Solved Enlightening comparison between NVMe and SATA SSD

Wisewiz · Mar 18, 2023

My OS and programs are on the disk that's imaged here. In the first report, Macrium is reading from NVMe and writing to SATA. In the second shot, Macrium is reading from SATA and writing to SATA. The evidence is that the speed of the slowest element determines the overall speed (like what happens when you mix slow and fast memory on the same machine), but note the difference between times taken between shot 1 and shot 2. Insignificant, because the write-to drive is the SATA on both, whereas the read-from disk is NVMe in shot 1, and SATA SSD in shot 2.

Shot 1:

Shot 2:

Interesting, eh?

cereberus · Mar 18, 2023

Wisewiz said:
My OS and programs are on the disk that's imaged here. In the first report, Macrium is reading from NVMe and writing to SATA. In the second shot, Macrium is reading from SATA and writing to SATA. The evidence is that the speed of the slowest element determines the overall speed (like what happens when you mix slow and fast memory on the same machine), but note the difference between times taken between shot 1 and shot 2. Insignificant, because the write-to drive is the SATA on both, whereas the read-from disk is NVMe in shot 1, and SATA SSD in shot 2.

Shot 1:
View attachment 55685
Shot 2:
View attachment 55686
Interesting, eh?

This is exactly as I would expect - the read/write figures are a bit meaningless as they are peak rates not an average rate.

So long as pc can read faster than it writes, the time is limited by the writing device. Think about it - if it reads data faster with nvme, all that happens is data gets held up in a queue until it is processed.

Also bear in mind there is a "compression" overhead that limits the write speeds as well.

What is interesting is if you repeat the test with no compression.
I was really gobsmacked by that on my laptop (nvme to nvme)!

Jamis · Mar 18, 2023

I got an eye opener with my new system. The old system was a W10, i5 3 Ghz 4 core with 6 Gb of RAM and a 1 TB HDD for the system drive. A Macrium Reflect full system image would run for 3 hours to a 2.5" USB SATA HDD connected to the old USB 2.0 type A port.
New system is W11, i7 2.1-4.9 Ghz 12 core with 16 Gb of RAM and 1 TB SSD system drive. A MR full system image runs in 23 minutes to the same USB HDD drive connected to a 3.1 type A USB port.
I figured it would be faster, just didn't expect that sort of faster.
Can't wait to see how long Norton's full system scan takes. On the old system, it took 12 1/2 hours to complete.

cereberus · Mar 20, 2023

Jamis said:
I got an eye opener with my new system. The old system was a W10, i5 3 Ghz 4 core with 6 Gb of RAM and a 1 TB HDD for the system drive. A Macrium Reflect full system image would run for 3 hours to a 2.5" USB SATA HDD connected to the old USB 2.0 type A port.
New system is W11, i7 2.1-4.9 Ghz 12 core with 16 Gb of RAM and 1 TB SSD system drive. A MR full system image runs in 23 minutes to the same USB HDD drive connected to a 3.1 type A USB port.
I figured it would be faster, just didn't expect that sort of faster.
Can't wait to see how long Norton's full system scan takes. On the old system, it took 12 1/2 hours to complete.

The single biggest improvement here is usb 2 versus usb 3.1.

Usb 2 is limited to max 60 MB/s, USB 3.1 will do a maximum of 1,250 MB/s.

In practice, you get about half those speeds at best. With a weaker CPU and limited RAM, you proably struggle to get more that 20MB/s. A usb 3.1 port will be around 20 x faster. At this point, you become limited by the write speed of hdd (probably around 100-120 MB/s).

Coupled with extra RAM, CPU power, a difference of approx 8:1 ratio is quite believable, although the numbers do indicate original pc was running slower than you might expect.

Wisewiz · Mar 20, 2023

cereberus said:
So long as pc can read faster than it writes, the time is limited by the writing device.

Yeah, so I'm putting in an NVMe as writing device. We'll see. I'm not sure it will reduce the total time used by MR for this imaging task.

Jamis · Mar 20, 2023

cereberus: Yes, I do believe the old system had an issue. Task Manager would show the C drive go 100% active but with no read/write/access figures for a long time. None of the usual diagnostics showed anything wrong with the drive and reloading the OS had no effect on the intermittent performance issue. After "flogging the dead 14 year old horse" with no effect for two weeks, I replaced it with a current one. Thanks for your analysis. Apologies for the bold text, can't get the iPad text editor to change it.

cereberus · Mar 20, 2023

Wisewiz said:
Yeah, so I'm putting in an NVMe as writing device. We'll see. I'm not sure it will reduce the total time used by MR for this imaging task.

Pretty sure it will

.

Bree · Mar 20, 2023

Jamis said:
Apologies for the bold text, can't get the iPad text editor to change it.

In the editor box click the [ ] (Toggle BB code) icon at the top right.
This greys out all the formatting icons, and anything you type will be posted as plain text.

Jamis · Mar 20, 2023

Bree said:
In the editor box click the [ ] (Toggle BB code) icon at the top right.
This greys out all the formatting icons, and anything you type will be posted as plain text.

View attachment 55861

Ah! Once again, Bree to the rescue!

SIW2 · Mar 20, 2023

It depends on many things, including the size of the files being copied. One large file nvme>nvmne should be much quicker than sata ssd >sata ssd. With many small files, the picture might not be as dramatic.

people often look at the top two rows of diskmark, in practice for most purposes, the bottom rows are more relevant.

I had a look at an nvme attached by pcie x1 adapter. In general use the difference between attaching via x4 adapter and x1 adapter wasn't obvious.

That linus youtube channel did an experiment where a few of his tech minded buddies blind tested pcs with different types of ssd. They all got it wrong, they couldn't tell which was which in general use.

Sir_George · Mar 20, 2023

Bree said:
In the editor box click the [ ] (Toggle BB code) icon at the top right.
This greys out all the formatting icons, and anything you type will be posted as plain text.

View attachment 55861

That was a beneficial piece of information for me. Thanks!

Wisewiz · Mar 20, 2023

SIW2 said:
people often look at the top two rows of diskmark, in practice for most purposes, the bottom rows are more relevant.

I don't understand that claim. What makes them more relevant than the top two rows?

SIW2 said:
I had a look at an nvme attached by pcie x1 adapter. In general use the difference between attaching via x4 adapter and x1 adapter wasn't obvious.

Mine's attached to an x16 (x4 negotiable), and its speeds are right up to the claims of Samsung for it. The "write-to" (target) NVMe I'm going to install will be a standard installation on the mobo. Just waiting for a chunk of Lenovo plastic that will make the mounting easier.

SIW2 · Mar 20, 2023

because mostly one is reading and writing small files. Windows writes small files constantly. The big figures in the top rows might be relevant when copying massive files, but most people do that rarely.

That linus youtube channel did an experiment where a few of his tech minded buddies blind tested pcs with different types of ssd, iirc it was a pcie3 nvme, a pcie4 nvme and a sata ssd. They all got it wrong, they couldn't tell which was which in general use. They probably would have been able to tell if they had attempted very large file copies. They even booted into windows on each of the pcs and still guessed it wrong.

It is a while since I saw that video and I cant remember the precise details, but I wasn't surprised.

Those fast test bench speeds in reviews are important for professional video editors perhaps, but not all that relevant for the average punter

Wisewiz · Mar 20, 2023

@SIW2

Thanks for the instruction. I hadn't thought about it that way, but I believe you're absolutely right.

And you said, "It is a while since I saw that video and I cant remember the precise details, but I wasn't surprised."

C'mon, now. NOTHING surprises you any more.