Arc A MobileLaptopAlchemistMOBILEoneAPI
Operating mode
Choose the operating mode for this hardware
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.
About this GPU for AI
The Arc A370M 4GB is Intel's entry-level Alchemist mobile GPU, found in affordable laptops and thin-and-light designs. Its 4 GB of GDDR6 severely constrains AI inference to small quantized models — it can handle 3B or 7B Q4 models only with some CPU offloading. As an entry point to Intel's oneAPI ecosystem on mobile, it is better suited for light AI workloads and experimentation than production inference. The Vulkan backend in llama.cpp provides a simpler setup path than the full oneAPI SYCL stack.
Beyond LLMs
AI Capability Matrix
What AI tasks this GPU can handle — from text generation to image and video creation.
| Capability | Status | Representative Model | Detail |
|---|
| LLM Chat (7B) | Won’t fit | Llama 3.1 8B Q4 | — |
| LLM Coding (30B) | Won’t fit | Qwen 3 30B Q4 | — |
| LLM Large (70B) |
laptop-gpubudget-friendlyoneapi-sycllimited-vram
Specifications
Compute
FP168 TFLOPS
INT864 TOPS
ArchitectureAlchemist
Memory
VRAM4 GB
Bandwidth112 GB/s
General
FamilyArc A Mobile
SegmentLaptop
InterconnectMOBILE
Compute PlatformONEAPI
Key Features
Intel Xe Matrix Extensions (XMX) for INT8/FP16 acceleration4 GB GDDR6 at 112 GB/s bandwidthSYCL/oneAPI and Vulkan backend support in llama.cpp64 TOPS INT8 computeMobile PCIe interfaceAlchemist (Xe HPG) entry-level mobile architecture
For AI Workloads
Strengths
- Enables discrete GPU inference on budget laptops that otherwise rely entirely on CPU
- Lower power consumption keeps laptop battery life manageable during inference
- Vulkan backend offers a simpler setup path for casual LLM use
- Entry point to Intel oneAPI ecosystem for experimentation
Considerations
- 4 GB VRAM is a hard constraint — most 7B models require CPU offloading, reducing speed significantly
- 112 GB/s memory bandwidth is very low, making token generation slow even for models that fit
- oneAPI ecosystem complexity amplified on laptops with hybrid GPU configurations
- Not practical for regular local LLM workflows; better suited as a CPU-assist than a standalone inference device
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.
Process: TSMC N6Platform: ONEAPIPrecisions: FP32, FP16, INT8
Recommendations by Workload
Qwen 3 1.7B 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.
Decode 23.8 tok/s · 16K ctx · llama.cppEST.
StarCoder2 3B is a specialized fit for Coding. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope.
Decode 34.7 tok/s · 56K ctx · llama.cppEST.
Just out of reach
Models you could run with an upgrade
High-quality models that need a bit more memory
30.5BTier 100Needs ~20.6 GB
397BTier 100Needs ~244.9 GB
123BTier 100Needs ~79.0 GB
1000BTier 100Needs ~615.0 GB
1000BTier 100Needs ~615.0 GB
Image & Video Generation
Diffusion Model Compatibility
1 of 52 models can generate images or video on your Intel Arc A370M 4GB
Upgrade paths
Upgrade from Intel Arc A370M 4GB
See what you unlock with more powerful hardware
Upgrade options
Upgrade options
Frequently Asked Questions
Intel Arc A370M 4GBCategory AvgRTX 2060 6GB
| Image Gen (SDXL) | Won't fit | SDXL 1.0 FP16 | ~~1m 4s per image |
| Image Gen (Flux) | Won't fit | Flux.1 Dev FP16 | ~~4m 46s per image |
| Image Gen (SD 3.5) | Won't fit | SD 3.5 Large FP16 | ~~5m 50s per image |
| Video Short (25f) | Won't fit | LTX Video 2B | ~~55.2s/frame |
| Video Long (100f) | Won't fit | Wan Video 14B | ~~2m 43s/frame |
StarCoder2 3B is a specialized fit for Agentic Coding. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope.
Decode 39.9 tok/s · 70K ctx · llama.cppEST.
ai21labs AI21 Jamba Reasoning 3B matches Reasoning 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.
Decode 34.7 tok/s · 56K ctx · llama.cppEST.
Qwen2.5 3B Instruct is viable for RAG, but is not the most specialized choice. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope.
Decode 39.9 tok/s · 70K ctx · llama.cppEST.
30.5B21.4 GB2 tok/s4K ctx
Image
| MAGI-1Video | 256×256 | ~2m 29s/frame | 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.
There are 4 upgrade path(s) from Intel Arc A370M 4GB: RTX 2060 6GB, Intel Arc A380 6GB. Upgrading would unlock larger models and faster inference speeds.
Buying advice
Should you buy Intel Arc A370M 4GB for local AI?
Usable for local AI with limits
Can run 2 of 50 top models, mostly smaller ones. Larger models need heavy quantization or won't fit.
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 93 additional models that do not fit on the current setup.
Want more headroom? RTX 2060 6GB (6.0 GB VRAM) is the next step up.
