INTL 3:770/1 3:770/3   
   REPLYADDR theom+news@chiark.greenend.org.uk   
   REPLYTO 3:770/3.0 UUCP   
   MSGID:  e01e127f   
   REPLY:  f040b2e3   
   PID: SoupGate-Win32 v1.05   
   Pancho  wrote:   
   > I have three 2242 NVMe, they work fine, apart from some versions of   
   > U-Boot boot loader (They actually worked in older versions, then stopped   
   > working). A couple of those are 256GB from a couple of years ago, due to   
   > the price low differential I would buy 512GB now.   
   >   
   > I'm thinking of getting a M.2 NVMe adapter for my rPI5, I'll probably   
   > get a Pimoroni one, because it take standard 2280 drives. Best to go   
   > with the flow.   
      
   Agreed, if you don't need the small size then I'd go 2280 - plenty more to   
   choose from.   
      
   > > I think they've got them around the wrong way.  Their ODM Biwin's 2230 has   
   > > more read than write IOPS:   
   > > https://droix.co.uk/product/biwin-2230/   
   > >   
   >   
   > Yeahbut...   
   >   
   >    
   >   
   > 650K  IOPS Max. Random Read 4K   
   > 800K  IOPS Max. Random Write 4K   
   >   
   >   
   > But as I said, I don't really understand what IOPS means. The same   
   > device quotes a faster Max Read than Max Write (presumably sustained   
   > read/write).   
      
   4K random read test:   
   If the disc contains N blocks of size 4K   
   Repeat:   
   - roll a dice between 0 and N to give you D   
   - read 4K block number D from the disc   
   - throw away the data that came back   
   IOPS = how many times you can do that in a second   
      
   IOPS is a function of how well the flash and controller can manage an   
   unpredictable workload.  It's also a function of transfer speed to some   
   extent - you still need to move the data.  That 800K IOPS is 3.2GB/s or   
   26.2Gbps.  The Pi's single PCIe lane is only officially rated at Gen2 or   
   5Gbps, which puts a hard limit of 4K IOPS of 150K (and PCIe transfers have   
   additional overhead on top of that)   
      
   A more interesting graph in reviews is 'performance consistency': if you   
   give it a sustained write workload like this, after many minutes eventually   
   the speed falls as all the on-drive buffers fill up.  How well engineered it   
   is will show whether it can hold its performance, degrade gradually, hop   
   around erratically, or fall off a cliff.   
      
   Theo   
      
   --- SoupGate-Win32 v1.05   
    * Origin: Agency HUB, Dunedin - New Zealand | Fido<>Usenet Gateway (3:770/3)   
   SEEN-BY: 10/0 1 90/1 103/705 105/81 106/201 124/5016 128/187 129/305   
   SEEN-BY: 153/757 7715 218/0 1 601 700 840 870 930 220/70 221/1 6 360   
   SEEN-BY: 226/17 30 100 227/114 229/110 111 114 200 206 300 317 400   
   SEEN-BY: 229/426 428 470 550 616 664 700 240/1120 266/512 267/800   
   SEEN-BY: 282/1038 291/111 292/854 301/1 113 812 310/31 320/219 322/757   
   SEEN-BY: 335/364 341/66 342/200 396/45 460/58 633/280 712/848 770/1   
   SEEN-BY: 770/3 100 330 340 772/210 220 230 5020/400 1042 5075/35   
   PATH: 770/3 1 218/840 221/6 301/1 218/700 229/426