GPU Processing Options
For researchers with a need to perform signal processing or image reconstruction using GPUs, Verasonics offers two options, GPU Toolkit and GPU Toolkit with GPU Direct, to accelerate Vantage Research Ultrasound System data transfer, which reduces latency and improves frame rate. These options also provide programming tools and capabilities that can assist with the overall development process.
Features
Available as licensable options, GPU Toolkit and GPU Toolkit with GPU Direct are also available in packages that include an NVIDIA GPU card, software license(s) and the Verasonics’ Specialty Host Controller.
Both options include the following features:
1) User selected memory allocation of the Vantage buffers utilized in the image reconstruction process. The current default for the system is the CPU, new choices include:
- CUDA allocated host memory – enabling up to 2X faster asynchronous memory copies between the host computer and the GPU device
- CUDA allocated unified memory – simplifying CUDA programming by utilizing NVIDIA’s unified memory architecture
2) An Event Profiling Tool that allows users to see timing information for each operation in the event sequence, which can help to identify and eliminate bottlenecks.
3) A 1-step compilation of CUDA source code into .mex files that does not require the MATLAB® Parallel Computing Toolbox.
4) Example scripts illustrating implementation of RF or IQ data processing. These examples can be tested to run synchronously or asynchronously with different memory allocation methods for comparison.
Verasonics understands that many researchers have already invested time and money in their own GPU hardware. A GPU diagnostic tool (gpuDiag), available starting with the Vantage 4.7.6 release, can be used by existing Vantage users to determine if their GPU card(s) are compatible with GPU Toolkit option features. Support for GPU Toolkit options vary, the Verasonics sales team can assist with customer questions.
The Fastest Route to GPU Processing
For Vantage users who are performing signal processing or their own image reconstructions on the GPU, consider GPU Toolkit with GPU Direct. This option allows researchers to move data directly from the Vantage System to the graphics card – increasing processing speeds yet again.
GPU Toolkit with GPU Direct
For researchers who have already implemented RF Data processing algorithms on the GPU or are planning to do so, GPU Toolkit with GPU Direct technology writes RF data directly to the GPU memory, bypassing the host controller CPU and RAM. This reduces overall latency by removing an extra operation. For processing algorithms such as 3D reconstruction and super-resolution imaging, this optimization reduces over all processing time. Please note that GPU Toolkit with GPU Direct technology is only relevant to writing RF Data directly to the GPU. Also, GPU Toolkit with GPU Direct will work on LINUX systems only.
11335 NE 122nd Way, Suite 100
Kirkland WA 98034