Is your feature request related to a problem? Please describe.
I am attempting to onboard a high-capacity enterprise AI machine to the FluxEdge marketplace, but I am hitting a hard architecture block. My machine is an ASUS Ascent GX10 featuring the Grace Blackwell Superchip (ARM64 CPU + Blackwell GPU) with 128GB of unified memory.
Currently, the Linux setup script explicitly downloads the fluxcore-linux-amd64 binary. When systemd attempts to execute it as a background service, it instantly fails with a status=126 architecture mismatch error. It is frustrating to have massive AI compute power available but be structurally blocked from supplying it to the network because there is no native client for enterprise ARM environments.
Describe the solution you'd like
Please compile and release a native arm64 binary for the Linux FluxCore client. This would allow enterprise ARM64 hardware to pass the bare-metal hardware validation algorithms and supply high-tier tensor compute and unified memory to FluxEdge tenants.
Describe alternatives you've considered
-
Box64 / FEX Emulation: I considered forcing the x86 binary to run via Box64. However, because FluxCore requires low-level NVML driver polling and bare-metal PCIe/memory bus scanning, the emulator fails to pass through these hardware calls accurately. Furthermore, doing so would likely flag the system's anti-spoofing algorithms due to the mismatched architecture reporting.
-
Hosting a FluxNode instead: While FluxNodes already natively support ARM64 architectures, this forces the hardware into an infrastructure role (CPU/RAM only) rather than allowing the massive integrated GPU and unified memory to be rented out via the Edge marketplace.
Additional context
The machine is running Ubuntu-based DGX OS on ARM64 architecture with the NVIDIA 580 enterprise driver branch. Given the influx of ARM-based AI superchips (like Grace Hopper and Grace Blackwell) entering the market, adding native ARM64 support to the compute client would open FluxEdge to a significantly higher tier of enterprise hardware.
Is your feature request related to a problem? Please describe.
I am attempting to onboard a high-capacity enterprise AI machine to the FluxEdge marketplace, but I am hitting a hard architecture block. My machine is an ASUS Ascent GX10 featuring the Grace Blackwell Superchip (ARM64 CPU + Blackwell GPU) with 128GB of unified memory.
Currently, the Linux setup script explicitly downloads the fluxcore-linux-amd64 binary. When systemd attempts to execute it as a background service, it instantly fails with a status=126 architecture mismatch error. It is frustrating to have massive AI compute power available but be structurally blocked from supplying it to the network because there is no native client for enterprise ARM environments.
Describe the solution you'd like
Please compile and release a native arm64 binary for the Linux FluxCore client. This would allow enterprise ARM64 hardware to pass the bare-metal hardware validation algorithms and supply high-tier tensor compute and unified memory to FluxEdge tenants.
Describe alternatives you've considered
Box64 / FEX Emulation: I considered forcing the x86 binary to run via Box64. However, because FluxCore requires low-level NVML driver polling and bare-metal PCIe/memory bus scanning, the emulator fails to pass through these hardware calls accurately. Furthermore, doing so would likely flag the system's anti-spoofing algorithms due to the mismatched architecture reporting.
Hosting a FluxNode instead: While FluxNodes already natively support ARM64 architectures, this forces the hardware into an infrastructure role (CPU/RAM only) rather than allowing the massive integrated GPU and unified memory to be rented out via the Edge marketplace.
Additional context
The machine is running Ubuntu-based DGX OS on ARM64 architecture with the NVIDIA 580 enterprise driver branch. Given the influx of ARM-based AI superchips (like Grace Hopper and Grace Blackwell) entering the market, adding native ARM64 support to the compute client would open FluxEdge to a significantly higher tier of enterprise hardware.