IN Intel Consumer Discrete GPU

Intel Arc B580 12GB

Runs downloadable LLMs up to ~14.5B at Q4, ~59-73.3 tok/s on an 8B model.

Intel Arc B580 12GB
12GB
VRAM
456
GB/s bandwidth
14.5B
Max model (Q4)
$249
MSRP

What you can run

At 8k context, FP16 KV. Full-speed tok/s holds up to "max ctx"; beyond it the context spills to system RAM (CPU/PCIe offload) and drops to the degraded figure.

Model sizeFP16Q8Q4
3B 44.1-54.7 t/s
≤54k ctx · then ~3.9-4.8 (slow)
88.2-109.4 t/s
≤100k ctx · then ~7.7-9.6 (slow)
157.4-195.4 t/s
≤120k ctx · then ~13.8-17.1 (slow)
7-8B won't run 33.1-41 t/s
≤12k ctx · then ~2.9-3.6 (slow)
59-73.3 t/s
≤39k ctx · then ~5.2-6.4 (slow)
13B won't run won't run 36.3-45.1 t/s
≤14k ctx · then ~3.2-4 (slow)
32B won't run won't run won't run
70B won't run won't run won't run
120B won't run won't run won't run
400B won't run won't run won't run

Estimate: tok/s ≈ bandwidth ÷ (params × bytes/param) × efficiency (Intel). Cyan = fits ~11 GB usable. Beyond "ctx", KV spills to system RAM (CPU/PCIe offload) (rough degraded speed shown).

What else you need for a full build

Minimum supporting components for a desktop built around the Intel Arc B580 12GB. Guidance — exact needs depend on your case, other parts and how much you offload to system RAM.

Power supply
≥ 550W · 80+ Gold
Headroom over the 190W card plus the rest of the build.
System RAM
≥ 16 GB · 32 GB recommended
At least the 12 GB of VRAM; 2× lets you CPU-offload models larger than VRAM.
Motherboard
1× free PCIe x16 slot (4.0/5.0)
A full-length x16 electrical slot; PCIe 4.0 is ample, 5.0 future-proofs.
CPU
Any modern multi-core
The GPU runs inference; more cores and memory channels only matter when offloading to RAM.
Storage
≥ 1 TB NVMe SSD
Weights are large — a 70B model at Q4 is ~40 GB; budget for a few.

Run popular models on the Intel Arc B580 12GB

Videos about the Intel Arc B580 12GB

Running DeepSeek R1 on Intel Arc B580: My Setup and Performance Experience

Gen AI on Intel Arc GPUs - Building a Dual Arc B580 LLM Inference Server! (24 GB VRAM!)

More hardware like the Intel Arc B580 12GB

About the Intel Arc B580 12GB

Intel Arc B580 12GB is a discrete GPU from Intel on Battlemage, 2024. For local genAI the numbers that matter are its 12 GB (how big a model + context fits) and 456 GB/s bandwidth (generation speed). Plan for a ~550W power supply.

Frequently asked questions

What size LLM can the Intel Arc B580 12GB run?

With ~11 GB usable it runs up to ~14.5B at Q4 (~4.5B at FP16) at 8k context. Bigger spills into system RAM (CPU/PCIe offload) and slows sharply.

How many tokens/sec does the Intel Arc B580 12GB generate?

Generation is bandwidth-bound. At 456 GB/s an 8B model at Q4 runs ~59-73.3 tok/s (estimate).

Why does long context slow it down?

The KV cache for the context must also fit fast memory; once weights+KV exceed 11 GB it spills to system RAM (CPU/PCIe offload) and throughput collapses. Each row shows the max usable context.

What else do I need to run the Intel Arc B580 12GB?

Besides the card: a ≥550W 80+ Gold power supply, at least 16GB of system RAM (2× the 12GB VRAM is ideal for CPU offload), a free PCIe x16 slot, any modern multi-core CPU, and a ≥1TB NVMe SSD for model weights.