FitMyLLM
ARM· BIFROST

ARM Mali-G52 MC1

Running LLMs on the ARM Mali-G52 MC1 — the long read: which models fit at which quantisation, and the settings worth changing. · Or what a budget buys

VRAM
2 GB
BUDGET
BANDWIDTH
8.5
GB/S
MODELS Q4
47/449
10%
7B Q4 SPEED
~1
SLOW
▸ MODEL COVERAGE @ Q410% OF ALL
▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

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

3B
1.7GB NEEDED
7B
3.9GB NEEDED
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
2 GB
BANDWIDTH
8.5 GB/s
FP16 COMPUTE
0.03 TFLOPS
TDP
1.5W
MEMORY
ARCHITECTURE
Bifrost
5
FAST MODELS · >30 TOK/S
Real-time chat speed
24
USABLE · >10 TOK/S
Comfortable for all tasks
47
TOTAL COMPATIBLE
Fit in VRAM at Q4
▸ DON’T WANT TO BUY?

Test ARM Mali-G52 MC1 (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· 47
S
Falcon-H1R Tiny 90M0.09B
FALCON·256K CTX· CHAT· REASONING
84
TOK/S · 27% VRAM
A
SmolLM2 135M0.135B
SMOLLM·2K CTX· CHAT
56
TOK/S · 29% VRAM
A
nomic-embed-text-v1.5 100M0.14B
EMBEDDING·8K CTX· CHAT
54
TOK/S · 29% VRAM
A
GPT-2 124M0.14B
GPT2·1K CTX· CHAT
54
TOK/S · 29% VRAM
B
SmolVLM 256M0.256B
SMOLLM·8K CTX· CHAT· VISION
30
TOK/S · 32% VRAM
B
Gemma 3 270M0.27B
GEMMA·32K CTX· CHAT
28
TOK/S · 33% VRAM
B
Snowflake Arctic Embed M v2.00.305B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
25
TOK/S · 34% VRAM
C
bge-large-en-v1.5 335M0.335B
EMBEDDING·1K CTX· CHAT
23
TOK/S · 35% VRAM
C
mxbai-embed-large-v10.335B
EMBEDDING·1K CTX· EMBEDDING
23
TOK/S · 35% VRAM
C
Snowflake Arctic Embed L0.335B
EMBEDDING·1K CTX· EMBEDDING
23
TOK/S · 35% VRAM
C
LFM2 350M0.35B
LFM·125K CTX· CHAT· TOOL_USE
22
TOK/S · 35% VRAM
C
SmolLM2 360M0.36B
SMOLLM·8K CTX· CHAT
21
TOK/S · 35% VRAM
C
GPT-2 Medium 345M0.38B
GPT2·1K CTX· CHAT
20
TOK/S · 36% VRAM
C
Falcon-H1 0.5B0.52B
FALCON·128K CTX· CHAT
15
TOK/S · 40% VRAM
C
Qwen 2.5 0.5B0.5B
QWEN·32K CTX· CHAT
15
TOK/S · 40% VRAM
C
SmolVLM 500M0.5B
SMOLLM·8K CTX· CHAT· VISION
15
TOK/S · 40% VRAM
D
Qwen3-Embedding 0.6B0.6B
EMBEDDING·32K CTX· EMBEDDING· MULTILINGUAL
13
TOK/S · 43% VRAM
D
Falcon-H1R Tiny 0.6B0.6B
FALCON·32K CTX· CHAT· REASONING
13
TOK/S · 43% VRAM
D
Falcon Perception 0.6B0.6B
FALCON·4K CTX· VISION
13
TOK/S · 43% VRAM
D
BGE-M30.568B
EMBEDDING·8K CTX· EMBEDDING
13
TOK/S · 42% VRAM
D
Snowflake Arctic Embed L v2.00.568B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
13
TOK/S · 42% VRAM
D
Qwen 1.5 0.5B0.62B
QWEN·32K CTX· CHAT
12
TOK/S · 43% VRAM
D
Qwen3 0.6B0.75B
QWEN·32K CTX· CHAT· REASONING
10
TOK/S · 47% VRAM
D
LFM2 700M0.74B
LFM·125K CTX· CHAT· TOOL_USE
10
TOK/S · 47% VRAM
D
Qwen 3.5 0.8B0.87B
QWEN·256K CTX· CHAT· CODING· MULTILINGUAL
9
TOK/S · 51% VRAM
D
GPT-2 Large 774M0.81B
GPT2·1K CTX· CHAT
9
TOK/S · 49% VRAM
D
InternVL3 1B0.94B
OTHER·32K CTX· CHAT· VISION
8
TOK/S · 53% VRAM
F
TinyLlama 1.1B1.1B
LLAMA·2K CTX· CHAT
7
TOK/S · 58% VRAM
F
MiniCPM5 1B1.08B
MINICPM·128K CTX· CHAT· REASONING· MULTILINGUAL
7
TOK/S · 57% VRAM
F
DeepSeek Coder 1.3B1.35B
DEEPSEEK·16K CTX· CODING
6
TOK/S · 66% VRAM
F
EXAONE-4.0-1.2B1.3B
EXAONE·64K CTX· CHAT
6
TOK/S · 64% VRAM
F
OPT 1.3B1.3B
OPT·2K CTX· CHAT
6
TOK/S · 64% VRAM
F
MiniCPM-V 4.61.3B
OTHER·256K CTX· CHAT· VISION
6
TOK/S · 64% VRAM
F
LFM2.5-1.2B-Thinking1.2B
LFM·122K CTX· CHAT· REASONING· TOOL_USE
6
TOK/S · 61% VRAM
F
Llama-3.2-1B1.2B
LLAMA·4K CTX· CHAT
6
TOK/S · 61% VRAM
F
LFM2 1.2B1.2B
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
6
TOK/S · 61% VRAM
F
Zamba2 1.2B1.2B
OTHER·4K CTX· CHAT
6
TOK/S · 61% VRAM
F
GPT-2 XL 1.5B1.61B
GPT2·1K CTX· CHAT
5
TOK/S · 74% VRAM
F
stablelm-2-1_6b1.6B
STABLELM·4K CTX· CHAT
5
TOK/S · 73% VRAM
F
Falcon-H1 1.5B1.55B
FALCON·128K CTX· CHAT· CODING
5
TOK/S · 72% VRAM
▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your ARM Mali-G52 MC1.

WHAT THIS CARD IS WORTH

ARM Mali-G52 MC1 holds 47 of the models in our catalogue and is, in practice, a Q4_K_M card — the largest it takes is Moondream2 1.9B 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
Moondream2 1.9B1.9B · Q4_K_M1.7 GB5 tok/sEST
Falcon3-1B1.67B · Q5_K_M1.8 GB5 tok/sEST
LFM2-VL 1.6B1.6B · Q5_K_M1.7 GB6 tok/sEST
Falcon-H1 1.5B1.55B · Q6_K1.7 GB5 tok/sEST
Qwen2.5-Coder-1.5B1.5B · Q6_K1.6 GB5 tok/sEST
Qwen2 Math 1.5B1.5B · Q6_K1.6 GB5 tok/sEST
Qwen 2.5 1.5B1.5B · Q6_K1.6 GB5 tok/sEST
Stella en 1.5B v51.5B · Q6_K1.6 GB5 tok/sEST
AGAINST CARDS OF SIMILAR MEMORY
M2 (8GB)5 GB$5992.8 tok/s per $100
M3 (8GB)5 GB$5992.8 tok/s per $100
M1 (8GB)5 GB$4992.2 tok/s per $100
RTX 2060 6GB6 GB$15026.7 tok/s per $100
ARM Mali-G52 MC12 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

ARM Mali-G52 MC1 — 2 GB VRAM.

ARM MALI-G52 MC1 SPEC
BRAND
Apple
VRAM
2 GB
BANDWIDTH
8.5 GB/s
FP16 COMPUTE
0.03 TFLOPS
FP32 COMPUTE
0.015 TFLOPS
TDP
1.5 W
ARCHITECTURE
Bifrost
▸ AI CAPABILITY
47/ 449 models @ Q4

With 2 GB VRAM and 8.5 GB/s bandwidth, this GPU handles models up to 1.61B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~1 tok/s.

§ 01TOP MODELS FOR ARM MALI-G52 MC1
47 FIT · SHOWING 20
MODELSIZEVRAM Q4TOK/SAVG
GPT-2 XL 1.5B1.61B1.5 GB55.1
stablelm-2-1_6b1.6B1.5 GB59.5
Falcon-H1 1.5B1.55B1.4 GB543.8
Qwen2.5-Coder-1.5B1.5B1.4 GB519.6
Qwen2 Math 1.5B1.5B1.4 GB519.6
Qwen 2.5 1.5B1.5B1.4 GB530.2
Yi Coder 1.5B1.5B1.4 GB514.6
Stella en 1.5B v51.5B1.4 GB5
Phi-1 1.3B1.42B1.4 GB57.2
Phi-1.5 1.3B1.42B1.4 GB57.2
DeepSeek Coder 1.3B1.35B1.3 GB616.8
EXAONE-4.0-1.2B1.3B1.3 GB618.9
OPT 1.3B1.3B1.3 GB65.3
MiniCPM-V 4.61.3B1.3 GB621.5
LFM2.5-1.2B-Thinking1.2B1.2 GB619.6
Llama-3.2-1B1.2B1.2 GB610.1
LFM2 1.2B1.2B1.2 GB615.6
Zamba2 1.2B1.2B1.2 GB641.5
TinyLlama 1.1B1.1B1.2 GB713.6
MiniCPM5 1B1.08B1.1 GB726.5