SSD SATA Recommendation

[Dann Corbit]

If you have a really fast CPU it matters a lot to use the m.2 drives.
My 3970x was definitely bottlenecked by using a 5MB/Sec thumb drive.
The 1 TB thumb drive would saturate at 100% and the NPS rate would go to 10% of the full NPS. Changing to ultra-fast m.2 SSDs I see no slowdown whatsoever and at over 100MB/sec in TB heavy positions, the drive is nowhere close to saturated.
I guess if you have a slower CPU you can buy a cheaper drive and it will not matter much.

I've seen my M.2 PCIe Ver.4 drive sustain 3.5Gb/sec for many minutes as a result of massive TB requests by the CPU. The drive is rated at 750,000 IOPS and 5Gb/sec. Even at this very fast pace it slows the NPS down by about 30% or so (90Mnps reduced to 60Mnps with heavy TB requests). The disk que was between 3 and 10 requests long (mostly around 5). In order to have as little effect as possible a 3 or 4 drive raid array with 10-15Gb/sec and ~3,000,000 IOPS sustained is likely needed. This is my next upgrade!

Regards,

Zenmastur

Because my machine has a PCIE 4.0 bus, I bought 4 of the new M.2 ssds and put them on a card. The card was inexpensive, and also easier to install than pulling out my two GPUs (WHY did they put the m.2 slots in such an inaccessible location? And the other one is even worse -- on the other side of the motherboard). Of course, the cost is the catch, with the 4 x 2 TB SSD gumsticks costing more than my entire 120 TB jbod disk array.

[Zenmastur]

If you have a really fast CPU it matters a lot to use the m.2 drives.
My 3970x was definitely bottlenecked by using a 5MB/Sec thumb drive.
The 1 TB thumb drive would saturate at 100% and the NPS rate would go to 10% of the full NPS. Changing to ultra-fast m.2 SSDs I see no slowdown whatsoever and at over 100MB/sec in TB heavy positions, the drive is nowhere close to saturated.
I guess if you have a slower CPU you can buy a cheaper drive and it will not matter much.

I've seen my M.2 PCIe Ver.4 drive sustain 3.5Gb/sec for many minutes as a result of massive TB requests by the CPU. The drive is rated at 750,000 IOPS and 5Gb/sec. Even at this very fast pace it slows the NPS down by about 30% or so (90Mnps reduced to 60Mnps with heavy TB requests). The disk que was between 3 and 10 requests long (mostly around 5). In order to have as little effect as possible a 3 or 4 drive raid array with 10-15Gb/sec and ~3,000,000 IOPS sustained is likely needed. This is my next upgrade!

Regards,

Zenmastur

Because my machine has a PCIE 4.0 bus, I bought 4 of the new M.2 ssds and put them on a card. The card was inexpensive, and also easier to install than pulling out my two GPUs (WHY did they put the m.2 slots in such an inaccessible location? And the other one is even worse -- on the other side of the motherboard). Of course, the cost is the catch, with the 4 x 2 TB SSD gumsticks costing more than my entire 120 TB jbod disk array.

Have you tested the performance with one of the benchmarks?

[Dann Corbit]

Yes, the SSDs measure very close to their performance spec.
I have an odd mix because I have (on the card) one Gigabyte and 3 Sabrent SSDs (all 2 TB). The Sabrent's were a little cheaper and almost as fast. I was worried that the odd mix might cause problems but I have not seen any.
I also have a 1 TB Gigabyte on the motherboard as my system disk, and standard rotating disk as well.

[Zenmastur]

a 3 or 4 drive raid array with 10-15Gb/sec and ~3,000,000 IOPS sustained is likely needed

so actually we are talking about software RAID-0 (under Windows it will be 99% built-in raid support, Aorus just GUI front-end)

Here they tested Samsung 983 DCT (1TB, NVME PCIE x4 version)

https://www.pugetsystems.com/labs/artic ... -Pro-1369/
https://www.samsung.com/semiconductor/m ... er/983dct/

Standalone Samsung SSD has 400-440 kIOPS (4KB, QD32)

Under software RAID IOPS degraded with some hardware to as low as 45...75k IOPS and best result peaked at 152k IOPS ( X299 chipset + ASUS HYPER M.2 x16 card)

That is less than 35% of standalone perfomance (if you are lucky with chipset).

And we do not mention yet CPU/RAM load during this operations.

You'll notice that Puget sounds test uses the Samsung 983 DCT 1TB M.2 SSD which is a PCIe V.3 drive. It's controller is only capable of 400,000 IOPS and it's write performance is terrible. Even with these disadvantages many of the tests show near 10Gb/sec read. The write performance suffers due to the very slow write performance of the drives used.

Just like I said, older and/or slower drives can still reach 10Gb/sec during reads.

If you substitute Sabrent Rocket NVMe PCIe Gen 4.0 M.2 SSD for the drives used in the test it's VERY easy to imagine read speeds around 15Gb/sec.

A quick review of the Sabrent drives is here:

They had trouble getting good write performance in this review. I've tested mine and I have no problem with consistent write performance or reaching advertised speed.

So, my previous statements seem to be justified.

Regards,

Zenmastur

[Zenmastur]

Yes, the SSDs measure very close to their performance spec.
I have an odd mix because I have (on the card) one Gigabyte and 3 Sabrent SSDs (all 2 TB). The Sabrent's were a little cheaper and almost as fast. I was worried that the odd mix might cause problems but I have not seen any.
I also have a 1 TB Gigabyte on the motherboard as my system disk, and standard rotating disk as well.

Which Adapter card did you buy?

[MikeB]

I need a recommendation for an internal 2TB SATA SSD.

I'm trying to consolidate all my EGTBs which I currently have on a 256Mb USB3 flash drive. I can't fit the Nav 6 man files on it though, but in total, all the EGTBs I have take up about 1.2TB

I saw this unit on Amazon (cheapest)
Crucial BX500 2TB 3D NAND SATA 2.5-Inch Internal SSD - CT2000BX500SSD1

I don't have much experience with this type of device.
Any recommendations?

This is what you want at relatively low cost- if you can get it to work in your setup. 2TB model is only $259.

[yurikvelo]

Even with these disadvantages many of the tests show

Tests show expected behavior : IOPS of software RAID-0 degrade with number of array members.
4 SSD are 4x times slower than any individual SSD
If you put 2 SSD into garbage bin, speed will increase twice

not a rocket science, but take 5 stages: denial, anger, bargaining, depression and acceptance

[Dann Corbit]

Yes, the SSDs measure very close to their performance spec.
I have an odd mix because I have (on the card) one Gigabyte and 3 Sabrent SSDs (all 2 TB). The Sabrent's were a little cheaper and almost as fast. I was worried that the odd mix might cause problems but I have not seen any.
I also have a 1 TB Gigabyte on the motherboard as my system disk, and standard rotating disk as well.

Which Adapter card did you buy?

I don't remember the name of it but I made sure it was PCIE 4.0. There was a similar one by the same manufacturer that was only PCIE 3, which would not have worked nearly so well.

[Dann Corbit]

Even with these disadvantages many of the tests show

Tests show expected behavior : IOPS of software RAID-0 degrade with number of array members.
4 SSD are 4x times slower than any individual SSD
If you put 2 SSD into garbage bin, speed will increase twice

not a rocket science, but take 5 stages: denial, anger, bargaining, depression and acceptance

They measured improvement:
https://www.pugetsystems.com/labs/artic ... -Pro-1369/
24 GB/sec:
https://cdi.liqid.com/lqd4500-performan ... gIewPD_BwE
20 GB/sec:
https://hothardware.com/news/gigabyte-a ... 9gbsec-ces

Now these sorts of speeds require threadripper 3 architecture. On the 3950 and below you have to put them on the CPU channel and the only benefit is 2x write speed. Everything else runs at the same speed as one SSD. But that's not bad. You still get more space.