https://www.eetimes.com/rv64x-a-free-open-source-gpu-for-risc-v/
2 Likes
Does proposing random new sets of (extension) instructions help? It would need to be adopted (ratified) as the standard, which may be some years away. There are a couple of related unratified extensions (SIMD, vector).
I would rather a core general purpose implementation that gives decent performance, in other words, an open CPU that I can actually use soon if not today, even if I have to use it with a blackbox GPU.
1 Like
ahh right. rather embarrassingly the actual article was written and published 2 years ago, re-published last month on eetimes as if it was up-to-date.
the article was inspired by a group led by Atif that included myself as part of LibreSOC before any of us knew the extent of the intense and violent dislike of Commercial Transparent Libre Projects within the RISCV Foundation. this cost Atif a considerable sum of money.
LibreSOC has since moved to OpenPOWER where the respect for innovation regardless of origin is genuine. note that i did not say it is “easier”! we still have an enormous amount of work to do and would welcome help, for which there are donations from NLnet available.
2 Likes
you cannot in any way do “random” extensions, it needs to be targetted and a huge amount of thought and effort put into it. two years of research so far.
that research showed very clearly and unequivocably that any given high-performance “compute” engine is NOT capable of delivering modern competitive 3D GPU Shader performance (only around 25% on a performance/watt metric).
the power efficiency of modern GPUs has run far ahead due to optimisation tricks that are acceptable for GPUs but completely unacceptable for HPC where accuracy is paramount.
i used to believe in this approach: do something now so that it results in gratification, earlier. the problem is that the costs involved (NREs) are exactly the same. 7nm is USD 16 million mask charges.
would you invest USD 16 million into something that is merely “decent” performance? especially if there was a way to invest in something with exceptional performance that had a possibility of a 100x return on investment?
2 Likes
I was speaking as a user, not as an investor but OK.
given that it takes money to create, the idea was to give you a perspective on what would be most likely to be funded into reality.
2 Likes