FitMyLLM
NVIDIA· Pascal

NVIDIA P102-101

Running LLMs on the P102-101 — the long read: which models fit at which quantisation, and the settings worth changing. · Or what a budget buys

VRAM
10 GB
Entry-level
Bandwidth
320
GB/s
Models Q4
238/498
48%
7B Q4 speed
~37
Fast
Model coverage at Q448% of all
Estimated speed· By model size at Q4

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

3B
~85
tok/s
7B
~37
tok/s
14B
~18
tok/s
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
10 GB
Bandwidth
320 GB/s
FP16 Compute
0.2 TFLOPS
TDP
250W
Memory
GDDR5
Architecture
Pascal
CUDA Cores
3,200
PCIe
Gen 3 x4
215
Fast models · >30 tok/s
Real-time chat speed
238
Usable · >10 tok/s
Comfortable for all tasks
238
Total compatible
Fit in VRAM at Q4
Don’t want to buy?

Test P102-101 (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· 238
S
Falcon-H1R Tiny 90M0.09B
Falcon·256K ctx· chat· reasoning
2844
tok/s · 5% VRAM
›
S
SmolLM2 135M0.135B
Smollm·2K ctx· chat
1896
tok/s · 6% VRAM
›
S
nomic-embed-text-v1.5 100M0.14B
Embedding·8K ctx· chat
1829
tok/s · 6% VRAM
›
S
GPT-2 124M0.14B
Gpt2·1K ctx· chat
1829
tok/s · 6% VRAM
›
S
SmolVLM 256M0.256B
Smollm·8K ctx· chat· vision
1000
tok/s · 6% VRAM
›
S
Gemma 3 270M0.27B
Gemma·32K ctx· chat
948
tok/s · 7% VRAM
›
S
Snowflake Arctic Embed M v2.00.305B
Embedding·8K ctx· embedding· multilingual
839
tok/s · 7% VRAM
›
S
bge-large-en-v1.5 335M0.335B
Embedding·1K ctx· chat
764
tok/s · 7% VRAM
›
S
mxbai-embed-large-v10.335B
Embedding·1K ctx· embedding
764
tok/s · 7% VRAM
›
S
Snowflake Arctic Embed L0.335B
Embedding·1K ctx· embedding
764
tok/s · 7% VRAM
›
S
LFM2 350M0.35B
Lfm·125K ctx· chat· tool_use
731
tok/s · 7% VRAM
›
S
SmolLM2 360M0.36B
Smollm·8K ctx· chat
711
tok/s · 7% VRAM
›
S
GPT-2 Medium 345M0.38B
Gpt2·1K ctx· chat
674
tok/s · 7% VRAM
›
S
Qwen 2.5 0.5B0.5B
Qwen·32K ctx· chat
512
tok/s · 8% VRAM
›
S
SmolVLM 500M0.5B
Smollm·8K ctx· chat· vision
512
tok/s · 8% VRAM
›
S
Falcon-H1 0.5B0.52B
Falcon·128K ctx· chat
492
tok/s · 8% VRAM
›
S
BGE-M30.568B
Embedding·8K ctx· embedding
451
tok/s · 8% VRAM
›
S
Snowflake Arctic Embed L v2.00.568B
Embedding·8K ctx· embedding· multilingual
451
tok/s · 8% VRAM
›
S
Qwen3-Embedding 0.6B0.6B
Embedding·32K ctx· embedding· multilingual
427
tok/s · 9% VRAM
›
S
Falcon-H1R Tiny 0.6B0.6B
Falcon·32K ctx· chat· reasoning
427
tok/s · 9% VRAM
›
S
Falcon Perception 0.6B0.6B
Falcon·4K ctx· vision
427
tok/s · 9% VRAM
›
S
Qwen 1.5 0.5B0.62B
Qwen·32K ctx· chat
413
tok/s · 9% VRAM
›
S
LFM2 700M0.74B
Lfm·125K ctx· chat· tool_use
346
tok/s · 9% VRAM
›
S
Qwen3 0.6B0.75B
Qwen·32K ctx· chat· reasoning
341
tok/s · 9% VRAM
›
S
GPT-2 Large 774M0.81B
Gpt2·1K ctx· chat
316
tok/s · 10% VRAM
›
S
Qwen 3.5 0.8B0.87B
Qwen·256K ctx· chat· coding· multilingual
294
tok/s · 10% VRAM
›
S
InternVL3 1B0.94B
Other·32K ctx· chat· vision
272
tok/s · 11% VRAM
›
S
MiniCPM5 1B1.08B
Minicpm·128K ctx· chat· reasoning· multilingual
237
tok/s · 11% VRAM
›
S
K2 Horizon 0.9B1.08B
Other·128K ctx· chat· coding· reasoning
237
tok/s · 11% VRAM
›
S
TinyLlama 1.1B1.1B
Llama·2K ctx· chat
233
tok/s · 12% VRAM
›
S
LFM2.5-1.2B-Thinking1.2B
Lfm·122K ctx· chat· reasoning· tool_use
213
tok/s · 12% VRAM
›
S
Llama-3.2-1B1.2B
Llama·4K ctx· chat
213
tok/s · 12% VRAM
›
S
LFM2 1.2B1.2B
Lfm·125K ctx· chat· tool_use· multilingual
213
tok/s · 12% VRAM
›
S
Zamba2 1.2B1.2B
Other·4K ctx· chat
213
tok/s · 12% VRAM
›
S
Ling 3.0 Tiny7.89BMoE
Other·128K ctx· chat· coding· reasoning
197
tok/s · 53% VRAM
›
S
EXAONE-4.0-1.2B1.3B
Exaone·64K ctx· chat
197
tok/s · 13% VRAM
›
S
OPT 1.3B1.3B
Opt·2K ctx· chat
197
tok/s · 13% VRAM
›
S
MiniCPM-V 4.61.3B
Other·256K ctx· chat· vision
197
tok/s · 13% VRAM
›
S
DeepSeek Coder 1.3B1.35B
Deepseek·16K ctx· coding
190
tok/s · 13% VRAM
›
S
Phi-1 1.3B1.42B
Phi·2K ctx· coding
180
tok/s · 14% VRAM
›
Next step

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your P102-101.

What this card is worth

P102-101 holds 238 of the models in our catalogue and is, in practice, a Q4_K_M card — the largest it takes is Mistral-Nemo 12.2B 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
Mistral-Nemo 12.2B12.2B · Q4_K_M9.0 GB23 tok/sest
Mellum2.1 12B-A2.5B Thinking12.15B · Q4_K_M8.4 GB112 tok/sest
Dolly v2 12B12B · Q4_K_S8.7 GB25 tok/sest
StableLM 2 12B12B · Q4_K_M9.0 GB23 tok/sest
Falcon2 11B11B · Q4_K_M8.5 GB26 tok/sest
SOLAR-10.7B10.7B · Q4_K_M8.1 GB26 tok/sest
Falcon3-10B10.3B · Q5_K_M8.8 GB23 tok/sest
GLM-4.1V 9B Thinking10.29B · Q4_K_M8.9 GB27 tok/sest
Against cards of similar memory
M3 Pro (18GB)12 GB$1,5991.5 tok/s per $100
M1 Pro (16GB)11 GB$9992.9 tok/s per $100
M2 Pro (16GB)11 GB$1,2992.2 tok/s per $100
M4 (16GB)11 GB$4994.0 tok/s per $100
P102-10110 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 P102-101 — 10 GB VRAM.

P102-101 spec
Brand
NVIDIA
VRAM
10 GB GDDR5
Bandwidth
320 GB/s
FP16 Compute
0.2 TFLOPS
FP32 Compute
10.7 TFLOPS
CUDA Cores
3,200
TDP
250 W
Architecture
Pascal
AI capability
238/ 498 models @ Q4

With 10 GB VRAM and 320 GB/s bandwidth, this GPU handles models up to 12.2B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~37 tok/s.

Rent it — live prices
Checking what the clouds are charging…
Top models for P102-101
238 fit · showing 20
ModelSizeVRAM Q4tok/sAvg
Mistral-Nemo 12.2B12.2B7.9 GB21129.3
Mellum2.1 12B-A2.5B Thinking12.15B7.9 GB10264.5
Dolly v2 12B12B7.8 GB216.4
StableLM 2 12B12B7.8 GB2121.3
Falcon2 11B11B7.2 GB2333.2
SOLAR-10.7B10.7B7.0 GB2428.2
Falcon3-10B10.3B6.8 GB2538.2
Bamba 9B v29.78B6.5 GB2626.1
Qwen 3.5 9B9.65B6.4 GB2749.2
RecurrentGemma 9B9.63B6.4 GB2735.0
glm-4-9b9.4B6.2 GB2720.5
MiniCPM-o 4.59.37B6.2 GB2781.9
gemma-2-9b9.2B6.1 GB28171.8
Yi 1.5 9B9B6.0 GB2830.3
Yi Coder 9B9B6.0 GB2835.8
K2 Horizon 7B9B6.0 GB2852.4
Ministral 3 8B8.92B5.9 GB2922.6
Ministral 3 8B Reasoning8.92B5.9 GB2969.0
NVIDIA-Nemotron-Nano-9B-v28.9B5.9 GB2944.2
Apertus 1.5 8B8.9B5.9 GB29—