SGI Prepares to Reboot; Intel Beams About Its Laser

By Michael Feldman

September 22, 2006

Lots of interesting news for the HPC crowd this week. SGI arranged for its return from bankruptcy; Intel made a splash with a breakthrough in silicon photonics; and two vendors introduced a couple of unique products. I got a chance to speak with a number of these companies about their news this week and they all had a good story to tell. I'll try to keep my kibitzing to a minimum and just give you the facts.

SGI, Take Two

After declaring bankruptcy last spring, Silicon Graphics is now just a few weeks away from joining the land of the solvent. On Tuesday, a U.S. bankruptcy court approved its Plan of Reorganization, setting the stage for the reemergence of the company sometime in early October. When the company restarts they'll have a new management team in place, headed by CEO Dennis McKenna, and a new board of directors.

With the backing of the new investors and some additional loans announced this week, the company plans to hit the ground running. “As we look at this emergence process, we will have completed, in essence, our financial re-engineering,” said McKenna. “We will have eliminated our legacy debt and our overall cap[italization] position will have strengthened significantly.”

“We expect to be profitable on an operation basis during our fiscal year 07,” he added.

No doubt some of the SGI's customers were shaken by the company's descent into insolvency, but apparently no one panicked. “At first there was confusion at what all this meant and what risks were there,” admitted McKenna. “The good news is that we haven't lost any customers through this process and they've been extremely supportive.”

The company's strategy will be changing from an engineering-centric organization that produces products to a market-focused organization that produces solutions. In practical terms, this means that they'll be using their various hardware platforms to carry “application bundles” to their target markets. This is not so different from Linux Networx' strategy to provide more application-focused solutions.

But unlike Linux Networx, which will remain focused in the high performance technical computing market, SGI also intends to expand into the much larger enterprise business management market, where it hopes to leverage its expertise in high performance systems and large memory footprint architectures. New target applications will include real-time data warehousing, data mining, large-scale database management and digital media archiving/content management. SGI estimates the enterprise business market is around $60 billion, which dwarfs the company's traditional markets in defense and intelligence ($3.7 billion) science application ($7.1 billion) and engineering applications ($13.7 billion). In that sense, SGI's new approach seems closer to that of IBM, which fuels its HPC business from its enterprise computing base.

The high-end Altix blade servers, which were traditionally used for HPC, will now be targeted to Oracle and SAP database applications. McKenna says they've achieved 10X performance improvements with their newest Itanium(Montecito)-based platforms on these types of workloads. For time-sensitive results, performance improvements of this magnitude can create a unique opportunity for SGI.

McKenna also notes that the company now offers a “more meaningful product portfolio” than just the high-end platforms of the past. SGI will be selling servers at all price points: high-end (Altix 4700 blade, Itanium-based), mid-range (Altix 450 blade, Itanium-based) and value (Altix XE cluster, Xeon-based). In particular the new Woodcrest-based Altix XE system represents SGI's entry into the commodity cluster space — a market which represents about half of all server sales. SGI intends to deliver optimized cluster systems with software tuned for particular application areas. SGI is also offering a wider range of storage hardware (NAS and SAN).

According to McKenna, their expanded offerings will enable the company to compete for 80 percent of their customers' IT budgets rather than just 20 percent. “In fact, today we have the most complete product portfolio in the company's history. And we've got a roadmap that further strengthens this portfolio to the strategic direction of the company.”

I can't resist a little kibitzing here. For the time being at least, SGI is going to remain an Intel-only shop — Itaniums or Xeons — while almost every other server/cluster vendor offers AMD Opteron-based systems somewhere in their product lineup. Will an Intel-only approach work? Maybe they should ask Michael Dell.

—–

Intel Goes Photonic

On Monday, Intel and the University of California, Santa Barbara (UCSB) announced that they have developed the first electrically pumped hybrid silicon laser, characterizing this as one of the last major hurdles in silicon photonics. By combining the light emitting properties of indium phosphide with high volume, low-cost features of silicon, they have constructed a laser that can be integrated onto a silicon substrate and drive chip-level optical communications.

Because of the current expense associated with with converting photons to electrons, and vice versa, optical communications are only cost-effective at distances greater than the scale of a computer system. The goal of the Intel/UCSB work is to “siliconize” photonics so that optical communications can achieve a cost basis that will allow it to be applied at the level of chips, boards and racks.

According to Mario Paniccia, director of the Photonics Technology Lab at Intel, the commercialization of this technology will give system designers the ability to converge communication and computing all on one platform. Integrating a CMOS silicon-based laser will enable the production of hundreds of lasers on a single chip. Intel and UCSB have currently demonstrated 10 Gb/s performance for a single laser. They are confident that they'll reach 40 Gb/s in the not-too-distant future. With 25 lasers at 40 Gb/s, that's a terabit of data per second. And that can be multiplexed into a single fiber.

“What this allows us to do is not only drive new ways of high performance computing (and/or computing in general) but also the ability to re-architect the systems,” said Paniccia. He noted that with terabit optical links, system designers would be free to move system components, like memory, further away from the CPU, enabling much more architectural freedom.

The Intel/UCSB team emphasized that this is a research breakthrough, not a commercial demonstration. A lot of work remains to bring it to commercial production. Paniccia estimated that this technology should be ready for prime time at some point in the early part of the next decade.

Intel isn't the only company working to siliconize optical communications. Luxtera, a fabless semiconductor startup company in Carlsbad California, has developed CMOS photonic transceivers. Its roadmap for bringing photonics to the chip is somewhat different than Intel's.

“Intel has taken the approach that this stuff won't work until the lasers are integrated into the silicon,” said Alex Dickinson co-founder, president and CEO of Luxtera. “We think that's a fabulous long-term goal and we applaud any efforts in that area. The challenge in that area is manufacturability.”

Dickinson believes integrating the indium phosphide lasers onto the chip may not be the best approach since this introduces a non-standard semiconductor manufacturing process into the mix. Luxtera's philosophy is to view the laser as you would a DC power supply: a source of energy, in this case photons. So they've mounted external lasers onto the silicon to drive their CMOS optical components. Luxtera has demonstrated their photonic solution, which is now being evaluated by Sun and other system vendors. They are looking at commercial deployments within the year.

But Dickinson admits the company is hedging its bets. Luxtera is currently involved in academic partnerships, with funding from DARPA's Electronic Photonic Integrated Chip (EPIC) program, in an effort to develop their own version of on-chip lasers. It's possible they will have an integrated silicon-based laser in the same timeframe as Intel's solution.

—–

Stretching Those Cluster Interconnects

Optical interconnects may be the future, but don't throw out those copper cables just yet. On Monday, Quellan Inc. unveiled analog semiconductor devices that can be incorporated into copper interconnects, significantly enhancing their reach. The devices increase signal strength and reduce noise, providing more throughput and reach. And since the devices are very low power, they use much less energy than a standard 10 Gb/s interconnect, either copper or optical. Quellan calls this their Q:Active cable technology.

“There've been a lot of breakthroughs in the data center — going from mainframes to blade servers, proprietary operating systems to open Linux, Intel to AMD, air cooling to water cooling, AC power distribution to DC power distribution,” said Tony Stelliga, Chairman and CEO of Quellan. “But the cable or the interconnect has never really had a breakthrough improvement. It still burns hundreds of kilowatts, is still limited in reach, and it still weighs two to three tons. It really is time for a breakthrough in that space.”

As data centers ramp up to 10 GbE and double data rate InfiniBand, the capital expense of the infrastructure and the operating expenses ramp up as well. A reach of ten meters using copper interconnects at single data rates (2.5 Gb/s) is reduced to five meters at double data rates (5 Gb/s).

“Going to fiber to reach 20 meters is just a waste of time, money and power, said Stelliga. “By embedding our active silicon, we can get that 10 meter radius to a 20 or 25 meter radius. As you can imagine, if you can double the reach of the cable you can get four times the compute density in that circle.”

According to Stelliga, if you don't need to reach these longer distances you can use their devices to drop down to smaller cable sizes — 30 gauge, which is the same diameter as CAT5 cable — reducing weight and cost and improving the bend ratio. The smaller diameter cabling also allows better data center airflow for cooling. When 10 GbE over twisted pair becomes a ratified standard within the next couple of years, it will run on CAT7. The Q:Active technology offers a 10 GbE solution today that is much thinner than CAT7 and uses a fraction of the power.

The Q:Active technology also claims lower latency and lower power than either 10BaseT or optical. The Quellan devices avoid both the extra power and latency involved for the 10BaseT solution and the electrical to optical conversion that takes place with fiber.

—–

PeakStream Goes With the Flow

A week after my commentary about the challenges of heterogeneous computing, a startup company named PeakStream unveiled its new software platform that targets simple hetero systems based on commodity multi-core and accelerator processors. Their “PeakStream Platform” is designed to provide programmers with a software abstraction for x86 multi-core, GPU and Cell architectures.

Like everyone else, PeakStream has noticed that both Intel and AMD have been busy adding more cores to their processors. This year they'll have quad-core processors, and octa-cores within two years. And by 2010, the chip makers are expected to have processors with 16 or perhaps even 32 cores. At the same time, HPC users are being tempted by the exceptional price/performance in commodity accelerator processors like GPUs and the Cell processor. These devices are low-cost and ship roughly in the same kinds of volumes as commodity general-purpose processors. But the floating point performance of these accelerators tend to be an order of magnitude greater than a general-purpose processor.

“The main issue is that nobody knows how to take advantage of this parallelism very easily — either on the main chip or the co-processor,” says Michael Mullany, VP of Marketing at PeakStream. “The second thing is that co-processors are making a very big comeback. There's been a lot of work on using co-processors in high performance computing in the last couple of years. We think that the commodity co-processors that are out there, particularly GPUs and the Cell processor design are particularly interesting.”

In fact, software platforms for multi-core processors are already beginning to go mainstream — from industry-standard solutions like OpenMP to a variety of vendor solutions. What PeakStream brings to the table is support for commodity accelerators, alongside of multi-core x86 processor support.

Actually, the first version of the PeakStream Platform will support only AMD/Intel x86 multi-core processors and ATI R580 based boards — Radeon X1900 XTX and variants. The company plans to add support for the Cell BE processor in a future release and expand GPU coverage over time. But PeakStream is clearly focused on commodity processor hardware. Things like the ClearSpeed co-processor are considered too low-volume for them at this point.

PeakStream's software abstraction is centered on (surprise) a stream programming model. It requires that the programmer use array objects and a parallel API within a conventional C/C++ environment. According to PeakStream, the API is modeled on standard HPC interface conventions, minimizing the learning curve for developers. A virtual machine is used to perform memory management and map the code onto the available parallel hardware. The model was designed to insulate the developer from hardware concerns. And the use of the virtual machine allows the resulting binaries to remain aloof from processor dependencies.

The company will mainly be targeting customers in defense (big on COTS technology), financial services, oil & gas and academia. The company believes their platform can be used to develop new HPC software as well for porting moderate-sized legacy codes (less than 100,000 SLOC) to the next generation of hardware.

“Pretty much every five to ten years, most of the code base gets rewritten to support a new processor,” said Mullany. “This whole market has gone from Crays in the early 80s to propriety array processors in the early 90, then has moved to RISC machines in the mid-90s and to clusters in the early 2000s. This is market that is used to moving code around.”

—–

Rocky Mountain High Performance Computing

On September 18th through the 20th, government and industry leaders converged in Denver to attend the HPC User Forum. In this week's issue, we feature interviews with three of the speakers at the meeting: Horst Simon and David Probst, who offer two different perspectives on petascale computing, and Richard Walsh who discusses how the next generation of processors are being applied to high performance computing. The interviews were conducted by contributing editor Steve Conway as the presenters gathered their thoughts for the September meeting. See our special HPC User Forum section, “Denver Dialogue,” in this week's edition of HPCwire.

—–

As always, comments about HPCwire are welcomed and encouraged. Write to me, Michael Feldman, at [email protected].

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!

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 XL — were added to the benchmark suite as MLPerf continues 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 power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device 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 the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa 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…

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…

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…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. 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…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. 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…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

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 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire