Samsung 970 EVO NVMe SSD Benchmarks On Ubuntu Linux

[arj239a]

Used to use the Samsung evo 970 series looking for a replacement for new builds... "speed doe snot matter if the device is planing to nuke itself with massive over access upon mounting the drive"
Also Samsung will not support Linux or Unix nor do they have a bug reporting support for Unix or Linux firmware based issues.
No Guidelines for best or optimal use offered even after explaining the process, but there are magic windoze directions everywhere.
New Evo should come with the Windoze logo only or the MacOs only logos... no longer Unix/Linux compatible
The Old firmware 2018 worked flawlessly with Mint 18.x 19.x the new 2020 EVO 970 units ship with Firmware that after formatting the drive for EXT4 there is extensive hyper access by the OS to the EVO drive even when there is no access to the device... HEADS UP NEW EVO is Windoze only, but they will not say that on the packaging.
1.) install EVO in external Enclosure ( make sure production date is 2020 ) - have seen issue with EVO internals as well production date 2020
2.) Gparted Create EXT4 partition
3.) navigate to the Drive via CLI or GUI
4.) watch the light on the enclosure just start flashing and it will not stop
5.) disconnect drive feel NVME controller it will be abnormally hot

* note all EVO drives tested and used with production dates of 2018 08 or earlier do not exhibit this behaviour

[drSeehas]

... What options exist for running nvme with fewer pcie3 lanes, or lesser pcie2 lanes?

I don't understand your question.

I am running a Samsung 950 Pro in a PCIe2x4 slot, which means max 2 GB/s throughput. During writing this is more than the SSD is capable. During reading it is measurable, but not noticeable.

[msroadkill612]

It is explained by the SSD having too little of those slow TLC chips. 512 GB and larger models have more chips and thus more bandwidth especially for writes. This one is rated for only 1500 MB/s writes but 512 GB model has double the capacity and double writing bandwidth.



EVO is not the best that Samsung has to offer. It is a very good SSD never the less.

I am aware of these details, but the fact remains that even the best rarely use >2GB/s, and then dont exceed it by much.

What options exist for running nvme with fewer pcie3 lanes, or lesser pcie2 lanes?

[Tomin]

Sequential reads here yield an impressive ~3400MB/s, utilising a decent portion of the ~4GB/s theoretical bandwidth of the scarce pcie 3 lanes allocated.

But no other result comes even close to using 2GB/s, which could be catered to by merely allocating 2 scarce lanes (or 4x pcie 2 lanes).

It is explained by the SSD having too little of those slow TLC chips. 512 GB and larger models have more chips and thus more bandwidth especially for writes. This one is rated for only 1500 MB/s writes but 512 GB model has double the capacity and double writing bandwidth.

Most of us, or even all of us most of the time, are ~wasting 2 precious pcie 3 lanes for each nvme drive, even the very best, like the evo here. Bandwidth is rarely the bottleneck.

EVO is not the best that Samsung has to offer. It is a very good SSD never the less.

[msroadkill612]

Sequential reads here yield an impressive ~3400MB/s, utilising a decent portion of the ~4GB/s theoretical bandwidth of the scarce pcie 3 lanes allocated.

But no other result comes even close to using 2GB/s, which could be catered to by merely allocating 2 scarce lanes (or 4x pcie 2 lanes).

Most of us, or even all of us most of the time, are ~wasting 2 precious pcie 3 lanes for each nvme drive, even the very best, like the evo here. Bandwidth is rarely the bottleneck.

Is there a solution? The latency advantages of direct to the cpu nvme (i.e. not high latency chipset nvme ports) , but using fewer lanes?

[Michael]

Michael, how do I run the tests on a specific drive? I have 5 drives in my computer and it automatically ran on the drive which the current folder was on. Do I need to move the entire test suite to the drive which I want to test, or can I just run it from an empty folder on the targeted drive?

Edit ~/.phoronix-test-suite/user-config.xml (or its /etc/phoronix-test-suite.xml as root) and change EnvironmentDirectory to wherever the desired drive is mounted.

[Azpegath]

Michael, how do I run the tests on a specific drive? I have 5 drives in my computer and it automatically ran on the drive which the current folder was on. Do I need to move the entire test suite to the drive which I want to test, or can I just run it from an empty folder on the targeted drive?

[edwaleni]

I just ran this test case against a new SanDisk Extreme SSD using USB 3 Gen 2 and a coaxial USB cable (not a .99 cheapie) with the Intel Haswell based USB controller . While it stunk on the database tests, it actually beat the Crucial and Intel drives on a few of the IO tests. I was surprised.

I will have to re-run this test on a ASM-2142 with a 4 lane PCIe v3 connection. It is interesting how far the bottom has come up.

[oleyska]

It was some EVO I had that died, still got it somewhere.... I tried RMA'ing it a few times but the Samsung website never worked. Mind sharing what URL you used for RMA'ing it from?

I used my e-tailer
In Norway by law we're allowed to do the RMA directly with retailer we purchased from and they do it to samsung and or other manufactures.
I receive my new drive before they've done RMA back to samsung so for me it was done in 5 days or so

[Michael]

I had a 960 evo for 2 hours and mine died, got an rma and it's still going strong

It was some EVO I had that died, still got it somewhere.... I tried RMA'ing it a few times but the Samsung website never worked. Mind sharing what URL you used for RMA'ing it from?