Overclock.net - An Overclocking Community - Reply to Topic

Thread: Samsung 840 EVO read speed drops on old-written data in the drive Reply to Thread
Title:
Message:

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in


  Additional Options
Miscellaneous Options

  Topic Review (Newest First)
03-08-2018 04:02 AM
5291Crash

the old test from 09-10-2014 (post 126)

Today's test attached along with smart data, room temp is 21.4, with no fan on the drives (no case sides in months)
07-10-2016 12:14 AM
TELVM
Quote:
Originally Posted by prosumer View Post

... I'm considering getting a new SSD 480-500Gb capacity. Any recommendations? ...

I'd steer clear from anything with planar TLC, and would seek only MLC or 3D TLC: 850 EVO, Extreme Pro, Reactor. Have a look here or here.


Quote:
Originally Posted by prosumer View Post

... Has the 850 evo so far been free of the problem that the 840 evo had? ...

Yes, as far as I know (have fiddled with both the 840 EVO and 850 EVO).
07-08-2016 10:52 PM
prosumer hi all, new to the forum here and hoping to find some answers for a problem I'm having with the 840 evo. After some research on the net, this thread seems to have the most knowledgeable posters. I bought my 840 evo in 11/2014, which was blazing fast initially until the performance degradation issue hit. I have since then updated to the latest firmware EXT0DB6Q, which subjectively seems to have fixed the problem. However, over time, the computer seems to slow down despite the latest firmware. In 4/25/2016, using the Magician's performance benchmark, my random read/write is 29K/23K and seq read/write in 548/528 I did an advanced performance optimization back then on the Magician software, which restored the random read/write to 90K/50K. Then yesterday, 7/7/16, the computer started slowing down significantly, and a recheck of SSD performance found that random read/write is back down to 29K/23K . I repeated the advanced performance optimization and benchmark found random read/write is back up to 90K/65K. However, the computer seems to have these periodic 0.5 sec freezes here and there which I want to attribute to the evo 840.

Is periodic advanced performance optimization needed every 2-3 months to maintain the optimal performance of the drive? Has that been the experience for other folks who own the 840 evo? Or did I get a bad batch? And are people happy with having to do advanced performance optimization every 2-3 months? Does Samsung consider that fixed?

I'm still under warranty until 11/2017. What has other folks' experience been with warranty service. It seems to me people just get another 840 evo refurb, which I really don't want. Has anyone had success with getting Samsung to give a 850 evo replacement instead?

I'm considering getting a new SSD 480-500Gb capacity. Any recommendations? Do folks pretty much agree with Sean's SSD buyers guide suggestions? It seems like other brands have their own issues as well, but so far I've not read too many problems with Crucial MX200, SanDisk Ultra II, SanDisk Extreme Pro. Has the 850 evo so far been free of the problem that the 840 evo had?

I'd appreciate any comments and suggestions. I'd be interested to hear the competing perspectives of malventano and darkhaze too.

Thanks
07-04-2016 08:19 PM
malventano
Quote:
Originally Posted by Cyclops View Post

So, it wasn't anal enough you're going double and triple anal for the nichest of niche reader that needs to know how their SSD performs on an atomic level.

I approve.

What's hilarious about all of this is that the end goal is to *reduce* the number of charts in a review. Sometimes to get the simplest yet most realistic data, you have to jump through all sorts of hoops before distilling down to those very simple results.
07-04-2016 08:17 PM
malventano
Quote:
Originally Posted by JackCY View Post

Do Windows APIs provide such access when reading data from drives? Or does it have to be somehow workaround and low level accessed without using Windows DLLs? Curious how these SSD testers are made.

Everything I'm doing is handled at the application level, relying on standard API calls for all IO. Any other way and it wouldn't be a 'correct' test, as it would bypass the typical path through the NVMe/AHCI driver. In this case, the test app has to be really damn good at pacing the IO's issued, and tracking how long each one of them takes to complete.
07-04-2016 07:45 PM
JackCY
Quote:
Originally Posted by malventano View Post

NOT GOOD ENOUGH!

(Sam Breakstone mode disabled)

Latency Percentile was only half of the story. When comparing SSDs with those results, we (everyone) hits an SSD with a saturation load at a specific QD. This means that in the Percentile comparison plots, you have differences in latency *and* IOPS. The reason we haven't posted any enterprise stuff since Latency Percentile was rolled out was that true comparisons can only be made if one of those other differences can be made equal across the tested SSDs. The solution is that I can now apply workloads at a precise paced IOPS and let the SSDs 'float' to whatever QD they need to service those requests. Now the Latency Percentile comparisons will eliminate a variable (IOPS), faster SSDs won't be working as hard, driving their latencies lower, meaning the percentiles can be used for better comparison without having to also account for varying IOPS in the sidebar. The IOPS data that used to be in the legend will be replaced with the average QD of each SSD in the test run.

...so yeah, basically I'm flipping the whole damn thing upside down. We used to have fixed QD that gave max IOPS. I'm doing fixed IOPS and obtaining the variable QD that results (as well as the Latency Percentile at that load). It's the only way I can imagine getting a true a vs. b comparison. It's taken so damn long because I had to do a lot of coding and qualifying tests to get there.
Do Windows APIs provide such access when reading data from drives? Or does it have to be somehow workaround and low level accessed without using Windows DLLs? Curious how these SSD testers are made.
07-04-2016 05:33 PM
Cyclops
Quote:
Originally Posted by malventano View Post

NOT GOOD ENOUGH!

(Sam Breakstone mode disabled)

Latency Percentile was only half of the story. When comparing SSDs with those results, we (everyone) hits an SSD with a saturation load at a specific QD. This means that in the Percentile comparison plots, you have differences in latency *and* IOPS. The reason we haven't posted any enterprise stuff since Latency Percentile was rolled out was that true comparisons can only be made if one of those other differences can be made equal across the tested SSDs. The solution is that I can now apply workloads at a precise paced IOPS and let the SSDs 'float' to whatever QD they need to service those requests. Now the Latency Percentile comparisons will eliminate a variable (IOPS), faster SSDs won't be working as hard, driving their latencies lower, meaning the percentiles can be used for better comparison without having to also account for varying IOPS in the sidebar. The IOPS data that used to be in the legend will be replaced with the average QD of each SSD in the test run.

...so yeah, basically I'm flipping the whole damn thing upside down. We used to have fixed QD that gave max IOPS. I'm doing fixed IOPS and obtaining the variable QD that results (as well as the Latency Percentile at that load). It's the only way I can imagine getting a true a vs. b comparison. It's taken so damn long because I had to do a lot of coding and qualifying tests to get there.

So, it wasn't anal enough you're going double and triple anal for the nichest of niche reader that needs to know how their SSD performs on an atomic level.

I approve.
07-04-2016 05:13 PM
malventano
Quote:
Originally Posted by Cyclops View Post

New test suite? Latency percentile wasn't good enough?

NOT GOOD ENOUGH!

(Sam Breakstone mode disabled)

Latency Percentile was only half of the story. When comparing SSDs with those results, we (everyone) hits an SSD with a saturation load at a specific QD. This means that in the Percentile comparison plots, you have differences in latency *and* IOPS. The reason we haven't posted any enterprise stuff since Latency Percentile was rolled out was that true comparisons can only be made if one of those other differences can be made equal across the tested SSDs. The solution is that I can now apply workloads at a precise paced IOPS and let the SSDs 'float' to whatever QD they need to service those requests. Now the Latency Percentile comparisons will eliminate a variable (IOPS), faster SSDs won't be working as hard, driving their latencies lower, meaning the percentiles can be used for better comparison without having to also account for varying IOPS in the sidebar. The IOPS data that used to be in the legend will be replaced with the average QD of each SSD in the test run.

...so yeah, basically I'm flipping the whole damn thing upside down. We used to have fixed QD that gave max IOPS. I'm doing fixed IOPS and obtaining the variable QD that results (as well as the Latency Percentile at that load). It's the only way I can imagine getting a true a vs. b comparison. It's taken so damn long because I had to do a lot of coding and qualifying tests to get there.
07-04-2016 05:00 PM
Cyclops
Quote:
Originally Posted by malventano View Post

I wish. This one is from another company. Hopefully Samsung passes me some enterprise samples once I roll out our new test suite.

New test suite? Latency percentile wasn't good enough?
07-04-2016 04:39 PM
malventano
Quote:
Originally Posted by Cyclops View Post

Let me hazard a guess.... It's from Samsung, it runs hot, and does a couple of GB/s Read and write.

I wish. This one is from another company. Hopefully Samsung passes me some enterprise samples once I roll out our new test suite.
This thread has more than 10 replies. Click here to review the whole thread.

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off