I’ve previously discussed our PC for acquiring images at 1.1 GB/s, and some of the problems we’ve had with that SSD array slowing down over time. I just got a second PC for high speed streaming acquisition, this one based on a Core i7 platform, rather than a Xeon platform. In addition to being about 60% of the cost of the Xeon system, I am also hoping that the TRIM support in the Intel RST drivers will prevent the slowing down of the array we’ve seen before.
The system has an Asus Maximus VI EXTREME Intel Z87 motherboard, a Core i7-4770K processor, and four 256 GB Samsung 840 Pro SSDs in RAID 0. The system was built for us by Central Computers. As measured by Crystal Disk Mark, the sequential write speed for the RAID array is about 1115 MB/s. This is about half the speed we see for our Xeon system with the Intel RAID card (2050 MB/s) and indicates that there is some bottleneck when using the Intel motherboard RAID support. However, the write speeds should be just fast enough to cope with the 1.1GB/s produced by an Andor Zyla operating at top speed, and is easily fast enough to handle the full data rate of 800 MB/s produced by a Hamamatsu Orca-Flash4.0.
To test TRIM support and slowing down of both RAID arrays, I then wrote 900GB of data to the drive, measured the speed, then erased this data and repeated this test twice. At the end of this test we’ve written about 2.8x more data than capacity of the drive, which I think means that it should need to erase pages on the drive. However, I don’t see a significant slowdown on either system – the sequential write speeds are essentially unchanged at the end of this test.
Clearly, there is something I don’t understand about where the slowing of write speeds over time is coming from. I’m going to continue monitoring the performance of both systems as we use them for data acquisition. However, for many high speed streaming applications, a Core i7-based system is likely to be fast enough, and save the expense of a RAID card.