Cray Brings Windows 7 to HPC
If you thought Windows 7 was just for mere mortals, think again. Microsoft’s latest OS is about to show up in Cray’s newest CX1 deskside supercomputer that puts a Windows workstation and a Windows HPC Server cluster into a single box. Called the CX1-iWS (for integrated workstation), the machines are to be sold exclusively through Dell and will range in price from $39K to $55K.
The idea behind the iWS is to retain the interactive experience of a personal workstation, but extend its computational power to that of a small HPC cluster. It’s generally aimed at technical computing users who have simply run out of compute headroom on their two-socket machines, but are loathe to give up the intimacy of the workstation environment. A generic CX1 can be configured to provide the equivalent capabilities, but the iWS is preconfigured to deliver this experience right out of the box.
Since Windows 7 and Windows HPC Server 2008 form a natural client-server relationship, cluster administration and job management becomes relatively seamless. In addition, since there is disk storage shared between the workstation and the cluster, data management becomes much more straightforward. As long as your data set fits in 4 TB, no data transfers back and forth between client and server will be necessary.
“We’ve done a lot of work in preparation for this, talking to the ISVs, and trying to make sure that it isn’t some kind of dreamed up usage model, but is actually something that is relevant to customers,” explains Ian Miller, senior vice president for the Cray’s Productivity Solutions Group. According to him, the company has received a lot of positive feedback on the new iWS from a number of software vendors, including ANSYS, SIMULIA, and The MathWorks. Since most of the people using these software package are already on Windows desktops, it’s not surprising the ISVs are happy that Cray is extending the CX1 into the comfort zone of workstation power users.
Hardware-wise, the CX1-iWS is a four-node machine: one for the workstation that runs Windows 7; the other three are allocated to the compute cluster, which runs Windows HPC Server 2008. One of the three cluster nodes does double-duty as a storage node, and drives four 1TB SATA drives that are shared between the workstation and the cluster. Each node consists of a dual-socket, quad-core Intel Nehalem CPU-based blade, which gives the entire machine 32 cores — 8 for the workstation, 24 for the cluster.
An NVIDIA Quadro GPU is included in the workstation node to drive up to two monitors. The GPU can also be used as a CUDA accelerator for visualization or more general-purpose computing, but only on the workstation. The cluster nodes don’t have a GPU option.
In fact, the option set is fairly limited. Unlike the general CX1 offering, where you can mix and match a range of components, the iWS model comes in just three configurations, which vary by Quadro graphics card (on the workstation node), and CPU and memory (on the cluster nodes). Cray calls the three configurations “Good,” “Better,” and “Best,” which are spec’ed as follows:
- Good: Quadro FX 380; 3 x dual-socket 2.26GHz Xeon L5520 CPUs; 3 x 12GB memory (MSRP $38,999)
- Better: Quadro FX 4800; 3 x dual-socket 2.66GHz Xeon L5550; 3 x 24GB memory (MSRP $48,499)
- Best: Quadro FX 5800; 3 x dual-socket 2.93GHz Xeon X5570; 3 x 24GB memory (MSRP $54,999)
The invariant components of workstation node include a dual-socket 2.26GHz Xeon E5520 board with 24GB of memory, and 250GB of local disk. System memory is of the DDR3 variety, 1066MHz DIMMs for the workstation and low-end cluster nodes, and 1333MHz DIMMs for the mid-range and high end cluster nodes.
Like the rest of the CX1 family, the iWS can also be used as a common HPC resource for a workgroup (or even a small department), via an integrated 16-port Gigabit Ethernet switch. Five of the switch ports hook up the blades inside the CX1-iWS, leaving 11 ports to connect additional storage or workstations.
The fact that the system can be used as a workgroup hub means Dell will have the opportunity to sell its own Precision workstations, storage, and other computer paraphernalia around the CX1-iWS. From Cray’s point of view, Dell provides a more direct channel to the workstation customers they’re going after, although it’s not clear why Cray gave Dell the exclusive deal. The generic CX1 machines are currently available from more than 30 resellers.
So where is this all headed? With the introduction of the CX1-iWS, along with the CX1-LC (light configuration) announced in July, Cray is certainly making a concerted effort to guide the technical workstation crowd into HPC. To pursue the personal HPC model a little further, it seems like the next logical step would be to offer a Windows-based SMP variant of the CX1. In other words, get rid of the client-server separation altogether and just make a big, multi-socketed workstation.
Since the Windows 7 kernel apparently already has the necessary plumbing to support up to 256 “logical processors” (not CPUs, but logical threads of control) in a multi-socket NUMA architecture, it’s certainly plausible that, for example, a 4-socket Nehalem EX blade could be the basis of a 32-core Windows machine for personal HPC use. Alternatively, a virtual SMP, using ScaleMP or even 3Leaf Systems technology (if and when these vendors add Windows support) could be used to deliver the same effect using a couple of two-socket x86 nodes. At the very least, this eliminates cluster management, dual operating systems, and in the physical SMP implementation, the node interconnects and network switch. Plus it’s easy to name: CX1-SMP.
Feeds by Topic
- Developer Tools
Feeds by Industry
April 24, 2014
April 23, 2014
- Forschungszentrum Jülich Joins OpenPOWER Foundation
- Tower International to Utilize Altair’s Compute Manager
- Cray to Discuss 1st Quarter 2014 Financial Results on April 29
- NCAR and CU-Boulder to Join Intel Parallel Computing Centers Program
- Altera Brings Floating Point DSP Blocks to FPGAs
- Supermicro Announces 3rd Quarter 2014 Financial Results
April 22, 2014
April 21, 2014
April 17, 2014
- Registration Open for NCSA PSP Meeting
- Intel Selects Georgia Tech as Site for Next Parallel Computing Center
- AltaSim Technologies Receives DOE Grant for Additive Manufacturing
- LANL to Host Supercomputing Challenge Expo
April 16, 2014
- ORNL’s John Wagner Receives E.O. Lawrence Award
- SC14 Submissions for Panels Due April 25
- NEC Selects Chelsio Adapters for Vector Supercomputer
- SGI Announces Partnership with GIS Federal and NVIDIA
- Mellanox Collaborates with Dell
April 15, 2014
Most Read Features
- Cloud Bolsters HPC/HTC Student Research
- Inside Major League Baseball’s “Hypothesis Machine”
- Big Science, Tiny Microservers: IBM Research Pushes 64-Bit Possibilities
- NVIDIA Highlights GPU Progress on Titan Supercomputer
- Mont Blanc Forges Cluster from Smartphone Chips
- NVIDIA Steers Roadmap Around GPU Bottlenecks
- DOE Exascale Roadmap Highlights Big Data
- Details Emerging on Japan’s Future Exascale System
- The Top Supercomputing Led Discoveries of 2013
- Reality Check on Liquid Cooling in the Data Center
- More Features…
Most Read Short Takes
- Programming for Unreliable Hardware
- Museum to Resurrect First-Gen Supercomputer
- Big Data Reaches to the Stratosphere
- How NASA Is Meeting the Big Data Challenge
- Making HPC and Big Data More Accessible
- Research Advances on Key Quantum Computing Elements
- Russia, India Take Aim at China's FLOPS Lead
- Big Data Versus Big Compute
- Moore's Law Versus the Laws of Physics
- AMD Developer Summit Spotlights Heterogeneous Computing
- More Short Takes…
Most Read Off The Wire
- EGI Community Forum 2014 to Take Place in Finland
- Convey Computer Announces Wolverine Family of Coprocessors
- University of Illinois Wins MEMOCODE Design Contest Using Convey HC-1
- IEEE Launches 400GbE Task Force
- EGI Community Forum Agenda is Now Online
- Fujitsu Announces New Edition of ETERNUS DX200 S3 Storage System
- Ethernet Alliance to Participate at OFC 2014
- Cirrascale Announces Improved Power Solution for BladeRack 2 F-Series
- Supermicro Debuts New UltraTwin Server
- IIT Professor Establishes New Mathematical Model for Reducing Data Access Delay
- More Off The Wire…
- Read more…
- Read more…
- Read more…
- Read more…
- Read more…
- Read more…
- 3/12/14 | HPCwire, EnterpriseTech, ScaleMP | In the first of our new Spotlight ON: series, we’re taking a look beyond the public sector to find out how end-users in markets from Read more…
- 2/24/14 | Altair | Australia’s National Computational Infrastructure (NCI) chose PBS Professional to manage workload for its ‘Raijin’ cluster from Fujitsu, the Read more…
- Read more…
- Read more…