I have an MX300 525GB M.2 drive and it performs just fine in "real world" usage. Better than any other SSD I've used anyway.
Dunno why..but 3500mb advertised reads are scoring lower in every aspect than my 950 pro in Anvil test...
Sequential tests are pointless. For a consumer, random read/write performance and QD1 is what matters. Your drive loses in both 4K QD16 tests as well as the 4K QD4 and 4MB sequential write tests. And WTF cares? Sequential read/write tests don't indicate real-world performance.
Sure, but was just surprised to see better overall read score with the old model(used 512gb model running the OS vs new 1tb one) where advertised read speed are 3500 vs 2500 mb/s for both models respectively
DStealth, because 99% of the purchases will be done because 3500>2500 and people will actually be happy they have faster drive, even thou, in reality, it is vice versa. sykozis, stuff might have changed indeed, crucial loves updating their drives indeed, can you run qd1,4k rr? If it beats 960 pro I'll be laughing my ass here
I'm seeing significantly better results across the board on other websites. HH is probably using default windows 10 driver where other sites are using a modified 950 pro NVMe driver. Yes it's 2TB but 2TB drives are a little slower than the 1TB ones. We should see even better improvements when they release a driver for it.
That custom driver thingy seems quite good - over 50mb woot, finally an ssd to break the magical barrier haha. Anyway writes to be > reads is totally possible... for a time. The write speed of a flash block is actually not all that bad, it is the erase that takes forever. All modern ssds, combined with modern OS are doing the delayed erase (i.e. when you erase a block it is not actually erased, but done at idle time) and generally all blocks are kept "free" so writes can go right in. The second block to the puzzle - caches. With random reads the ssd can't predict what data you want next, but for writes it can just cache write requests, respond "OK" and do the actual writes as late as it needs Third write optimization - let's say the ssd got itself 32 banks of flash memory and you send a command to write 10 blocks of data - instead of writing them one after another, it will write them to random 10 banks, all at once. This is the very reason bigger ssds of the same model have drastic better write (and to some extend sequental read) speeds. Add to the mix windows's own caches that stores write operations and sending them to the ssd in batches This all works to the point the ssd is getting full/out of clean/erased blocks ofc, but you need to torture it quite a lot to get to that state. In general you should not worry about ssd's write speed much, because in pretty much all of the daily work the data that needs to be written will be "fast enough" - that is - cached by windows and done with, then it will send to the ssd in background, the ssd itself will issue background write, stuff you won't really feel.
I don't see anything about a modfied NVME driver in any reviews I read/searched. You mean you found the mentioned driver and it's indeed a modified version of Samsungs NVME driver for 950 PRO? Mine says it's the version 1.4.7.17 btw in Magician. If you have indeed found a modified version of that driver, could you please share the link? I would like to try it out. :thumbup:
Its just a driver forced installed via have disk, inf maybe has the device id added. The actual driver is the same though. Do you have a 960 pro?
Yeah I just quoted A01, forcing 950 driver will probably work perfectly. INF modification will be a bit more problematic as you need to turn off windows signature verification (on 8 and onward) or give me the modified driver to sign (yeah I do have ms authenticode certificates).
I don't(950Pro owner). DrunkenDonkey quoted DStealth which both were referring to your post. For some reason I though DrunkenDonkey had a 950 PRO and he found a modified Sammy NVME driver from the reviews and he used it and got improved speeds. lol Poor reading comprehension from my part. He was just referring the LR score(4K=53.87MB/s) you posted.
In other sad news, the only thing that could really make storage faster - intel's 3d xpoint memory (supposedly released this year) is delayed till the end of 2019 probably. And this is server, then an year or 2 for consumer... meh I'm getting old.
Good - yeah, possible - nope Something's wrong with the test for sure. Try with another, for example http://crystalmark.info/software/CrystalDiskMark/index-e.html