llmrun

NVIDIA · Turing

Best AI Models for NVIDIA GeForce GTX 1650 (4 GB)

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

4 GB GDDR5128 GB/s6 TFLOPS FP1675 WTuring 2019
Buyabout $149

4 GB at 128 GB/s: small models only (~3B at Q4), slow decode. A later GDDR6 version of the card has 192 GB/s. No Tensor Cores.

What LLMs can the GTX 1650 run?

#ModelSpeedllmrun ScoreQuant · VRAM
1
Qwen3.5 4BAlibabaHow to run →
Q4_K_M · 3.4 GB · 25.5 t/s
25.5 tok/sQ4_K_M · 3.4 GB
2
Qwen3.5 2BAlibaba
Q4_K_M · 1.9 GB · 47.0 t/s
47.0 tok/sQ4_K_M · 1.9 GB
3
MiniCPM5 2BOpenBMB
Q4_K_M · 2.5 GB · 43.8 t/s
43.8 tok/sQ4_K_M · 2.5 GB
4
Gemma 4 E2B ITGoogle
Q4_K_M · 3.5 GB · 24.3 t/s
24.3 tok/sQ4_K_M · 3.5 GB
5
LFM3 1.7BLiquid AI
Q4_K_M · 2.1 GB · 57.8 t/s
57.8 tok/sQ4_K_M · 2.1 GB
See all 151 models on the GTX 1650 →Best quant that fits · speeds estimated on this card

About the GTX 1650

The GTX 1650 is an entry card in the 4–6 GB class: it runs only small models well. It is more efficient per watt than most 4–6 GB cards, and the best speed per dollar of the cards compared at ≈$94 used. Turing generation, released 2019.

Speed in 4–6 GB#5128 GB/s
Per watt#30.31 tok/s per W on Qwen3.5 9B (9B params)
Per dollar#124.7 tok/s per $100 on Qwen3.5 9B (9B params)
GenerationTuringreleased 2019

GTX 1650 strengths and weaknesses

Strengths

  • Low power draw: 75 W board power; a 150 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 14.2 s for the first token
  • No FP8 or NVFP4: Turing predates both; use GGUF, AWQ or GPTQ builds

Speed per dollar and per watt

On Qwen3.5 9B (9B params) at Q4_K_M. New prices are compared with new, used with used. The last two columns are the speed you get per $100 of price and per 100 W of board power; higher is better. The GTX 1650 has no new price, so the New view shows it at its MSRP.

CardNew pricetok/stok/s per $100tok/s per 100 W
GTX 16504 GB
about $14923.2
15.6
30.9
RTX 50608 GBBest new value
about $46045.8
10.0
31.6
RTX 507012 GB
about $84068.7
8.2
27.5
RTX 40608 GB
about $52027.8
5.3
24.2
CardUsed pricetok/stok/s per $100tok/s per 100 W
GTX 16504 GB
about $9423.2
24.7
30.9
RTX 3060 12GB12 GB
about $32936.8
11.2
21.6
RTX 50608 GBBest used value
about $43045.8
10.7
31.6
RTX 407012 GB
about $57551.5
9.0
25.8
RTX 40608 GB
about $33527.8
8.3
24.2
RTX 4070 SUPER12 GB
about $67551.5
7.6
23.4
RTX 4060 Ti 8GB8 GB
about $40029.4
7.4
18.4
RTX 4070 Ti12 GB
about $72251.5
7.1
18.1

Cost per token: the GTX 1650 against 4–12 GB cards

On Qwen3.5 9B (9B params, Q4_K_M, 4K context). Speed, cost per token and power per token depend on the model.

4 GB12 GBthis cardat MSRP
$0.00$0.50$1.00$1.50$0$1,000$ per 1M output tokens · lower is betterused price →RTX 3060 12GB · 12 GB · ≈$329 · $0.76 per 1M tokensRTX 4060 Ti 8GB · 8 GB · ≈$400 · $1.09 per 1M tokensRTX 5060 · 8 GB · ≈$430 · $0.73 per 1M tokensRTX 5070 · 12 GB · ≈$736 · $0.83 per 1M tokensRTX 4070 SUPER · 12 GB · ≈$675 · $1.01 per 1M tokensRTX 4070 Ti · 12 GB · ≈$722 · $1.12 per 1M tokensRTX 4060 · 8 GB · ≈$335 · $0.94 per 1M tokensRTX 4070 · 12 GB · ≈$575 · $0.87 per 1M tokensGTX 1650 · 4 GB · ≈$94 · $0.39 per 1M tokensGTX 1650RTX 5060RTX 5070RTX 4070 SUPERRTX 4070 TiRTX 4070RTX 3060 12GB
$0.00$0.50$1.00$1.50$0$1k$ per 1M output tokens · lower is betterused price →RTX 3060 12GB · 12 GB · ≈$329 · $0.76 per 1M tokensRTX 4060 Ti 8GB · 8 GB · ≈$400 · $1.09 per 1M tokensRTX 5060 · 8 GB · ≈$430 · $0.73 per 1M tokensRTX 5070 · 12 GB · ≈$736 · $0.83 per 1M tokensRTX 4070 SUPER · 12 GB · ≈$675 · $1.01 per 1M tokensRTX 4070 Ti · 12 GB · ≈$722 · $1.12 per 1M tokensRTX 4060 · 8 GB · ≈$335 · $0.94 per 1M tokensRTX 4070 · 12 GB · ≈$575 · $0.87 per 1M tokensGTX 1650 · 4 GB · ≈$94 · $0.39 per 1M tokensGTX 1650RTX 5060RTX 5070
Cost per 1M output tokenson Qwen3.5 9B · card over 3 years + power
$0.39
Card, used price
$0.26
Power
$0.13
3-year cost of ownership$143
Or rent an RTX 3080 Ti in the cloud$0.13/hr

Renting costs as much as buying after ≈700 hours (≈0.5 years at 4 h a day).

Include card price

Each dot is a card; hover or tap it for its numbers, click to open it. Cards without a price for the chosen basis are left out. How we calculate cost per token

Similar GPUs

Cards people weigh against the GTX 1650, each with the one reason it is worth a look. Speeds on Qwen3.5 9B.

4 GB more and 27% fasterRTX 4060 Ti 8GBNVIDIA · 8 GB · Ada Lovelace
▲ 27% speed▲ 4 GBabout $399rent $0.15/hr
8 GB more and 59% fasterRTX 3060 12GBNVIDIA · 12 GB · Ampere
▲ 59% speed▲ 8 GBabout $329rent $0.06/hr
4 GB more and nearly 2× the speedRTX 5060NVIDIA · 8 GB · Blackwell
▲ 97% speed▲ 4 GBabout $460
50% faster, same 4 GBRTX 3050 Laptop GPUNVIDIA · 4 GB · Ampere
▲ 50% speedsame 4 GB
2 GB more and 71% fasterGTX 1660 SUPERNVIDIA · 6 GB · Turing
▲ 71% speed▲ 2 GBabout $229
4 GB more and 74% fasterA750Intel · 8 GB · Alchemist
▲ 74% speed▲ 4 GBabout $289

GTX 1650 tokens per second by model size

Output speed depends on memory bandwidth (128 GB/s). Grey rows need more than 4 GB.

  1. 9BQwen3.5 9B · not at 4-bit
    needs 6.6 GB
  2. 14BQwen3 14B · not at 4-bit
    needs 11.9 GB
  3. 27–31BQwen3.8 27B · not at 4-bit
    needs 17.9 GB
  4. 35B MoEQwen3.6 35B A3B · not at 4-bit
    needs 22.0 GB
  5. 70BLlama 3.3 70B Instruct · not at 4-bit
    needs 51.3 GB
  6. 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 GTX 1650

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 GTX 1650 — ranked by Score, recency and Fit
ModelVRAMScoreFit
WeVisOne VL 4B
IQ3_M·31.3 t/s
4.0 GBC40, Usable
Chandra Ocr 2
Q4_K_M·22.8 t/s
3.8 GBC38, Usable
Granite Docling 258M
Q4_K_M·166.4 t/s
0.6 GBS98, Runs great
LFM2.5 8B A1B
IQ3_XS·94.1 t/s
4.0 GBC44, Usable
Dots.ocr
Q4_K_M·38.2 t/s
2.6 GBA74, Runs well
Gemma 3 270M IT
Q4_K_M·462.2 t/s
1.2 GBS100, Runs great
Qwen3 VL 2B Instruct
Q4_K_M·46.0 t/s
3.5 GBA70, Runs well
Gemma 4 E2B
Q4_K_M·24.3 t/s
3.5 GBB56, Decent
Phi 2
Q4_K_M·31.5 t/s
2.6 GBA71, Runs well
Qwen2.5 VL 3B Instruct
Q4_K_M·31.6 t/s
3.2 GBA71, Runs well
Gemma 3 1B IT
Q4_K_M·126.1 t/s
2.1 GBS95, Runs great
Qwen3.5 0.8B Base
Q4_K_M·95.6 t/s
0.9 GBS90, Runs great
Typhoon Ocr1.5 2B
Q4_K_M·46.0 t/s
3.5 GBA70, Runs well
MiMo V2.6 Distill Qwen 9B
IQ2_S·23.7 t/s
3.8 GBC42, Usable
Llama 3.2 3B Instruct
Q4_K_S·41.8 t/s
4.0 GBC44, Usable

GTX 1650 specs, power and PSU

Specs

Architecture
Turing
VRAM
4 GB GDDR5
Memory bandwidth
128.0 GB/s
FP16 (dense)
5.97 TFLOPS
CUDA cores
896
Tensor cores
0
Compute capability
7.5
TDP
75 W
Released
2019-04-23
MSRP
$149

GTX 1650 FAQ

Is the GTX 1650 still worth it in 2026?

At ≈$94 used you get 24.7 tok/s per $100 on Qwen3.5 9B, which ranks 1st of the 8 cards compared here; it is also the 5th fastest out of 8 cards with 4–6 GB. It runs only small models well; 7–9B doesn't fit in 4 GB.

What is the largest model the GTX 1650 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 25.5 tok/s.

Can the GTX 1650 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 GTX 1650 give?

About 25.5 tok/s on Qwen3.5 4B at Q4_K_M, with the first token after 14.2 s on a 4K-token prompt.

What power supply does the GTX 1650 need?

Plan a 150 W PSU (2× the 75 W TDP).

Is renting cheaper than buying a GTX 1650?

An RTX 3080 Ti (12 GB) rents from $0.13/hr. Renting costs as much as buying after ≈700 hours (≈0.5 years at 4 h a day).

Related

Best coding LLMs8 GB VRAMGPUs with 4 GB or moreRTX 3070 TiA580A770 8GB