 |
VOOZH | about |
AMD
Operating mode
Use this to bias workload recommendations toward responsiveness, background autonomy, lighter serving, or multi-GPU scale-out.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
The RX 7900 XT 20GB is one of the high-end RDNA 3 consumer GPUs with official ROCm support. AMD lists it alongside the 7900 XTX as officially supported, meaning ROCm installers work without workarounds. The 20 GB of GDDR6 VRAM enables 13B models at FP16 and 34B+ models at Q4 — making it one of the most capable consumer AMD cards for local AI inference with a proper ROCm software stack.
Beyond LLMs
What AI tasks this GPU can handle — from text generation to image and video creation.
| Capability | Status | Representative Model | Detail |
|---|---|---|---|
| LLM Chat (7B) | Runs natively | Llama 3.1 8B Q4 | — |
| LLM Coding (30B) | Needs offload | Qwen 3 30B Q4 | — |
Architecture
RDNA 3 is AMD's chiplet-based GPU architecture, combining a 5nm Graphics Compute Die (GCD) with 6nm Memory Cache Dies (MCDs). It introduces AI accelerators and a new unified compute unit design.
AI Relevance
ROCm support for RDNA 3 is maturing but lags behind NVIDIA's CUDA ecosystem. AI accelerator units provide some inference acceleration, but lack the dedicated Tensor Core equivalent found in NVIDIA GPUs.
Cost vs cloud API
Assumes 4 hours/day of active inference at 61 tok/s, RX 7900 XT 20GB amortized over 36 months, US residential electricity ($0.15/kWh), blended cloud pricing at $10 per 1M tokens (GPT-4o / Claude Sonnet tier).
26.2M
Tokens/month at this pace
$30.6
Monthly local cost
$262
Same tokens on cloud API
$1.17
Local $/1M tokens
Break-even: pays for itself in 3.5 months vs cloud API at this workload. Price reference: $899 MSRP.
Chat
SQwen 3 14B matches Chat and keeps a practical fit profile. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Coding
SQwen 3.5 9B is a specialized fit for Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Agentic Coding
SJust out of reach
High-quality models that need a bit more memory
Image & Video Generation
39 of 52 models can generate images or video on your RX 7900 XT 20GB
| Model | Max Resolution | Gen Time | Grade |
|---|---|---|---|
| SD TurboImage | 512×512 | ~1s | S |
| Stable Diffusion 1.5Image | 512×768 | ~2s | S |
| Realistic Vision v5.1Image | 512×768 | ~2s | S |
| DreamShaper 8Image | 512×768 | ~2s | S |
| LCM DreamShaper v7 |
Upgrade paths
See what you unlock with more powerful hardware
Upgrade options
Unlocks 17 additional models that do not fit on the current setup.
~$2,499 MSRP
Unlocks 22 additional models that do not fit on the current setup.
~$599 MSRP
Unlocks 28 additional models that do not fit on the current setup.
~$1,899 MSRP
Unlocks 67 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 139%.
~$8,000 MSRP
Compare with similar
Related guides
| Won’t fit |
| Llama 3.1 70B Q4 |
| — |
| Image Gen (SDXL) | Runs natively | SDXL 1.0 FP16 | ~~8s per image |
| Image Gen (Flux) | Very constrained | Flux.1 Dev FP16 | ~~36s per image |
| Image Gen (SD 3.5) | Tight fit | SD 3.5 Large FP16 | ~~43.9s per image |
| Video Short (25f) | Runs natively | LTX Video 2B | ~~20.8s/frame |
| Video Long (100f) | Won't fit | Wan Video 14B | ~~20.4s/frame |
Qwen 3.5 9B is a specialized fit for Agentic Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Reasoning
SQwen 3 14B matches Reasoning and keeps a practical fit profile. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
RAG
AGranite 4.1 8B matches RAG and keeps a practical fit profile. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama.
| 512×768 |
| 600ms |
| S |
| PixArt-SigmaImage | 1024×1024 | ~8s | S |
| FramePack I2VVideo | 256×256 | ~14.7s/frame | S |
| SDXL TurboImage | 512×512 | ~1s | S |
| SDXL LightningImage | 1024×1024 | ~3s | S |
| Stable Diffusion XL 1.0Image | 1024×1024 | ~8s | S |
| Playground v2.5Image | 1024×1024 | ~12s | S |
| RealVisXL v5.0Image | 1024×1024 | ~9s | S |
| DreamShaper XLImage | 1024×1024 | ~9s | S |
| Juggernaut XL v9Image | 1024×1024 | ~9s | S |
| Animagine XL 3.1Image | 1024×1024 | ~9s | S |
| Pony Diffusion V6 XLImage | 1024×1024 | ~9s | S |
| Animagine XL 4.0Image | 1024×1024 | ~9s | S |
| Illustrious XLImage | 1024×1024 | ~9s | S |
| Wan Video 2.1 1.3BVideo | 256×256 | ~5.8s/frame | S |
| Stable Diffusion 3.5 MediumImage | 1024×1024 | ~14s | S |
| Flux.2 Klein 4BImage | 1024×1024 | ~2.4s | S |
| LTX Video 2BVideo | 512×512 | ~20.8s/frame | S |
| KolorsImage | 1024×1024 | ~16s | S |
| Stable CascadeImage | 1024×1024 | ~20s | S |
| AuraFlow v0.3Image | 1536×1536 | ~36s | A |
| Stable Diffusion 3.5 LargeImage | 1024×1024 | ~43.9s | A |
| Stable Diffusion 3.5 Large TurboImage | 1024×1024 | ~8s | A |
| CogVideoX 2BVideo | 256×256 | ~20.8s/frame | B |
| HunyuanVideoVideo | 256×256 | ~14.7s/frame | B |
| ChromaImage | 256×256 | ~8s | B |
| Z-Image TurboImage | 256×256 | ~16.5s | B |
| Flux.1 DevImage | 256×256 | ~36s | D |
| Flux.1 SchnellImage | 256×256 | ~7s | D |
| LTX Video 13BVideo | 256×256 | ~14.7s/frame | D |
| Flux.1 Kontext DevImage | 256×256 | ~39.9s | D |
| AnimateDiff v1.5.3Video | 512×768 | ~3.6s/frame | D |
| Cosmos Diffusion 7BVideo | 256×256 | ~22.1s/frame | D |
| CogVideoX 5BVideo | 256×256 | ~21s/frame | D |
| Wan2.2 TI2V 5BVideo | 256×256 | ~21s/frame | D |
| Flux.2 Klein 9BImage | 256×256 | ~4s | F |
| Flux.1 Fill DevImage | 256×256 | ~34s | F |
| Mochi 1 PreviewVideo | 256×256 | ~13.2s/frame | F |
| HunyuanVideo 1.5Video | 256×256 | ~12.3s/frame | F |
| Helios 14BVideo | 256×256 | ~15.1s/frame | F |
| SkyReels V2 14BVideo | 256×256 | ~15.1s/frame | F |
| Wan Video 2.1 14BVideo | 256×256 | ~15.1s/frame | F |
| Wan Video 2.2 14BVideo | 256×256 | ~15.1s/frame | F |
| Qwen ImageImage | 256×256 | ~13.5s | F |
| Qwen Image EditImage | 256×256 | ~13.5s | F |
| Flux.2 DevImage | 256×256 | ~6m 18s | F |
| MAGI-1Video | 256×256 | ~18.7s/frame | F |
| HunyuanImage 3.0Image | 256×256 | ~23.7s | F |
Image models estimated at 1024×1024 (28 steps, FP16). Video models estimated at 768×512 (25 frames, 30 steps, FP16). Actual performance varies with runtime and system load.
Buying advice
Good for local AI
Handles 21 of 50 top models. Smaller and mid-size models run comfortably.
20.0 GB
VRAM
$899
MSRP
$45/GB
Cost per GB VRAM
Best models for this GPU
What will limit you first
This setup is broadly balanced for this model.
No major red flags
This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.
Best upgrade itinerary
Unlocks 17 additional models that do not fit on the current setup.
Want more headroom? MacBook Pro M1 Max 32GB (32.0 GB unified memory) is the next step up.
