SSD CrystalDiskMark Benchmark sub-thread

Peter_Brosdahl

Moderator
Staff member
Joined
May 28, 2019
Messages
7,960
Points
113
Hi, all,

This as the title says is a subthread for various SSD benchmarks. It doesn't matter if its Gen3, 4, or 5, and whatever comes after, so feel free to post your results here. A number of us are currently focusing on the 4KRND metrics as those could have a more meaningful impact on gaming.
 
My main OS drive a 960 EVO 500GB
 

Attachments

  • CrystalDiskMark_960 EVO 500GB.png
    CrystalDiskMark_960 EVO 500GB.png
    32.7 KB · Views: 1
Here's some data from my rig. This first post will be my NVMe drive: Sabrent Rocket 4 on PCI 4 slot (Asus B550 platform)

First - I think it's somewhat important to post some concrete drive data. I picked a random program (HDDScan) so it can show actual model number and FW revisions. There's probably a better tool for this, but I think this shows the relevant data.

1673549652336.png

This is with my normal third-party cache enabled (PrimoCache)

1673549720480.png


Same drive, a couple minutes later, with cache disabled

1673550052703.png

I should have done that second run w/o cache after a fresh reboot, not sure if it would affect the numbers, but didn't think about it until after it was almost done.
 
Last edited:
Second drive, a SATA3 Samsung 870 EVO

1673550536136.png

With the cache

1673551211619.png

Without cache running

1673550891084.png
 
Last drive, a WDC 10T spinner

1673551272685.png

With cache:

1673551559360.png

Without cache:

1673551949318.png
 
My takeaway:

Cache is effective, as long as you are just caring about Read (which, as a typical gamer/consumer, is 90% of what I do), and what you are doing can fit inside the cache. There is an option to enable write caching, but I do not use it as I fear about data loss situations, and I don't do a lot of writes the disk anyway.

The cache effectively makes it so Read is just testing the RAM cache moreso than the drive -- there's probably a way to tweak the test to account for that and run the cache out, but I'm just using out-of-the-box settings on CrystalDisk, and that's why I did the second run with the cache disabled so you can see the drive on it's own. Kind of neat seeing a platter drive come in about the same as a PCI 4 NMVe, but I do realize that it isn't actually hitting the drive, it's just the cache.

In real world use, I can say: if I play a game from the spinner; the first time it will be sluggish. Times after that, much better. Not quite as good as SSD - as the cache isn't 100% effective in all cases, but pretty darn close. I was surprised that it was even effectively caching the PCI 4 drive with some good effect.

Other than that, with the cache disabled the drives should more or less come in similarly to the same drives on other platforms - I don't think there are any surprises.
 
Here's a quick run for my Crucial P5 Plus m.2 PCIe 4.0 2TB that is on a Gen4 slot. I got this on sale sometime last summer for just under $200 and I'm pretty happy with it. Not the fastest but not horrible either.

1673879214071.png


and here are the scores for the Sabrent m.2 1 TB drive. It is a Gen4 drive but in a Gen3 slot. I do find it interesting in how its 4K scores traded blows with the Gen4 slot and was even faster on one test.

1673879573993.png
 
So if we're going to build a system to do m.2 nvme testing... what motherboard offers the most m.2 slots at the highest pcie rating?

I'm thinking a server board that supports nvme blades that can house stand alone m.2 drives. Issue would be cooler heights for drives that come with a heat sink.

Maybe a high end thread ripper or xeon desktop motherboard? A workstation class system so it will have all of the needed direct pcie lanes?
 
If anyone's interested, I tested the drives currently in my system:

ADATA SX8200 Pro | 2TB | PCIe Gen 3 x4 | TLC

adata_CrystalDiskMark_20230206104133.png

Crucial P3 | 4TB | PCIe Gen 3 x4 | QLC

crucial_CrystalDiskMark_20230206104726.png

Intel 660p | 1TB | PCIe Gen 3 x4 | QLC

Intel_CrystalDiskMark_20230206103524.png
 
Since I was finally able to get a couple of Samsung's flagship 980 Pro's on sale for the new build I'll try to get around to posting my scores for them this week. The motherboard I'm using is a budget ($119) board so there could be some limitations there.

What I can tell you is that I used a Sabrent SATA III>USB 3 adapter so I could do a quick transfer from an old Toshiba OCZ 1 TB SSD to my game drive that's installed in the Gen4 slot and it breezed through nearly 500 GB in under an hour at roughly 400 MB/s. That just happened to be the limit of the read speed, task manager showed the source drive pegged at 100% and I think the Samsung was something like 40% write. While this was happening I even had GOG installing CB 2077 (64 GB) and it was managing around a gigabit (I know I'm as shocked as anyone that GOG did that).
 
Task manager is not reliable, it often says 100% disk usage, when it's not even near that in my experience.
I've wondered about that. I also noticed something interesting last night while installing more games. Read speeds by one game launcher (didn't matter which between Steam, Epic, GOG, Rockstar) would top at ~470 MB/s and the drive can obviously do more. As an experiment I opened a 2nd launcher to verify a different install and it too topped at that bringing the combined total to nearly 1000 MB/s. Seem's like something is capping per app. I'll have to look more into it.

I'm almost done with my installs and then I'll run CrystalDiskMark and see what it says.
 
I've wondered about that. I also noticed something interesting last night while installing more games. Read speeds by one game launcher (didn't matter which between Steam, Epic, GOG, Rockstar) would top at ~470 MB/s and the drive can obviously do more. As an experiment I opened a 2nd launcher to verify a different install and it too topped at that bringing the combined total to nearly 1000 MB/s. Seem's like something is capping per app. I'll have to look more into it.

I'm almost done with my installs and then I'll run CrystalDiskMark and see what it says.
That's interesting... is it an OS throttle or something else?
 
That's interesting... is it an OS throttle or something else?
That's what I'm not sure about. I'm going to google it later to find out. It was definitely being throttled per app and at consistent speeds though, 460-470 MB/s. At first, I didn't think much of it but when I had multiple running at the same time, doing the same thing, it really got me curious. I'm also going to check the BIOS to see if there's anything there that needs to be turned on but I kind of doubt it.
 
That's what I'm not sure about. I'm going to google it later to find out. It was definitely being throttled per app and at consistent speeds though, 460-470 MB/s. At first, I didn't think much of it but when I had multiple running at the same time, doing the same thing, it really got me curious. I'm also going to check the BIOS to see if there's anything there that needs to be turned on but I kind of doubt it.
I'll say this a few minutes of google fu turned up nothing.... maybe I'll see if Bard AI can help. hummm...

Nope BARD can not help with this one. LOL.
 
Lol, there's an old thread from 2021 on [H] about this, but only with Epic and they hit a higher cap than me, whereas I saw it happen with Steam, Epic, GOG, and Rockstar. Some folks there got confused that the OP was talking about download speeds but they were actually talking about the read/write speeds during the install. I'm literally speaking about the read speeds during verification. Another person posted at the end of the thread about doing a test file transfer and saw the same thing but with speeds closer to what I was seeing.

 
So here's my scores for the 980 Pro's. The first is for the 2TB drive in the Gen4 slot that I use for games and just another reminder that this is on a budget 550 board.

1684282980103.png


and here's the 1 TB OS drive in the Gen3 slot. Still impressive other than the Seq read/write.

1684282651050.png

So I guess the bottom line take away is what we already know. Synthetic benchmarks don't always equate to all real-world scenarios. At some point I'll do more testing in transferring files between the drives but that'll be another day.
 
Sorry about the blurrycam - appears it focused on the dudes in the background and not the benchies... Is Raid 0 fair? This is from the MSI booth at Computex.


20230531_162303.jpg20230531_162300.jpg
 
Become a Patron!
Back
Top