Los Gatos, California. , – CacheQ Techniques, Inc. It introduced GPU assist for its QCC acceleration platform. It’s a heterogeneous computing improvement setting that gives sooner efficiency and diminished improvement time for pc architectures incl multi-core processorsGPUs and the sphere Programmable gate matrices (FPGA).
“The demand for {hardware} acceleration with GPUs and different heterogeneous computing {hardware} is rising exponentially,” notes Clay Johnson, CEO and co-founder of CacheQ Techniques, a developer of heterogeneous acceleration options. Our purpose is to simplify the high-performance information middle and develop edge computing functions. The QCC accelerator platform achieves this purpose and can allow new options throughout quite a lot of functions.
GPU deployment has progressed at a fast tempo previously 5 years. The $25 billion yearly business is predicted to proceed to develop at a compound annual progress price of roughly 33% via 2028.
The benefit of the QCC accelerator platform
Heterogeneous computing techniques akin to multi-core processors and GPUs in addition to FPGAS hooked up to those processing techniques relied on software program instruments supported by {hardware} distributors and open supply social communication. These instruments have historically relied on software program builders to move data to compilers. That is to precise parallelism of their code via {hardware} APIs akin to CUDA from NVIDIA, HIP from AMD, and oneAPI from Intel.
Different efforts try to assist built-in pragmas in C, C++, and Fortran via OpenACC, OpenMP, and OpenCL. All of them require deep data of the goal {hardware} to regulate reminiscence copying and synchronization occasions. As well as, to create groups from threads, manually take away the loop load dependencies, race circumstances, and add abstracts. The aim is to attain efficiency and proper code conduct on parallel compute items.
CacheQ QCC is the primary compiler platform to robotically extract parallelism from customary C, C++, and Fortran code. It doesn’t require the developer to explicitly talk parallelism to the compiler. AQCC robotically accelerates functions utilizing quite a lot of units, outperforming pragma-based strategies. It could additionally deal with manually coded API options with minimal {hardware} data. This enables the developer to put in writing generic code and goal high-performance {hardware} at compile time with out refactoring, or refactoring in a means that doesn’t goal particular {hardware} and is definitely functionally verifiable.
Based mostly on the CacheQ Digital Machine (CQVM), the QCC Acceleration Platform is a heterogeneous computing improvement setting that converts high-level serial language (HLL) code right into a parallel illustration in lower than 30 seconds for probably the most advanced designs. It helps code profiling, utilization estimations, efficiency simulation, reminiscence configuration, and partitioning throughout quite a lot of compute engine processors together with GPUs, x86, Arm, and RISC-Vand FPGAs earlier than creating an executable computation.
Options:
Options embody a improvement setting with unified drivers, protected containers, and assist for a number of boards from a number of distributors. Its design evaluation provides profiling, Efficiency simulation and reminiscence exercise studies. The optimization functionality provides user-driven decoding, reminiscence configuration, and computerized, user-directed partitioning.
An FPGA implementation features a useful resource estimator, pre-made wrappers, a number of boards and elements, and implementation instrument automation. The reminiscence implementation helps computerized integration, multi-port/multi-access and stripe.
Availability and pricing
The QCC acceleration platform ships now in restricted portions with basic availability within the undertaking in late 2023. Model 0.18 helps GPUs from nVidia and AMD acceleration boards, Xilinx FPGAs and CPUs from Intel, AMD, Arm, Apple and RISC-V.
Pricing is offered upon request.
In the meantime, go to CacheQ web site For added data, demo requests or early entry to the QCC accelerator platform.