#riscv — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #riscv, aggregated by home.social.
-
Coincidentally I was also looking for information on the SpacemiT-K3's AI capabilities and came across another of Bruno's articles. Interesting to see the performance figures here, and it's a real shame that - for now - it seems you have to ue prebuilt binaries to access the AI instructions with llama.cpp on Bianbu. But it does work. I have a local qwen2.5-coder running on mine. Just need a bit more work to let it do changes on the file system ...
#AI #RISCV #OpenSource
https://dev.to/gounthar/benchmarking-llamacpp-on-spacemit-k3-risc-v-ai-cores-vs-standard-rvv-part-4-10mc -
Coincidentally I was also looking for information on the SpacemiT-K3's AI capabilities and came across another of Bruno's articles. Interesting to see the performance figures here, and it's a real shame that - for now - it seems you have to ue prebuilt binaries to access the AI instructions with llama.cpp on Bianbu. But it does work. I have a local qwen2.5-coder running on mine. Just need a bit more work to let it do changes on the file system ...
#AI #RISCV #OpenSource
https://dev.to/gounthar/benchmarking-llamacpp-on-spacemit-k3-risc-v-ai-cores-vs-standard-rvv-part-4-10mc -
Coincidentally I was also looking for information on the SpacemiT-K3's AI capabilities and came across another of Bruno's articles. Interesting to see the performance figures here, and it's a real shame that - for now - it seems you have to ue prebuilt binaries to access the AI instructions with llama.cpp on Bianbu. But it does work. I have a local qwen2.5-coder running on mine. Just need a bit more work to let it do changes on the file system ...
#AI #RISCV #OpenSource
https://dev.to/gounthar/benchmarking-llamacpp-on-spacemit-k3-risc-v-ai-cores-vs-standard-rvv-part-4-10mc -
Coincidentally I was also looking for information on the SpacemiT-K3's AI capabilities and came across another of Bruno's articles. Interesting to see the performance figures here, and it's a real shame that - for now - it seems you have to ue prebuilt binaries to access the AI instructions with llama.cpp on Bianbu. But it does work. I have a local qwen2.5-coder running on mine. Just need a bit more work to let it do changes on the file system ...
#AI #RISCV #OpenSource
https://dev.to/gounthar/benchmarking-llamacpp-on-spacemit-k3-risc-v-ai-cores-vs-standard-rvv-part-4-10mc -
From FUZxxl:
> SIMD programming challenge: take a vector v and a mask m segmenting it into k segments. Given a (variable) permutation π with k entries, shuffle the segments of v into the order given by π (segment-wise shuffle). How would you approach this problem?
My approach is to run-length-encode the offset of the start of each segment, apply the permutation, decode the RLE-offsets and finally fixup the offsets for the second to last element of each segment. This gives you the shuffle you can apply to the input.
Here is the RVV implementation: https://github.com/camel-cdr/rvv-playground/blob/main/segment-vrgather.c (https://godbolt.org/z/dMhxz84hz)
I've attached an image of an execution log to help visualize this.
-
From FUZxxl:
> SIMD programming challenge: take a vector v and a mask m segmenting it into k segments. Given a (variable) permutation π with k entries, shuffle the segments of v into the order given by π (segment-wise shuffle). How would you approach this problem?
My approach is to run-length-encode the offset of the start of each segment, apply the permutation, decode the RLE-offsets and finally fixup the offsets for the second to last element of each segment. This gives you the shuffle you can apply to the input.
Here is the RVV implementation: https://github.com/camel-cdr/rvv-playground/blob/main/segment-vrgather.c (https://godbolt.org/z/dMhxz84hz)
I've attached an image of an execution log to help visualize this.
-
The RISC-V ISA is crossing the chasm from hobbyist microcontrollers to datacenter-grade CPUs. SiFive, Ventana, and Rivos are shipping serious silicon while the software stack (LLVM, Linux, Debian) catches up. The open model works - no license fees, custom extensions, and a global talent pool driving innovation. #RISCV #OpenSource #Hardware #Linux #Embedded #OpenISA
-
The RISC-V ISA is crossing the chasm from hobbyist microcontrollers to datacenter-grade CPUs. SiFive, Ventana, and Rivos are shipping serious silicon while the software stack (LLVM, Linux, Debian) catches up. The open model works - no license fees, custom extensions, and a global talent pool driving innovation. #RISCV #OpenSource #Hardware #Linux #Embedded #OpenISA
-
RISC-V Is Inevitable, State of the Union Keynote Argues - EE Times
https://www.eetimes.com/risc-v-is-inevitable-state-of-the-union-keynote-argues/
-
RISC-V Is Inevitable, State of the Union Keynote Argues - EE Times
https://www.eetimes.com/risc-v-is-inevitable-state-of-the-union-keynote-argues/
-
Running DOOM is cool. Building the CPU that runs it is cooler.
This write-up follows the journey from designing a custom RISC-V processor to FPGA implementation and using DOOM to validate the design.
Excellent hardware/software co-design read.
-
Running DOOM is cool. Building the CPU that runs it is cooler.
This write-up follows the journey from designing a custom RISC-V processor to FPGA implementation and using DOOM to validate the design.
Excellent hardware/software co-design read.
-
Keeping the 2*8 cores (not threads!) of this SpacemiT-K3 busy #RISCV
-
Keeping the 2*8 cores (not threads!) of this SpacemiT-K3 busy #RISCV
-
Having fun with alternate #ISA like #RISC_V #RISCV #ARM #PowerPC #LoongArch is fun but also revealing how bad people code these days even aside from SLOP, when something like #Discord or #Fluxer absolutely murders your little dev board or in general how popular using web apps over a framework is, just bloody sucks.
-
EMiX extends FPGA emulation beyond the limits of a single board by partitioning RISC-V SoCs across multiple interconnected FPGAs. The authors demonstrate a Linux-capable 64-core system spanning 8 FPGAs and plan to open-source the framework.
https://arxiv.org/abs/2604.27012
#RISCV #FPGA -
EMiX extends FPGA emulation beyond the limits of a single board by partitioning RISC-V SoCs across multiple interconnected FPGAs. The authors demonstrate a Linux-capable 64-core system spanning 8 FPGAs and plan to open-source the framework.
https://arxiv.org/abs/2604.27012
#RISCV #FPGA -
-
-
New #riscv Fast-Track Proposal: Zijf (Far jumps and calls):
https://github.com/riscv/riscv-isa-manual/pull/3269
A) should've been defined way earlier
B) remember the Qualcomm slides (https://lists.riscv.org/g/tech-profiles/attachment/400/0/AOSP%20Compression.pdf) -
New #riscv Fast-Track Proposal: Zijf (Far jumps and calls):
https://github.com/riscv/riscv-isa-manual/pull/3269
A) should've been defined way earlier
B) remember the Qualcomm slides (https://lists.riscv.org/g/tech-profiles/attachment/400/0/AOSP%20Compression.pdf) -
-
-
-
-
-
-
-
Fixing #RISCV cross-compilation and usage of @gotosocial on @nixos_org 🚀 https://github.com/NixOS/nixpkgs/pull/544570
-
Fixing #RISCV cross-compilation and usage of @gotosocial on @nixos_org 🚀 https://github.com/NixOS/nixpkgs/pull/544570
-
You need a 1234 GiB GPU to run your program?
Let me play you the tiniest... #microcontroller breakout board!The 10-cent CH32V003 microcontroller and a single capacitor make a complete breakout board. Honestly, I didn't even need the PCB. The capacitor could be soldered directly on the pins.
I flashed it with the #Arduino . Not so "Nano" in comparison! #ESP32 for scale.
And it has #Rust support!
-
You need a 1234 GiB GPU to run your program?
Let me play you the tiniest... #microcontroller breakout board!The 10-cent CH32V003 microcontroller and a single capacitor make a complete breakout board. Honestly, I didn't even need the PCB. The capacitor could be soldered directly on the pins.
I flashed it with the #Arduino . Not so "Nano" in comparison! #ESP32 for scale.
And it has #Rust support!
-
You need a 1234 GiB GPU to run your program?
Let me play you the tiniest... #microcontroller breakout board!The 10-cent CH32V003 microcontroller and a single capacitor make a complete breakout board. Honestly, I didn't even need the PCB. The capacitor could be soldered directly on the pins.
I flashed it with the #Arduino . Not so "Nano" in comparison! #ESP32 for scale.
And it has #Rust support!
-
You need a 1234 GiB GPU to run your program?
Let me play you the tiniest... #microcontroller breakout board!The 10-cent CH32V003 microcontroller and a single capacitor make a complete breakout board. Honestly, I didn't even need the PCB. The capacitor could be soldered directly on the pins.
I flashed it with the #Arduino . Not so "Nano" in comparison! #ESP32 for scale.
And it has #Rust support!
-
You need a 1234 GiB GPU to run your program?
Let me play you the tiniest... #microcontroller breakout board!The 10-cent CH32V003 microcontroller and a single capacitor make a complete breakout board. Honestly, I didn't even need the PCB. The capacitor could be soldered directly on the pins.
I flashed it with the #Arduino . Not so "Nano" in comparison! #ESP32 for scale.
And it has #Rust support!
-
Christopher Barnatt of ExplainingComputers.com reviews the most powerful RISC-V SBC he's yet tested:
Milk-V Jupiter 2: RISC-V RVA23 SBC
https://www.youtube.com/watch?v=vSKA6thLI1U
tl;dw: ahead of RasPi 5 in some ways, behind it in others, hardware acceleration works well with the vendor OS (Bianbu, a patched Ubuntu derivative)
-
Christopher Barnatt of ExplainingComputers.com reviews the most powerful RISC-V SBC he's yet tested:
Milk-V Jupiter 2: RISC-V RVA23 SBC
https://www.youtube.com/watch?v=vSKA6thLI1U
tl;dw: ahead of RasPi 5 in some ways, behind it in others, hardware acceleration works well with the vendor OS (Bianbu, a patched Ubuntu derivative)
-
Christopher Barnatt of ExplainingComputers.com reviews the most powerful RISC-V SBC he's yet tested:
Milk-V Jupiter 2: RISC-V RVA23 SBC
https://www.youtube.com/watch?v=vSKA6thLI1U
tl;dw: ahead of RasPi 5 in some ways, behind it in others, hardware acceleration works well with the vendor OS (Bianbu, a patched Ubuntu derivative)
-
Christopher Barnatt of ExplainingComputers.com reviews the most powerful RISC-V SBC he's yet tested:
Milk-V Jupiter 2: RISC-V RVA23 SBC
https://www.youtube.com/watch?v=vSKA6thLI1U
tl;dw: ahead of RasPi 5 in some ways, behind it in others, hardware acceleration works well with the vendor OS (Bianbu, a patched Ubuntu derivative)
-
Christopher Barnatt of ExplainingComputers.com reviews the most powerful RISC-V SBC he's yet tested:
Milk-V Jupiter 2: RISC-V RVA23 SBC
https://www.youtube.com/watch?v=vSKA6thLI1U
tl;dw: ahead of RasPi 5 in some ways, behind it in others, hardware acceleration works well with the vendor OS (Bianbu, a patched Ubuntu derivative)
-
Sooner or later I'll buy an Risc-V SBC board, maybe with NPU.. #riscv #sbc #opensource #
https://youtu.be/vSKA6thLI1U -
Sooner or later I'll buy an Risc-V SBC board, maybe with NPU.. #riscv #sbc #opensource #
https://youtu.be/vSKA6thLI1U -
Sooner or later I'll buy an Risc-V SBC board, maybe with NPU.. #riscv #sbc #opensource #
https://youtu.be/vSKA6thLI1U -
Sooner or later I'll buy an Risc-V SBC board, maybe with NPU.. #riscv #sbc #opensource #
https://youtu.be/vSKA6thLI1U -
Sooner or later I'll buy an Risc-V SBC board, maybe with NPU.. #riscv #sbc #opensource #
https://youtu.be/vSKA6thLI1U -
Update: I added a sound card :3
The timings are a bit fucked, I'll fix that at some point.
The way it works is slightly cursed, since protoflux doesn't have a way to output arbitrary audio. It drives the frequencies and amplitudes of 20 sine waves to reconstruct the sound.
The CPU is also really weak (running at 100 KHz here), all the audio data is preprocessed and all it has to do is copy it out of ROM and into the audio chip's memory, it is just barely able to do that x3
-
Update: I added a sound card :3
The timings are a bit fucked, I'll fix that at some point.
The way it works is slightly cursed, since protoflux doesn't have a way to output arbitrary audio. It drives the frequencies and amplitudes of 20 sine waves to reconstruct the sound.
The CPU is also really weak (running at 100 KHz here), all the audio data is preprocessed and all it has to do is copy it out of ROM and into the audio chip's memory, it is just barely able to do that x3
-
Update: I added a sound card :3
The timings are a bit fucked, I'll fix that at some point.
The way it works is slightly cursed, since protoflux doesn't have a way to output arbitrary audio. It drives the frequencies and amplitudes of 20 sine waves to reconstruct the sound.
The CPU is also really weak (running at 100 KHz here), all the audio data is preprocessed and all it has to do is copy it out of ROM and into the audio chip's memory, it is just barely able to do that x3
-
Update: I added a sound card :3
The timings are a bit fucked, I'll fix that at some point.
The way it works is slightly cursed, since protoflux doesn't have a way to output arbitrary audio. It drives the frequencies and amplitudes of 20 sine waves to reconstruct the sound.
The CPU is also really weak (running at 100 KHz here), all the audio data is preprocessed and all it has to do is copy it out of ROM and into the audio chip's memory, it is just barely able to do that x3
-
Update: I added a sound card :3
The timings are a bit fucked, I'll fix that at some point.
The way it works is slightly cursed, since protoflux doesn't have a way to output arbitrary audio. It drives the frequencies and amplitudes of 20 sine waves to reconstruct the sound.
The CPU is also really weak (running at 100 KHz here), all the audio data is preprocessed and all it has to do is copy it out of ROM and into the audio chip's memory, it is just barely able to do that x3