Supported Hardware & Languages¶
Makora runs your code on real GPUs and accelerators in the cloud. This page lists all available devices and programming languages.
Device Table¶
| Vendor | Device | API Name (generate) |
Profiler available? | Default Language | Supported Languages |
|---|---|---|---|---|---|
| NVIDIA | H100 | H100 |
Yes |
CUDA | CUDA, Triton, CuteDSL |
| NVIDIA | H200 | H200 |
No |
CUDA | CUDA, Triton, CuteDSL |
| NVIDIA | B200 | B200 |
No |
CUDA | CUDA, Triton, CuteDSL |
| NVIDIA | L40S | L40S |
No |
CUDA | CUDA, Triton, CuteDSL |
| AMD | MI300X | MI300X |
No |
HIP | HIP, Triton |
| Qualcomm | Adreno 830 | Adreno 830 |
No |
OpenCL | OpenCL |
| Qualcomm | Adreno 750 | Adreno 750 |
No |
OpenCL | OpenCL |
| Qualcomm | Hexagon v79 | Hexagon v79 |
No |
Ripple | Ripple |
| Qualcomm | Hexagon v75 | Hexagon v75 |
No |
Ripple | Ripple |
The --device Flag¶
You can use --device or -d
Use the device enum name directly:
makora generate --file problem.py --device H100
makora generate --file problem.py --d MI300X
makora evaluate problem.py solution.py --device "Adreno 830"
makora check problem.py --device L40S
Programming Languages¶
| Language | Value | Used With |
|---|---|---|
| CUDA | cuda |
NVIDIA GPUs (H100, H200, B200, L40S) |
| Triton | triton |
NVIDIA GPUs, AMD MI300X |
| CuteDSL | cutedsl |
NVIDIA GPUs (H100, H200, B200, L40S) |
| HIP | hip |
AMD MI300X |
| OpenCL | opencl |
Qualcomm Adreno GPUs |
| Ripple | ripple |
Qualcomm Hexagon NPUs |
Selecting a Language¶
If you don't specify --language, Makora uses the device's default (shown in the device table above).
To explicitly choose a language:
# Use Triton on H100 (instead of default CUDA)
makora generate --file problem.py --device H100 --language triton
# Use Triton on MI300X (instead of default HIP)
makora generate --file problem.py --device MI300X --language triton
Makora will reject language/device combinations that aren't supported (e.g., CUDA on MI300X).