Our team has engineered an advanced upgrade of our Direct Memory Access (DMA) engine, achieving a transfer speed that is twice as fast as current fourth-generation digitizer products.Considering system configuration options, one could efficiently integrate dual ADQ35 units with a single GPU to reach transfer rates of roughly 28 Gbyte/s within a single system. In the perform tests, we used two ADQ35 digitizers to stream data to a single GPU as well as two GPUs respectively. Notably, the cost of the (PC) system remains under $2000, not including the GPU. This performance is facilitated through the use of a x16 PCIe Gen3 link. The tests were performed both on Linux (topmost figure below) and Windows (bottom figure) operating systems.When streaming to a single GPU from both digitizers, the achieved aggregate transfer rate approached 28 Gbyte/s, as indicated by the blue traces in the figure below. Upon incorporating a second GPU, the corresponding aggregate data rate increased to 30 Gbyte/s, as depicted by the orange trace. Thus, it was the GPU that served as the limiting factor in achieving the rate when utilizing a single device. Notably, the transfer rate tends to be higher on Linux platforms due to NVIDIA GPUs providing genuine peer-to-peer streaming capabilities for this operating system. Conversely, on Windows, streaming is conducted through pinned buffers, which may occasionally marginally restrict performance and elevate CPU/DRAM load on the host PC.
