Did they forget to say the VIM3 was the only one of the bunch with worldly prices,
and very low lead times ?
It was nice to see Khadas VIM3 Pro get the recognition, as people often to fail to mention it when talking about the N2+.
On the other hand the article also hilights that there’s nothing affordable for us Arm-Linuxers in 2021, that delivers a performance class beyond 2015’s cortex A72.
So we’re in the 6th year of the linux-ARM drought. Maybe the biblical 7th year will end it?
The RK3588 is just around the corner (it still 2018’s A76 cores), but on paper its nothing like we have ever seen before. Hopefully it will be a good release
There’s no news about RK3588 at all. We still don’t have an idea of the specs of the GPU. We haven’t heard any information about reason for the 1+ year delay in release. It’s not good.
Amlogic has been in radio silence about post-A73 SoC’s, as far as I can tell.
Some people are expecting a mediatek 8172 release this year, and that is looking most likely to me. Sadly not many shader cores on that one either.
The ideal hardware for the current panfrost drivers would be something like a G72-MP12 or MP16. This could deliver GPU shader performance on the order of the Nvidia Jetson Nano.
iirc that last SoC I saw that had a crazy GPU horsepower with a Mali GPU has to be the hisilicon kirin 970, sporting the G72-mp12, I am not sure about the exact number of shader cores it had, but it was a chip with amazing hardware on paper
alas silicon fabrication has come to a stale point in current times, latest news was that Rockchip isn’t able to fab any new chips because of the global fab shortage, amlogic has settled under the dust for now and allwinner has hardware that is deemed too old in terms of what is available today,
we should really focus on improving the software support for what we have now, the A311D is much more capable than we can perceive, and its only current software setbacks that are limiting what we can do, so lets just work with what we have and patiently wait for what the future shall present
Good advice, Electr1.
To get an idea of relative graphics performance between systems i like to run the glslsandbox-player with some of the lighter shaders.
Some samples for VIM3 with shellscript randplay.sh http://0x0.st/-Tip.zip
The geometry rendering of these portable GPUs is quite good, but relative to old graphics cards, the pixel shaders are weak. Simple shaders I write for emulators and games target 2012’s Intel HD4000 as baseline. Jetson Nano manages this level of performance.
These same shaders currently run 10-15x slower on Mali G52 MP4 with Panfrost. There will not be driver optimizations by Panfrost developers which deliver 10-15x more performance. We could see perhaps +10-15% by end-of-year.
My goal is to help get Linux-ARM to a performance, fun and usability of linux 2000-2010-era desktop/laptop experience. This doesn’t require much faster single threaded CPU performance but rather GPU shader performance closer to 2012’s Intel HD4000 or desktop GPU performance of Geforce/Radeon around 2004-2005.
There are Android-only Kirin, Samsung chips with Mali G72 in MP12 and MP16 versions but mainline linux support is 404-notfound. Samsung did have a short-lived linux on Dex program for the Exynos 9xxx but this was also ended quickly.
Our sights for the next iteration of open-source low-power general purpose computing should be focused by the constraints of feasibility. Since the Panfrost project is the only game in town for performant open-source ARM GPU drivers for linux, I would like efforts focused on chipsets that Panfrost can realistically drive.
Existing hardware such as Exynos 9810 sports a Mali-G72 MP18, and Kirin 970 the MP12. It seems we do not have a lack chip designs for the next iteration of ARM SBCs but rather software (linux mainline kernel support).
I think attention should be focused on getting those SoC’s with G72 to run mainline linux. (If there is still production capacity or stockpiles of those chipsets lying around)