![]() At a queue depth of 32 and using 4 threads, the reads and writes are both quoted as 500,000 IOPS. Random 4K performance is quoted at 15,000 IOPS and 55,000 IOPS for reads and writes respectively at a queue depth of one and using a single thread. The 512GB drive has the same read figure with writes dropping to 2,300MB/s. Official sequential performance figures for the 1TB drive are 3,500MB/s for reads and 2,700MB/s for writes. Samsung hasn't released much in the way of info about the new controller apart from one of its five cores is dedicated to communications between the controller and the system host and that the clock speeds are faster than the Polaris controller of the previous generation SSD 960 series. The drive uses Samsung's latest 5 core controller called Phoenix. Coming in at £599.99 it is certainly not cheap, but can the performance justify that hefty price tag?Īt its launch the 970 PRO comes in just two capacities: 512GB and 1TB, and instead of the TLC NAND that the 970 EVO uses, the 970 PRO comes equipped with Samsung's latest 64-layer MLC V-NAND (3D NAND). It also makes absolutely no economic sense unless you lan on populating the the remaining slots.Įven then, an OWC Envoy Express TB3 enclosure ($79) plus your choice of 2, 4 or 8TB NVMe blade will yield better performance (~1500 MBs) unless you RAID 3 or 4 slots in the 4M2.Launched at the same time as the 970 EVO, the 970 PRO is Samsung's 3rd generation flagship NVMe consumer drive, taking over the mantle from the legend that is the 960 PRO. While I am not getting into whether that is fast enough for sample streaming (probably is but you will see a big difference in load speeds depending on sample player), it is hardly an apples to apples comparison to the 5000 MBs+ internal drives. A single NVMe blade in one of these boxes will likely yield speeds similar to SATA SSD's. These chassis are designed to run in RAID 0 (striped) - 4 x 700MBs. The problem is TB3/4 only allocate approx 2800 MBs for data (the rest reserved for video streams) and it is impossible to simultaneously run 4 NVMe devices (JBOD) at anywhere close to the blade speed maximums. *rant incoming* The online commentators who went "oh look you can replace the SSD's" should have figured this out before putting out really misleading information!)Ĭlick to expand.I don't know for certain, but I strongly suspect that the Express 4M2, like most external TB 4 drive enclosures, only use 1 PCie lane per drive. (not to be confused with the two sets of NAND chips for the internal storage of new MacBook pro's and Mac studios, which are controlled by a single chip and NOT raid. And for reasons it doesn't suffer the same issues as running a software raid on a Mac. My old (not really that old) Glyph M2R 4TB thunderbolt SSB *is* internally running raid 0. And once you see those numbers, you might end up diving into deep rabbit holes.Īnd yes - Raid 0 is generally not great for sample libraries. Oh - and even good ol activity monitor is good for seeing real-time what a drive is doing. SSDs are fine.ĪTTO make an awesome drive benchmarking tool - which can sequentially test different size read/writes and display on a graph. you don't really need NVMe M.2 drives for audio. don't worry about filling up your drives Note that the random 4k performance is still pretty good at 205MB/s - still 3x faster than 4 M.2 drives in a RAID0! And actually they're still pretty fast even though the drive is almost full!Īnd just for comparison, here's an old Samsung 860 EVO SSD in a BlackMagic Multidock. The above drive is 97% full but notice how only the WRITE speeds are affected. READ speed is still extremely fast even on a full drive. How do those results vary on a full drive? We often hear about needing to leave 10-20% free on SSDs to preserve their speed, but this only affects WRITE speed. Still great contiguous read speed at 6.7GB/s but look at the random 4k result - 649MB/s - 10x faster than the RAID0! Wow! 13.8GB/s read speed! Great for video, but look at the random 4k result - only 64.47MB/s! RAID0 just isn't a good choice for sample libraries. ![]() These will test the all-important random 4k read speeds which matter for library streaming. The BlackMagic disk benchmark only measures reading and writing of large contiguous files and isn't much use for audio tests. They are mainly of interest to people doing video work, where you work with very large contiguous files with extremely high data rates.Īudio streaming is all about loading (tens of) thousands of tiny audio files in a fairly random order. The high figures (500MB/s, 3.5GB/s, 7GB/s etc) advertised by SSD/NVMe manufacturers aren't actually that relevant to audio work.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |