Question Samsung 970 EVO Plus 2TB Speeds

[0nion1999]

My M.2 drive has been loading me into games slower than my friends on regular SATA SSDs, so I downloaded CrystalDiskMark and let it run. I have looked up images of this SSDs results online and it seems like i should be getting much higher readings for the third category.

Here are the results from my test:

https://imgur.com/a/jvceiQE

View: https://imgur.com/a/jvceiQE


Is there something wrong with my drive?

Thanks.

 

[Colif]

Samsung 970 Evo Plus 1tb

Careful when reading results as it could be the people you are comparing to may be running slightly different parameters, if you run it with Q32T16 the score can be much higher. I had to redo mine once I saw the differences in our tests.

The default for that test is RND4K Q32T16 - It is interpreted as Random 4 Kibibyte Test for 32 processes in queue on 16 threads. You are only looking at 1 thread, should be looking at 16.

I think I had mine on defaults for below tests - you can change them in settings.

https://www.thanalysis.com/2020/11/crystaldiskmark-and-interpretation-of-test-results.html

what are specs of your PC? Was pc idling when you ran tests?

what score you get if you just run the RND4K test by itself? you can click button next to the test to run it alone.
Just checks to see if it might be thermal throttling. What temp does nvme hit in test? Mine was on 66c

what does Magician show for your drive? Its often better to use the Makers tests as they may interpret things differently

 

[0nion1999]

Motherboard: MSI B450 Gaming Plus Max
CPU: AMD Ryzen 5 3600
GPU: NVIDIA GeForce GTX 970
RAM: 2x8GB Corsair Vengeance LPX DDR4 3000MHz
PSU: AeroCool Intergrator 600W 80 Plus Bronze
Boot SSD: Samsung 750 Evo 250GB
HDD 1: Seagate Barracuda 1TB ST1000DM003
HDD 2: Seagate Barracuda 1TB ST1000DM010
Game SSD: Samsung 970 EVO Plus 2TB

My PC was idling when I did the original test, but it was after hours of gaming.

I ran some more tests after booting up this morning, with very different results, the first one with the same parameters as before:

https://imgur.com/0yTnmuR

View: https://imgur.com/0yTnmuR


After that, I installed Samsung Magician, ran their benchmark, then followed it with CrystalDiskMark using RND4K Q32T16. Temp peaked at 69°C during Crystal:

https://imgur.com/lFz3gYl

View: https://imgur.com/lFz3gYl


Running the RND4K test on its own, Q32T16 first, peaked at 56°C:

https://imgur.com/wJtnGGk

View: https://imgur.com/wJtnGGk


Running RND4K Q32T1 on its own, also peaked at 56°C:

https://imgur.com/luBuKXp

View: https://imgur.com/luBuKXp

 

[Colif]

its clearly not consistent. Temperatures seem to affect it.
We have same CPU

nvme load times shouldn't be slower than an ssd, if anything they should be the same. There is no benefit to a faster nvme just for gaming really. But there also shouldn't be a negative

Are you using the Samsung Nvme drivers?

Your results look better, I don't know what an amazing score is for Q32T1 - i can find one guy with 660. It can depend on how much free space is used on Drive too. If its empty you get best scores.

the order of the tests is in size of files being tested.
The first two tests run a file that is 1,048,576 bytes in size
the last two run a file that is 4096 bytes

It is much easier for an nvme to move 1 big file a lot of times than lots of small files. Its why the top scores are close to max speed of PCIe3 whereas the bottom 2 look so bad.

So it depends on if game is lots of small files or bigger ones as to if you notice... but ssd should be slower in these tests.

I don't think there is anything wrong with your drive. Its only had 4tb written, mine has been running for 2 years and is on 25.3tb written but it is C drive.

 

[0nion1999]

I was using the Microsoft drivers, I've now installed the Samsung NVMe drivers which seems to have given it a small boost, however it might just be the inconsistency. Drive temperature peaked at 71°C

https://imgur.com/tjkDspo

View: https://imgur.com/tjkDspo


I will keep a check on speeds & temps, if it is because of thermal throttling, I guess I'll have to get a heatsink for it.

 

[Colif]

max temp for the drive is 80c so you not going that bad.
Looking at hd Sentinal, the temp I got today was highest its recorded on my nvme - 66c
its average temp range is 52 to 56 most days. lowest its ever had was 17c but that was after starting pc in middle of winter

I am using heatsink that came with my Motherboard

These are meant to be good - https://www.ekwb.com/shop/ek-m-2-nvme-heatsink-black

 

[0nion1999]

OK so I've just been gaming for a good while, closed it all down so my PC is idling for a few minutes, then tested it again.

https://imgur.com/0QkzPJ1

View: https://imgur.com/0QkzPJ1


Instead of being in the thousands, it wont even break 100MB/s. As you can see it wasn't the thermals... I don't get it...

 

[Colif]

try running this and see what scores you get
https://www.techspot.com/downloads/6014-as-ssd-benchmark.html
just another way to look at drive

As-SSD has 4 fields to help you gauge your drives performance.

1: Sequential is your drive's fastest transfer rates. This applies to sequential files which get written/read in whole. (500+ is common for modern Sata SSD's)

2: 4k. This is basically random access performance and is the slowest for drives. (score above 35read/80write are typical)

3: 4k-64 is random reads with 64 other requests waiting to process. Since 4k is slow a fast SSD can service multiple requests at the same time, this also tests the SSD's flash controllers performance as well as the flash memory itself. Common speed is roughly about 70-75% of sequential.

4: Acc Time. This is Access Time. How long it took the drive to be ready with the request. Read would be the drive ready to start sending the data and write would be the drive starting to take the data from the PC.

I don't know why its dropped. Could be windows decided to run some process mid test. It doesn't always do what you want.

there is this but not sure it applies to read speeds - https://www.techspot.com/news/90998-samsung-swapping-parts-their-970-evo-plus-ssds.html = or how many drives were effected or if any 2tb drives are involved. It may have happened before 2tb drive was released.

 

[0nion1999]

Here is the results from the new test:

https://imgur.com/AebrKzg

View: https://imgur.com/AebrKzg


This was done ~1hr 45 after the last test, I'm not sure that would be windows still running something. Especially as it was doing the same thing yesterday too.

 

[Colif]

I believe the 4k represents how fast the ssd can respond to 1 request of a 4k chunk of data.

4k-64thrd represents 64 simultaneous requests of 4k chunks of data.

https://www.overclock.net/threads/understanding-4k-and-4k-64thrd.1173805/

There shouldn't be such a dramatic difference between our PC. I have an X570 motherboard but the Nvme runs at pcie3 speed regardless of fact its in a pcie 4 slot, so that shouldn't be reason

Do you have latest BIOS on motherboard?
try updating chipset drivers - https://www.amd.com/en/support/chipsets/amd-socket-am4/b450 (you don't need raid or stormi)

 

[0nion1999]

I've now updated my chipset drivers, and my BIOS.

I still get the same low results.

 

[falcon291]

Same disk, taken when the disk was almost empty right after the installation:


------------------------------------------------------------------------------
CrystalDiskMark 8.0.4 x64 (C) 2007-2021 hiyohiyo
Crystal Dew World: https://crystalmark.info/
------------------------------------------------------------------------------

• MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
• KB = 1000 bytes, KiB = 1024 bytes

[Read]
SEQ 1MiB (Q= 8, T= 1): 3454.292 MB/s [ 3294.3 IOPS] < 2421.64 us>
SEQ 128KiB (Q= 32, T= 1): 3496.182 MB/s [ 26673.8 IOPS] < 1161.74 us>
RND 4KiB (Q= 32, T=16): 1644.083 MB/s [ 401387.5 IOPS] < 1234.57 us>
RND 4KiB (Q= 1, T= 1): 46.619 MB/s [ 11381.6 IOPS] < 87.69 us>

[Write]
SEQ 1MiB (Q= 8, T= 1): 3067.548 MB/s [ 2925.4 IOPS] < 2727.23 us>
SEQ 128KiB (Q= 32, T= 1): 3077.848 MB/s [ 23482.1 IOPS] < 1343.92 us>
RND 4KiB (Q= 32, T=16): 882.227 MB/s [ 215387.5 IOPS] < 2300.79 us>
RND 4KiB (Q= 1, T= 1): 154.746 MB/s [ 37779.8 IOPS] < 26.32 us>

Profile: Default
Test: 1 GiB (x5) [C: 9% (173/1862GiB)]
Mode: [Admin]
Time: Measure 5 sec / Interval 5 sec
Date: 2021/11/28 17:52:42
OS: Windows 11 Professional [10.0 Build 22000] (x64)

Three points to note:

1. It matters a lot if the disk is empty or not.
2. The temperature is also important, so if you keep testing and testing, the result will get worse.
3. It is strange that that M2 drive results fluctate a lot. HDD results so SSD results are a lot more consistent.

My results seem consistent with Colif. Please be sure that you are doing the same tests, your tests do not seem the same.

 

[0nion1999]

Turned on PC, ran CrystalDiskMark only RND4K Q32T1 until I got the first result, which was a good result (2000+).
Launched a game for a few minutes, closed everything down fully.

Drive temperature before test: 44°C.

------------------------------------------------------------------------------
CrystalDiskMark 8.0.4 x64 (C) 2007-2021 hiyohiyo
Crystal Dew World: https://crystalmark.info/
------------------------------------------------------------------------------

• MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
• KB = 1000 bytes, KiB = 1024 bytes

[Read]
SEQ 1MiB (Q= 8, T= 1): 3199.682 MB/s [ 3051.5 IOPS] < 2293.02 us>
SEQ 1MiB (Q= 1, T= 1): 3029.612 MB/s [ 2889.3 IOPS] < 345.90 us>
RND 4KiB (Q= 32, T=16): 53.850 MB/s [ 13147.0 IOPS] < 38705.65 us>
RND 4KiB (Q= 1, T= 1): 50.238 MB/s [ 12265.1 IOPS] < 81.41 us>

[Write]
SEQ 1MiB (Q= 8, T= 1): 3330.934 MB/s [ 3176.6 IOPS] < 2353.91 us>
SEQ 1MiB (Q= 1, T= 1): 3180.135 MB/s [ 3032.8 IOPS] < 329.51 us>
RND 4KiB (Q= 32, T=16): 290.391 MB/s [ 70896.2 IOPS] < 7211.98 us>
RND 4KiB (Q= 1, T= 1): 201.043 MB/s [ 49082.8 IOPS] < 20.24 us>

Profile: Default
Test: 1 GiB (x5) [G: 49% (908/1863GiB)]
Mode: [Admin]
Time: Measure 5 sec / Interval 5 sec
Date: 2022/10/18 17:52:28
OS: Windows 10 Professional [10.0 Build 19044] (x64)

Peak temp during test 69°C.

I can see some parameters are different, even though that's not the one that's the problem, I ran a second test with exactly the same parameters just to be sure:

Temperature before test: 40°C.

------------------------------------------------------------------------------
CrystalDiskMark 8.0.4 x64 (C) 2007-2021 hiyohiyo
Crystal Dew World: https://crystalmark.info/
------------------------------------------------------------------------------

• MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
• KB = 1000 bytes, KiB = 1024 bytes

[Read]
SEQ 1MiB (Q= 8, T= 1): 3201.493 MB/s [ 3053.2 IOPS] < 2291.78 us>
SEQ 128KiB (Q= 32, T= 1): 2595.063 MB/s [ 19798.8 IOPS] < 1565.22 us>
RND 4KiB (Q= 32, T=16): 51.700 MB/s [ 12622.1 IOPS] < 40293.24 us>
RND 4KiB (Q= 1, T= 1): 48.349 MB/s [ 11804.0 IOPS] < 84.59 us>

[Write]
SEQ 1MiB (Q= 8, T= 1): 3326.482 MB/s [ 3172.4 IOPS] < 2203.21 us>
SEQ 128KiB (Q= 32, T= 1): 2645.700 MB/s [ 20185.1 IOPS] < 1535.12 us>
RND 4KiB (Q= 32, T=16): 290.532 MB/s [ 70930.7 IOPS] < 7207.58 us>
RND 4KiB (Q= 1, T= 1): 200.348 MB/s [ 48913.1 IOPS] < 20.33 us>

Profile: Default
Test: 1 GiB (x5) [G: 49% (908/1863GiB)]
Mode: [Admin]
Time: Measure 5 sec / Interval 5 sec
Date: 2022/10/18 18:07:13
OS: Windows 10 Professional [10.0 Build 19044] (x64)

Peak temperature during test: 65°C.

 

[Colif]

you didn't ask but this shows why 4k writing is faster than reading - https://superuser.com/questions/1168014/nvme-ssd-why-is-4k-writing-faster-than-reading

In the Read version of the test the drive controller, now recognising that it is under very constant heavy load, stops preloading data, possibly avoids the buffer and instead switches to a "raw" read mode, again approaching the sequential read speed.

Basically the drive controller can do something to make a 4K write more efficient, especially if a cluster of them arrive at a similar time, while it can't do anything to make a single 4K read more efficient, especially if it is trying to optimise dataflow by pre-loading data into the cache.

from same link

it appears your drives ability to read really small files (which is majority of files) is


you could ask on here, not sure how far you get though - https://us.community.samsung.com/t5...g-monitors-and-memory/label-name/970 evo plus