When will the project get Android support?

Probably never. Not everything that runs on Windows, Linux or Mac can successfully be ported to Android.

If it was possible to easily port over, the computing time would probably be in the order of several days to maybe to a week plus for tasks that would run for several hours on a desktop computer.

---

Thanks for the response.

Might not be possible to utilize the Adreno mobile GPU's, but the latest Snapdragon 8 Elite chip has performance on par, if not better, than the latest Iphones.

And since you wrote, in the recent post about Apple Silicon support, that the M1 chip, and Apple Silicon CPU's, LOVE the tasks from this project, from the one application that now has support, I would imagine that performance on the latest Android chip, like in my Samsung Galaxy S25, would also be surprisingly good, not to mention very efficient.

Well, there are 3 other working Android BOINC projects at the moment, with more WU's that will most likely ever be processed on the platform, but hopefully Minecraft@Home will get Android support one day in the future :)

I built the Repeating decoration seeds OpenCL for Android

The good thing is that the app doesn’t require FP64 support. However, despite strong theoretical peak performance, mobile Adreno GPUs struggle with compiling more complex code due to their minimalistic cache system. So I had to lower the intensity and remove some inlining; otherwise, the app wouldn’t build at all.

The results aren't impressive at all

                                                                  runtime         fp32          int32        memory
nvidia t600 (10CU, turing)                                        41.169s         2.1tflops   2.1tiops      145gbps
9950x igpu (2CU rdna 2, shared mem)                              299.616s         0.5tflops   0.13tiops      60gbps
ardeno 750 (6CU)                                             216 000.000s         2.5tflops   0.5tiops       50gbps

There are certainly ways to optimize the code for those GPUs to a reasonable performance, but you get the point...

41 seconds seems too fast for a T600 on the repeating decoration seeds tasks. was this with a shorter than production test task?

@ian It's just a benchmark ./repeatdecor-cl from https://github.com/BoySanic/repeatdecor-cl

I can run tests with real work units, but it’ll probably take forever, and checkpoints are still broken.

./repeatdecor-cl -s 1800780143971532800 -e 1800788940064555008
=== OpenCL Platform Detection ===
Found 1 OpenCL platform(s)

Platform 0:
  Name: QUALCOMM Snapdragon(TM)
  Vendor: QUALCOMM
  Version: OpenCL 3.0 QUALCOMM build: 0762.33.2

Using platform 0

=== OpenCL Device Detection ===
Found 1 device(s)

  OpenCL Version: OpenCL 3.0 Adreno(TM) 750
  OpenCL C Version: OpenCL C 3.0 Adreno(TM) 750
Device 0:
  Name: QUALCOMM Adreno(TM) 750
  Vendor: QUALCOMM
  Driver Version: OpenCL 3.0 QUALCOMM build: 0762.33.2 Compiler E031.45.02.23
  Device Version: OpenCL 3.0 Adreno(TM) 750
  Type: GPU
  Compute Units: 6
  Max Work Group Size: 256
  Global Memory: 5.41 GB
  Local Memory: 32.00 KB
  Max Clock Frequency: 1 MHz

Selected device 0

Max work group size: 256
=== OpenCL Context and Queue Creation ===
Context created successfully
Command queue created successfully

=== Kernel Loading and Compilation ===
Kernel source loaded successfully (7383 bytes)
Program created successfully
Building program...
Program built successfully
Kernel created successfully

=== Buffer Creation ===
Results buffer created (32768 bytes)
Results count buffer created (4 bytes)
Checksum buffer created (4 bytes)

=== Starting Processing ===
Work size: 8388608
Total iterations: 1048576
Batch size: 8388608
Progress: 0.0% (361/1048576) ETA: 78h 14m 13s

whereas the T600 can complete it in an hour and "fast" GPUs in a few minutes.

I tried using an anonymous platform, but the server returned a 500 internal server error. Not sure if that’s due to a bug or if it’s intentionally blocked....

anonymous apps are not enabled here.