FitMyLLM
NVIDIA· Maxwell 2.0

NVIDIA GeForce GTX 950

Running LLMs on the GeForce GTX 950 — the long read: which models fit at which quantisation, and the settings worth changing. · Or what a budget buys

VRAM
2 GB
Budget
Bandwidth
106
GB/s
Models Q4
48/498
10%
7B Q4 speed
~12
Usable
Model coverage at Q410% of all
Estimated speed· By model size at 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
106 GB/s
FP16 Compute
1.8 TFLOPS
TDP
90W
Memory
GDDR5
Architecture
Maxwell 2.0
CUDA Cores
768
PCIe
Gen 3 x16
48
Fast models · >30 tok/s
Real-time chat speed
48
Usable · >10 tok/s
Comfortable for all tasks
48
Total compatible
Fit in VRAM at Q4
Don’t want to buy?

Test GeForce GTX 950 (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· 48
S
Falcon-H1R Tiny 90M0.09B
Falcon·256K ctx· chat· reasoning
942
tok/s · 27% VRAM
›
S
SmolLM2 135M0.135B
Smollm·2K ctx· chat
628
tok/s · 29% VRAM
›
S
nomic-embed-text-v1.5 100M0.14B
Embedding·8K ctx· chat
606
tok/s · 29% VRAM
›
S
GPT-2 124M0.14B
Gpt2·1K ctx· chat
606
tok/s · 29% VRAM
›
S
SmolVLM 256M0.256B
Smollm·8K ctx· chat· vision
331
tok/s · 32% VRAM
›
S
Gemma 3 270M0.27B
Gemma·32K ctx· chat
314
tok/s · 33% VRAM
›
S
Snowflake Arctic Embed M v2.00.305B
Embedding·8K ctx· embedding· multilingual
278
tok/s · 34% VRAM
›
S
bge-large-en-v1.5 335M0.335B
Embedding·1K ctx· chat
253
tok/s · 35% VRAM
›
S
mxbai-embed-large-v10.335B
Embedding·1K ctx· embedding
253
tok/s · 35% VRAM
›
S
Snowflake Arctic Embed L0.335B
Embedding·1K ctx· embedding
253
tok/s · 35% VRAM
›
S
LFM2 350M0.35B
Lfm·125K ctx· chat· tool_use
242
tok/s · 35% VRAM
›
S
SmolLM2 360M0.36B
Smollm·8K ctx· chat
236
tok/s · 35% VRAM
›
S
GPT-2 Medium 345M0.38B
Gpt2·1K ctx· chat
223
tok/s · 36% VRAM
›
S
Qwen 2.5 0.5B0.5B
Qwen·32K ctx· chat
170
tok/s · 40% VRAM
›
S
SmolVLM 500M0.5B
Smollm·8K ctx· chat· vision
170
tok/s · 40% VRAM
›
S
Falcon-H1 0.5B0.52B
Falcon·128K ctx· chat
163
tok/s · 40% VRAM
›
S
BGE-M30.568B
Embedding·8K ctx· embedding
149
tok/s · 42% VRAM
›
S
Snowflake Arctic Embed L v2.00.568B
Embedding·8K ctx· embedding· multilingual
149
tok/s · 42% VRAM
›
S
Qwen3-Embedding 0.6B0.6B
Embedding·32K ctx· embedding· multilingual
141
tok/s · 43% VRAM
›
S
Falcon-H1R Tiny 0.6B0.6B
Falcon·32K ctx· chat· reasoning
141
tok/s · 43% VRAM
›
S
Falcon Perception 0.6B0.6B
Falcon·4K ctx· vision
141
tok/s · 43% VRAM
›
S
Qwen 1.5 0.5B0.62B
Qwen·32K ctx· chat
137
tok/s · 43% VRAM
›
S
LFM2 700M0.74B
Lfm·125K ctx· chat· tool_use
115
tok/s · 47% VRAM
›
S
Qwen3 0.6B0.75B
Qwen·32K ctx· chat· reasoning
113
tok/s · 47% VRAM
›
S
GPT-2 Large 774M0.81B
Gpt2·1K ctx· chat
105
tok/s · 49% VRAM
›
S
Qwen 3.5 0.8B0.87B
Qwen·256K ctx· chat· coding· multilingual
97
tok/s · 51% VRAM
›
S
InternVL3 1B0.94B
Other·32K ctx· chat· vision
90
tok/s · 53% VRAM
›
S
MiniCPM5 1B1.08B
Minicpm·128K ctx· chat· reasoning· multilingual
79
tok/s · 57% VRAM
›
S
K2 Horizon 0.9B1.08B
Other·128K ctx· chat· coding· reasoning
79
tok/s · 57% VRAM
›
S
TinyLlama 1.1B1.1B
Llama·2K ctx· chat
77
tok/s · 58% VRAM
›
S
LFM2.5-1.2B-Thinking1.2B
Lfm·122K ctx· chat· reasoning· tool_use
71
tok/s · 61% VRAM
›
S
Llama-3.2-1B1.2B
Llama·4K ctx· chat
71
tok/s · 61% VRAM
›
S
LFM2 1.2B1.2B
Lfm·125K ctx· chat· tool_use· multilingual
71
tok/s · 61% VRAM
›
S
Zamba2 1.2B1.2B
Other·4K ctx· chat
71
tok/s · 61% VRAM
›
S
EXAONE-4.0-1.2B1.3B
Exaone·64K ctx· chat
65
tok/s · 64% VRAM
›
S
OPT 1.3B1.3B
Opt·2K ctx· chat
65
tok/s · 64% VRAM
›
S
MiniCPM-V 4.61.3B
Other·256K ctx· chat· vision
65
tok/s · 64% VRAM
›
S
DeepSeek Coder 1.3B1.35B
Deepseek·16K ctx· coding
63
tok/s · 66% VRAM
›
S
Phi-1 1.3B1.42B
Phi·2K ctx· coding
60
tok/s · 68% VRAM
›
S
Phi-1.5 1.3B1.42B
Phi·2K ctx· chat· coding
60
tok/s · 68% VRAM
›
Next step

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your GeForce GTX 950.

What this card is worth

GTX 950 holds 48 of the models in our catalogue and is, in practice, a Q4_K_M card — the largest it takes is LFM3-VL 2B 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
LFM3-VL 2B1.92B · Q4_K_M1.7 GB65 tok/sest
Moondream2 1.9B1.9B · Q4_K_M1.7 GB66 tok/sest
LFM3 1.7B1.73B · Q6_K1.8 GB54 tok/sest
Falcon3-1B1.67B · Q5_K_M1.8 GB64 tok/sest
LFM2-VL 1.6B1.6B · Q5_K_M1.6 GB67 tok/sest
Falcon-H1 1.5B1.55B · Q6_K1.7 GB60 tok/sest
MiniCPM-V 4.7 1B1.55B · Q5_K_M1.6 GB69 tok/sest
Qwen2.5-Coder-1.5B1.5B · Q6_K1.6 GB62 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 9502 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 950 — 2 GB VRAM.

GeForce GTX 950 spec
Brand
NVIDIA
VRAM
2 GB GDDR5
Bandwidth
106 GB/s
FP16 Compute
1.8 TFLOPS
FP32 Compute
1.8 TFLOPS
CUDA Cores
768
TDP
90 W
Architecture
Maxwell 2.0
AI capability
48/ 498 models @ Q4

With 2 GB VRAM and 106 GB/s bandwidth, this GPU handles models up to 1.61B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~12 tok/s.

Top models for GeForce GTX 950
48 fit · showing 20
ModelSizeVRAM Q4tok/sAvg
GPT-2 XL 1.5B1.61B1.5 GB535.1
stablelm-2-1_6b1.6B1.5 GB539.5
Falcon-H1 1.5B1.55B1.4 GB5543.8
Qwen2.5-Coder-1.5B1.5B1.4 GB5719.6
Qwen2 Math 1.5B1.5B1.4 GB5719.6
Qwen 2.5 1.5B1.5B1.4 GB5730.2
Yi Coder 1.5B1.5B1.4 GB5714.6
Stella en 1.5B v51.5B1.4 GB5752.4
Phi-1 1.3B1.42B1.4 GB607.2
Phi-1.5 1.3B1.42B1.4 GB607.2
DeepSeek Coder 1.3B1.35B1.3 GB6316.8
EXAONE-4.0-1.2B1.3B1.3 GB6522.6
OPT 1.3B1.3B1.3 GB655.3
MiniCPM-V 4.61.3B1.3 GB6528.5
LFM2.5-1.2B-Thinking1.2B1.2 GB7122.3
Llama-3.2-1B1.2B1.2 GB7110.1
LFM2 1.2B1.2B1.2 GB7115.6
Zamba2 1.2B1.2B1.2 GB7141.5
TinyLlama 1.1B1.1B1.2 GB7713.6
MiniCPM5 1B1.08B1.1 GB7930.8