Hi Anton, I understand your almost totally negative review. But it seems it is entirely based on the documentation around the board. However: what do you think about the hardware? Is the hardware any good? I am an almost total beginner and want to learn risc-v assembly with this board, which is why I bought it. At some point I want to use the board for what it is meant for: machine learning, computer vision and so on. Building a cat door that only lets my cats through and not any other cats would be a nice project. Also: I have the Milk-V Duo S, with 512MB RAM and risc-v vector instructions. So what I really want to ask: is this totally hopeless for a beginner, or should I just keep going, try to follow all your links, translate all the chinese with google translate or chatgpt and so on? By the way: great T-shirt
Hi! Thank you! Nobody expects...:) If we talk about hardware as "SOPHGO SG2002" - then it's not bad hardware. An example would be MaixCAM. People are happy with it. If we talk about the Milk-V Duo S platform without an AI component - it's a typical board. The main thing is not to do anything related to AI. If we talk about which board is best for AI in the same price range and quality, it seems that the best are RV1106/RV1103 boards. But AI on Milk-V is hopeless, in my opinion.
Thanks for the informative video and one of the few youtube videos that are not overly positive. I wanted to try using RTOS for example , but ... i realized I had to rebuild RTOS to implement what I want, realize how to upload it to 2nd kernel (can linux do it? does it have two operating systems in bootloader?) and I need to cross-compile c program that will feed data into RTOS or use shared memory somehow and synchronize updates. It could be fun. But there are better ways how to spend two weeks and not having compiler on this board and having to download files all the time via scp, netcat of web ... no, thanks. Good stuff is there's GDB and strace and I was able to write two very simple programs in assembly for it (I have DuoS)
Is there any solution on how to get highest quality speech to text LLMs available right now and open source to run relatively fast on just the CPU or just by using just a 10-12 yeard old integrated GPU? Thank you!
I highly doubt it 1) There was no example of a single transformer in the documentation. 2) Most of the code is 2-3 years old. Transformers were not very popular then. 3) Very little memory But I haven't tried it. And maybe there is some way to formally satisfy this criterion:)
@@AntonMaltsev Anton, what board do you recommend for a beginner? I understand that categorically Milk-V is out, RK3568 from Radxa will be suitable (I saw tutorial in your channel)? I am thinking of a rover with custom computer vision...
@@marcinszymusiak5422 All RockChip boards are almost the same. In my opinion RK3588>RK3576>RK3568>RK3566> RV... Regular Linux is available for all RK series. Therefore, it's wise to select a board that fits within your budget, ensuring a responsible and prudent decision. Speaking of not RockChip boards: *Jetosns are the beast, but they are a little expensive *Intel-based ones are great, but for the cheap ones, they are pretty slow (N100), and for fast ones, they are power-consuming and expensive. * RPi5 is pretty fast. And sometimes it's enough for Computer Vision, for classification tasks espettially. Also, you can connect some acceleratior (like Hailo)
RockChip хороши. VeriSilicon не хуже других. Ещё было/есть несколько вменяемых плат. Так что не все так однозначно:) Паршивых плат их других частей мира тоже много.
Hi Anton,
I understand your almost totally negative review. But it seems it is entirely based on the documentation around the board.
However: what do you think about the hardware? Is the hardware any good?
I am an almost total beginner and want to learn risc-v assembly with this board, which is why I bought it. At some point I want to use the board for what it is meant for: machine learning, computer vision and so on. Building a cat door that only lets my cats through and not any other cats would be a nice project.
Also: I have the Milk-V Duo S, with 512MB RAM and risc-v vector instructions.
So what I really want to ask: is this totally hopeless for a beginner, or should I just keep going, try to follow all your links, translate all the chinese with google translate or chatgpt and so on?
By the way: great T-shirt
Hi! Thank you!
Nobody expects...:)
If we talk about hardware as "SOPHGO SG2002" - then it's not bad hardware. An example would be MaixCAM. People are happy with it.
If we talk about the Milk-V Duo S platform without an AI component - it's a typical board. The main thing is not to do anything related to AI.
If we talk about which board is best for AI in the same price range and quality, it seems that the best are RV1106/RV1103 boards.
But AI on Milk-V is hopeless, in my opinion.
awesome🔥 Thank you
I like it because it has to do with the milk v duo
Thanks for the informative video and one of the few youtube videos that are not overly positive. I wanted to try using RTOS for example , but ... i realized I had to rebuild RTOS to implement what I want, realize how to upload it to 2nd kernel (can linux do it? does it have two operating systems in bootloader?) and I need to cross-compile c program that will feed data into RTOS or use shared memory somehow and synchronize updates. It could be fun. But there are better ways how to spend two weeks and not having compiler on this board and having to download files all the time via scp, netcat of web ... no, thanks.
Good stuff is there's GDB and strace and I was able to write two very simple programs in assembly for it (I have DuoS)
Is there any solution on how to get highest quality speech to text LLMs available right now and open source to run relatively fast on just the CPU or just by using just a 10-12 yeard old integrated GPU? Thank you!
very good, thanks for this info, ill stick to my luckfox boards.
Can it run llm? Or something like that?
I highly doubt it
1) There was no example of a single transformer in the documentation.
2) Most of the code is 2-3 years old. Transformers were not very popular then.
3) Very little memory
But I haven't tried it. And maybe there is some way to formally satisfy this criterion:)
@3:14 I don't understand. Is CV181xC somehow equal to SG200X? Milk-V Duo 256M use SG2002
Yes. It's archetecture benith SG200x. Don't ask me why:) probably can be different CV181 versions for the same SG. But not sure.
@@AntonMaltsev Anton, what board do you recommend for a beginner? I understand that categorically Milk-V is out, RK3568 from Radxa will be suitable (I saw tutorial in your channel)? I am thinking of a rover with custom computer vision...
@@marcinszymusiak5422 All RockChip boards are almost the same. In my opinion RK3588>RK3576>RK3568>RK3566>
RV...
Regular Linux is available for all RK series. Therefore, it's wise to select a board that fits within your budget, ensuring a responsible and prudent decision.
Speaking of not RockChip boards:
*Jetosns are the beast, but they are a little expensive
*Intel-based ones are great, but for the cheap ones, they are pretty slow (N100), and for fast ones, they are power-consuming and expensive.
* RPi5 is pretty fast. And sometimes it's enough for Computer Vision, for classification tasks espettially. Also, you can connect some acceleratior (like Hailo)
Thanks, very informative 🤗 I know great guy "Ai Flux" here which also very deep in the topic.
Thank you! Will check
Китайцы что-нибудь толковое сделали когда-нибудь? Они могут только копировать. И то не всегда нормально получается.
RockChip хороши. VeriSilicon не хуже других. Ещё было/есть несколько вменяемых плат. Так что не все так однозначно:)
Паршивых плат их других частей мира тоже много.
This video could've been 3 mins long. Instead we get a 14min github text-to-speech
Greate job, thanks!