NVIDIA® has released the Tesla C2050® GPU computing processor. The Tesla C2050® is based on the next generation CUDA architecture, codenamed Fermi, that adds several unique features to their 10-series GPU offerings. Improvements include the addition of ECC memory for enhanced results accuracy and scalability, and up to a 7x increase in double precision floating point performance over their 10 series GPUs.
- Form Factor / Interface: PCIe x16 Gen2
- CUDA Cores: 448
- Core Frequency: 1.15 GHz
- Peak Double Precision Floating Point Performance: 515 Gflops
- Peak Single Precision Floating Point Performance: 1.03 Tflops
- Dedicated Memory: 3 GB GDDR5 (ECC on results in 12.5% reduction in usable memory, or 2.625 GB)
- Memory Speed: 1.5 GHz
- Memory Interface: 384 bit
- Memory Bandwidth: 144 GB/s
- Power Consumption: 247W TDP
The addition of ECC memory is a big improvement for high performance computing. When ECC is turned on, the Tesla C2050 protects the data in memory and enhances data integrity and application reliability. Register files, L1 and L2 cache, shared memory, and DRAM are all protected when ECC is enabled. The system can also be ran without ECC protection if that is desired.
Additional improvements include the NVIDIA Parallel Datacache™, which can accelerate algorithms such as physics solvers, ray-tracing, and sparse matrix multiplication where data addresses are not know beforehand. This includes a configurable L1 cache per Streaming Multiprocessor block and a unified L2 cache for all of the processor cores.
The NVIDIA Gigathread™ engine is another significant improvement to the NVIDIA architecture, speeding up context switching 10 times compared to previous architectures, and allowing concurrent kernel execution and improved thread block scheduling. Scheduled for release in Q3 2010, the Tesla C2070 GPU Computing Processor will increase the on-board memory per GPU to 6 GB.
Aspen
Quad GPU System
Aspen can provide you with Tesla C2050 GPU computing processors in several different form factors, depending on your needs. If you require a smaller system, perhaps beside your desk or in an office environment, our 4U 4 GPU integrated compute server can provide over 2 Tflops of double precision floating point performance, right at your fingertips. This configuration puts a super computer at your desk side, bringing extremely high performance computing into the single system realm. If desired, this system can be rack mounted and equipped with high speed interconnects.
The 4U 4 GPU integrated server is also quite popular for testing and developing GPU codes, as the system provides a significant amount of compute power for scaling tests as well as the development environment, all in a single package that can reside with the researcher and doesn't require data center access or expensive infrastructure.
Aspen
Dual GPU 1U System
The number of GPUs per CPU is a common question that must be answered based on your code performance. In a GPU computing environment, some code is normally ran on the general purpose CPU, such as sequential segments or data pre- and post-processing, and every code may work best with different mixes of GPU and CPU ratios. In data center and multi-use environments, a proper mix of GPU/CPU computing can be difficult, as hybrid codes may be implemented in different ways based on what portions of the codes are left on the CPU and what portions are parallelized and ran on the GPU. For the greatest density, useful for large GPU compute systems, Aspen offers a 1U rack mounted integrated GPU server. This system can provide over 1Tflop of double precision floating point performance in a 1U package!
Tesla
S2050 System
For the greatest flexibility, Aspen offers the Tesla™ S2050 1U GPU Computing System, which can be paired with CPU computing nodes based on your desired density. This system allows you to tailor your GPU/CPU mix in several different ways, and can increase your GPU computing density to 4 Tflops of double precision floating point performance in 3U! Tesla S2050 systems utilize host interface cards ( HICs ) to connect one or more Aspen general purpose CPU platforms to one or more Tesla S2050 1U platforms, depending on your requirements.
Tesla S2050 GPU Configurations: 1:4 & 1:2 CPU/GPU ratios shown
Whatever your GPU Computing questions or needs are, you can rest assured that Aspen can help! Our experienced sales engineers can help you with your GPU computing questions, and design a highly reliable and scalable GPU Computing solution for you today. Call us, we're here to help!
Contact Aspen Systems sales at 1-800-992-9242 for more information about our high performance GPU computing clusters built with NVIDIA® Tesla™ C2050 GPU Computing Processors and S2050 1U GPU Computing Systems today!
