Sequential/Random Read and Write
We use Anvil's Storage Utilities 1.0.51 to measure random and sequential read and write speeds of our SSDs. The synthetic benchmark integrated into this software suite provides a great overview of the tested products by checking out their key speed characteristics. The results you will see here refer to the FOB performance (fresh out-of-box, i.e. non-degraded) of the SSDs. It must also be noted that we use incompressible data for this test.
The first results of the synthetic benchmark suggest that the Samsung 840 EVO is an excellent drive indeed. It is fast because it is superior to all the opponents at random-address reading with a short request queue, which is the most important parameter for consumer-class SSDs. As for other usage scenarios, the 840 EVO is only relatively weak at random-address operations with a long request queue. That’s not a typical load for the disk subsystem of a regular desktop computer, though.
So Samsung seems to have successfully reached the initial goal. The new SSD with TLC NAND flash memory is indeed comparable to flagship products from other brands in terms of performance. There is only one thing we should be aware of. The high writing performance of the Samsung 840 EVO is ensured by the TurboWrite technology, which is only efficient when rather small chunks of data are being written continuously. To illustrate this fact, we can show you how the linear write speed of the 840 EVO changes as the drive is being filled with data.
First, the Samsung 840 EVO really delivers an outstanding linear write speed, which is limited by the SATA 6 Gbit/s bandwidth. It cannot maintain that speed for long, though. The 250GB model we are testing today has a TurboWrite cache which is only 3 GB large. So after writing 3 gigabytes, the drive's speed sags to the bandwidth of 8-channel TLC NAND flash with 2-way interleaving on each channel. To be precise, the initial write speed is up to 495.2 MB/s but then it goes down to about 280 MB/s.
Hence yet another fact: synthetic benchmarks may disagree in their estimate of the Samsung 840 EVO's write speed depending on what amount of data they use for testing. For example, we use 8GB arrays of data, so the results are far from the possible maximum. To check this out, we additionally measured how the write speed changed depending on the amount of data.
It is the speed of sequential and random-address writing with a long request queue that depends the most on the amount of data. These parameters are limited by the flash memory bandwidth, so the TurboWrite technology is the most beneficial for them. When the request queue is short, the speed of random-address writing depends on the controller's performance rather than on flash memory, so it doesn't change much as the amount of written data increases.