Raspberry Pi 500

[geerlingguy]

Basic information

• Board URL (official): https://www.raspberrypi.com/products/raspberry-pi-500/
• Board purchased from: Provided by Raspberry Pi
• Board purchase date: Nov 27, 2024
• Board specs (as tested): 8GB RAM
• Board price (as tested): $90 ($120 if purchased in Kit)

Linux/system information

# output of `screenfetch`
         _,met$$$$$gg.           pi@pi500
      ,g$$$$$$$$$$$$$$$P.        OS: Debian 12 bookworm
    ,g$$P""       """Y$$.".      Kernel: aarch64 Linux 6.6.51+rpt-rpi-2712
   ,$$P'              `$$$.      Uptime: 8m
  ',$$P       ,ggs.     `$$b:    Packages: 1579
  `d$$'     ,$P"'   .    $$$     Shell: bash 5.2.15
   $$P      d$'     ,    $$P     Disk: 4.9G / 33G (16%)
   $$:      $$.   -    ,d$$'     CPU: ARM Cortex-A76 @ 4x 2.4GHz
   $$\;      Y$b._   _,d$P'      GPU: 
   Y$$.    `.`"Y$$$$P"'          RAM: 861MiB / 8048MiB
   `$$b      "-.__              
    `Y$$                        
     `Y$$.                      
       `$$b.                    
         `Y$$b.                 
            `"Y$b._             
                `""""

# output of `uname -a`
Linux pi500 6.6.51+rpt-rpi-2712 #1 SMP PREEMPT Debian 1:6.6.51-1+rpt3 (2024-10-08) aarch64 GNU/Linux

Benchmark results

CPU

• Geekbench 6: (836 single / 1743 multi - https://browser.geekbench.com/v6/cpu/9325556) (run as of Dec 7, 2024 - pre-launch Pi OS)
• Geekbench 6: (892 single / 2112 multi - https://browser.geekbench.com/v6/cpu/9325828) (run as of Dec 9, 2024 - launch-day Pi OS)
• 29.812 Gflops / 2.59 Gflops/W (geerlingguy/top500-benchmark HPL result)
• 35.586 Gflops at 11W, for 3.24 Gflops/W (after December 9 Pi OS update with NUMA faking and SDRAM tweaks)

Power

• Idle power draw (at wall): 3.1 W
• Maximum simulated power draw (stress-ng --matrix 0): 9 W
• During Geekbench multicore benchmark: TODO W
• During top500 HPL benchmark: 11.5 W

Disk

Raspberry Pi A2 microSD Card (32GB)

Benchmark	Result
iozone 4K random read	9.58 MB/s
iozone 4K random write	2.65 MB/s
iozone 1M random read	88.08 MB/s
iozone 1M random write	21.99 MB/s
iozone 1M sequential read	88.93 MB/s
iozone 1M sequential write	22.71 MB/s

Also consider running PiBenchmarks.com script.

Network

Built-in Ethernet

iperf3 results:

• iperf3 -c $SERVER_IP: 938 Mbps
• iperf3 -c $SERVER_IP --reverse: 940 Mbps
• iperf3 -c $SERVER_IP --bidir: 933 Mbps up, 436 Mbps down

Built-in WiFi

iperf3 results:

• iperf3 -c $SERVER_IP: 268 Mbps
• iperf3 -c $SERVER_IP --reverse: 235 Mbps
• iperf3 -c $SERVER_IP --bidir: 145 Mbps up, 90.4 Mbps down

GPU

glmark2

glmark2-es2 / glmark2-es2-wayland results:

=======================================================
    glmark2 2023.01
=======================================================
    OpenGL Information
    GL_VENDOR:      Broadcom
    GL_RENDERER:    V3D 7.1.10.2
    GL_VERSION:     OpenGL ES 3.1 Mesa 24.2.4-1~bpo12+1~rpt1
    Surface Config: buf=32 r=8 g=8 b=8 a=8 depth=24 stencil=0 samples=0
    Surface Size:   800x600 windowed
=======================================================
[build] use-vbo=false: FPS: 2558 FrameTime: 0.391 ms
[build] use-vbo=true: FPS: 3365 FrameTime: 0.297 ms
[texture] texture-filter=nearest: FPS: 2827 FrameTime: 0.354 ms
[texture] texture-filter=linear: FPS: 2789 FrameTime: 0.359 ms
[texture] texture-filter=mipmap: FPS: 2842 FrameTime: 0.352 ms
[shading] shading=gouraud: FPS: 2725 FrameTime: 0.367 ms
[shading] shading=blinn-phong-inf: FPS: 2471 FrameTime: 0.405 ms
[shading] shading=phong: FPS: 2104 FrameTime: 0.475 ms
[shading] shading=cel: FPS: 2039 FrameTime: 0.491 ms
[bump] bump-render=high-poly: FPS: 1376 FrameTime: 0.727 ms
[bump] bump-render=normals: FPS: 3026 FrameTime: 0.331 ms
[bump] bump-render=height: FPS: 2832 FrameTime: 0.353 ms
[effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 1170 FrameTime: 0.855 ms
[effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 478 FrameTime: 2.093 ms
[pulsar] light=false:quads=5:texture=false: FPS: 2925 FrameTime: 0.342 ms
[desktop] blur-radius=5:effect=blur:passes=1:separable=true:windows=4: FPS: 288 FrameTime: 3.476 ms
[desktop] effect=shadow:windows=4: FPS: 1076 FrameTime: 0.930 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 521 FrameTime: 1.920 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=subdata: FPS: 504 FrameTime: 1.987 ms
[buffer] columns=200:interleave=true:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 619 FrameTime: 1.617 ms
[ideas] speed=duration: FPS: 2247 FrameTime: 0.445 ms
[jellyfish] <default>: FPS: 1208 FrameTime: 0.828 ms
[terrain] <default>: FPS: 75 FrameTime: 13.440 ms
[shadow] <default>: FPS: 367 FrameTime: 2.727 ms
[refract] <default>: FPS: 119 FrameTime: 8.455 ms
[conditionals] fragment-steps=0:vertex-steps=0: FPS: 3272 FrameTime: 0.306 ms
[conditionals] fragment-steps=5:vertex-steps=0: FPS: 2292 FrameTime: 0.436 ms
[conditionals] fragment-steps=0:vertex-steps=5: FPS: 3227 FrameTime: 0.310 ms
[function] fragment-complexity=low:fragment-steps=5: FPS: 2739 FrameTime: 0.365 ms
[function] fragment-complexity=medium:fragment-steps=5: FPS: 1865 FrameTime: 0.536 ms
[loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 2637 FrameTime: 0.379 ms
[loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: 2625 FrameTime: 0.381 ms
[loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 1774 FrameTime: 0.564 ms
=======================================================
                                  glmark2 Score: 1907 
=======================================================

NOTE: I had to switch to wayfire (instead of labwc) to get GLMark2 to run. (Inside sudo raspi-config > Advanced > Wayland).

GravityMark

GravityMark results:

1. Download the latest version of GravityMark: https://gravitymark.tellusim.com
2. Run `chmod [downloaded_file.run]`
3. Run `sudo ./[downloaded_file.run]` and press `y` to accept the terms.
4. Open the link it prints, and run the Benchmark defaults, changing to 720p resolution and 50,000 asteroids.

Note: These benchmarks require an active display on the device. Not all devices may be able to run glmark2-es2, so in that case, make a note and move on!

Ollama

ollama LLM model inference results:

Device	CPU/GPU	Model	Speed	Power (Peak)
Pi 500 - 8GB	CPU	llama3.2:3b	4.53 Tokens/s	10.4 W
Pi 500 - 8GB	CPU	llama3.2:8b	1.90 Tokens/s	10.6 W

While running Ollama, I experienced a few thermal throttling / frequency capping events. The performance was consistent, but the SoC did end up hitting 80-85°C, and power consumption was steady at 10.5-10.7W for about 5 minutes, then it would drop to 9.8W. The clock speeds varied between 2.2 and 2.4 GHz.

TODO: See this issue for discussion about a full suite of standardized GPU benchmarks.

Memory

tinymembench results:

Click to expand memory benchmark result

tinymembench v0.4.10 (simple benchmark for memory throughput and latency)

==========================================================================
== Memory bandwidth tests                                               ==
==                                                                      ==
== Note 1: 1MB = 1000000 bytes                                          ==
== Note 2: Results for 'copy' tests show how many bytes can be          ==
==         copied per second (adding together read and writen           ==
==         bytes would have provided twice higher numbers)              ==
== Note 3: 2-pass copy means that we are using a small temporary buffer ==
==         to first fetch data into it, and only then write it to the   ==
==         destination (source -> L1 cache, L1 cache -> destination)    ==
== Note 4: If sample standard deviation exceeds 0.1%, it is shown in    ==
==         brackets                                                     ==
==========================================================================

 C copy backwards                                     :   4737.9 MB/s (0.8%)
 C copy backwards (32 byte blocks)                    :   4768.7 MB/s
 C copy backwards (64 byte blocks)                    :   4772.9 MB/s
 C copy                                               :   5601.0 MB/s (0.2%)
 C copy prefetched (32 bytes step)                    :   5556.8 MB/s
 C copy prefetched (64 bytes step)                    :   5603.2 MB/s (1.0%)
 C 2-pass copy                                        :   5265.7 MB/s (0.4%)
 C 2-pass copy prefetched (32 bytes step)             :   5563.8 MB/s (0.3%)
 C 2-pass copy prefetched (64 bytes step)             :   5556.6 MB/s
 C fill                                               :  14031.8 MB/s (0.3%)
 C fill (shuffle within 16 byte blocks)               :  14001.3 MB/s (1.8%)
 C fill (shuffle within 32 byte blocks)               :  14000.6 MB/s (0.5%)
 C fill (shuffle within 64 byte blocks)               :  13967.4 MB/s (0.2%)
 NEON 64x2 COPY                                       :   5544.8 MB/s
 NEON 64x2x4 COPY                                     :   5557.9 MB/s (0.1%)
 NEON 64x1x4_x2 COPY                                  :   5572.6 MB/s
 NEON 64x2 COPY prefetch x2                           :   5031.2 MB/s (0.2%)
 NEON 64x2x4 COPY prefetch x1                         :   5083.5 MB/s
 NEON 64x2 COPY prefetch x1                           :   5033.5 MB/s (0.3%)
 NEON 64x2x4 COPY prefetch x1                         :   5087.0 MB/s (0.1%)
 ---
 standard memcpy                                      :   5557.5 MB/s (0.2%)
 standard memset                                      :  14013.8 MB/s (0.4%)
 ---
 NEON LDP/STP copy                                    :   5560.4 MB/s (0.1%)
 NEON LDP/STP copy pldl2strm (32 bytes step)          :   5571.1 MB/s
 NEON LDP/STP copy pldl2strm (64 bytes step)          :   5571.3 MB/s
 NEON LDP/STP copy pldl1keep (32 bytes step)          :   5552.8 MB/s
 NEON LDP/STP copy pldl1keep (64 bytes step)          :   5554.1 MB/s
 NEON LD1/ST1 copy                                    :   5554.0 MB/s
 NEON STP fill                                        :  13993.2 MB/s (0.7%)
 NEON STNP fill                                       :  14008.9 MB/s (0.6%)
 ARM LDP/STP copy                                     :   5594.4 MB/s (0.2%)
 ARM STP fill                                         :  14011.5 MB/s (0.5%)
 ARM STNP fill                                        :  14033.4 MB/s (0.2%)

==========================================================================
== Framebuffer read tests.                                              ==
==                                                                      ==
== Many ARM devices use a part of the system memory as the framebuffer, ==
== typically mapped as uncached but with write-combining enabled.       ==
== Writes to such framebuffers are quite fast, but reads are much       ==
== slower and very sensitive to the alignment and the selection of      ==
== CPU instructions which are used for accessing memory.                ==
==                                                                      ==
== Many x86 systems allocate the framebuffer in the GPU memory,         ==
== accessible for the CPU via a relatively slow PCI-E bus. Moreover,    ==
== PCI-E is asymmetric and handles reads a lot worse than writes.       ==
==                                                                      ==
== If uncached framebuffer reads are reasonably fast (at least 100 MB/s ==
== or preferably >300 MB/s), then using the shadow framebuffer layer    ==
== is not necessary in Xorg DDX drivers, resulting in a nice overall    ==
== performance improvement. For example, the xf86-video-fbturbo DDX     ==
== uses this trick.                                                     ==
==========================================================================

 NEON LDP/STP copy (from framebuffer)                 :   1942.5 MB/s (0.3%)
 NEON LDP/STP 2-pass copy (from framebuffer)          :   1731.0 MB/s
 NEON LD1/ST1 copy (from framebuffer)                 :   1954.7 MB/s (0.2%)
 NEON LD1/ST1 2-pass copy (from framebuffer)          :   1738.8 MB/s
 ARM LDP/STP copy (from framebuffer)                  :   1895.8 MB/s (0.1%)
 ARM LDP/STP 2-pass copy (from framebuffer)           :   1730.5 MB/s (0.2%)

==========================================================================
== Memory latency test                                                  ==
==                                                                      ==
== Average time is measured for random memory accesses in the buffers   ==
== of different sizes. The larger is the buffer, the more significant   ==
== are relative contributions of TLB, L1/L2 cache misses and SDRAM      ==
== accesses. For extremely large buffer sizes we are expecting to see   ==
== page table walk with several requests to SDRAM for almost every      ==
== memory access (though 64MiB is not nearly large enough to experience ==
== this effect to its fullest).                                         ==
==                                                                      ==
== Note 1: All the numbers are representing extra time, which needs to  ==
==         be added to L1 cache latency. The cycle timings for L1 cache ==
==         latency can be usually found in the processor documentation. ==
== Note 2: Dual random read means that we are simultaneously performing ==
==         two independent memory accesses at a time. In the case if    ==
==         the memory subsystem can't handle multiple outstanding       ==
==         requests, dual random read has the same timings as two       ==
==         single reads performed one after another.                    ==
==========================================================================

block size : single random read / dual random read
      1024 :    0.0 ns          /     0.0 ns 
      2048 :    0.0 ns          /     0.0 ns 
      4096 :    0.0 ns          /     0.0 ns 
      8192 :    0.0 ns          /     0.0 ns 
     16384 :    0.0 ns          /     0.0 ns 
     32768 :    0.0 ns          /     0.0 ns 
     65536 :    0.0 ns          /     0.0 ns 
    131072 :    1.1 ns          /     1.5 ns 
    262144 :    1.6 ns          /     2.0 ns 
    524288 :    1.9 ns          /     2.2 ns 
   1048576 :    8.3 ns          /    11.2 ns 
   2097152 :   12.1 ns          /    14.5 ns 
   4194304 :   53.1 ns          /    81.0 ns 
   8388608 :   83.8 ns          /   115.1 ns 
  16777216 :   99.2 ns          /   126.9 ns 
  33554432 :  108.9 ns          /   133.4 ns 
  67108864 :  114.5 ns          /   137.2 ns 

sbc-bench results

Run sbc-bench and paste a link to the results here:

wget https://raw.githubusercontent.com/ThomasKaiser/sbc-bench/master/sbc-bench.sh
sudo /bin/bash ./sbc-bench.sh -r

Phoronix Test Suite

Results from pi-general-benchmark.sh:

Pre-launch (Pi OS as of Dec 7 2024)

• pts/encode-mp3: 11.775 sec
• pts/x264 4K: 4.12 fps
• pts/x264 1080p: 17.41 fps
• pts/phpbench: 433941
• pts/build-linux-kernel (defconfig): 2486.594 sec

Launch day (Pi OS with NUMA faking and SDRAM tweaks)

• pts/encode-mp3: 11.924 sec
• pts/x264 4K: 4.17 fps
• pts/x264 1080p: 18.38 fps
• pts/phpbench: 434379
• pts/build-linux-kernel (defconfig): DNF (kept getting The test quit with a non-zero exit status.)

[srcshelton]

Any clues yet as to what the populated 2-pin header just below the GPIO connector might be?

I was perhaps hoping for a way to connect an RTC battery/super-capacitor, but a UART port for the Debug Probe would also be useful!

(P.S. 62°C at 3.0GHz whilst compiling gcc-14.2 from scratch...)

[geerlingguy]

@srcshelton 99% certain that's UART for like the Pi Debug Probe. And there's an RTC battery header space to the right of the Keyboard connector.