FitMyLLM
▸ INTEL· XE-HPG

Intel Arc Pro A50

Running LLMs on the Arc Pro A50 — the long read: which models fit at which quantisation, and the settings worth changing. · Or what a budget buys

VRAM
6 GB
BUDGET
BANDWIDTH
192
GB/S
MODELS Q4
130/449
29%
7B Q4 SPEED
~9
USABLE
▸ MODEL COVERAGE @ Q429% OF ALL
▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

Average speeds at Q4 quantization. Actual performance varies by model architecture and context length.

3B
~20
TOK/S
7B
~9
TOK/S
14B
—
7.9GB NEEDED
32B
—
18.0GB NEEDED
70B
—
39.4GB NEEDED
▸ MEASURED RIG REPORTS

We rent the machine and time every model on it: decode, VRAM peak, concurrency, watts and cost per million tokens. Yours may already be one of them — and two are free to read in full.

SEE THE REPORTS →
▸ SPECIFICATIONS
VRAM
6 GB
BANDWIDTH
192 GB/s
FP16 COMPUTE
9.6 TFLOPS
TDP
75W
MEMORY
GDDR6
ARCHITECTURE
Xe-HPG
COMPUTE UNITS
64
PCIE
Gen 4 x8
56
FAST MODELS · >30 TOK/S
Real-time chat speed
113
USABLE · >10 TOK/S
Comfortable for all tasks
130
TOTAL COMPATIBLE
Fit in VRAM at Q4
▸ DON’T WANT TO BUY?

Test Arc Pro A50 (or anything bigger) without committing. Pay by the second, cancel anytime.

Spin up in ~60s. Pay by the second. Cancel anytime.

Some links are affiliate links — we may earn a small commission at no extra cost to you. This helps keep FitMyLLM free and independent.

▸ COMPATIBLE MODELS· 130
S
Falcon-H1R Tiny 90M0.09B
FALCON·256K CTX· CHAT· REASONING
683
TOK/S · 9% VRAM
›
S
SmolLM2 135M0.135B
SMOLLM·2K CTX· CHAT
455
TOK/S · 10% VRAM
›
S
nomic-embed-text-v1.5 100M0.14B
EMBEDDING·8K CTX· CHAT
439
TOK/S · 10% VRAM
›
S
GPT-2 124M0.14B
GPT2·1K CTX· CHAT
439
TOK/S · 10% VRAM
›
S
SmolVLM 256M0.256B
SMOLLM·8K CTX· CHAT· VISION
240
TOK/S · 11% VRAM
›
S
Gemma 3 270M0.27B
GEMMA·32K CTX· CHAT
228
TOK/S · 11% VRAM
›
S
Snowflake Arctic Embed M v2.00.305B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
201
TOK/S · 11% VRAM
›
S
bge-large-en-v1.5 335M0.335B
EMBEDDING·1K CTX· CHAT
183
TOK/S · 12% VRAM
›
S
mxbai-embed-large-v10.335B
EMBEDDING·1K CTX· EMBEDDING
183
TOK/S · 12% VRAM
›
S
Snowflake Arctic Embed L0.335B
EMBEDDING·1K CTX· EMBEDDING
183
TOK/S · 12% VRAM
›
S
LFM2 350M0.35B
LFM·125K CTX· CHAT· TOOL_USE
176
TOK/S · 12% VRAM
›
S
SmolLM2 360M0.36B
SMOLLM·8K CTX· CHAT
171
TOK/S · 12% VRAM
›
S
GPT-2 Medium 345M0.38B
GPT2·1K CTX· CHAT
162
TOK/S · 12% VRAM
›
S
Qwen 2.5 0.5B0.5B
QWEN·32K CTX· CHAT
123
TOK/S · 13% VRAM
›
S
SmolVLM 500M0.5B
SMOLLM·8K CTX· CHAT· VISION
123
TOK/S · 13% VRAM
›
S
Falcon-H1 0.5B0.52B
FALCON·128K CTX· CHAT
118
TOK/S · 13% VRAM
›
S
BGE-M30.568B
EMBEDDING·8K CTX· EMBEDDING
108
TOK/S · 14% VRAM
›
S
Snowflake Arctic Embed L v2.00.568B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
108
TOK/S · 14% VRAM
›
S
Qwen3-Embedding 0.6B0.6B
EMBEDDING·32K CTX· EMBEDDING· MULTILINGUAL
102
TOK/S · 14% VRAM
›
S
Falcon-H1R Tiny 0.6B0.6B
FALCON·32K CTX· CHAT· REASONING
102
TOK/S · 14% VRAM
›
S
Falcon Perception 0.6B0.6B
FALCON·4K CTX· VISION
102
TOK/S · 14% VRAM
›
S
Qwen 1.5 0.5B0.62B
QWEN·32K CTX· CHAT
99
TOK/S · 14% VRAM
›
S
LFM2 700M0.74B
LFM·125K CTX· CHAT· TOOL_USE
83
TOK/S · 16% VRAM
›
S
Qwen3 0.6B0.75B
QWEN·32K CTX· CHAT· REASONING
82
TOK/S · 16% VRAM
›
S
GPT-2 Large 774M0.81B
GPT2·1K CTX· CHAT
76
TOK/S · 16% VRAM
›
S
Qwen 3.5 0.8B0.87B
QWEN·256K CTX· CHAT· CODING· MULTILINGUAL
71
TOK/S · 17% VRAM
›
S
InternVL3 1B0.94B
OTHER·32K CTX· CHAT· VISION
65
TOK/S · 18% VRAM
›
A
MiniCPM5 1B1.08B
MINICPM·128K CTX· CHAT· REASONING· MULTILINGUAL
57
TOK/S · 19% VRAM
›
A
TinyLlama 1.1B1.1B
LLAMA·2K CTX· CHAT
56
TOK/S · 19% VRAM
›
A
LFM2.5-1.2B-Thinking1.2B
LFM·122K CTX· CHAT· REASONING· TOOL_USE
51
TOK/S · 20% VRAM
›
A
Llama-3.2-1B1.2B
LLAMA·4K CTX· CHAT
51
TOK/S · 20% VRAM
›
A
LFM2 1.2B1.2B
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
51
TOK/S · 20% VRAM
›
A
Zamba2 1.2B1.2B
OTHER·4K CTX· CHAT
51
TOK/S · 20% VRAM
›
A
EXAONE-4.0-1.2B1.3B
EXAONE·64K CTX· CHAT
47
TOK/S · 21% VRAM
›
A
OPT 1.3B1.3B
OPT·2K CTX· CHAT
47
TOK/S · 21% VRAM
›
A
MiniCPM-V 4.61.3B
OTHER·256K CTX· CHAT· VISION
47
TOK/S · 21% VRAM
›
A
DeepSeek Coder 1.3B1.35B
DEEPSEEK·16K CTX· CODING
46
TOK/S · 22% VRAM
›
A
Phi-1 1.3B1.42B
PHI·2K CTX· CODING
43
TOK/S · 23% VRAM
›
A
Phi-1.5 1.3B1.42B
PHI·2K CTX· CHAT· CODING
43
TOK/S · 23% VRAM
›
A
granite-4.0-h-tiny 6.9B6.9BMoE
GRANITE·128K CTX· CHAT
41
TOK/S · 78% VRAM
›
▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your Arc Pro A50.

WHAT THIS CARD IS WORTH

Arc Pro A50 holds 130 of the models in our catalogue and is, in practice, a Q4_K_M card — the largest it takes is Falcon-H1 7B at Q4_K_M.

TOKENS/SEC PER $100
—
8B at Q4_K_M, so cards compare like for like
VRAM PER $100
—
what memory costs on this card
THE BIGGEST IT TAKES
Falcon-H1 7B7.59B · Q4_K_M5.4 GB7 tok/sEST
Falcon-H1R 7B7.59B · Q4_K_M5.4 GB7 tok/sEST
Falcon Mamba 7B7.27B · Q4_K_M5.0 GB8 tok/sEST
WizardLM 2 7B7B · Q4_K_M5.3 GB8 tok/sEST
StarCoder2 7B7B · Q4_K_M5.0 GB8 tok/sEST
Dolly v2 7B6.9B · Q4_K_M5.3 GB8 tok/sEST
granite-4.0-h-tiny 6.9B6.9B · Q4_K_M5.0 GB38 tok/sEST
ChatGLM2 6B6.24B · Q5_K_M5.1 GB8 tok/sEST
AGAINST CARDS OF SIMILAR MEMORY
RTX 3080 10GB10 GB$42917.7 tok/s per $100
Arc B57010 GB$2198.2 tok/s per $100
Radeon RX 670010 GB$29912.0 tok/s per $100
Radeon RX 6750 GRE 10 GB10 GB$22915.7 tok/s per $100
Arc Pro A506 GB——

Neighbours in memory rather than in price: memory decides whether a card can do the job at all, so two cards of the same size at different prices is the comparison you are making. Ordered by memory, then bandwidth — not by the value column, which is worked out from bandwidth and price alone and therefore rewards a cheap card whatever its software stack does to that bandwidth in practice. Read it as one input, not as a ranking.

▸ DEVICE UNDER TEST

Intel Arc Pro A50 — 6 GB VRAM.

▸ ARC PRO A50 SPEC
BRAND
Intel
VRAM
6 GB GDDR6
BANDWIDTH
192 GB/s
FP16 COMPUTE
9.6 TFLOPS
FP32 COMPUTE
4.8 TFLOPS
TDP
75 W
ARCHITECTURE
Xe-HPG
▸ AI CAPABILITY
130/ 449 models @ Q4

With 6 GB VRAM and 192 GB/s bandwidth, this GPU handles models up to 7B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~22 tok/s.

RENT IT — LIVE PRICES
Checking what the clouds are charging…
§ 01TOP MODELS FOR ARC PRO A50
130 FIT · SHOWING 20
MODELSIZEVRAM Q4TOK/SAVG
Alpaca 7B7B4.8 GB927.7
Baichuan2 7B7B4.8 GB921.5
Vicuna 7B7B4.8 GB922.0
MPT-7B7B4.8 GB97.8
Orca 2 7B7B4.8 GB926.1
WizardLM 2 7B7B4.8 GB926.1
StarCoder2 7B7B4.8 GB917.0
WizardCoder Python 7B7B4.8 GB953.7
WizardLM 7B7B4.8 GB915.5
OLMo 3.1 RLZero 7B Code7B4.8 GB921.8
OLMo 3.1 RLZero 7B Math7B4.8 GB921.8
Dolly v2 7B6.9B4.7 GB97.0
granite-4.0-h-tiny 6.9B6.9B4.7 GB4149.2
Llama 2 7B6.74B4.6 GB921.1
CodeLlama 7B6.74B4.6 GB928.1
LLaMA 1 7B6.74B4.6 GB930.8
DeepSeek Coder 6.7B6.7B4.6 GB923.6
OPT 6.7B6.7B4.6 GB918.5
ChatGLM2 6B6.24B4.3 GB1020.7
ChatGLM3 6B6.24B4.3 GB1042.7