A Guide to Getting the Best Performance with Large Datasets

Issue link: https://geospatial.trimble.com/en/resources/i/1415711

Contents of this Issue


Page 9 of 12

Trimble Software: Working With Large Datasets 10 The Graphics Processing Unit (GPU) Specifically for Trimble scanning software, the speed of the CPU, disk drives and RAM have a bigger impact on the overall speed of processing in the software than the GPU speed. In other words, using a faster GPU has a smaller impact on the overall performance than a faster CPU and/or RAM. If everything else is optimized (i.e., the CPU, disk drives, and RAM), including a higher speed GPU will help the overall performance slightly. However, the high cost of a fast graphics card is unlikely to be worth the expense for the performance improvements it provides with Trimble scanning software. With that said, the use of integrated graphics (on-chip) is not optimal for processing large scan and image data. A dedicated GPU is highly recommended. Multi-Threading More points can be visualized on-screen in Trimble software by increasing the VRAM. CUDA and other GPU parallel processing techniques offer advantages when the software has been written to take advantage of them. Today Trimble's scan processing software makes minimal use of these parallel processing APIs when working with the GPU. OpenCL is used to provide GPU parallel processing in our software. While this is likely to benefit AMD users more, it is still recommended today that our users work with NVidia Quadro graphics systems. We suggest this because most of our developers are working on Quadro-equipped PCs. This means any bugs that are related to the GPU are more likely to be found and fixed. Even though a gaming configured GPU can provide more muscle for a given price, an NVidia Quadro-based graphics card is recommended over an NVidia GeForce-based system (Pro versus Gaming). Data Drives (HDD, SSD, NVMe, RAID Arrays) As a general rule, Hard Disk Drives are slower at reading and writing data than Solid State Drives. Newer NVMe drives can be faster still. A significant factor in identifying the fastest drives is the number of PCIe lanes that are available to pass data from and to the drives. PCIe is a high speed serial bus standard and a lane is composed of two differential signaling pairs, with one pair for receiving data and the other for transmitting. Thus, each lane is composed of four wires or signal traces. The chipsets supported by the CPU impact the number of PCIe lanes that NVMe drives can utilize. Additionally the Generation of PCIe that is utilized has an impact. In June 2020 AMD utilizes PCIe Gen 4 whereas the latest Intel products use PCIe Gen 3 (or earlier generation). The tradeoffs between number of lanes and speed of the cores is apparent in the choice between manufacturers' "Pro" products versus their "Gaming" products. For example, in the Pro product series AMD's Threadripper and Intel's Xeon (HEDT) products tend to have slower cores and more lanes than products in the AMD Ryzen and Intel Core Ix series, which are often thought of as Gaming products today. The Gaming products tend to have faster cores but fewer lanes. This equates to slower RAID operations. To summarize the different factors that impact disk drive speed: the number of lanes is roughly one third of the equation, the types of lanes comprise another third, and drive speed is the remaining third.

Articles in this issue

Links on this page

view archives of Whitepapers - A Guide to Getting the Best Performance with Large Datasets