Phison e18 to be Fastest PCIe Gen 4x4 NVMe SSD Controller going for 7400 MB/s

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Nov 9, 2020.

  1. Hilbert Hagedoorn

    Hilbert Hagedoorn Don Vito Corleone Staff Member

    Messages:
    41,083
    Likes Received:
    9,346
    GPU:
    AMD | NVIDIA
  2. NightWind

    NightWind Master Guru

    Messages:
    269
    Likes Received:
    252
    GPU:
    Asus ROG 2060 Super
    ''phison knows what you need''
    How come? :rolleyes:
     
    Tim24 likes this.
  3. Dazz

    Dazz Master Guru

    Messages:
    918
    Likes Received:
    102
    GPU:
    ASUS STRIX RTX 2080
    So Samsung 980 Pro was short lived.
     
  4. wavetrex

    wavetrex Maha Guru

    Messages:
    1,423
    Likes Received:
    1,061
    GPU:
    Zotac GTX1080 AMP!
    970 Evo Plus is faster than 970 Pro... so won't be surprised if they release another 980 version.

    I spy with my little eye...

    Come to daddy !!!

    (Not that I need it ... but... damn, that's fast !)
     

  5. Locoyote

    Locoyote Member Guru

    Messages:
    173
    Likes Received:
    12
    GPU:
    GTX1060 6GB
    Phison knows I need a molecatcher.
     
  6. barbacot

    barbacot Master Guru

    Messages:
    533
    Likes Received:
    464
    GPU:
    Asus 3080 Strix OC
    The Things That I need: A smoke. A whiskey. For the sun to shine. I want to sleep, to forget. Peace on earth. No more hunger and diseases.

    Your move Phison???
     
  7. mdk77777

    mdk77777 New Member

    Messages:
    3
    Likes Received:
    1
    GPU:
    AMD 5500xt
    OLD news.
    Let me know when a product launches that I can buy.
    I've already built my paper mache computer.
     
    Picolete likes this.
  8. bluedevil

    bluedevil Master Guru

    Messages:
    363
    Likes Received:
    11
    GPU:
    Kfa2 RTX 2060 6gb
    7400MB/s is for sequential read q depth 32.You won't see this kind of speed in practice as a normal consumer.

    If my understanding is correct data is stored on multiple flash chips . Each flash chip has a max speed so in the best case scenario you access all the flash chips at once and read sequentially from each chip. Sequential reads from a flash chip is a lot faster then random read.
    Now you just multiply the number of chips times the sequential read speed of each chip and you get the 7400MB/s.
    The bigger the SSD capacity the more flash chips it has to read from in parallel.
    In theory the SSD controller would just need to split up a file into 4k pieces across as many flash chips as possible to maximize performance. Of course if you have a file that is say 4k times 100 and 10 flash chips you would put the first 10 pieces on chips 1-10 then the next 10 pieces and so on but in a sequential manner so they are next to each other on the chip.
    Each flash chip is divided into blocks and each block can hold a certain number of 4k pieces. This is what i mean that the 4k pieces need to be next to each other, they need to be on the same block so we maximize performance.
    Because files are not exactly divisible by the number of chips and blocks on each chip (ex:4k times 87) it would be impossible to equally divide each file so that it's divided equally between the flash chips.

    Also because of how flash chips operate when deleting and writing information on them we get another factor that messes up how files are divided between the chips.
    Each flash chip is divided into a certain number of blocks and we can only delete a block in it's entirety. This means that if we need to keep some of the information from the block because it's part of another file we need to write it first to another block, preferably to a block where it can fit, then we can delete the whole block .

    As the drive gets fuller more and more flash chips get full so when we write new information out of the 10 initial chips we now have say only 4 available to split up the file. This means we get less performance on fuller drives.
    This is why we need to "Optimize" the SSD using software after some time. to spread files over all the flash chips and try to put the pieces of the file next to each other on the same chip.
    This is different from de-fragmenting on old HDD'S but the result is the same, more performance.

    Like i said in the beginning, if my understanding is correct this is why 4k q depth of 1 is what really matters as it's the closest metric to reality for a normal user and by normal i mean someone that does not do data server type stuff.
     

Share This Page