FitMyLLM
▸ NVIDIA· KEPLER

NVIDIA GeForce GTX 670MX

Running LLMs on the GeForce GTX 670MX — the long read: which models fit at which quantisation, and the settings worth changing. · Or what a budget buys

VRAM
3 GB
BUDGET
BANDWIDTH
67
GB/S
MODELS Q4
81/449
18%
7B Q4 SPEED
~8
USABLE
▸ MODEL COVERAGE @ Q418% OF ALL
▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

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

3B
~18
TOK/S
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
3 GB
BANDWIDTH
67 GB/s
FP16 COMPUTE
1.2 TFLOPS
TDP
75W
MEMORY
GDDR5
ARCHITECTURE
Kepler
CUDA CORES
960
PCIE
Gen 3 x16
51
FAST MODELS · >30 TOK/S
Real-time chat speed
81
USABLE · >10 TOK/S
Comfortable for all tasks
81
TOTAL COMPATIBLE
Fit in VRAM at Q4
▸ DON’T WANT TO BUY?

Test GeForce GTX 670MX (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· 81
S
Falcon-H1R Tiny 90M0.09B
FALCON·256K CTX· CHAT· REASONING
596
TOK/S · 18% VRAM
›
S
SmolLM2 135M0.135B
SMOLLM·2K CTX· CHAT
397
TOK/S · 19% VRAM
›
S
nomic-embed-text-v1.5 100M0.14B
EMBEDDING·8K CTX· CHAT
383
TOK/S · 19% VRAM
›
S
GPT-2 124M0.14B
GPT2·1K CTX· CHAT
383
TOK/S · 19% VRAM
›
S
SmolVLM 256M0.256B
SMOLLM·8K CTX· CHAT· VISION
209
TOK/S · 21% VRAM
›
S
Gemma 3 270M0.27B
GEMMA·32K CTX· CHAT
199
TOK/S · 22% VRAM
›
S
Snowflake Arctic Embed M v2.00.305B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
176
TOK/S · 22% VRAM
›
S
bge-large-en-v1.5 335M0.335B
EMBEDDING·1K CTX· CHAT
160
TOK/S · 23% VRAM
›
S
mxbai-embed-large-v10.335B
EMBEDDING·1K CTX· EMBEDDING
160
TOK/S · 23% VRAM
›
S
Snowflake Arctic Embed L0.335B
EMBEDDING·1K CTX· EMBEDDING
160
TOK/S · 23% VRAM
›
S
LFM2 350M0.35B
LFM·125K CTX· CHAT· TOOL_USE
153
TOK/S · 23% VRAM
›
S
SmolLM2 360M0.36B
SMOLLM·8K CTX· CHAT
149
TOK/S · 24% VRAM
›
S
GPT-2 Medium 345M0.38B
GPT2·1K CTX· CHAT
141
TOK/S · 24% VRAM
›
S
Qwen 2.5 0.5B0.5B
QWEN·32K CTX· CHAT
107
TOK/S · 26% VRAM
›
S
SmolVLM 500M0.5B
SMOLLM·8K CTX· CHAT· VISION
107
TOK/S · 26% VRAM
›
S
Falcon-H1 0.5B0.52B
FALCON·128K CTX· CHAT
103
TOK/S · 27% VRAM
›
S
BGE-M30.568B
EMBEDDING·8K CTX· EMBEDDING
94
TOK/S · 28% VRAM
›
S
Snowflake Arctic Embed L v2.00.568B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
94
TOK/S · 28% VRAM
›
S
Qwen3-Embedding 0.6B0.6B
EMBEDDING·32K CTX· EMBEDDING· MULTILINGUAL
89
TOK/S · 29% VRAM
›
S
Falcon-H1R Tiny 0.6B0.6B
FALCON·32K CTX· CHAT· REASONING
89
TOK/S · 29% VRAM
›
S
Falcon Perception 0.6B0.6B
FALCON·4K CTX· VISION
89
TOK/S · 29% VRAM
›
S
Qwen 1.5 0.5B0.62B
QWEN·32K CTX· CHAT
86
TOK/S · 29% VRAM
›
S
LFM2 700M0.74B
LFM·125K CTX· CHAT· TOOL_USE
72
TOK/S · 31% VRAM
›
S
Qwen3 0.6B0.75B
QWEN·32K CTX· CHAT· REASONING
71
TOK/S · 32% VRAM
›
S
GPT-2 Large 774M0.81B
GPT2·1K CTX· CHAT
66
TOK/S · 33% VRAM
›
S
Qwen 3.5 0.8B0.87B
QWEN·256K CTX· CHAT· CODING· MULTILINGUAL
62
TOK/S · 34% VRAM
›
A
InternVL3 1B0.94B
OTHER·32K CTX· CHAT· VISION
57
TOK/S · 35% VRAM
›
A
MiniCPM5 1B1.08B
MINICPM·128K CTX· CHAT· REASONING· MULTILINGUAL
50
TOK/S · 38% VRAM
›
A
TinyLlama 1.1B1.1B
LLAMA·2K CTX· CHAT
49
TOK/S · 39% VRAM
›
A
LFM2.5-1.2B-Thinking1.2B
LFM·122K CTX· CHAT· REASONING· TOOL_USE
45
TOK/S · 41% VRAM
›
A
Llama-3.2-1B1.2B
LLAMA·4K CTX· CHAT
45
TOK/S · 41% VRAM
›
A
LFM2 1.2B1.2B
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
45
TOK/S · 41% VRAM
›
A
Zamba2 1.2B1.2B
OTHER·4K CTX· CHAT
45
TOK/S · 41% VRAM
›
A
EXAONE-4.0-1.2B1.3B
EXAONE·64K CTX· CHAT
41
TOK/S · 43% VRAM
›
A
OPT 1.3B1.3B
OPT·2K CTX· CHAT
41
TOK/S · 43% VRAM
›
A
MiniCPM-V 4.61.3B
OTHER·256K CTX· CHAT· VISION
41
TOK/S · 43% VRAM
›
A
DeepSeek Coder 1.3B1.35B
DEEPSEEK·16K CTX· CODING
40
TOK/S · 44% VRAM
›
B
Phi-1 1.3B1.42B
PHI·2K CTX· CODING
38
TOK/S · 45% VRAM
›
B
Phi-1.5 1.3B1.42B
PHI·2K CTX· CHAT· CODING
38
TOK/S · 45% VRAM
›
B
Qwen2.5-Coder-1.5B1.5B
QWEN·32K CTX· CHAT· TOOL_USE· CODING
36
TOK/S · 47% VRAM
›
▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your GeForce GTX 670MX.

WHAT THIS CARD IS WORTH

GTX 670MX holds 81 of the models in our catalogue and is, in practice, a Q4_K_M card — the largest it takes is Falcon3-3B 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
Falcon3-3B3.23B · Q4_K_M2.6 GB25 tok/sEST
granite-4.0-h-micro 3.2B3.2B · Q4_K_M2.6 GB26 tok/sEST
Llama-3.2-3B3.2B · Q4_K_M2.7 GB26 tok/sEST
Falcon-H1 3B3.15B · Q5_K_M2.7 GB22 tok/sEST
Qwen 2.5 3B3.1B · Q5_K_M2.6 GB23 tok/sEST
SmolLM3-3B3.1B · Q4_K_M2.5 GB27 tok/sEST
Ministral 3B3B · Q4_K_M2.6 GB27 tok/sEST
StarCoder2 3B3B · Q5_K_M2.5 GB24 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
GTX 670MX3 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

NVIDIA GeForce GTX 670MX — 3 GB VRAM.

▸ GEFORCE GTX 670MX SPEC
BRAND
NVIDIA
VRAM
3 GB GDDR5
BANDWIDTH
67 GB/s
FP16 COMPUTE
1.2 TFLOPS
FP32 COMPUTE
1.2 TFLOPS
CUDA CORES
960
TDP
75 W
ARCHITECTURE
Kepler
▸ AI CAPABILITY
81/ 449 models @ Q4

With 3 GB VRAM and 67 GB/s bandwidth, this GPU handles models up to 3.1B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~8 tok/s.

§ 01TOP MODELS FOR GEFORCE GTX 670MX
81 FIT · SHOWING 20
MODELSIZEVRAM Q4TOK/SAVG
Qwen 2.5 3B3.1B2.4 GB1737.2
SmolLM3-3B3.1B2.4 GB1730.5
Ministral 3B3B2.3 GB1829.6
StarCoder2 3B3B2.3 GB189.5
Granite 4.1 3B3B2.3 GB1816.6
xLAM-2 3B Function-Calling3B2.3 GB18—
Jamba 2 3B3B2.3 GB18—
Dolly v2 3B2.8B2.2 GB195.6
StableLM Zephyr 3B2.79B2.2 GB1914.9
Zephyr 3B2.79B2.2 GB1914.4
OPT 2.7B2.7B2.1 GB2028.0
Phi-2 2.7B2.7B2.1 GB2024.1
Zamba2 2.7B2.7B2.1 GB2048.0
Granite 3.0 2B2.63B2.1 GB2035.8
gemma-2-2b2.6B2.1 GB2122.9
LFM2 2.6B2.6B2.1 GB2116.3
Granite 3.1 2B2.53B2.0 GB2137.8
Granite 3.3 2B2.53B2.0 GB2120.5
Gemma 1 2B2.51B2.0 GB2120.2
CodeGemma 2B2.51B2.0 GB2122.9