[SOLVED] M.2 NVMe SSD shopping: Is there more to speed than just read and write speeds?

Feb 8, 2020
4
0
10
Hi there everyone
I'm currently working on my dream build PC and want it to be reasonably fast. (Ideal scenario is able to record guitar and have a few effects applied without causing noticeable delay to output speakers)
I'm looking around on newegg to get a sense of things and thought that looking for higher read and write speeds would help with speed performance but looking at the Seagate BarraCuda 510
which has listed specs as
Max Sequential Read - Up to 3400 MBps

Max Sequential Write - Up to 3000 MBps

but there are multiple reviews saying that it isn't so great performance wise compared to the SAMSUNG 970 EVO
but looking at the specs for that it says
Max Sequential Read - Up to 3400 MBps

Max Sequential Write - Up to 2500 MBps

which is only the same for read speed and even slower in the write speed.
What am I missing here?
Wisdom tells me to go with user reviews, but logic tells me to go for the higher listed specs. I'm hoping someone can help me find the logic behind their comments and resolve the battle.
 
Solution
you can trust the specs up to a point - up to 3400 may only mean for the 1TB version and not the 250GB for example. it may only occur using the ATTO benchmark vs say AS-SSD.

a fast SSD may slow down significantly if it reaches a steady state from long sustained transfer - that's where things like DRAM cache and controller matter. and all SSD drop in performance as they fill up. a 5-yr wty may matter more than 100MB/s for 30 secs - but it may matter less if you have to RMA 3 or 4 times and have to recover the data.

Oh dang yeah I was a little worried by the "up to" part. (;=~=)
So basically can't trust any of the spec numbers?
Seagate Baracuda M.2 SSD listed "Up to 600,000 IOPS"
compared to
Samsung 970 EVO
"QD32: Up to 500,000...
Looking at an article https://www.gamingpcbuilder.com/best-m-2-nvme-ssd/
12 of the Fastest M.2 PCIe NVMe SSDs in 2020
the Seagate BarraCuda 510 M.2 wasn't even there even though it has a faster listed write speed than 4 of them, a faster read speed than 10 of them and a faster ISOP than 9 of them.
I'm so confused... why do the speeds listed not seem to have any effect on performance?
All my life numbers reliably meant something ...until now. My world is chaos now. (@O@)?
 
Hi there everyone
I'm currently working on my dream build PC and want it to be reasonably fast. (Ideal scenario is able to record guitar and have a few effects applied without causing noticeable delay to output speakers)
I'm looking around on newegg to get a sense of things and thought that looking for higher read and write speeds would help with speed performance but looking at the Seagate BarraCuda 510
which has listed specs as
Max Sequential Read - Up to 3400 MBps

Max Sequential Write - Up to 3000 MBps

but there are multiple reviews saying that it isn't so great performance wise compared to the SAMSUNG 970 EVO
but looking at the specs for that it says
Max Sequential Read - Up to 3400 MBps

Max Sequential Write - Up to 2500 MBps

which is only the same for read speed and even slower in the write speed.
What am I missing here?
Wisdom tells me to go with user reviews, but logic tells me to go for the higher listed specs. I'm hoping someone can help me find the logic behind their comments and resolve the battle.
In real work, you'd never notice difference in speed between those two
 
  • Like
Reactions: T H A L L
Random IOPs matter also. to obtain the sequential speeds you have to be reading from a faster device or working with a good deal of content.

SSDs list "up to" speeds so it's not a sustained value in all cases and capacities

the new NVMes with Phison controllers can be close enough to an 970 not to matter. Sabrent, Inland and Corsair all fit the bill


Looking at an article https://www.gamingpcbuilder.com/best-m-2-nvme-ssd/
12 of the Fastest M.2 PCIe NVMe SSDs in 2020
the Seagate BarraCuda 510 M.2 wasn't even there even though it has a faster listed write speed than 4 of them, a faster read speed than 10 of them and a faster ISOP than 9 of them.
I'm so confused... why do the speeds listed not seem to have any effect on performance?
All my life numbers reliably meant something ...until now. My world is chaos now. (@O@)?
 
Last edited:
  • Like
Reactions: T H A L L
Random IOPs matter also. to obtain the sequential speeds you have to be reading from a faster device or working with a good deal of content.

SSDs list "up to" speeds so it's not a sustained value in all cases and capacities

the new NVMes with Phison controllers can be close enough to an 970 not to matter. Sabrent, Inland and Corsair all fit the bill
Oh dang yeah I was a little worried by the "up to" part. (;=~=)
So basically can't trust any of the spec numbers?
Seagate Baracuda M.2 SSD listed "Up to 600,000 IOPS"
compared to
Samsung 970 EVO
"QD32: Up to 500,000 IOPS
QD1: Up to 15,000 IOPS"

I'm not trying to avoid Samsung, I just thought "why would I pay more for SSD that has lower listed specs?" but now I know the specs are basically useless, I'll go and read through reviews to try and find the best one.
Thanks for your help Mr.Spock (^o^)/ I've almost completed my dream build!
 
I'm not trying to avoid Samsung, I just thought "why would I pay more for SSD that has lower listed specs?" but now I know the specs are basically useless, I'll go and read through reviews to try and find the best one.
Thanks for your help Mr.Spock (^o^)/ I've almost completed my dream build!
Given similar performance, then you look at warranty and reliability.
 
you can trust the specs up to a point - up to 3400 may only mean for the 1TB version and not the 250GB for example. it may only occur using the ATTO benchmark vs say AS-SSD.

a fast SSD may slow down significantly if it reaches a steady state from long sustained transfer - that's where things like DRAM cache and controller matter. and all SSD drop in performance as they fill up. a 5-yr wty may matter more than 100MB/s for 30 secs - but it may matter less if you have to RMA 3 or 4 times and have to recover the data.

Oh dang yeah I was a little worried by the "up to" part. (;=~=)
So basically can't trust any of the spec numbers?
Seagate Baracuda M.2 SSD listed "Up to 600,000 IOPS"
compared to
Samsung 970 EVO
"QD32: Up to 500,000 IOPS
QD1: Up to 15,000 IOPS"

I'm not trying to avoid Samsung, I just thought "why would I pay more for SSD that has lower listed specs?" but now I know the specs are basically useless, I'll go and read through reviews to try and find the best one.
Thanks for your help Mr.Spock (^o^)/ I've almost completed my dream build!
 
Solution
The key thing about SSD speeds is that MB/s is the inverse of how we perceive speed - how much time we have to wait for a disk access operation to complete. Say you need to read 1 GB of data.

125 MB/s hard drive = 8 sec
250 MB/s SATA 2 SSD = 4 sec
500 MB/s SATA 3 SSD = 2 sec
1 GB/s PCIe SSD = 1 sec
2 GB/s NVMe SSD = 0.5 sec
3 GB/s newer NVMe SSD = 0.33 sec

Notice how every time the MB/s speed doubles, the time saved is half that of the previous doubling. The bulk of your time saved comes from the jump from a HDD to a SATA 3 SSD speeds (6 sec). The additional time saved for going to NVMe is tiny in comparison (1.5 to 1.67 sec).

On top of this, small file speeds are still stuck down at around 35-200 MB/s even on NVMe SSDs. So the SATA interface isn't a bottleneck. Say you need to read 1 GB of sequential data and 200 MB of small file data. Which is faster? A NVMe SSD with 3 GB/s sequential speeds and 40 MB/s small file speeds? Or a SATA SSD with 500 MB/s sequential speeds and 70 MB/s small file speeds?

NVMe: (1 GB) / (3 GB/s) + (200 MB) / (40 MB/s) = 0.33 sec + 5 sec = 5.33 sec
SATA: (1 GB) / (500 MB/s) + (200 MB) / (70 MB/s) = 2 sec + 2.87 sec = 4.87 sec

Surprise! The SATA SSD is faster despite there being 5x more sequential data than small file data, the NVMe drive having 600% higher sequential speeds, and the SATA drive only having 75% higher small file speeds. Because small file speeds take more time than sequential file speeds, they will make up the majority of your wait time even if the amount of small file data is a lot less than the sequential data. This is why you can't tell the difference between a M.2 NVMe SSD and a SATA SSD in real-world use most of the time. (The exception is if you're doing something completely reliant on sequential file read/writes, like real-time video editing. Then the higher sequential speeds matter.)

So what you should look for in order of importance:
  1. Small file read/write speeds. (Which is usually synonymous with high IOPS). Ideally 50+ MB/s reads, 100+ MB/s writes.
  2. Small file read/write speeds
  3. Small file read/write speeds
  4. Small file read/write speeds
  5. Sequential speeds
In fact I usually tell people to just ignore the sequential speeds. The 970 EVO is better because it scores about 50-55 MB/s 4k reads, 170-220 MB/s 4k writes. The Barracuda 510 only scores only 45 MB/s 4k reads, 130 MB/s 4k writes.