SpacemiT K3 Pico-ITX Review: 60-TOPS RISC-V Powerhouse for LLMs?

How to install an NVMe SSD

The K3 Pico-ITX Mini PC includes an M.2 Key M slot for a 2280 NVMe SSD, but if you have a shorter card like a 2230 or 2242, no problem—you can still use it with an M.2 NVMe SSD length extension adapter. For this test, we used our Fanxiang S700 500GB with an extension adapter.

Checking for drives and partitions

In our setup, we installed a Fanxiang S700 500GB NVMe SSD, and Bianbu OS successfully recognized it as the device /dev/nvme0n.

Listing details of all available block devices, including hard drives, SSDs, and their partitions.

tzah@tzah-spacemit:/$ lsblk -o NAME,SIZE,TYPE,MOUNTPOINT
NAME        SIZE TYPE MOUNTPOINT
sda       119.3G disk
├─sda1      256M part /run/media/tzah/ESP
├─sda2      256M part /boot
└─sda3    118.8G part /
mtdblock0     8M disk
mtdblock1   128K disk
mtdblock2   512K disk
mtdblock3    64K disk
mtdblock4     1M disk
mtdblock5   384K disk
mtdblock6   5.9M disk
nvme0n1   465.8G disk <<< Our Fanxiang S700 500GB NVMe SSD

Performance benchmarks

⚡Fio benchmark with a 10 GB test file (sequential read/write): Testing the Fanxiang S700 NVMe

tzah@tzah-spacemit:/mnt/nvme$ sudo fio --name=seq --directory=/mnt/nvme --size=10G --bs=1M --rw=readwrite --ioengine=io_uring --iodepth=32 --numjobs=1 --time_based --runtime=20
seq: (g=0): rw=rw, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=io_uring, iodepth=32
fio-3.41
Starting 1 process
Jobs: 1 (f=1): [M(1)][100.0%][r=699MiB/s,w=698MiB/s][r=699,w=698 IOPS][eta 00m:00s]
seq: (groupid=0, jobs=1): err= 0: pid=1457151: Mon May 25 21:43:33 2026
  read: IOPS=951, BW=951MiB/s (998MB/s)(18.6GiB/20010msec)
    slat (usec): min=4, max=6776, avg=333.70, stdev=302.42
    clat (nsec): min=250, max=1654.7M, avg=12924141.23, stdev=68796525.77
     lat (usec): min=170, max=1654.8k, avg=13257.84, stdev=68805.99
    clat percentiles (nsec):
     |  1.00th=[       334],  5.00th=[       502], 10.00th=[       708],
     | 20.00th=[      1256], 30.00th=[     76288], 40.00th=[    505856],
     | 50.00th=[   5013504], 60.00th=[   8224768], 70.00th=[  10027008],
     | 80.00th=[  11468800], 90.00th=[  16449536], 95.00th=[  66322432],
     | 99.00th=[  91750400], 99.50th=[ 102236160], 99.90th=[1652555776],
     | 99.95th=[1652555776], 99.99th=[1652555776]
   bw (  MiB/s): min=  302, max= 2144, per=100.00%, avg=1033.00, stdev=448.27, samples=36
   iops        : min=  302, max= 2144, avg=1033.00, stdev=448.27, samples=36
  write: IOPS=952, BW=952MiB/s (999MB/s)(18.6GiB/20010msec); 0 zone resets
    slat (usec): min=17, max=6209, avg=175.72, stdev=252.88
    clat (usec): min=6, max=351109, avg=19037.78, stdev=28599.45
     lat (usec): min=369, max=351150, avg=19213.49, stdev=28585.35
    clat percentiles (usec):
     |  1.00th=[  1434],  5.00th=[  6259], 10.00th=[  8717], 20.00th=[ 10421],
     | 30.00th=[ 11731], 40.00th=[ 13304], 50.00th=[ 14222], 60.00th=[ 14877],
     | 70.00th=[ 15533], 80.00th=[ 16581], 90.00th=[ 20055], 95.00th=[ 41681],
     | 99.00th=[191890], 99.50th=[210764], 99.90th=[350225], 99.95th=[350225],
     | 99.99th=[350225]
   bw (  MiB/s): min=  314, max= 2206, per=100.00%, avg=1032.94, stdev=449.46, samples=36
   iops        : min=  314, max= 2206, avg=1032.94, stdev=449.46, samples=36
  lat (nsec)   : 500=2.31%, 750=3.06%, 1000=2.54%
  lat (usec)   : 2=4.50%, 4=0.97%, 10=0.04%, 20=0.03%, 50=0.41%
  lat (usec)   : 100=1.59%, 250=2.17%, 500=2.44%, 750=0.95%, 1000=0.34%
  lat (msec)   : 2=1.44%, 4=2.74%, 10=17.81%, 20=47.10%, 50=3.89%
  lat (msec)   : 100=4.16%, 250=1.36%, 500=0.07%, 2000=0.08%
  cpu          : usr=8.03%, sys=52.80%, ctx=7428, majf=0, minf=17
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=99.9%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.1%, 64=0.0%, >=64=0.0%
     issued rwts: total=19037,19055,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=32

Run status group 0 (all jobs):
   READ: bw=951MiB/s (998MB/s), 951MiB/s-951MiB/s (998MB/s-998MB/s), io=18.6GiB (20.0GB), run=20010-20010msec
  WRITE: bw=952MiB/s (999MB/s), 952MiB/s-952MiB/s (999MB/s-999MB/s), io=18.6GiB (20.0GB), run=20010-20010msec

Disk stats (read/write):
  nvme0n1: ios=42738/71067, sectors=21881856/36282440, merge=0/4, ticks=154469/2725114, in_queue=2879614, util=65.54%

Full Benchmark Summary — 10 GB Sequential Read/Write Test

Fanxiang S700 NVMe fio benchmark results on the K3 Pico ITX.

⚡Fio Sequential Write Speed Test: Writing (creating) a 10 GB file on our UFS

tzah@home-spacemit:~$ fio --name=write --filename=ssdtest --rw=write --bs=1M --size=10G --iodepth=16
write: (g=0): rw=write, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=psync, iodepth=16
fio-3.41
Starting 1 process
note: both iodepth >= 1 and synchronous I/O engine are selected, queue depth will be capped at 1
Jobs: 1 (f=1): [W(1)][98.3%][eta 00m:01s]
write: (groupid=0, jobs=1): err= 0: pid=268204: Wed May 20 15:09:49 2026
  write: IOPS=175, BW=176MiB/s (184MB/s)(10.0GiB/58310msec); 0 zone resets
    clat (usec): min=252, max=4590.4k, avg=5678.33, stdev=120813.39
     lat (usec): min=264, max=4590.4k, avg=5692.21, stdev=120813.35
    clat percentiles (usec):
     |  1.00th=[    277],  5.00th=[    281], 10.00th=[    281],
     | 20.00th=[    285], 30.00th=[    285], 40.00th=[    289],
     | 50.00th=[    293], 60.00th=[    297], 70.00th=[    306],
     | 80.00th=[    420], 90.00th=[  11207], 95.00th=[  11338],
     | 99.00th=[  15664], 99.50th=[  23200], 99.90th=[2432697],
     | 99.95th=[3506439], 99.99th=[4462740]
   bw (  KiB/s): min= 2048, max=2125824, per=100.00%, avg=420508.73, stdev=462855.44, samples=49
   iops        : min=    2, max= 2076, avg=410.65, stdev=452.01, samples=49
  lat (usec)   : 500=86.68%, 750=0.33%, 1000=0.48%
  lat (msec)   : 2=0.35%, 4=0.04%, 10=0.10%, 20=11.25%, 50=0.66%
  lat (msec)   : >=2000=0.11%
  cpu          : usr=0.21%, sys=5.66%, ctx=1277, majf=0, minf=12
  IO depths    : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,10240,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=16

Run status group 0 (all jobs):
  WRITE: bw=176MiB/s (184MB/s), 176MiB/s-176MiB/s (184MB/s-184MB/s), io=10.0GiB (10.7GB), run=58310-58310msec

Disk stats (read/write):
  sda: ios=9/9560, sectors=392/19491360, merge=29/18, ticks=333/3442620, in_queue=3442991, util=97.91%

Full Benchmark Summary (10 GB Sequential Write Test)

⚡Fio Sequential Read Speed Test: Reading a 10 GB file on our UFS

tzah@home-spacemit:~$ fio --name=read --filename=ssdtest --rw=read --bs=1M --size=10G --iodepth=16
read: (g=0): rw=read, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=psync, iodepth=16
fio-3.41
Starting 1 process
note: both iodepth >= 1 and synchronous I/O engine are selected, queue depth will be capped at 1
Jobs: 1 (f=1): [R(1)][100.0%][r=871MiB/s][r=870 IOPS][eta 00m:00s]
read: (groupid=0, jobs=1): err= 0: pid=269981: Wed May 20 15:12:12 2026
  read: IOPS=845, BW=845MiB/s (886MB/s)(10.0GiB/12114msec)
    clat (usec): min=170, max=12842, avg=1130.56, stdev=977.49
     lat (usec): min=171, max=12842, avg=1130.79, stdev=977.40
    clat percentiles (usec):
     |  1.00th=[  176],  5.00th=[  178], 10.00th=[  178], 20.00th=[  180],
     | 30.00th=[  188], 40.00th=[  190], 50.00th=[  857], 60.00th=[ 2024],
     | 70.00th=[ 2040], 80.00th=[ 2040], 90.00th=[ 2057], 95.00th=[ 2057],
     | 99.00th=[ 2540], 99.50th=[ 3392], 99.90th=[ 6063], 99.95th=[ 6718],
     | 99.99th=[ 9241]
   bw (  KiB/s): min=868352, max=921600, per=100.00%, avg=903257.04, stdev=16167.04, samples=23
   iops        : min=  848, max=  900, avg=882.09, stdev=15.79, samples=23
  lat (usec)   : 250=49.78%, 500=0.21%, 1000=0.01%
  lat (msec)   : 2=2.64%, 4=47.05%, 10=0.30%, 20=0.01%
  cpu          : usr=0.19%, sys=27.98%, ctx=5140, majf=0, minf=266
  IO depths    : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=10240,0,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=16

Run status group 0 (all jobs):
   READ: bw=845MiB/s (886MB/s), 845MiB/s-845MiB/s (886MB/s-886MB/s), io=10.0GiB (10.7GB), run=12114-12114msec

Disk stats (read/write):
  sda: ios=10026/2, sectors=20529152/48, merge=0/4, ticks=34204/53, in_queue=34260, util=71.09%

Full Benchmark Summary — 10 GB Sequential READ Test

What do p80, p90, p95, and p99 latency mean?

These are latency percentiles — they tell you how long I/O operations take at different points in the distribution.

Think of them as “how fast the SSD responds for X% of operations”.

Test Results In plain language:

• The UFS is quite reliable.
• 99% of reads complete in under 3 ms
• Only 1% are slower (due to caching, garbage collection, controller behavior).
• Stable Reads: Conversely, the read latency is rock-solid. Even its absolute worst-case scenario is just 12.8 ms, ensuring quick response times when retrieving files.
• The performance is quite impressive for a 128 GB embedded storage.