NVIDIA® Tesla™ GPU computing products are designed for high-performance computing at the workstation and in the data center. There are a number of CUDA and OpenCL software features that are designed specifically for GPGPU and are exclusively supported only on Tesla products. These are summarized in the table below.
Software Application |
Description |
Support Matrix |
Download |
Performance Driver for Windows: TCC driver |
| • |
Tesla driver for Windows |
| • |
Enables remote desktop, Windows services |
| • |
Reduces kernel launch overhead |
| • |
Works for CUDA C/C++ |
|
| • |
C-series GPUs: Windows Vista, Windows 7 |
| • |
M-series and S-series products: Windows HPC Server 2008 and 2008 R2 |
|
|
GPU Monitoring: nvsmi |
| • |
GPU temperature |
| • |
Fan speed |
| • |
ECC information |
|
| • |
M-series and S-series products |
| • |
Windows and Linux |
|
|
GPU Cluster Management |
| • |
Exclusive mode: Enables application-exclusive access to a particular GPU |
| • |
GPU visible devices: Environment variable to enable cluster management software to limit the Tesla GPUs an application can use |
|
| • |
M-series and S-series products |
| • |
Windows and Linux |
|
|
NVIDIA GPUDirect™ |
| • |
Tesla and InfiniBand driver that enables faster communication between GPU and InfiniBand |
|
| • |
M-series and S-series products |
| • |
Redhat Enterprise Linux (RHEL) 5.4 only |
|
Download
| • |
Tesla driver |
| • |
RHEL kernel patch |
| • |
InfiniBand OFED drivers |
| • |
Instructions for installation |
|
TCC DRIVER FOR WINDOWS
The TCC (Tesla Computer Cluster) driver is a Windows driver for CUDA C/C++ that enables remote desktop, services, and reduces the CUDA kernel launch overhead on Windows. Note that the TCC driver disables graphics on the Tesla products.
GPU MONITORING
The GPU monitoring software for Tesla is available using the nvsmi tool. This tool currently gives the GPU temperature, the fan speed, and also the ECC information. nvsmi will continue to evolve as we add more GPU monitoring features.
GPU CLUSTER MANAGEMENT
NVIDIA works with several cluster management software providers, who already support GPU-based systems:
| – |
Bright Computing |
| – |
ClusterCorp Rocks |
| – |
Platform Computing |
In addition to this, two important features supported in the CUDA drivers are
| – |
Exclusive mode: Enables an application to get exclusive access to a GPU |
| – |
GPU Visible Devices: Enables the cluster management software to manage GPU resources by controlling which GPUs an application can use |
NVIDIA GPUDIRECT™
NVIDIA GPUDirect™ is a new technology from NVIDIA that enables faster communication between the GPU and other devices on the PCIe bus by removing the overhead due to the CPU. The first device NVIDIA has focused on is InfiniBand, where NVIDIA worked with Mellanox to make communication between the GPU and InfiniBand faster with fewer memory copies in the main memory.
 |
The memory copy from buffer 1 to 2 is eliminated using NVIDIA GPUDirect |
