 |
VOOZH | about |
Intel
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 Arc A770 16GB is Intel's flagship Alchemist GPU and the most capable consumer Arc card for AI inference. Its 16 GB of GDDR6 — more than many competing cards at the same price — allows it to run 7B models at FP16 or 13B models at Q4 quantization entirely on-GPU. llama.cpp's SYCL backend supports it natively, and performance has improved significantly since launch with driver and oneAPI stack maturation. At roughly 37 tokens/second on LLaMA-2-7B Q4, it offers meaningful throughput for local inference at an accessible price.
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) | Won’t fit | Qwen 3 30B Q4 | — |
| LLM Large (70B) |
Architecture
Alchemist is Intel's first discrete GPU architecture under the Arc brand, using Xe-HPG cores manufactured on TSMC's N6 process. It features XMX (Xe Matrix Extensions) engines for AI acceleration.
AI Relevance
XMX engines provide some AI inference acceleration via oneAPI/SYCL. However, the software ecosystem for LLM inference on Intel Arc is still developing, with limited runtime support compared to CUDA.
Cost vs cloud API
Assumes 4 hours/day of active inference at 49 tok/s, Intel Arc A770 16GB amortized over 36 months, US residential electricity ($0.15/kWh), blended cloud pricing at $10 per 1M tokens (GPT-4o / Claude Sonnet tier).
21.3M
Tokens/month at this pace
$13.7
Monthly local cost
$213
Same tokens on cloud API
$0.645
Local $/1M tokens
Break-even: pays for itself in 1.7 months vs cloud API at this workload. Price reference: $349 MSRP.
Chat
SQwen 3.5 9B 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
31 of 52 models can generate images or video on your Intel Arc A770 16GB
| Model | Max Resolution | Gen Time | Grade |
|---|---|---|---|
| SD TurboImage | 512×512 | ~3.1s | S |
| Stable Diffusion 1.5Image | 512×768 | ~6.3s | S |
| Realistic Vision v5.1Image | 512×768 | ~6.3s | S |
| DreamShaper 8Image | 512×768 | ~6.3s | S |
| LCM DreamShaper v7 |
Upgrade paths
See what you unlock with more powerful hardware
Upgrade options
Unlocks 2 additional models that do not fit on the current setup.
~$1,099 MSRP
Unlocks 36 additional models that do not fit on the current setup.
~$599 MSRP
Unlocks 68 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 176%.
~$15,000 MSRP
Unlocks 81 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 280%.
~$8,000 MSRP
Compare with similar
| Won’t fit |
| Llama 3.1 70B Q4 |
| — |
| Image Gen (SDXL) | Runs natively | SDXL 1.0 FP16 | ~~25.2s per image |
| Image Gen (Flux) | Won't fit | Flux.1 Dev FP16 | ~~1m 53s per image |
| Image Gen (SD 3.5) | Runs with sequential offload | SD 3.5 Large FP16 | ~~6m 14s per image |
| Video Short (25f) | Runs natively | LTX Video 2B | ~~21.9s/frame |
| Video Long (100f) | Won't fit | Wan Video 14B | ~~1m 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 3.5 9B 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 |
| ~1.9s |
| S |
| PixArt-SigmaImage | 1024×1024 | ~25.2s | S |
| FramePack I2VVideo | 256×256 | ~46.2s/frame | S |
| SDXL TurboImage | 512×512 | ~3.1s | S |
| SDXL LightningImage | 1024×1024 | ~9.4s | S |
| Stable Diffusion XL 1.0Image | 1024×1024 | ~25.2s | S |
| Playground v2.5Image | 1024×1024 | ~37.8s | S |
| RealVisXL v5.0Image | 1024×1024 | ~28.3s | S |
| DreamShaper XLImage | 1024×1024 | ~28.3s | S |
| Juggernaut XL v9Image | 1024×1024 | ~28.3s | S |
| Animagine XL 3.1Image | 1024×1024 | ~28.3s | S |
| Pony Diffusion V6 XLImage | 1024×1024 | ~28.3s | S |
| Animagine XL 4.0Image | 1024×1024 | ~28.3s | S |
| Illustrious XLImage | 1024×1024 | ~28.3s | S |
| Wan Video 2.1 1.3BVideo | 256×256 | ~18.4s/frame | S |
| Stable Diffusion 3.5 MediumImage | 256×256 | ~2m 12s | S |
| Flux.2 Klein 4BImage | 256×256 | ~17s | S |
| LTX Video 2BVideo | 256×256 | ~21.9s/frame | S |
| KolorsImage | 256×256 | ~2m 14s | A |
| Stable CascadeImage | 1024×1024 | ~1m 3s | B |
| AuraFlow v0.3Image | 256×256 | ~3m 44s | B |
| Stable Diffusion 3.5 LargeImage | 256×256 | ~6m 14s | B |
| Stable Diffusion 3.5 Large TurboImage | 256×256 | ~1m 8s | B |
| CogVideoX 2BVideo | 256×256 | ~21.9s/frame | D |
| HunyuanVideoVideo | 256×256 | ~46.2s/frame | D |
| ChromaImage | 256×256 | ~25.2s | D |
| Z-Image TurboImage | 256×256 | ~52s | D |
| Flux.1 DevImage | 256×256 | ~1m 53s | F |
| Flux.1 SchnellImage | 256×256 | ~22s | F |
| LTX Video 13BVideo | 256×256 | ~46.2s/frame | F |
| Flux.1 Kontext DevImage | 256×256 | ~2m 6s | F |
| AnimateDiff v1.5.3Video | 512×768 | ~11.5s/frame | F |
| Cosmos Diffusion 7BVideo | 256×256 | ~36.1s/frame | F |
| CogVideoX 5BVideo | 256×256 | ~31.5s/frame | F |
| Wan2.2 TI2V 5BVideo | 256×256 | ~31.5s/frame | F |
| Flux.2 Klein 9BImage | 256×256 | ~12.6s | F |
| Flux.1 Fill DevImage | 256×256 | ~1m 47s | F |
| Mochi 1 PreviewVideo | 256×256 | ~41.6s/frame | F |
| HunyuanVideo 1.5Video | 256×256 | ~38.6s/frame | F |
| Helios 14BVideo | 256×256 | ~47.6s/frame | F |
| SkyReels V2 14BVideo | 256×256 | ~47.6s/frame | F |
| Wan Video 2.1 14BVideo | 256×256 | ~47.6s/frame | F |
| Wan Video 2.2 14BVideo | 256×256 | ~47.6s/frame | F |
| Qwen ImageImage | 256×256 | ~42.4s | F |
| Qwen Image EditImage | 256×256 | ~42.4s | F |
| Flux.2 DevImage | 256×256 | ~19m 51s | F |
| MAGI-1Video | 256×256 | ~59.1s/frame | F |
| HunyuanImage 3.0Image | 256×256 | ~1m 15s | 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
Usable for local AI with limits
Can run 11 of 50 top models, mostly smaller ones. Larger models need heavy quantization or won't fit.
16.0 GB
VRAM
$349
MSRP
$22/GB
Cost per GB VRAM
Best models for this GPU
What will limit you first
The raw memory story may look fine, but the software ecosystem is still a constraint here.
Runtime ecosystem is narrower than CUDA
Intel GPUs can look attractive on memory per dollar, but local AI tooling, kernels, and model coverage are still broader and easier on CUDA today.
Best upgrade itinerary
Prefer CUDA if you want the path of least resistance
If your goal is maximum runtime coverage, easier troubleshooting, and better support for new local AI releases, CUDA is usually still the safer upgrade path.
Unlocks 2 additional models that do not fit on the current setup.
Want more headroom? MacBook Pro M3 24GB (24.0 GB unified memory) is the next step up.
