- Intel Arc Pro16 GB · runs 13–14B models at 4-bit
Full specifications
- GPU
- Intel Xe2 (Battlemage)
- Xe cores
- 16
- Memory
- 16 GB GDDR6
- Memory interface
- 128-bit, 224 GB/s
- Interface
- PCIe 5.0 x8
- Board power
- 70 W (powered from the slot)
- Display outputs
- 4 × Mini DisplayPort 2.1
- Form factor
- Low-profile, dual-slot (full-height bracket included)
Good for
- Small workstations and servers without spare power cables
- Models up to 14B, embeddings and other small AI jobs
- Multi-monitor CAD and design desks
- Machines that stay on all day and need low running power
Not ideal for
- 30B-class models or larger
- Fast generation on bigger models
- Gaming
- CUDA-only software
Intel
Intel Arc Pro B50
The Arc Pro B50 is a low-profile workstation card with 16 GB of memory that draws all its power from the slot. It fits small workstations and compact servers where a full-size card will not go.
AED 1,209Charged in US dollars at checkout: $329. Import duties and taxes included.In stock. Dispatched in 1 to 3 business days.
- 30-day returns
- Free shipping to the UAE
- Ships from our warehouse in Malaysia
- Import duties and taxes included
- Shipping, returns and warranty
What it can run
Up to 13–14B at 4-bit, 7–8B at 8-bit (approximate)
Based on 16 GB of GPU memory. See every machine by model size
Guides for this product
Local AI · 5 min read
How Much Memory Do You Need to Run a Local LLM?
How much memory a local LLM needs: 7B to 120B+ models at 4-bit and 8-bit, how unified memory differs from VRAM, and which machines hold each size.
Read the guide →Graphics cards · 5 min read
Choosing a GPU for Local AI: RTX 5090, R9700, B60
Choose a GPU for local AI by memory, software and bandwidth: RTX 5090, Radeon AI PRO R9700, Arc Pro B60 and the 48 GB and 96 GB workstation cards.
Read the guide →About the Intel Arc Pro B50
The Intel Arc Pro B50 is a compact 16 GB workstation card that draws only 70 W from the slot, with no power cable. It runs models up to 14B at 4-bit and 8B at 8-bit, and fits small workstations and servers where a full-size card will not. Buy it for GPU acceleration in a small or low-power machine; for 30B models you need a 24 GB card or larger.
Key features
- 16 GB GDDR6 with ECC, 128-bit bus, 224 GB/s
- 16 Xe2 cores with 128 XMX engines, 170 TOPS INT8
- 10.65 TFLOPS FP32, up to 2.6 GHz
- 70 W board power, no power connector, 280 W PSU minimum
- Small form factor: dual slot, 6.6 in long
- 4 × DisplayPort 2.1, up to 8K at 60 Hz
- PCIe 5.0 x8
What it runs
About 90% of 16 GB is usable, which covers 13–14B models at 4-bit (about 10 GB) and 7–8B models at 8-bit. 30B-class models need about 20 GB even at 4-bit, so they do not fit. That makes the B50 a good match for small chat and coding models, embedding models, speech-to-text and image tools that fit in 16 GB. Check exact sizes in our model fit tool.
Performance context
Memory bandwidth is 224 GB/s. Small models still generate at a comfortable pace, but this is an entry card: the 24 GB Arc Pro B60 has about twice the bandwidth and 8 GB more memory. Where the B50 wins is power and size, not speed.
Size and power
The B50 draws all of its 70 W from the PCIe slot, so it needs no power cable, and Intel's minimum power supply recommendation is just 280 W. It is a small-form-factor, dual-slot card, 6.6 inches (about 168 mm) long and weighing 330 g. That suits compact workstations and small servers that cannot take a long card or lack spare power cables.
It also suits machines that run all day, such as a homelab server that needs GPU help for media work or a small model, without adding much to the power bill.
Outputs
Four DisplayPort 2.1 outputs drive up to four displays, up to 8K at 60 Hz, which suits multi-monitor CAD and design desks. ECC memory support is useful for long-running professional work.
Software
Intel provides Arc Pro drivers for Windows and Linux, with oneAPI and OpenVINO for AI work. PyTorch supports Intel GPUs through its XPU backend, and llama.cpp runs through its SYCL or Vulkan backends. It is not a CUDA card, so check that your tools support Intel GPUs before you buy.
What to pair it with
Any machine with a free PCIe x8 or x16 slot and room for a dual-slot card, including compact towers with small power supplies. For bigger models, step up to the Arc Pro B60 (24 GB) or a 32 GB card such as the Radeon AI PRO R9700. Our guide to memory for local LLMs helps you size the card to the model, and our graphics card guide for local AI covers the wider range.
Good for
- Small workstations and servers without spare power cables
- Models up to 14B, embeddings and other small AI jobs
- Multi-monitor CAD and design desks
- Machines that stay on all day and need low running power
Not ideal for
- 30B-class models or larger
- Fast generation on bigger models
- Gaming
- CUDA-only software
No power cable is needed, so it suits small-form-factor machines. 16 GB runs 7B to 14B models at 4-bit; anything larger needs a bigger card.
Who it's for
Small-workstation users and homelabbers who want GPU acceleration in a compact, low-power card.
Full specifications
- GPU
- Intel Xe2 (Battlemage)
- Xe cores
- 16
- Memory
- 16 GB GDDR6
- Memory interface
- 128-bit, 224 GB/s
- Interface
- PCIe 5.0 x8
- Board power
- 70 W (powered from the slot)
- Display outputs
- 4 × Mini DisplayPort 2.1
- Form factor
- Low-profile, dual-slot (full-height bracket included)
Reviews
What reviewers say
Puget Systems
Compact, low power and nearly silent, with good value for an entry-level workstation. NVIDIA's CUDA software still gives it the edge in some apps.
“Overall, we were very impressed with the B50.”
Read the review
Atomrig customer reviews
No customer reviews yet. Bought this from us? Tell us how it runs.
Related products

NVIDIA
NVIDIA GeForce RTX 5090
Runs up to 30–34B at 4-bit
- 32 GB
- NVIDIA Blackwell (GB202)
32 GB runs 30B-class models comfortably at 4-bit, but a 70B model will not fit on one card. For 70B, plan for two cards or a large unified-memory system.
AED 7,339
PowerColor
PowerColor AMD Radeon AI PRO R9700 32GB
Runs up to 30–34B at 4-bit
- 32 GB
- AMD RDNA 4 (Navi 48)
Check that your tools support ROCm on your operating system before buying. PyTorch and llama.cpp do; some niche tools still assume CUDA.
AED 4,769
ASRock
ASRock Intel Arc Pro B60 Creator 24GB
Runs up to 30–34B at 4-bit
- 24 GB
- Intel Xe2 (Battlemage)
This is a single-GPU card with 24 GB. It uses a PCIe 5.0 slot wired at x8 and also works in PCIe 4.0 slots; check ASRock's spec sheet for the exact length and power connector before you order.
AED 2,199
NVIDIA
NVIDIA RTX PRO 6000 Blackwell Workstation Edition (96 GB)
Runs up to 120B+ at 4-bit
- 96 GB
- NVIDIA Blackwell (GB202)
This is the 600 W Workstation Edition with a double flow-through cooler. It needs a 16-pin power cable, a case that takes a 12-inch card and good airflow.
AED 53,599