Last year, users discovered a problem with Samsung's 840 EVO SSD that caused dramatic slowdowns when reading older data. Samsung attributed the issue to an algorithmic error in the management routine that tracks the status of cells over time. A firmware fix and accompanying Performance Restoration utility were issued in October, and they seemed to do the trick. However, new evidence suggests that the problem persists.
A couple of TR readers (thanks Horia and Richard) pointed me to recent entries in the original Overclock.net thread complaining of slow read performance. Those reports come from drives running the supposedly fixed EXT0CB6Q firmware, and they prompted me to test an EVO I've been saving for just such an occasion. The results don't bode well for the TLC drive.
When the initial fix was issued, I patched our 840 EVO 250GB SSD and then filled it with a mix of movies, MP3s, images, and other files. That drive spent more than three months on the shelf before being called up for a round of read speed tests. Here's HD Tach's assessment:
35MB/s on a modern SSD? Yeah, that ain't right.
Next, I ran SSD Read Speed Tester, which tabulates read speeds based on the age of the files. This benchmark indicates that everything on the drive was almost 15 weeks old when the test was run.
The average here is 54MB/s, with some files reading at well over 100MB/s. That's faster than in HD Tach, but it's still a far cry from what the drive can do with fresh data.
Slow read speeds didn't just afflict targeted benchmarks, either. Transfer rates were extremely slow when I copied the EVO's contents to a secondary SSD in the same system. I then formatted the EVO, loaded it with the original data, and ran the same tests again.
That's more like it. The drive averages 430MB/s in HD Tach, a 12X increase. The EVO is about 10X faster according to SSD Read Speed Tester, which reports a 529MB/s average.
SSDs typically aren't left unpowered for months at a time, so it's possible that hiatus contributed to the slow read speeds exhibited by our sample. However, the other recent reports of read slowdowns come from drives that have been in service since the patch was applied. The issue doesn't appear to be confined to unused drives.
Bruno, our resident coder, has an 840 EVO in his personal machine. He agreed to run a few tests for me, and his drive isn't substantially slower to access older files. But NTFS compression was recently enabled for much of the drive's contents, so I'm hesitant to draw any conclusions based on those results. Compressing the data should effectively refresh the contents of the NAND cells even if there's no change in the stated age of the files.
We've notified Samsung of our findings and are awaiting an official comment from the company. In the meantime, I'm curious if any TR readers have experienced similar slowdowns. Have any of you noticed any read speed issues with patched 840 EVO SSDs?
116 comments — Last by LarryM7 at 12:26 PM on 04/09/15
|Samsung's 960 Pro 2TB SSD reviewedHoly crap||125|
|Toshiba's OCZ VX500 512GB SSD reviewedA19 flash bids adieu||33|
|Adata's Premier SP550 480GB SSD reviewedTaking aim at the budget segment||36|
|Samsung's Portable SSD T3 reviewed2TB in the palm of your hand||15|
|Crucial's MX300 SSD reviewedThe MX series enters the third dimension||57|
|Toshiba's OCZ RD400 512GB SSD reviewedNVMe inches towards attainability||24|
|Mushkin's Reactor 1TB SSD reviewedA familiar one-two punch||31|
|Adata's XPG SX930 240GB SSD reviewedAnother 16-nm Micron MLC challenger appears||24|
|AMD drops prices on the Radeon RX 460 and RX 470||31|
|Reports: Radeon RX 470D is a budget Polaris card for China||6|
|Examining reports of slow write speeds on the 32GB iPhone 7||21|
|Cellular Insights dissects iPhone 7 Plus modem performance||11|
|Deals of the week: scads of high-performance storage and more||8|
|Tobii's Eye Tracker 4C knows where your head is||1|
|GeForce driver 375.57 is prepared for Titanfall 2||7|
|Phanteks Eclipse P400 gets a tempered glass option||0|
|Radeon 16.10.2 drivers add support for October's big games||10|
|A real "console monitor" would be 720p @ 30 Hz ;P||+58|