Real-World Software Performance Testing
PCMark 8 Real-World Software Performance
Using time-to-complete as a performance measurement, even the Adobe Photoshop Heavy Test shows very little separation. All of these drives and configurations fare almost identically.
What we do see, though, is that striped arrays often take longer to finish these tasks. Client workloads simply aren't taxing enough to stack commands the way our RAID arrays would need to outshine single drives.
To really exploit the extra bandwidth, the software needs to push the storage subsystem harder. Even small CAD and CAM workloads shove most of the data into DRAM. It isn't until you work with extremely large data sets that RAID can improve upon this level of baseline performance. If anything, the biggest benefit you get from two drives striped together is a larger volume to work from.
PCMark 8 Advanced Workload Performance
This set of metrics illustrates performance during and after steady state conditions. Most desktop users should focus on the Recovery portion, where the drives have idle time to execute their garbage collection and background flash management activities. The wear from the Degrade tests also simulates what you'll encounter with a lot of data on your SSD.
Frankly, we still don't fully understand the ins and outs of these drives yet. One variable we haven't worked out is the way thermal throttling works. In our first test with Lenovo's AHCI-attached SM951, we ran into throttling conditions, even with a fan blowing air over Samsung's controller. The SM951 AHCI we received from RamCity had newer firmware and didn't throttle as much. Lenovo released a firmware update to increase the point at which the SSD throttles. Then we fielded another SM951 AHCI from Lenovo (one we're using today) with an even newer software version that isn't available online yet.
Samsung's SM951 is purchased by a number of system integrators, and we suspect that most request specific firmware enhancements. Where, exactly, a drive throttles is an important variable to set, since M.2 drives are predominantly used in notebooks. All of this is to say that our array of 950 Pros enjoy a slight advantage over the SM951-NVMe array. We suspect throttling and firmware play a part in this. When the load lightens, both pairs deliver almost identical performance.
Total Access Time
The access time test also favors the striped 950 Pros, but again only by a small margin. The best single-drive access time goes to the SM951-NVMe. This crisscross happens throughout our comparison, and it may cause you to question which models are truly better. For the most part, they're evenly matched. Although one model shows better in one test, it comes up short in another.