Milk-V Jupiter #47
Comments
|
After speaking with Milk-V about some of their RISC-V boards, they wanted to send over a Jupiter for me to test. I plan on running it through my usual tests, and also seeing how the board behaves as a regular ol' PC ITX board, testing front panel connectors, USB 3.0 and other IO. It also comes with a heatsink/fan (a pretty small one, mind you), and an IO shield. I haven't unboxed it fully but will hopefully do that tomorrow or Monday! |
|
First test round will be with https://github.com/milkv-jupiter/jupiter-ubuntu-build/releases - Ubuntu 23.10 desktop release 1.0.8. I will flash it to a microSD card first. First challenge: I can't determine what pins to use for a front panel power button connection on the F_PANEL header, so I opened the linked issue to see if someone can help with that.
|
|
Apparently there's some amount of standardization in the world... got this from Milk-V:
It does seem to boot up immediately upon getting power. To hard-power-off, hold down the power button for 5 seconds (or maybe 6 or so). To boot up again, press the power button for 1-2s. |
|
I can boot and log in via serial console; it also has an Ethernet connection, at 1 Gbps, and I can access it via UART or SSH, but so far I don't have an HDMI output working. |
|
The docs for the Jupiter describe using |
|
I was originally wondering if the fan or fan connector was dead, but no—just the heatsink seems to be good enough for 99% of the time—the fan will only spin up deep into a multicore test, and briefly at that. It's noticeable but not annoying the few times it does spin up. |
|
I'm able to get the HDMI output to feed into OBS through my CamLink 4K, so it's probably an incompatibility with that HP monitor: 
|
|
So far I'm unable to activate Bluetooth or see any WiFi settings, and I'm wondering if that's due to the Ubuntu image not including drivers or something. Need to do a little more exploration there. I just noticed there's a newer release of the 23.10 desktop: https://github.com/milkv-jupiter/jupiter-ubuntu-build/releases/tag/v1.0.9 (As well as Bianbu). Fedora also has a RISC-V task force building images for the Jupiter—see the downloadable images here: https://mirror.iscas.ac.cn/fedora-riscv/dl/Milk-V/Jupiter/images/latest/ |
|
Notes on Bianbu OS minimal install:
As with the Ubuntu image, I'm not seeing the WiFi device, just the two Ethernet ports and a Bluetooth looks to be present, at least: I opened the issue on the Milk-V community forum WiFi under Ubuntu not working. |
|
I'm also re-running Geekbench 6 under the Bianbu minimal install to see if there's any difference between it and the Ubuntu release. [Edit: Heh, identical result, 78/356.] |
|
I've also asked on the Community forum about the It would be nice to provide a couple fan headers so I can run a case fan off the ITX board, but I'm not sure if those are wired up to any PWM controls. |
|
Fedora 40 with Gnome runs stable, if slow. No GPU drivers present in that build, it seems—everything goes to the CPU and runs a bit slowish. YouTube can kinda play back at 240p, whereas on Bianbu and Ubuntu I can hit 720p passably. 
I'm also not able to enable Bluetooth (which seems to be present) or WiFi (which is not, like on the other OSes). |
|
I don't seem to be getting any GPU acceleration under Ubuntu or Bianbu, at least not by default... it seems like they should both be able to render a lot better using the built-in GPU (see Banana Pi BPI-F3 here). |
|
Kioxia XG8 over PCIe: |
|
Testing my 32GB eMMC module that I used before with Rock 5 B: Adding the results for that and NVMe up in the OP. |
|
I also tested Docker on this board, since that's something called out in the Jupiter's documentation. Installing was easy (install For many images, like Redis, they say It feels like the Wrote up a blog post on it: The state of Docker on popular RISC-V platforms. |
|
Testing an old AMD R5 230 2GB GPU, I plugged it into the board while powered off, and was surprised to see the fan spin up—thus I opened PCIe slot powered even when Jupiter is powered off. The system boots with the card installed, and Then there's I am going to try this k1-gpu branch to see if I can get the AMD GPU to work in Bianbu Linux. That's a fork of Bianbu Linux 6.1, and the patch seems to perform a few of the write-combining hacks that we had to do for |
|
For GPU testing, the command provided by SpaceMIT is as follows: |
Our default logic is to turn on the fan when the core temperature reaches above 60 degrees. We will update to PWM speed regulation in later versions. |
You can also watch the gpu status: |
|
@carbonfix - Thanks! I was able to get a much more reasonable score of 459 with that! |
|
Regarding WiFi, under Bianbu I'm seeing the following in dmesg: Upon @carbonfix's suggestion I also dumped i2c data: |
|
Memory performance (especially latency) is not that great (well, bad). Are the Milk-V guys aware of this and might provide 'better' DRAM initialization code? |
|
@ThomasKaiser - They are watching this thread, would be good to hear their thoughts—not sure if this is the first board they're using the Spacemit SoC with, but I wonder if it's an issue with Spacemit's controller or with something more under their control. |
Just why? By looking at https://browser.geekbench.com/v6/cpu/compare/6979805?baseline=5769184 it's 100% obvious that Geekbench on RISC-V is not capable to generate any meaningful numbers that would represent anything 'real-world performance'. It's just nonsense. |
|
@geerlingguy Currently it's needed to replace Mesa with a genuine copy from a generic distro to use external GPU (because the Mesa shipped with Bianbu is polluted by PowerVR driver, and its non-PowerVR-driver part seems to be broken). I didn't try this (because I just replaced the whole distro rootfs to test GPU support). |
@geerlingguy it should be under their control since we can compare with another K1 equipped thingy limited to 1.6 GHz though. When looking at memory latency accesses at 1M or above are lower on the Jupiter than on the BPi F3 while below it's the other way around (internal cache and higher clockspeeds). |
Because people will ask regardless, and I like having the data, regardless of the fitness of Geekbench 6 as a universal benchmarking tool :) The reason I run a few dozen other tests too is because:
To that point, I still can't run a few common benchmarks (like PTS tests) because they aren't set up for RISC-V. Whether that reflects real-world performance ('0' vs some metric that is above '0') is immaterial—instead, it reflects reality: if you buy one of these boards today, the performance might as well be 0 if you can't compile or run software the same as you can on x86 and Arm currently. At least, if you're not willing to roll up your sleeves and patch makefiles, compile newer versions than are available in repositories, build your own containers, etc. |
|
I have a new board today, and here are a few notes: First, idle power draw as tested with a USB-C power adapter is slightly higher from the wall (maybe slightly less efficient supply), measuring 5W idle, 10W under stress-ng all cores: 
Second, WiFi works (yay!), so now I can run my network performance tests with WiFi instead of wired. |
|
YouTube video: The PC industry is changing: RISC-V goes mainstream. |
|
Try this benchmark on your systems. Computing primes to 10000000 on 4 threads for 5 seconds. davepl_par;64;5.07541;4;algorithm=base,faithful=yes,bits=1 It will be interesting to see the results I got the idea to run it on my system after watching https://www.youtube.com/watch?v=7lYX0Eo4cbE&t=1s Make sure to set the CPU to performance mode |
Hi @geerlingguy , thanks a lot for the detailed work on this board. I am yet to receive my board which is on the way. |
|
Hi @geerlingguy! I know this is a little bit off-topic, but I've been building a site for hosting CI/CD runners powered by RISC-V architecture. I was in the need of it and noticed there wasn't one out there that was easy to setup. Here's the site: For now it only supports Gitlab Runners, but I plan to add support for other platforms too! If you want to try it without paying, check out this Gitlab Project: https://gitlab.com/riscvrunners/runner-test That project is open to the public, so you can push code and run CI pipelines to see if it works as you expect. |
Basic information
Linux/system information
Benchmark results
CPU
Power
stress-ng --matrix 0): 8 Wtop500HPL benchmark: 10.6 W(Note: Tests were performed using a 12V 8A power adapter. When I switched to a 27W Aergon PWR GaN adapter, idle draw was measured at 10W, and
stress-ng --matrix 0at 10W)Disk
MakerDisk A1 128GB microSD
Kioxia XG8 2TB NVMe SSD
32GB SanDisk eMMC Module V1.2
Run benchmark on any attached storage device (e.g. eMMC, microSD, NVMe, SATA) and add results under an additional heading.
Also consider running PiBenchmarks.com script.
Network
iperf3results:Built-in 1 Gbps LAN
iperf3 -c $SERVER_IP: 939 Mbpsiperf3 -c $SERVER_IP --reverse: 885 Mbpsiperf3 -c $SERVER_IP --bidir: 942 Mbps up, 136 Mbps downBuilt-in WiFi 6
So far, I have been unable to bring up this interface. See WiFi not working under Ubuntu or Bianbu.
iperf3 -c $SERVER_IP: 362 Mbpsiperf3 -c $SERVER_IP --reverse: 313 Mbpsiperf3 -c $SERVER_IP --bidir: 228 Mbps up, 140 Mbps downGPU
glmark2-es2-waylandresults:Note: This benchmark requires 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!TODO: See this issue for discussion about a full suite of standardized GPU benchmarks.
Memory
tinymembenchresults:Click to expand memory benchmark result
sbc-benchresultsRun sbc-bench and paste a link to the results here: ThomasKaiser/sbc-bench#96 / https://0x0.st/X904.bin
Phoronix Test Suite
Results from pi-general-benchmark.sh:
The text was updated successfully, but these errors were encountered: