The most reliable SSD: results of the experiment one and a half years

Modern SSD-drives reliable enough, and considering that the price per 1 GB (in dollars) gradually decreases, then use SSD in many cases even more efficient than working with the HDD. But what SSD choose?

Eighteen months ago, journalist Tech Report decided to conduct an experiment to identify the most reliable SSD . He took six drive models: Corsair Neutron GTX, Intel 335 Series, Kingston HyperX 3K, Samsung 840, Samsung 840 Pro, and put all six on the cyclical process of reading / writing. The memory capacity of each drive was 240-256 GB, depending on the model.

Immediately it should be said that all six models have successfully passed the load declared by the manufacturer. Moreover, most of the models survived longer read-write cycles than is stated by developers.

However, 4 of 6 models surrendered before reaching volume 1 PB "pumped" through the disc information. But the two models of those that participated in this attraction "death railway» (Kingston and Samsung 840 Pro) withstood even 2 PB, and then refused. Of course, the sample of 6 SSD is not a measure of work for all, without exception, SSD, but some representation in this sample is still there. The procedure for cyclic read-write, too, is not a perfect measure, because the drives can fail for various reasons. But the test results are very interesting.

One of the conclusions: the producers enough sensitive approach to the issue of choice of work limit their drives - as already mentioned above, all SSD withstood put limit the volume of recorded information.

As for the models, the first out of order Intel 335 Series In this model SSD is one feature - they will stop working as soon as there are bad sectors. Immediately after that, the drive enters the reading, and then completely transformed into a "brick". If it were not for the instruction "stop on failure" may, SSD would have worked longer. The problems started with the disc after passing the mark of 700 TB. The information on the disk remain legible until the reboot, and then drive turned into a piece of iron.

Samsung 840 Series successfully reached the mark of 800 TB, but began to show a lot of errors, starting with 900 TB, and dismissed without warning, before reaching petabytes.

Should abandon Kingston HyperX 3K - the model also has instructions to stop work when a number of bad sectors. By the end of operation, the device began to issue notification of problems, allowing to understand that the end is near. After the mark of 728 TB drive moved to the reading mode, and after rebooting stopped responding.

Corsair Neutron GTX was the next victim, passing the mark of 1.1 SDS. But the drive already has thousands of bad sectors, the unit began to issue a large number of warnings about problems. Even after eschee 100 TB disk allowing you to record data. But after a device is no longer even rebuta determined by the system.

There are only two models Kingston and Samsung 840 Pro, who heroically continued to work, even reaching 2 PB.

Kingston Hyper X uses data compression is possible, but the tester began writing incompressible data for purity test. To do this, we used the Anvil's Storage Utilities, serving to perform tests in reading, writing data.

The disc showed good results, although the gap between the 900 and the 1 TB PB has errors occurred, plus bad sectors. Errors were only two, but it's still a problem. Once a disk fails to 2.1 PB, he ceased to be determined by the system after rebuta.

The last fallen soldier iron in this battle was the Samsung 840 Pro

The drive passed the mark of 2, 1 PB and refused.

Well, there is no happy ending, all died, unfortunately. But otherwise it could not be - in the sense of the test and, wait for refusing to participate in it all models of SSD.


The tester had a series of tests to determine the performance of the models participating in the test. Here are the results:

As you can see, a lot of work, which gives an indication of the reliability and stability of modern drives. It is worth repeating that no sample or test conducted are not the most representative, but some conclusions can be made. Whether or not the author of the experiment to ask additional questions about the article?
Yes, I'm in the comments zaposchu your question - send it to the author, pozhalusta ​​td> No, quite the information that already exists Voted 170 people. Excused 174 people. Only registered users can vote in polls. Sign , please.



See also

New and interesting