FitMyLLM
NVIDIA· Pascal

NVIDIA P104-101

Running LLMs on the P104-101 — the long read: which models fit at which quantisation, and the settings worth changing. · Or what a budget buys

VRAM
4 GB
Budget
Bandwidth
256
GB/s
Models Q4
105/498
21%
7B Q4 speed
~29
Good
Model coverage at Q421% of all
Estimated speed· By model size at Q4

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

3B
~68
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
4 GB
Bandwidth
256 GB/s
FP16 Compute
0.1 TFLOPS
TDP
75W
Memory
GDDR5
Architecture
Pascal
CUDA Cores
2,560
PCIe
Gen 1 x4
105
Fast models · >30 tok/s
Real-time chat speed
105
Usable · >10 tok/s
Comfortable for all tasks
105
Total compatible
Fit in VRAM at Q4
Don’t want to buy?

Test P104-101 (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· 105
S
Falcon-H1R Tiny 90M0.09B
Falcon·256K ctx· chat· reasoning
2276
tok/s · 14% VRAM
›
S
SmolLM2 135M0.135B
Smollm·2K ctx· chat
1517
tok/s · 14% VRAM
›
S
nomic-embed-text-v1.5 100M0.14B
Embedding·8K ctx· chat
1463
tok/s · 14% VRAM
›
S
GPT-2 124M0.14B
Gpt2·1K ctx· chat
1463
tok/s · 14% VRAM
›
S
SmolVLM 256M0.256B
Smollm·8K ctx· chat· vision
800
tok/s · 16% VRAM
›
S
Gemma 3 270M0.27B
Gemma·32K ctx· chat
759
tok/s · 16% VRAM
›
S
Snowflake Arctic Embed M v2.00.305B
Embedding·8K ctx· embedding· multilingual
671
tok/s · 17% VRAM
›
S
bge-large-en-v1.5 335M0.335B
Embedding·1K ctx· chat
611
tok/s · 17% VRAM
›
S
mxbai-embed-large-v10.335B
Embedding·1K ctx· embedding
611
tok/s · 17% VRAM
›
S
Snowflake Arctic Embed L0.335B
Embedding·1K ctx· embedding
611
tok/s · 17% VRAM
›
S
LFM2 350M0.35B
Lfm·125K ctx· chat· tool_use
585
tok/s · 18% VRAM
›
S
SmolLM2 360M0.36B
Smollm·8K ctx· chat
569
tok/s · 18% VRAM
›
S
GPT-2 Medium 345M0.38B
Gpt2·1K ctx· chat
539
tok/s · 18% VRAM
›
S
Qwen 2.5 0.5B0.5B
Qwen·32K ctx· chat
410
tok/s · 20% VRAM
›
S
SmolVLM 500M0.5B
Smollm·8K ctx· chat· vision
410
tok/s · 20% VRAM
›
S
Falcon-H1 0.5B0.52B
Falcon·128K ctx· chat
394
tok/s · 20% VRAM
›
S
BGE-M30.568B
Embedding·8K ctx· embedding
361
tok/s · 21% VRAM
›
S
Snowflake Arctic Embed L v2.00.568B
Embedding·8K ctx· embedding· multilingual
361
tok/s · 21% VRAM
›
S
Qwen3-Embedding 0.6B0.6B
Embedding·32K ctx· embedding· multilingual
341
tok/s · 21% VRAM
›
S
Falcon-H1R Tiny 0.6B0.6B
Falcon·32K ctx· chat· reasoning
341
tok/s · 21% VRAM
›
S
Falcon Perception 0.6B0.6B
Falcon·4K ctx· vision
341
tok/s · 21% VRAM
›
S
Qwen 1.5 0.5B0.62B
Qwen·32K ctx· chat
330
tok/s · 22% VRAM
›
S
LFM2 700M0.74B
Lfm·125K ctx· chat· tool_use
277
tok/s · 24% VRAM
›
S
Qwen3 0.6B0.75B
Qwen·32K ctx· chat· reasoning
273
tok/s · 24% VRAM
›
S
GPT-2 Large 774M0.81B
Gpt2·1K ctx· chat
253
tok/s · 25% VRAM
›
S
Qwen 3.5 0.8B0.87B
Qwen·256K ctx· chat· coding· multilingual
235
tok/s · 26% VRAM
›
S
InternVL3 1B0.94B
Other·32K ctx· chat· vision
218
tok/s · 27% VRAM
›
S
MiniCPM5 1B1.08B
Minicpm·128K ctx· chat· reasoning· multilingual
190
tok/s · 29% VRAM
›
S
K2 Horizon 0.9B1.08B
Other·128K ctx· chat· coding· reasoning
190
tok/s · 29% VRAM
›
S
TinyLlama 1.1B1.1B
Llama·2K ctx· chat
186
tok/s · 29% VRAM
›
S
LFM2.5-1.2B-Thinking1.2B
Lfm·122K ctx· chat· reasoning· tool_use
171
tok/s · 31% VRAM
›
S
Llama-3.2-1B1.2B
Llama·4K ctx· chat
171
tok/s · 31% VRAM
›
S
LFM2 1.2B1.2B
Lfm·125K ctx· chat· tool_use· multilingual
171
tok/s · 31% VRAM
›
S
Zamba2 1.2B1.2B
Other·4K ctx· chat
171
tok/s · 31% VRAM
›
S
EXAONE-4.0-1.2B1.3B
Exaone·64K ctx· chat
158
tok/s · 32% VRAM
›
S
OPT 1.3B1.3B
Opt·2K ctx· chat
158
tok/s · 32% VRAM
›
S
MiniCPM-V 4.61.3B
Other·256K ctx· chat· vision
158
tok/s · 32% VRAM
›
S
DeepSeek Coder 1.3B1.35B
Deepseek·16K ctx· coding
152
tok/s · 33% VRAM
›
S
Phi-1 1.3B1.42B
Phi·2K ctx· coding
144
tok/s · 34% VRAM
›
S
Phi-1.5 1.3B1.42B
Phi·2K ctx· chat· coding
144
tok/s · 34% VRAM
›
Next step

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your P104-101.

What this card is worth

P104-101 holds 105 of the models in our catalogue and is, in practice, a Q4_K_M card — the largest it takes is InternLM2 5B 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
InternLM2 5B4.5B · Q4_K_M3.6 GB53 tok/sest
Ministral 3 3B Reasoning4.25B · Q4_K_M3.3 GB56 tok/sest
Qwen3 4B4B · Q4_K_M3.3 GB59 tok/sest
Qwen3-4B Instruct 25074B · Q4_K_M3.3 GB59 tok/sest
Qwen3-Embedding 4B4B · Q4_K_M3.3 GB59 tok/sest
Nemotron 3 Nano 4B3.97B · Q5_K_M3.2 GB51 tok/sest
Ministral 3 3B3.85B · Q5_K_M3.5 GB53 tok/sest
phi-3-mini-4k 3.8B3.8B · Q4_K_M3.4 GB62 tok/sest
Against cards of similar memory
RTX 3070 Ti8 GB$49912.6 tok/s per $100
RTX 3060 Ti GDDR6X8 GB$39915.8 tok/s per $100
RTX 3070 Ti 8 GB GA1028 GB$59910.5 tok/s per $100
Arc A7508 GB$19910.1 tok/s per $100
P104-1014 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 P104-101 — 4 GB VRAM.

P104-101 spec
Brand
NVIDIA
VRAM
4 GB GDDR5
Bandwidth
256 GB/s
FP16 Compute
0.1 TFLOPS
FP32 Compute
8.6 TFLOPS
CUDA Cores
2,560
TDP
75 W
Architecture
Pascal
AI capability
105/ 498 models @ Q4

With 4 GB VRAM and 256 GB/s bandwidth, this GPU handles models up to 4.25B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~29 tok/s.

Top models for P104-101
105 fit · showing 20
ModelSizeVRAM Q4tok/sAvg
Ministral 3 3B Reasoning4.25B3.1 GB4860.1
Qwen 1.5 4B4B2.9 GB5112.6
Qwen3 4B4B2.9 GB5140.7
Qwen3-4B Instruct 25074B2.9 GB5137.2
Qwen3-Embedding 4B4B2.9 GB5168.5
Nemotron 3 Nano 4B3.97B2.9 GB5260.1
Ministral 3 3B3.85B2.8 GB5318.3
Phi-3.5 Mini 3.8B3.82B2.8 GB5446.6
phi-3-mini-4k 3.8B3.8B2.8 GB5430.5
Phi-4-mini 3.8B3.8B2.8 GB5449.0
Granite 4.2 3B3.66B2.7 GB5638.0
Cogito 3B3.61B2.7 GB5722.1
Falcon3-3B3.23B2.5 GB6325.7
granite-4.0-h-micro 3.2B3.2B2.4 GB6427.3
Llama-3.2-3B3.2B2.4 GB6417.9
Falcon-H1 3B3.15B2.4 GB6549.5
LFM2.5-VL 3B3.12B2.4 GB6653.3
Qwen 2.5 3B3.1B2.4 GB6637.2
SmolLM3-3B3.1B2.4 GB6635.1
Ministral 3B3B2.3 GB6823.7