Hardware Recommendations
RAM | Minimum: 64 GB (32 GB DSM/True Ortho) |
---|---|
CPU | Recommended: Intel Xeon/i9/i7 or AMD Threadripper/Epyc |
GPU (optional) | Minimum: none *GPU (CUDA) processing is supported when using a GPU with Compute Capability 3.2 or higher. **SURE uses more memory dynamically to improve performance |
Storage | Distributed Processing: Single Node Processing: |
Operating System | Microsoft Windows 7 or newer, Windows Server 2012 or newer, recent Linux distributions including a glibc version >= 2.23 (e.g. Ubuntu 16.04 or newer, Amazon Linux 2, CentOS 8 or newer). |
General Remarks
SURE will automatically benefit from additional RAM if available, due to processing in larger parts and less data transfer.
If your RAM significantly exceeds recommended RAM specifications, manually setting a larger tile size can increase performance. See also Performance Optimization for a table of recommended tile sizes.
Additional benefits can be expected from a fast communication between CPU and RAM - use all RAM slots (e.g. 8x16GB instead of 4x32GB) if possible to get maximum throughput on the memory lanes
Processing on graphics card is beneficial for hardware setups with an average CPU.
Avoid the use of USB and network drives. These type of drives do not enable caching, which leads to a strong performance loss.
Set the Windows energy plan to Ultimate Performance or other maximum settings by the CPU vendor, as this can have a strong impact on performance
Disable Antivirus software as file reading and writing can be slowed down - including Windows Defender and its realtime protection
Use operating system that can handle memory, e.g. Windows for Workstations and update latest chipset drivers and BIOS - particularly for AMD processors
Some server infrastructures provide more than one strong GPU or grouped CPU kernels. However, running multiple SURE instances in parallel on these infrastructures would not scale the processing throughput in the same way, since the large amount of data traffic (e.g. memory transport) between CPU / GPU and hard disks during the processing would reveal physical limitations in transfer rates between RAM, CPU, GPU and Disk. When targeting cost efficiency, we thus recommend using separate medium-priced hardware instances for each processing Node.
Please follow our best practices and always use distributed processing. Doing so, ensures the processing nodes stay within the 2TB storage recommendation, permitting the use of less expensive hardware. Also note that Master nodes store all the input data and final results, thus they require larger storage capacity.
RAM Minimum Requirements
Image Resolution | Images | Min. RAM | Recommended RAM |
---|---|---|---|
Up to 60 MP | <1000 | 16 GB | 32 GB |
Up to 60 MP | >1000 | 32 GB | 64 GB |
Up to 60 MP | >5000 | 64 GB | 128 GB |
< 400 MP - DSM & True Ortho | <5000 | 32 GB | 64 GB |
Unlimited MP - All Products | <10000 | 64 GB | 128 GB |
Unlimited MP - All Products | <60000 | 128 GB | 128 GB |
Storage Hardware Examples
Node Storage | 2TB Disk space: One 2TB SSD connected via PCIe/M2/NVMe with 3500MB/s transfer speed (e.g. Samsung 970 EVO 2TB SSD) |
---|---|
Master Storage | RAID controller in RAID 0 (or RAID 5) with 6 or 8 SAS spinning drives:
|