Synthetic Performance Testing
We tested two Hyperstone models against a number of off-the-shelf USB flash drives with and without an onboard DRAM buffer. The first U9 tested uses pSLC programming with 2-bit per cell (MLC) flash. The second U9 just uses MLC flash without the pSLC-programmed cache.
Sequential Read Performance
The performance is flat with all of the drives as we increase the load through queue depth for all of the drives. The Hyperstone U9 with pSLC programming trails the model without the cache programming in sequential reads in our testing but will have a large advantage writing data.
The Corsair Force GT leads this group in sequential read performance. The drive has been a favorite with performance enthusiasts for several years and has a strong reputation for portable storage. It uses a DRAM buffer to lead this test, but this will be the only one where it sits at the top.
Sequential Write Performance
The pSLC programming on the more advanced Hyperstone allows it to charge past the other products in our sequential write test. Again we don't see much performance variation as we scale the queue depth with any of the drives.
The Hyperstone writing data in two bits per cell delivers nearly identical performance to the Corsair Force GT with a DRAM buffer. This is quite an accomplishment given the memory restrictions for the Hyperstone. The drive is able to make up the performance gap through superior firmware and a streamlined data path.
Sustained Sequential Mixed Workloads
In this test, we focus on the mixed workloads, where both data reads and writes exist. This is closer to the way these products work in IoT devices since the parts are the only data storage component in the system.
The Hyperstone drive with pSLC shows strong performance in the three mixed workload tests and outperforms the other products by a sizable margin.
The MLC, only Hyperstone, also performs well. This model often outperforms other USB drives with a DRAM buffer in every mix.
Both Hyperstone drives show very impressive results in writing sequential data. The performance in this area is laser-like consistent, and for many applications, this is very important. One of the most demanding areas that require sequential write consistency is in video applications like dash and other video cameras. Inconsistent performance forces cameras to drop frames and even lock up the device altogether.
Random Read Performance
Industrial and other IoT type devices generally see long streams of random data that's collected, processed and often sent to the cloud. Random data is much harder on flash memory due to the read, modify, and write program cadence. We eight non-Hyperstone drives for this article but only half survived testing. The other four prematurely expired during the random write or applications tests.
The Hyperstone drives really stand out in the random performance tests. The pSLC drive stands above all others in our random read test using 4KB data. The non-pSLC model even outperforms the Flash Voyager GT with a DRAM buffer in this test.
Random Write Performance
The Hyperstone pSLC drive really stands out in the random write test where it laps the rest of the field at all queue depths. Again, we see the non-pSLC Hyperstone drive compete with the Corsair drive with the DRAM buffer.
Sustained Random Mixed Workloads
The last synthetic test mixes random reads and writes at different weights. The random mixed tests are very similar to what your data logging IoT device faces in use. The two ADATA and single Corsair drives fall flat here with the IO pinned to the bottom of the chart. Over large portions of the test, those drives don't even register a single IO for several seconds. The SanDisk Ultra Flair USB 3.0 looks like it does better but can't sustain a 100% write random workload.
Neither of the Hyperstone drives failed to register an IO in our test. The pSLC-enabled model seemed to accelerate with the random mixed portions of the test with a gradual increase from 100% reads to 25% reads with IOPS between 3,100 and 4,000. The most impressive part of these results is the consistency of the IO. The non-pSLC Hyperstone drive also delivers impressive results in the random mixed test. This drive favors less writes from the host device, making it ideal for industrial applications with less data logging.