RAID 0 SSD vs HDD for Scratch only drive for Resolve

Hello folks, before I set my problem scenario and question, allow me to start by saying I have read a chunk on the threads regarding RAID 0 SSD vs HDD but they end up to just go the technical deep-end and not the practical direction, it is difficult to make a conclusion that can help me. Therefore, I am starting this new topic because I want to ask and hear from those who may have the same experience and issues I am encountering and their practical decision and outcome thereof. So if this post sound redundant to others, I appreciate if they just ignore the thread and not reply with information that has been posted in others. I have read them.

I am also specifically limiting this discussion on Resolve only scratch drive, RAID 0, SDD or HDD setup. Again, I will emphasize scratch because I have read other posts where folks have used RAID 0, 5, or 10 for both Scratch and Data... again without value to my inquiry here and just run into the technical deep-end back and forth without practical value for my need. SO PLEASE, just stay on topic - RAID 0 (2 Disc) SSD vs HDD for SCRATCH ONLY drive for RESOLVE. Thanks and now I can start with the problem scenario followed with my question.

I have always used 2 SDD in RAID 0 configuration as a dedicated scratch drive. I don't care about back-up or data loss because this is a scratch drive. A lot of you may have done the same. It has been efficient for my workflow to date, playback is decent even when I'm editing 4K DCI RAW. The power to the SSDs are stable (very little to no spike at all) so I am confident that it is normally functioning when in operation. However, every 4-5 months, I will have a fault on the SDDs for which the only option is to replaced them. I've used the Samsung EVO, and other high end SSD like the Intel, as well as consumer ones like the Crucial. Regardless of the brand, it will eventually fail after heavy scratch use in 4-5 months, lucky enough if it goes 8 months. In 2 years, I have changed the drives 4 times.

I have read a lot about RAID 0 SSD having high failure of rates over RAID 0 HDD. An article in Tom's hardware even discuss why this is and that the favorable option is to use HDD with a RAID 0 configuration. The explanation seem sound enough that I want to try this out since I am once again getting to that point where my scratch SSD is exhibiting issues time and again.

So I have two simple questions and if you have encountered the same problem, I hope you will be kind to share your answers to my questions and anything that you have learned from the experience.

Q1. Understanding that HDD is mechanical and slower than SDD, will a RAID 0 HDD for scratch deliver reasonable performance as a cache for Resolve?

Q2. What HDD drive will you recommend? I'm thinking of the WD Black 7200 RPM or the Seagate Barracuda 7200 RPM.

Thank you and looking forward to make a decision based on responses from this thread. Cheers!

We are using a 4 x 512GB Samsung 960 Pro raid0 for a year now and with no problems what so ever. We overprovisioned them 20% so we have a space of around 1.6 TB. The reason we picked the pro and not the EVO was the sustained speed and the TBW(which is about twice the size compared to TLC nand). The overprovisioning is done to get the sustained speed for the full 1.6 TB (7 GByte/s R/W from 8 PCIe3-lanes) and when a piece of the nand breaks down you can still read it and gets replaced with a fresh piece of nand from the overprovisoning.

Regarding spinning drives look for a test that mentions both max. and min write speeds.
When you have a 6..8-drives raid0 just pack it with 1TB 7200 rpm drives (something like seagate barracuda or WD-blue for around $45 each). These kind of drives have a 100-145-190MB/s min-avg-max R/W-speed.

Worst case with 4 drives is around 4x100= 400MB/s (drives almost full) and 4x190=760MB/s (drives are empty).

We are using a 4 x 512GB Samsung 960 Pro [RAID-0] for a year now and with no problems...

The reason we picked the [PRO] and not the EVO was the sustained speed and the TBW(which is about twice the size compared to TLC nand).

1) The acronyms TBW and TLC NAND escaped me. What do they mean?

The overprovisioning is done to get the sustained speed for the full 1.6 TB (7 GByte/s R/W from 8 PCIe3-lanes) and when a piece of the nand breaks down you can still read it and gets replaced with a fresh piece of nand from the overprovisoning.

2) Is "overprovisioning" a thing one actually performs on a set of drives, during the process of setting them up as a RAID?

3) Without being too technical and without the use of undefined acronyms, kindly explain to us how "overprovisoning" is done and what tool is used to do so?

4) Finally, please also explain in practical terms how one can readily identify when a piece of NAND [memory] breaks down -- and also how they can *recover* from such a NAND failure?

Thanks, Misha -- and anyone *else* who can, in layman's terns, unravel these matters for us SSD newbies.

[Re]Pete


ResolveStudio 14.3...014 | MiniMonitor | DTV 10.9.7 | Win8.1 x64 | ROG G751JL, 2.8GHz Intel i7HQ, 24GB DDR4, 1TB HD, 500GB EVO 850, 2GB GTX965M | Mackie MCU Pro | Softube Console 1 Mkii | Shuttle Pro 2 | more...

1) The acronyms TBW and TLC NAND escaped me. What do they mean?

TBW is the write endurance.
The TBW of the 970 pro 1 TB is claimed to be 1200 Tera Bytes Writen (MLC -> 2 bits per cell)
The TBW of the 970 evo 1 TB is claimed to be 600 Tera Bytes Written (TLC -> 3 bits per cell)

2) Is "overprovisioning" a thing one actually performs on a set of drives, during the process of setting them up as a RAID?

When you have a drive (any drive) of lets say 1 TB gross and when you format it in windows you only assign 0.8 TB to be used you have it overprovisoned by 20%. I overprovisioned it with Samsung software, but there are many other tools. When a SSD is overprovisioned it can do background garbage collection so that you always have enough nand cell to write to.

3) Without being too technical and without the use of undefined acronyms, kindly explain to us how "overprovisoning" is done and what tool is used to do so?

See 2)

without overprovisioning


~19.000 iops worst case

With 25% overprovisioning


~55.000 iops worst cast

4) Finally, please also explain in practical terms how one can readily identify when a piece of NAND [memory] breaks down -- and also how they can *recover* from such a NAND failure?

Have a look overhere https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=14&cad=rja&uact=8&ved=2ahUKEwim4PL-3rDdAhUvsKQKHUKNDSoQFjANegQIChAC&url=https%3A%2F%2Fwww.micron.com%2F~%2Fmedia%2Fdocuments%2Fproducts%2Ftechnical-note%2Fnand-flash%2Ftn2959_bbm_in_nand_flash.pdf&usg=AOvVaw23eY7pw4tZAY66STWvbPPC

Or just use Google or Duckduckgo

Thanks, Misha -- and anyone *else* who can, in layman's terns, unravel these matters for us SSD newbies.