Skip to content
Local AI Wizard
Your machine
Theme

GTX 1650 · 4 GB VRAM

NVIDIA · GTX 1650

15 text models run well · 128 GB/s bandwidth

Models on this hardware

Start here

Llama 3.2 1B InstructRuns wellPicked by LMArena · text rating · 1,111

File Q4_K_M · 0.8 GB · bartowski/Llama-3.2-1B-Instruct-GGUF · What is Q4?

Memory Needs ~1 GB · ~3 GB on GTX 1650 · 4 GB VRAM, including ~2 GB the system keeps

ollama run hf.co/bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M
  1. Install Ollama
  2. Run the command above
  3. Ask: Explain what a context window is, in two sentences.

Text

Within 4 GB: 15 run well · 14 tight · 2 reviewed · 129 more run slowly (19 of them run well with 6 GB)

Text · LMArena

rating · higher is better

Source: LMArena 2026-10-22 · collected 2026-10-25 · dataset · CC BY 4.0 · Models without a rating are left out, not scored as zero.

Rows run well first, then tight fits; within each, reviewed models, then Hub downloads over 30 days, not quality. Sort a score column to rank by that source alone. All 19 text models that would run well with 6 GB but not here already run here, slowly.
Text models that run well within 4 GB
Llama 3.2 1B InstructMeta · 1.23B · LMArena 1,111Q4 · published params · source1,111 ± 8~3 GBQ4 · published paramssource
Qwen3-0.6BQwen · 0.75BQ4 · tensor count · sourceCatalog record: not reviewed for LMArena~3 GBQ4 · tensor countsource
SmolLM2-135MHuggingFaceTB · 0.13BQ4 · tensor count · sourceCatalog record: not reviewed for LMArena~3 GBQ4 · tensor countsource
SmolLM2-135M-InstructHuggingFaceTB · 0.13BQ4 · tensor count · sourceCatalog record: not reviewed for LMArena~3 GBQ4 · tensor countsource
Llama-3.2-1Bmeta-llama · 1.24BQ4 · tensor count · sourceCatalog record: not reviewed for LMArena~3 GBQ4 · tensor countsource
Qwen3Guard-Gen-0.6BQwen · 0.75BQ4 · tensor count · sourceCatalog record: not reviewed for LMArena~3 GBQ4 · tensor countsource
SmolLM2-360M-InstructHuggingFaceTB · 0.36BQ4 · tensor count · sourceCatalog record: not reviewed for LMArena~3 GBQ4 · tensor countsource
SmolLM2-360MHuggingFaceTB · 0.36BQ4 · tensor count · sourceCatalog record: not reviewed for LMArena~3 GBQ4 · tensor countsource
8 of 29

No model within 4 GB has a LiveBench score, so the table has no LiveBench column. LMArena ratings in this table: LMArena 2026-10-22 · collected 2026-10-25 · dataset · CC BY 4.0

Tight fit: it runs, but leaves less than 1 GB free beyond the 2 GB the system keeps (the memory figure already includes that reserve).

Every reviewed text model here 35

Every one that runs here, well or slowly, scored or not (2 run well, 33 slowly); 17 of 35 have a LiveBench or LMArena result. Run well first, then tight fits, then slow; within each, by Hub downloads over 30 days, read 2026-10-25, not by quality.

Reviewed text models that run on GTX 1650 · 4 GB VRAM
Llama 3.2 1B InstructMeta · 1.2BNot in LiveBench 2026-06-251,111~3 GB0.8Llama 3.23.5M
Qwen3 1.7BAlibaba · 1.7BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~4 GBTight fit1.1Apache 2.04.1M
Qwen3 14BAlibaba · 14.8BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~14 GBRuns slowly9.0Apache 2.010.4M
Gemma 4 31BGoogle · 30.7BNot in LiveBench 2026-06-251,452~26 GBRuns slowly18.3Apache 2.08.4M
Gemma 4 26B A4BGoogle · 25.2BNot in LiveBench 2026-06-251,437~21 GBRuns slowly16.9Apache 2.08.2M
Qwen3 8BAlibaba · 8.2BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~9 GBRuns slowly5.0Apache 2.08.1M
Qwen3.8 27BAlibaba · 27B75.31,437~23 GBRuns slowly16.5Apache 2.07.8M
Qwen3.5 9BAlibaba · 9BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~9 GBRuns slowly5.7Apache 2.06.1M
8 of 35
Also runs slowly96 models

Above the 4 GB of VRAM, the layers that do not fit run on the CPU from PC RAM, up to 28 GB in all. 32 GB of PC RAM assumed: this configuration lists none. Planning estimate, not a speed figure. The 33 reviewed models that run slowly are in the table above.

6 of 96

Browse Text models

Adjust the estimate: quantization, context, reserveloads the live catalog

Vision

Within 4 GB: 29 run well · 13 tight · 91 more run slowly (29 of them run well with 6 GB) · 6 more with 6 GB
Rows run well first, then tight fits; within each, reviewed models, then Hub downloads over 30 days, not quality. Sort a score column to rank by that source alone. Vision models listed by the memory they need on this hardware; no comparable vision benchmark is tracked.
Vision models that run well within 4 GB
Falcon-OCRtiiuae · 0.27B · Image to textQ4 · Hugging Face count · source~3 GBQ4 · Hugging Face countsource
Florence-2-basemicrosoft · 0.2B · Image to textQ4 · Hugging Face count · source~3 GBQ4 · Hugging Face countsource
Florence-2-largemicrosoft · 0.8B · Image to textQ4 · Hugging Face count · source~3 GBQ4 · Hugging Face countsource
GLM-OCRzai-org · 1.33B · Image to textQ4 · Hugging Face count · source~3 GBQ4 · Hugging Face countsource
GOT-OCR2_0stepfun-ai · 0.72B · Image to textQ4 · Hugging Face count · source~3 GBQ4 · Hugging Face countsource
granite-docling-258Mibm-granite · 0.3B · Image to textQ4 · Hugging Face count · source~3 GBQ4 · Hugging Face countsource
LightOnOCR-1B-1025lightonai · 1.16B · Image to textQ4 · Hugging Face count · source~3 GBQ4 · Hugging Face countsource
LightOnOCR-2-1Blightonai · 1.01B · Image to textQ4 · Hugging Face count · source~3 GBQ4 · Hugging Face countsource
8 of 42

Tight fit: it runs, but leaves less than 1 GB free beyond the 2 GB the system keeps (the memory figure already includes that reserve).

Also runs slowly91 models

Above the 4 GB of VRAM, the layers that do not fit run on the CPU from PC RAM, up to 28 GB in all. 32 GB of PC RAM assumed: this configuration lists none. Planning estimate, not a speed figure.

6 of 91

With 6 GB you also run… 6

The next memory size on this ladder runs these well. Figures are the requirement on this hardware type.

12 more vision models in the catalog have no usable memory figure or parameter count for this profile. Browse Vision models

Image

9 within 4 GB · 7 more with 6 GB

Image · Prompt alignment · GenEval

0–1 · higher is better

Source: GenEval · 2023 · reviewed 2026-10-05 · Published quality in the original benchmark preset; not measured on this hardware.

Image models within 4 GB
Stable Diffusion XL 1.0Stability AI · 2.6B · Text to imagePlatform path documented · stable-diffusion.cpp · source0.55Not on the LMArena text-to-image leaderboard of 2026-10-07Documented path without a memory figurePlatform path documentedstable-diffusion.cpp · source
Stable Diffusion 1.5Stability AI · 0.98B · Text to imagePlatform path documented · stable-diffusion.cpp · source0.43Not on the LMArena text-to-image leaderboard of 2026-10-07Documented path without a memory figurePlatform path documentedstable-diffusion.cpp · source
Qwen-ImageQwen · 27.1B · Text to image · Image to image · LMArena 1,057Conditional path · CUDA · sourceNo comparable published result1,057 ± 3~4 GBConditional pathCUDA · source
FLUX.2-small-decoderblack-forest-labs · 0.1B · Image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~3 GB16-bit est. · paramssource
OmniGen V2VectorSpaceLab · 7B · Text to imageConditional path · CUDA · sourceNo comparable published resultNot on the LMArena text-to-image leaderboard of 2026-10-07~3 GBConditional pathCUDA · source
shap-e-img2imgopenai · 0.3B · Image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~3 GB16-bit est. · paramssource
FLUX.1 [dev]Black Forest Labs · 12B · Text to imagePlatform path documented · Diffusers / PyTorch · sourceNo comparable published resultArena rated the FP8 build (flux-1-dev-fp8); this entry is the original FLUX.1 [dev] weightsDocumented path without a memory figurePlatform path documentedDiffusers / PyTorch · source
FLUX.1 [schnell]Black Forest Labs · 12B · Text to imagePlatform path documented · Diffusers / PyTorch · sourceNo comparable published resultNot on the LMArena text-to-image leaderboard of 2026-10-07Documented path without a memory figurePlatform path documentedDiffusers / PyTorch · source
8 of 9

LMArena ratings in this table: LMArena 2026-10-07 · collected 2026-10-25 · dataset · CC BY 4.0

With 6 GB you also run… 7

The next memory size on this ladder. Figures are the requirement on this hardware type.

6 of 7

17 more image models in the catalog have no usable memory figure or parameter count for this profile. Browse Image models

Video

3 within 4 GB · 6 more with 6 GB
Video models within 4 GB
animatediff-motion-adapter-v1-5-2guoyww · 0.5B · Text to video16-bit est. · params · source~4 GB16-bit est. · paramssource
URSA-0.6B-FSQ320BAAI · 0.7B · Text to video16-bit est. · params · source~4 GB16-bit est. · paramssource
Wan2.1 T2V 14BWan-AI · 14B · Text to videoPlatform path documented · Wan2.1 / PyTorch ≥2.4.0 · sourceDocumented path without a memory figurePlatform path documentedWan2.1 / PyTorch ≥2.4.0 · source

How this list works Documented execution paths on this hardware first, then publisher minimums (CUDA or Apple MPS).

With 6 GB you also run… 6

The next memory size on this ladder. Figures are the requirement on this hardware type.

20 more video models in the catalog have no usable memory figure or parameter count for this profile. Browse Video models

Audio

48 within 4 GB · 22 more with 6 GB

Audio · Word error rate · Open ASR

% · lower is better

Source: Word error rate · Open ASR · reviewed 2026-10-20 · Published quality in the original benchmark preset; not measured on this hardware.

Audio models within 4 GB
Whisper large-v3-turboOpenAI · 0.81B · Speech to textPlatform path documented · whisper.cpp / CUDA · source7.83Documented path without a memory figurePlatform path documentedwhisper.cpp / CUDA · source
Whisper large-v3OpenAI · 1.55B · Speech to textPlatform path documented · whisper.cpp / CUDA · source7.44Documented path without a memory figurePlatform path documentedwhisper.cpp / CUDA · source
canary-1b-flashnvidia · 0.8B · Speech to text16-bit est. · params · sourceNo comparable published result~4 GB16-bit est. · paramssource
chatterboxResembleAI · 0.5B · Text to speech16-bit est. · params · sourceNo comparable published result~4 GB16-bit est. · paramssource
ChatterboxResemble AI · 0.5B · Text to speech16-bit est. · params · sourceNo comparable published result~4 GB16-bit est. · paramssource
CosyVoice2-0.5BFunAudioLLM · 0.5B · Text to speech16-bit est. · params · sourceNo comparable published result~4 GB16-bit est. · paramssource
distil-large-v3distil-whisper · 0.8B · Speech to text16-bit est. · params · sourceNo comparable published result~4 GB16-bit est. · paramssource
distil-large-v3.5distil-whisper · 0.8B · Speech to text16-bit est. · params · sourceNo comparable published result~4 GB16-bit est. · paramssource
8 of 48

With 6 GB you also run… 22

The next memory size on this ladder. Figures are the requirement on this hardware type.

6 of 22

33 more audio models in the catalog have no usable memory figure or parameter count for this profile. Browse Audio models

Embeddings

62 run well within 4 GB · 7 more with 6 GB

Embeddings · MTEB (Multilingual, v2) · mean(task)

0–100 · higher is better

  • Color = publisher

Source: MTEB results (curated rows, sourced per model)

Embeddings models that run well within 4 GB
Harrier OSS v1 (0.6B)Microsoft8-bit · published params · source69.0~3 GB8-bit · published paramssource
Qwen3-Embedding 0.6BAlibaba (Qwen)8-bit · published params · source64.3~3 GB8-bit · published paramssource
Multilingual E5 Large Instructintfloat8-bit · published params · source63.2~3 GB8-bit · published paramssource
EmbeddingGemma 2Google8-bit · published params · source61.4~3 GB8-bit · published paramssource
EmbeddingGemma (300M)Google8-bit · published params · source61.1~3 GB8-bit · published paramssource
BGE-M3BAAI8-bit · published params · source59.6~3 GB8-bit · published paramssource
Nomic Embed Text v2 (MoE)Nomic AI8-bit · published params · source57.6~3 GB8-bit · published paramssource
Multilingual E5 Baseintfloat8-bit · published params · source57.0~3 GB8-bit · published paramssource
8 of 62

With 6 GB you also run… 7

The next memory size on this ladder runs these well. Figures are the requirement on this hardware type.

6 of 7

Browse Embeddings models

Runs well within 4 GB of VRAM · planning estimates, not measured speed bundled catalog copy · 1,777 repositories · checked

Measurements

Speed on this hardware

no published measurement yetWe have not yet included a verifiable measurement for this exact configuration. The memory estimates above do not indicate tokens per second or time to first token.

Specifications

15 sourced fields; the source is named once per group, and a field from another source links to it. Official specifications

GPU VRAM4 GBDedicated · system RAM does not add · source
Memory bandwidth128 GB/sManufacturer spec · not model speed · source
Launch price (MSRP)USD 149.00US · launched · source · not a tracked offer

Marker: the lightest text model that does not run well here but runs well with 6 GB. No text model that runs well here has a LiveBench score. Q4 planning estimate, not a measured fit.

Reviewed · Manufacturer source

Graphics processor
NVIDIA GeForce GTX 1650
GPU runtime status
Not stated by the source
Memory capacity
4 GB
Memory kind
Dedicated VRAM
Power connectors
6-pin
GPU architecture
Turing
CUDA cores
896
GPU memory type
GDDR5 · GDDR6
GPU power
75 W
Memory interface
128 bit
Minimum power supply
300 W
PCIe interface
Gen 3
Release date
2019-04-23
Slot width
2
Introduced
2019

NVIDIA GeForce GTX 1650 with 4 GB VRAM. Runtime support and local inference speed are not verified.

Photo: ASUS partner-board photo of a GeForce GTX 1650 · other partner boards vary. · Manufacturer source

Memory bandwidth is not measured model speed; verify runtime and checkpoint compatibility. Hardware specification reviewed: 2026-10-13.

How estimates work
Candidates
Reviewed pages use published parameter counts; automatic text candidates use official repository tensor counts (packed quantized weights and unsupported architectures excluded).
Planning
Q4 weights + 20% allowance + reserve: 25% of unified or system memory (at least 2 GB, at most 8 GB; 2 GB here) or 2 GB per dedicated GPU. Adjustable in “Adjust the estimate”. No KV cache is calculated from context.
Fit tiers
Runs well: the estimate fits the fast memory (VRAM, or unified memory within the macOS GPU share). Runs slowly: within the machine's memory but outside the fast part; a graphics card offloads layers to PC RAM (the listed RAM, or 32 GB assumed). Doesn't run: above both. Tiers are planning estimates, never a speed.
Vision
Vision-language models use the Q4 planning estimate from the parameter count (published by the model page, or read from Hugging Face). Classic vision models (classify, detect, segment) use a 16-bit estimate. No vision benchmark is tracked, so rows carry no score.
Media and embeddings
Media rows use documented execution paths (exact hardware or platform) and publisher minimums; a published requirement above this memory moves a model to the next tier. Without either, a 16-bit estimate from the published parameter count (× 1.2 + 2 GB), labelled as such. Embeddings use 8-bit weights + 20% + 2 GB.

Local AI basics: parameters, Q4 and memory · Full model catalog · Data and methodology

Alternatives

Other machines on the same memory ladder (VRAM); each page lists the models that fit it.