Los Gatos, California. , – CacheQ Techniques, Inc. It introduced GPU assist for its QCC acceleration platform. It’s a heterogeneous computing growth atmosphere that gives quicker efficiency and lowered growth time for laptop architectures incl multi-core processorsGPUs and the sector 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 purposes. The QCC accelerator platform achieves this purpose and can allow new options throughout a wide range of purposes.
GPU deployment has progressed at a speedy tempo previously 5 years. The $25 billion yearly business is anticipated to proceed to develop at a compound annual progress fee of roughly 33% via 2028.
The benefit of the QCC accelerator platform
Heterogeneous computing techniques comparable to multi-core processors and GPUs in addition to FPGAS connected 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 go data to compilers. That is to specific parallelism of their code via {hardware} APIs comparable 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 information of the goal {hardware} to manage 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 habits on parallel compute items.
CacheQ QCC is the primary compiler platform to mechanically extract parallelism from customary C, C++, and Fortran code. It doesn’t require the developer to explicitly talk parallelism to the compiler. AQCC mechanically accelerates purposes utilizing a wide range of gadgets, outperforming pragma-based strategies. It could possibly additionally deal with manually coded API options with minimal {hardware} information. This permits the developer to put in writing generic code and goal high-performance {hardware} at compile time with out refactoring, or refactoring in a approach that doesn’t goal particular {hardware} and is well functionally verifiable.
Primarily based on the CacheQ Digital Machine (CQVM), the QCC Acceleration Platform is a heterogeneous computing growth atmosphere that converts high-level serial language (HLL) code right into a parallel illustration in lower than 30 seconds for probably the most complicated designs. It helps code profiling, utilization estimations, efficiency simulation, reminiscence configuration, and partitioning throughout a wide range of compute engine processors together with GPUs, x86, Arm, and RISC-Vand FPGAs earlier than creating an executable computation.
Options:
Options embrace a growth atmosphere with unified drivers, protected containers, and assist for a number of boards from a number of distributors. Its design evaluation affords 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 components, and implementation software 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 mission 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 accessible upon request.
In the meantime, go to CacheQ web site For extra data, demo requests or early entry to the QCC accelerator platform.