Modern ARM CPU have a GPU, which is capable of OPENCL. You can have a look into the ARM developer documentation, which is open available!
https://developer.arm.com/documentation/
Mali-G57 MC2 (950 MHz) GPU
My Tablet, has two GPU Cores with 950Mhz and computes about 486 GFLOPS
Copyright © 2025 Einstein@Home. All rights reserved.
FWIW Mesa's Rusticl has been
)
FWIW Mesa's Rusticl has been adding OpenCL support for various GPU's used in ARM based SoCs for a bit now:
Qualcomm Adreno Rusticl-Based OpenCL Merged For Mesa 24.3
Mesa 24.3 Allows Rusticl On Asahi Gallium3D By Default - Apple Silicon
Rust-Written OpenCL Driver Now Works On Raspberry Pi GPUs - Broadcom V3D
OpenCL support in the Panfrost Mesa driver used for Mali GPUs (see Panfrost — The Mesa 3D Graphics Library latest documentation) will be included in Mesa 25.1:
Arm Mali Panfrost Driver Lands OpenCL C Support In Mesa 25.1
Given that Mesa 25.0 is working through 'rc' builds at the moment (see https://lists.freedesktop.org/archives/mesa-dev/2025-February/ ), it's going to be a bit before Mesa 25.1 hits OS distributions.
OpenCL support for Vivante GCxxx GPUs (See Etnaviv Wiki) used in some Freescale, Marvell, and Rockchip SoCs is also being worked on.
rusticl is irrelevant for
)
rusticl is irrelevant for android, but it may be useful for Qualcomm laptops running Linux.
I tested the OpenCL version on the Qualcomm Adreno™ 750 (one of the top mobile GPUs). BRP4 task took about 8 minutes which is good. I also tested PG (ap+gfn), but the performance with these apps was terrible.
The Mali GPU didn't work, but I have an old model. I believe it could be fixed with some dev effort.
Forget about Broadcom, RPis, or SOCs with even slower GPUs. Their performance is not viable for computing.
It is not possible to use the
)
It is not possible to use the gpu under android! We need hardware with modern ARM CPU and GPU, which is able to run linux or Windows. I say again, that the gpu of a mobile phone, could be used for crunching.
Even Nintendo switch with linux works good, because of the gpu!
The Mali G57 GPU computes about 486 GFLOPS ! A normal PC (multi cpu) is at 13 GFLOPS.
Ardeno 750
)
Ardeno 750 - 2,29TFLOPS, 20925p 3dmark, 460s / WU https://einsteinathome.org/cs/task/1729954059
Mali G57 - 0.28TFLOPS, 4495p 3dmark (the performance is doubled only in FP16 mode which no BOINC application utilizes. BRP is FP32)
so, your gpu should process a task in 3762s = about an hour.
Feel free to compare it with your CPU runtimes, and you'll see that it's nowhere near 37 times faster. While your CPU has more cores, you only have one GPU, and using it will limit your CPU’s performance due to shared power and thermal constraints.
RPI Videocore IV - 0.024 TFLOPS = 12,2 hours / WU. Can 4 CPU cores do better? I think so.
You can't directly compare gaming and compute performance, and peak performance alone doesn't tell the whole story. If you're expecting miraculous speedups on these GPUs, you'll be disappointed and the Einstein's app has been the best case I've tested so far.
ahorek's team schrieb: Feel
)
Okay, i believe you. you are right.