NVIDIA · Ampere

Best AI Models for NVIDIA GeForce RTX 2050 (4 GB)

Runs only small models well; 7–9B doesn't fit in 4 GB.

4 GB GDDR6112 GB/s12 TFLOPS FP1645 WAmpere 2022

Laptop-only 4 GB Ampere chip (GA107) with a 64-bit bus - despite the name it is NOT Turing. Its 4 GB limits it to ~3B-class Q4 models; bandwidth is only 112 GB/s.

What LLMs can the RTX 2050 run?

#	Model	Speed	Quant · VRAM
1	Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 22.3 t/s	22.3 tok/s	Q4_K_M · 3.4 GB
2	Qwen3.5 2BAlibaba Q4_K_M · 1.9 GB · 41.1 t/s	41.1 tok/s	Q4_K_M · 1.9 GB
3	MiniCPM5 2BOpenBMB Q4_K_M · 2.5 GB · 38.3 t/s	38.3 tok/s	Q4_K_M · 2.5 GB
4	Gemma 4 E2B ITGoogle Q4_K_M · 3.5 GB · 21.2 t/s	21.2 tok/s	Q4_K_M · 3.5 GB
5	LFM3 1.7BLiquid AI Q4_K_M · 2.1 GB · 50.6 t/s	50.6 tok/s	Q4_K_M · 2.1 GB

#	Model	Speed ↓	Quant · VRAM
1	Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 22.3 t/s	22.3 tok/s	Q4_K_M · 3.4 GB
2	LFM3 1.7BLiquid AI Q4_K_M · 2.1 GB · 50.6 t/s	50.6 tok/s	Q4_K_M · 2.1 GB
3	Qwen3.5 2BAlibaba Q4_K_M · 1.9 GB · 41.1 t/s	41.1 tok/s	Q4_K_M · 1.9 GB
4	MiniCPM5 2BOpenBMB Q4_K_M · 2.5 GB · 38.3 t/s	38.3 tok/s	Q4_K_M · 2.5 GB
5	Gemma 4 E2B ITGoogle Q4_K_M · 3.5 GB · 21.2 t/s	21.2 tok/s	Q4_K_M · 3.5 GB

#	Model	Speed	Quant · VRAM
1	Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 22.3 t/s	22.3 tok/s	Q4_K_M · 3.4 GB
2	Qwen3.5 2BAlibaba Q4_K_M · 1.9 GB · 41.1 t/s	41.1 tok/s	Q4_K_M · 1.9 GB
3	LFM3 Omni 2BLiquid AI Q4_K_M · 2.4 GB · 42.8 t/s	42.8 tok/s	Q4_K_M · 2.4 GB
4	MiniCPM V 4.7 1BOpenBMB Q4_K_M · 1.6 GB · 56.9 t/s	56.9 tok/s	Q4_K_M · 1.6 GB
5	Cosmos Reason2 2BNVIDIA Q4_K_M · 3.5 GB · 45.2 t/s	45.2 tok/s	Q4_K_M · 3.5 GB

See all 151 models on the RTX 2050 →Best quant that fits · speeds estimated on this card

About the RTX 2050

The RTX 2050 runs only small models well. Ampere generation, released 2022.

GenerationAmperereleased 2022

RTX 2050 strengths and weaknesses

Strengths

Low power draw: 45 W board power; a 100 W PSU is enough

Weaknesses

7–9B models don't fit: Qwen3.5 9B needs 6.6 GB at Q4 vs 4 GB here
Slow prompt reading: A 4K-token prompt on Qwen3.5 4B waits 7.0 s for the first token
No FP8 or NVFP4: Ampere predates both; use GGUF, AWQ or GPTQ builds

Similar GPUs

Cards people weigh against the RTX 2050, each with the one reason it is worth a look. Speeds on Qwen3.5 9B.

4 GB more and 45% fasterRTX 4060 Ti 8GBNVIDIA · 8 GB · Ada Lovelace

▲ 45% speed▲ 4 GB$399rent $0.15/hr

8 GB more and 81% fasterRTX 3060 12GBNVIDIA · 12 GB · Ampere

▲ 81% speed▲ 8 GB$329rent $0.06/hr

4 GB more and 2.3× the speedRTX 5060NVIDIA · 8 GB · Blackwell

▲ 126% speed▲ 4 GB$460

71% faster, same 4 GBRTX 3050 Laptop GPUNVIDIA · 4 GB · Ampere

▲ 71% speedsame 4 GB

2 GB more and nearly 2× the speedGTX 1660 SUPERNVIDIA · 6 GB · Turing

▲ 95% speed▲ 2 GB$229

4 GB more and nearly 2× the speedA750Intel · 8 GB · Alchemist

▲ 99% speed▲ 4 GB$289

RTX 2050 tokens per second by model size

Output speed depends on memory bandwidth (112 GB/s). Grey rows need more than 4 GB.

9BQwen3.5 9B · not at 4-bit
needs 6.6 GB
14BQwen3 14B · not at 4-bit
needs 11.9 GB
27–31BQwen3.8 27B · not at 4-bit
needs 17.9 GB
35B MoEQwen3.6 35B A3B · not at 4-bit
needs 22.0 GB
70BLlama 3.3 70B Instruct · not at 4-bit
needs 51.3 GB
120B MoEGPT OSS 120B · not at 4-bit
needs 70.8 GB

Estimated speeds: Q4_K_M, 4K context, one user. How we estimate

All 151 models on the RTX 2050

Featured only

Best quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works

LLM models compatible with NVIDIA GeForce RTX 2050 — ranked by Score, recency and Fit
Model	Quant	VRAM	Speed	Context	Fit
WeVisOne VL 4B4.8BVisionFunctions IQ3_M·27.4 t/s	IQ3_M	4.0 GB100%	27.4 t/s	262K	C37, Usable
LFM2.5 8B A1B8.5BChatFunctions IQ3_XS·83.6 t/s	IQ3_XS	4.0 GB100%	83.6 t/s	128K	C44, Usable
Granite Docling 258M258MVisionChatCodeMath Q4_K_M·145.6 t/s	Q4_K_M	0.6 GB16%	145.6 t/s	8K	S97, Runs great
Chandra Ocr 25.3BVisionFunctions Q4_K_M·19.9 t/s	Q4_K_M	3.8 GB96%	19.9 t/s	262K	C35, Usable
Dots.ocr3.0BVision Q4_K_M·33.4 t/s	Q4_K_M	2.6 GB65%	33.4 t/s	131K	A72, Runs well
Gemma 3 270M IT268MChat Q4_K_M·404.4 t/s	Q4_K_M	1.2 GB29%	404.4 t/s	—	S100, Runs great
Qwen3 VL 2B Instruct2.1BVisionFunctions Q4_K_M·40.2 t/s	Q4_K_M	3.5 GB86%	40.2 t/s	262K	A68, Runs well
Phi 22.8BChatCode Q4_K_M·27.6 t/s	Q4_K_M	2.6 GB66%	27.6 t/s	2K	A68, Runs well
Gemma 3 1B IT1000MChat Q4_K_M·110.3 t/s	Q4_K_M	2.1 GB52%	110.3 t/s	33K	S93, Runs great
Qwen2.5 VL 3B Instruct3.8BVision Q4_K_M·27.7 t/s	Q4_K_M	3.2 GB79%	27.7 t/s	128K	A68, Runs well
Gemma 4 E2B5.1BChat Q4_K_M·21.2 t/s	Q4_K_M	3.5 GB87%	21.2 t/s	131K	B52, Decent
Qwen3.5 0.8B Base873MVision Q4_K_M·83.7 t/s	Q4_K_M	0.9 GB23%	83.7 t/s	262K	S88, Runs great
Typhoon Ocr1.5 2B2.1BVisionFunctions Q4_K_M·40.2 t/s	Q4_K_M	3.5 GB86%	40.2 t/s	262K	A68, Runs well
Llama 3.2 3B Instruct3.2BChatFunctions Q4_K_S·36.6 t/s	Q4_K_S	4.0 GB100%	36.6 t/s	131K	C42, Usable
MiMo V2.6 Distill Qwen 9B9.4BVisionCodeFunctions IQ2_S·20.7 t/s	IQ2_S	3.8 GB95%	20.7 t/s	262K	C38, Usable

RTX 2050 specs, power and PSU

Specs

Architecture: Ampere
VRAM: 4 GB GDDR6
Memory bandwidth: 112.0 GB/s
FP16 (dense): 12.10 TFLOPS
CUDA cores: 2,048
Tensor cores: 64
Compute capability: 8.6
TDP: 45 W
Released: 2022-01-03

RTX 2050 FAQ

What is the largest model the RTX 2050 can run?

Qwen3.5 4B (4.7B) is the largest at 4-bit or better: at Q4_K_M it uses 3.4 GB of 4 GB, and runs at about 22.3 tok/s.

Can the RTX 2050 run 7–9B models?

No. Qwen3.5 9B needs 6.6 GB at Q4, more than the 4 GB on this card.

How many tokens per second does the RTX 2050 give?

About 22.3 tok/s on Qwen3.5 4B at Q4_K_M, with the first token after 7.0 s on a 4K-token prompt.

What power supply does the RTX 2050 need?

Plan a 100 W PSU (2× the 45 W TDP).

Is renting cheaper than buying an RTX 2050?

An RTX 3080 Ti (12 GB) rents from $0.13/hr. With regular daily use, owning pays off.

Best coding LLMs 8 GB VRAM GPUs with 4 GB or more RTX 3070 Ti A580 A770 8GB