rocmate — gfx1101 (RX 7800/7700 XT)

Axolotl ✅ tested ROCm 6.2

RX 7800 XT / RX 7700 XT — QLoRA of 7B models fits in 16 GB. Same setup as gfx1100.

Same install as gfx1100. Use gradient_checkpointing: true in your YAML for memory savings.

ComfyUI ✅ tested ROCm 6.2

RX 7800 XT / RX 7700 XT — same setup as gfx1100. 16 GB VRAM limits Flux.1; SDXL and SD 1.5 run well.

ExLlamaV2 ✅ tested ROCm 6.2

RX 7800 XT / RX 7700 XT — works well. 16 GB VRAM suits 13B Q4 or 7B Q8 models.

faster-whisper 🟡 partial ROCm 6.2

RX 7800 XT / RX 7700 XT — same PyTorch ROCm approach as gfx1100. Transcription speed similar to gfx1100.

pip install torch torchaudio --index-url https://download.pytorch.org/whl/rocm6.2

llama.cpp ✅ tested ROCm 6.2

RX 7800 XT / RX 7700 XT — same HIP build as gfx1100. 16 GB VRAM; Q4_K_M models up to 32B. Vulkan build is an alternative if ROCm gives trouble.

Same HIP build as gfx1100.
Vulkan alternative: cmake -B build -DGGML_VULKAN=ON && cmake --build build --config Release -j$(nproc)

Ollama ✅ tested ROCm 6.3

RX 7800 XT / RX 7700 XT — works out of the box on Linux with ROCm 6.x, same as gfx1100.

Stable Diffusion WebUI ✅ tested ROCm 6.2

RX 7800 XT / RX 7700 XT — same setup as gfx1100. 16 GB VRAM; SDXL runs well, Flux.1 tight.

vLLM ✅ tested ROCm 6.2

RX 7800 XT / RX 7700 XT — works on Linux. 16 GB VRAM limits to <=13B models.