With the world’s first teraflop many-core processor, the NVIDIA® Tesla™ S1070 computing system speeds the transition to energy-efficient parallel computing. With 960 processor cores and a standard C compiler that simplifies application development, Tesla S1070 scales to solve the world’s most important computing challenges—more quickly and accurately.

Technical Specifications
Product	Tesla S1070
# of Tesla Processors # of Computing Cores Floating Point Precision Total Dedicated Memory Memory Interface Memory Bandwidth Typical Power Consumption System Interface Programming environment	4 960 (240 per processor) IEEE 754 single & double 16 GB (organized as 4.0 GB per GPU) 4x 512-bit GDDR3 memory interface (organized as a 512-bit interface per GPU) 408 GB/sec (102 GB/s per GPU to local memory) 700 W PCIe x16 or x8 CUDA
PNY Part Numbers for Tesla S1070	Description
TCSS1070-BU	Tesla S1070 (Base Unit)*
SHAX16	Host Bus Adapter x16 PCIE
SHAX8	Host Bus Adapter x8 PCIE
S-1/2M-CBL	0.5M Interface Cable
S-2M-CBL	2.0M Interface Cable
*Tesla S1070 Base Unit does not include Host Bus Adapters or interface cables, they must be ordered separately.
Features and Benefits
Four 1-Teraflop Processors in a High Density 1U System	Delivers up to 4 teraflops of performance in a 1U rack-mount system for unmatched performance in high density rack systems
Massively-Parallel Many-Core Architecture	240 computing cores per processor that can execute thousands of concurrent threads.
Scales to Multi-GPU Computing	Scale to thousands of processor cores to solve large-scale problems by splitting the problem across multiple GPUs.
Program in NVIDIA CUDA™: C for GPU	Programmable using CUDA, the world's leading application development platform for many core solutions.
IEEE 754 Floating Point	Ensures your results meet industry standard precision including optional features to ensure accuracy.
Double-Precision Floating Point Support	Meets the precision requirements of your most demanding applications with IEEE 64-bit precision.
Asynchronous Data Transfer	Turbocharges system performance because data transfers can be executed even while the computing cores are busy.
16 GB Ultra-fast Memory	Enables larger datasets to be stored locally with 4 GB dedicated for each processor to maximize performance and minimize data movement around the system.
4x 512-bit Memory Interface	Delivers 408 GB/s peak memory bandwidth for blistering data transfer as a 512-bit interface dedicated to each processor
High-Speed, PCI-Express 2.0 Data Transfer	With low latency and high bandwidth, computing applications benefit from the highest data transfer rate possible through standard PCI-Express architecture
Single-screw Rail Mounting	Single-screw rail design is quick to install like a tool-less design, but with the extra security and rigidity from a single screw to secure the rail to the rack.
System Monitoring Features	Easy management and monitoring post-installation helps your IT staff manage systems with minimal effort. Remote capabilities and status lights on the front and rear of the unit ensure your staff can see the status whether they are on the other side of the rack, or the other side of the world.
Dual PCI-Express 2.0 Cable Connections	Maximizes bandwidth between the host processor and the Tesla processors with up to 12.8 GB/s transfer rates (up to 6.4 GB/s per PCI Express connection)
Small-form-factor (SFF) host adapter card	The low power host adapter card enables Tesla systems to work with virtually any PCIe compliant host system with an open PCI Express slot (x8 or x16).
Tesla S1070 computing system has three software components (Windows support scheduled for Q4 2008): NVIDIA CUDA driver (required to run and develop CUDA applications) Linux (32-bit and 64-bit) CUDA Toolkit (required to run or develop CUDA applications) Linux (32-bit and 64-bit) CUDA SDK (contains code samples and utilities) Linux (32-bit and 64-bit)