with the first shipments expected by second quarter of 2010 (following Intel’s release of the Nehalem EX CPUs). Initial customers include the University of Tennessee (1024 cores, 4 TB memory), the North German Supercomputing Alliance, known as HLRN (two systems, 4,352 cores, 18 TB memory), CALMIP in France (128 cores, 1 TB memory), and the University of Hokkaido (180 cores, 360 GB memory). A number of UV systems have also been purchased by the federal government for certain “defense applications” (which shall remain nameless). SGI is not making UV pricing public, but potential buyers can always obtain a quote under NDA.

Although many customers using Itanium Altix systems will undoubtedly transition to the x86 UV platform, Noer says SGI will continue to offer the Altix 450 and 4700 systems. And even though they are not publicly divulging specific plans for future Itanium-based shared memory machines, Noer did have this to offer: “It’s important not to look at Altix UV as a direct replacement for the 4700…. We are working with Intel on next-generation processor technologies as well.  For those customers that are getting the benefits out of the larger address space and benefits with the 4700, they absolutely don’t need to switch to Altix UV.”

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

ARM, Fujitsu Targeting Open-source Software for Power Efficiency in 2-nm Chip

July 19, 2024

Fujitsu and ARM are relying on open-source software to bring power efficiency to an air-cooled supercomputing chip that will ship in 2027. Monaka chip, which will be made using the 2-nanometer process, is based on the Read more…

SCALEing the CUDA Castle

July 18, 2024

In a previous article, HPCwire has reported on a way in which AMD can get across the CUDA moat that protects the Nvidia CUDA castle (at least for PyTorch AI projects.). Other tools have joined the CUDA castle siege. AMD Read more…

Quantum Watchers – Terrific Interview with Caltech’s John Preskill by CERN

July 17, 2024

In case you missed it, there's a fascinating interview with John Preskill, the prominent Caltech physicist and pioneering quantum computing researcher that was recently posted by CERN’s department of experimental physi Read more…

Aurora AI-Driven Atmosphere Model is 5,000x Faster Than Traditional Systems

July 16, 2024

While the onset of human-driven climate change brings with it many horrors, the increase in the frequency and strength of storms poses an enormous threat to communities across the globe. As climate change is warming ocea Read more…

Researchers Say Memory Bandwidth and NVLink Speeds in Hopper Not So Simple

July 15, 2024

Researchers measured the real-world bandwidth of Nvidia's Grace Hopper superchip, with the chip-to-chip interconnect results falling well short of theoretical claims. A paper published on July 10 by researchers in the U. Read more…

Belt-Tightening in Store for Most Federal FY25 Science Budets

July 15, 2024

If it’s summer, it’s federal budgeting time, not to mention an election year as well. There’s an excellent summary of the curent state of FY25 efforts reported in AIP’s policy FYI: Science Policy News. Belt-tight Read more…

SCALEing the CUDA Castle

July 18, 2024

In a previous article, HPCwire has reported on a way in which AMD can get across the CUDA moat that protects the Nvidia CUDA castle (at least for PyTorch AI pro Read more…

Aurora AI-Driven Atmosphere Model is 5,000x Faster Than Traditional Systems

July 16, 2024

While the onset of human-driven climate change brings with it many horrors, the increase in the frequency and strength of storms poses an enormous threat to com Read more…

Shutterstock 1886124835

Researchers Say Memory Bandwidth and NVLink Speeds in Hopper Not So Simple

July 15, 2024

Researchers measured the real-world bandwidth of Nvidia's Grace Hopper superchip, with the chip-to-chip interconnect results falling well short of theoretical c Read more…

Shutterstock 2203611339

NSF Issues Next Solicitation and More Detail on National Quantum Virtual Laboratory

July 10, 2024

After percolating for roughly a year, NSF has issued the next solicitation for the National Quantum Virtual Lab program — this one focused on design and imple Read more…

NCSA’s SEAS Team Keeps APACE of AlphaFold2

July 9, 2024

High-performance computing (HPC) can often be challenging for researchers to use because it requires expertise in working with large datasets, scaling the softw Read more…

Anders Jensen on Europe’s Plan for AI-optimized Supercomputers, Welcoming the UK, and More

July 8, 2024

The recent ISC24 conference in Hamburg showcased LUMI and other leadership-class supercomputers co-funded by the EuroHPC Joint Undertaking (JU), including three Read more…

Generative AI to Account for 1.5% of World’s Power Consumption by 2029

July 8, 2024

Generative AI will take on a larger chunk of the world's power consumption to keep up with the hefty hardware requirements to run applications. "AI chips repres Read more…

US Senators Propose $32 Billion in Annual AI Spending, but Critics Remain Unconvinced

July 5, 2024

Senate leader, Chuck Schumer, and three colleagues want the US government to spend at least $32 billion annually by 2026 for non-defense related AI systems.  T Read more…

Atos Outlines Plans to Get Acquired, and a Path Forward

May 21, 2024

Atos – via its subsidiary Eviden – is the second major supercomputer maker outside of HPE, while others have largely dropped out. The lack of integrators and Atos' financial turmoil have the HPC market worried. If Atos goes under, HPE will be the only major option for building large-scale systems. Read more…

Everyone Except Nvidia Forms Ultra Accelerator Link (UALink) Consortium

May 30, 2024

Consider the GPU. An island of SIMD greatness that makes light work of matrix math. Originally designed to rapidly paint dots on a computer monitor, it was then Read more…

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock_1687123447

Nvidia Economics: Make $5-$7 for Every $1 Spent on GPUs

June 30, 2024

Nvidia is saying that companies could make $5 to $7 for every $1 invested in GPUs over a four-year period. Customers are investing billions in new Nvidia hardwa Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

AMD Clears Up Messy GPU Roadmap, Upgrades Chips Annually

June 3, 2024

In the world of AI, there's a desperate search for an alternative to Nvidia's GPUs, and AMD is stepping up to the plate. AMD detailed its updated GPU roadmap, w Read more…

Some Reasons Why Aurora Didn’t Take First Place in the Top500 List

May 15, 2024

The makers of the Aurora supercomputer, which is housed at the Argonne National Laboratory, gave some reasons why the system didn't make the top spot on the Top Read more…

Intel’s Next-gen Falcon Shores Coming Out in Late 2025 

April 30, 2024

It's a long wait for customers hanging on for Intel's next-generation GPU, Falcon Shores, which will be released in late 2025.  "Then we have a rich, a very Read more…

Leading Solution Providers

Contributors

Google Announces Sixth-generation AI Chip, a TPU Called Trillium

May 17, 2024

On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.  The chip, essentially a TPU v6, is the company's l Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

IonQ Plots Path to Commercial (Quantum) Advantage

July 2, 2024

IonQ, the trapped ion quantum computing specialist, delivered a progress report last week firming up 2024/25 product goals and reviewing its technology roadmap. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

The NASA Black Hole Plunge

May 7, 2024

We have all thought about it. No one has done it, but now, thanks to HPC, we see what it looks like. Hold on to your feet because NASA has released videos of wh Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire