SSD CrystalDiskMark Benchmark sub-thread

David_Schroth said:
At least they didn't have a pair of Teamgroup's 120mm SSD AIOs hooked up?
Click to expand...
So, so, true. Man, the things I'm already blocking out of my head. I'd already forgotten about those. the DUO right? Yep, it's Gen4 for me, for a while to come.
 
Ran some speed test numbers on some enterprise NVME 4 SSD's direct to disk.

Note I ran 16Gib tests just because I wanted to remove the caching advantage the SSD Raid had in this config. 8 gig cache.

1714425524966.png

Now some of these may be because the software controller in the Dell BIOS is trash. For one... no Raid 5 support.

And looking at the results raid 10 across 6 disks isn't much faster than raid 0 across 2. Plenty of PCIE Lane's to go around.

Will be replacing this with a system with dual raid controllers to sperate out the SSD and the NVME into distinct hardware raid controllers.

SO this is what 20k gets you. ;)
 
Some notes... NVME Raid 1 by the OS as opposed to motherboard is MUCH faster... for the sequential, pretty in line for the random read and writes.
 
MadMummy76 said:
CORSAIR MP600 PRO NH | 8TB | PCIe Gen 4 x4* | TLC

View attachment 2476

*It's tested in a X299 Gen 3.0 system, so sequential is bottlenecked due to that.
Click to expand...
Lets not forget you're probably exceeding the consumer cache in the drive with a 4 gig test block. Most I know of cap out at 1 gig some 2 gig.
 
Samsung 980 Pro W heatsink on current system NVME 4 Ryzen 9 5900x x570 chipset.

1742178177690.png

single disk in system non quiesced.
 
It's the bump in random write speeds that has me scratching my head the most.
 
It helps if you switch CrystalDiskmark to do NVME speed testing... sadly I don't have my 980's results done correctly.

1742494754275.png
 
Grimlakin said:
It helps if you switch CrystalDiskmark to do NVME speed testing... sadly I don't have my 980's results done correctly.

View attachment 3643
Click to expand...
Wow.

I never noticed there was a separate mode for NVMe either.

I'm going to have to check that out.

The high queue depth RND4K (Q32T16) read numbers for this drive are beastly, but the low queue depth RND4k Q1T1 appear somewhat stagnant. They have been at 85MB/s or thereabouts for 2-3 generations now.

It has long been my theory that the RND4K Q1T1 are what make the Optane Drives so amazing, as they have them and no other drives do, and they tend to perform in a class of their own when it comes to responsiveness, but I have never tested anything with this kind of massive higher queue depth random numbers, so I am really curious how they compare.
 
Zarathustra said:
Wow.

I never noticed there was a separate mode for NVMe either.

I'm going to have to check that out.

The high queue depth RND4K (Q32T16) read numbers for this drive are beastly, but the low queue depth RND4k Q1T1 appear somewhat stagnant. They have been at 85MB/s or thereabouts for 2-3 generations now.

It has long been my theory that the RND4K Q1T1 are what make the Optane Drives so amazing, as they have them and no other drives do, and they tend to perform in a class of their own when it comes to responsiveness, but I have never tested anything with this kind of massive higher queue depth random numbers, so I am really curious how they compare.
Click to expand...
Just looking at the pure NVME testing it's clear that the drive has more bandwidth than even a single thread can saturate. I'm tempted to try a test with random 4k Q64T32 just to see if it's even MORE insane.

Test done.... not worth it.
 
Grimlakin said:
Just looking at the pure NVME testing it's clear that the drive has more bandwidth than even a single thread can saturate. I'm tempted to try a test with random 4k Q64T32 just to see if it's even MORE insane.

Test done.... not worth it.
Click to expand...

yeah, the RND4k Q1T1 tests are usually very limited by CPU core speed. Stick an Optane in an older system, and it can't really take advantage of it. Even in my Threadripper 3960x, I can't fully take advantage of all of the performance of the Optane DC p5800x in 4kRND Q1T1.
 
Zarathustra said:
yeah, the RND4k Q1T1 tests are usually very limited by CPU core speed. Stick an Optane in an older system, and it can't really take advantage of it. Even in my Threadripper 3960x, I can't fully take advantage of all of the performance of the Optane DC p5800x in 4kRND Q1T1.
Click to expand...
Well when I did the test with 32 threads and a queue depth of 64 I was pretty much using 80-90% of a 9950x3d.
 
You must log in or register to reply here.
Become a Patron!
Back
Top