Build and Maintain Your Graphics System One Module at a Time
A new snap-together approach to system architecture allows you to configure--and reconfigure--systems that match your requirements precisely. Just choose the modules that your applications require:

InfinitePerformance™ Graphics | InfiniteReality4™ Graphics
Compute | NUMAlink | Memory Expansion | PCI-X Expansion | PCI Expansion

Graphics Module for InfinitePerformance™ Graphics
The SGI® Onyx® 350 compute graphics module packs the power of two or four MIPS® R16000™ processors, the high-performance SGI® IRIX® operating system, up to 8GB of memory, and one integrated InfinitePerformance™ graphics pipe into an ultradense, 3.5-inch (2U), rackmountable chassis. Each compute graphics module supports:

• Two or four 600 or 700 MHz MIPS® CPUs with 4MB caches
• Up to 8GB RAM
• One InfinitePerformance graphics pipeline via XIO™ connection
• 2.4GB/sec peak I/O bandwidth

• Scalability up to 141 million polygons per second and 3.8 billion pixels per second per display channel
• Precise color (12-bit per component)
• Hardware-assisted image processing
• High-definition output with real-time colorspace conversion for film and video applications (optional)

• Manage 50, 500, or 5000GB data sets
• Visualize huge polygonal models interactively

• Scalable graphics pipelines
• Multimode usage
• Support of a wide range of virtual reality devices

Graphics Module for InfiniteReality4™ Graphics
Each graphics module can hold one 2RM (Raster Manager) and one 4RM InfiniteReality® family graphics pipe. SGI® Onyx® 350 InfiniteReality4™ systems can support a total of up to eight fully configured graphics pipes, each with a Geometry Engine® board, four Raster Managers, and a display generator with an output option.

Incredible visual realism:

• Interactive photo-realistic surfaces
• Interactive volume visualization
• 8 subsample full scene anti-aliasing
• Precise color information
• Hardware-assisted image processing
• High-definition output with real-time colorspace conversion for film and video applications (optional)

• Unlimited fly-over of geospecific terrain
• Interaction with huge medical and geophysical data

• Scalable graphics pipelines
• Multimode usage
• Support of a wide range of virtual reality devices

Compute Module
The SGI Onyx 350 compute module packs the power of two or four MIPS R16000 processors, the high-performance SGI IRIX operating system, and up to 8GB of memory into an ultradense, 3.5-inch (2U), rackmountable chassis. Each compute module supports:

• Two or four 600 or 700 MHz MIPS CPUs with 4MB caches
• Up to 8GB RAM
• 2.4GB/sec peak I/O bandwidth

NUMAlink® Module
The NUMAlink module allows SGI Onyx 350 with InfinitePerformance graphics to combine up to eight compute graphics modules, compute modules, memory expansion modules, or PCI-X expansion modules.

• Includes eight NUMAlink ports in 3.5-inch (2U) enclosure
• Allows scalability up to 32 CPUs and 64GB memory
• Allows scalability up to eight InfinitePerformance graphics pipes with 16 or 32 CPUs
• Allows scalability up to 60 PCI slots on 30 PCI buses with eight or 16 CPUs

Memory Expansion Module
The memory expansion module provides up to 8GB of additional RAM plus four additional PCI-X slots over two separate buses without needing to add CPUs.

• Eight memory slots for up to 8GB of additional RAM
• Additional four 64-bit/100 MHz PCI-X slots
• Two buses, two slots/bus, 3.5-inch (2U) enclosure
• 420MB/sec sustained, 512MB/sec peak bandwidth per bus

PCI-X Expansion Module
The PCI-X module provides four additional PCI-X slots over two separate buses.

• Additional four 64-bit/100 MHz PCI-X slots
• Two buses, two slots/bus, 3.5-inch (2U) enclosure
• 420MB/sec sustained, 512MB/sec peak bandwidth per bus

PCI Expansion Module
The PCI expansion module provides 12 additional PCI slots over the four standard PCI slots in each compute module.

• Additional 12 hot-pluggable 64-bit/66 MHz PCI slots
• Six buses, two slots/bus, 7-inch (4U) enclosure
• 400MB/sec sustained, 512MB/sec peak bandwidth per bus
• 2.4GB/sec aggregate sustained throughput