32GB VRAM

RTX 5090 — what LLMs can it run?

90 of 112 indexed models fit comfortably in 32GB at 4K context, each at the highest-quality quant that still leaves headroom.

The short answer for RTX 5090

32 GB, 1,792 GB/s GDDR7. 90 of 112 models in this index fit comfortably at 4K context; 91 load at all.

Biggest that fits: Nemotron 3.5 Bolt 74B-A9B — 23.1 GB at IQ2_XXS — 9.0 GB spare at 4K, so longer context comes out of a thin margin.
Room to grow: GLM-4.7-Flash — 19.2 GB at Q4_K_M — under 60% of the card, which leaves 12.8 GB for a long context window or a second process.
Speed ceiling: Llama 3.1 8B Instruct at Q4_K_M reads 4.6 GB of weights per token, so 1,792 GB/s puts a hard ceiling near 388 tok/s. That is arithmetic on two published numbers, not a benchmark — real throughput lands below it. No run on this card has been measured here, so there is nothing to compare the ceiling against.
The wall you will hit: Capacity. 90 of 112 models clear 32 GB comfortably at 4K context, and the ones that do not are short by whole gigabytes rather than by a rounding error — at 1,792 GB/s the memory system is not what holds this card back. Moving up a size class is what changes the list, not a different quant.

Best local LLM for 32GB

70B+ · 2

RTX 5090 — what LLMs can it run? — 70B+
Model	Quant	Est. VRAM	Headroom	tok/s
Nemotron 3.5 Bolt 74B-A9B74B-A9B	IQ2_XXS	23.05 GB	+9 GB	≤705Ceiling
Jamba 1.5 Mini52B-A12B	AWQ INT4	27.24 GB	+4.8 GB	≤314Ceiling

32B · 30

RTX 5090 — what LLMs can it run? — 32B
Model	Quant	Est. VRAM	Headroom	tok/s
Mixtral 8x7B Instruct47B MoE	AWQ INT4	24.95 GB	+7.1 GB	≤292Ceiling
Seed-OSS 36B Instruct36B	Q5_K_M	27.81 GB	+4.2 GB	≤74Ceiling
Qwen3.6 35B-A3B35B-A3B	Q4_K_M	23.38 GB	+8.6 GB	≤1004Ceiling
Command R 35B35B	Q4_K_M	22.86 GB	+9.1 GB	≤89Ceiling
Yi 1.5 34B Chat34B	Q4_K_M	22.82 GB	+9.2 GB	≤90Ceiling
Qwen3 32B Instruct32B	Q4_K_M	21.85 GB	+10.1 GB	55–73Derived
Qwen2.5 32B Instruct32B	Q4_K_M	21.69 GB	+10.3 GB	56–74Derived
Qwen2.5-Coder 32B Instruct32B	Q4_K_M	21.69 GB	+10.3 GB	56–74Derived
DeepSeek-R1-Distill-Qwen-32B32B	Q4_K_M	21.69 GB	+10.3 GB	56–74Derived
Nemotron 3.5 Lightning 30B-A3B30B-A3B	Q4_0	19.79 GB	+12.2 GB	≤1050Ceiling
Gemma 4 31B IT31B	Q4_K_M	21.08 GB	+10.9 GB	59–78Derived
Qwen3 30B-A3B Instruct30B-A3B	Q5_K_M	23.04 GB	+9 GB	≤886Ceiling
Qwen3-Coder 30B-A3B Instruct30B-A3B	Q5_K_M	23.04 GB	+9 GB	≤886Ceiling
Qwen3-VL 30B-A3B Instruct30B-A3B	Q4_K_M	19.73 GB	+12.3 GB	≤1037Ceiling
North Mini Code 1.0 30B-A3B30B-A3B	Q4_K_M	20.01 GB	+12 GB	≤1022Ceiling
GLM-4.7-Flash30B-A3B	Q4_K_M	19.19 GB	+12.8 GB	≤1037Ceiling
Granite 4.2 30B30B	Q6_K	26.19 GB	+5.8 GB	46–61Derived
Muse Glimmer 30B30B	Q6_K	24.61 GB	+7.4 GB	47–62Derived
Qwen3.8 27B27B	Q4_K_M	17.47 GB	+14.5 GB	67–88Derived
Qwen3.6 27B27B	Q5_K_M	22.19 GB	+9.8 GB	52–69Derived
Gemma 3 27B IT27B	Q4_K_M	18.37 GB	+13.6 GB	66–88Derived
Gemma 2 27B Instruct27B	Q5_K_M	21.76 GB	+10.2 GB	57–75Derived
Gemma 4 26B-A4B IT26B-A4B	Q4_K_M	16.39 GB	+15.6 GB	≤778Ceiling
Mistral Small 24B Instruct24B	EXL2 4.65bpw	15.26 GB	+16.7 GB	≤135Ceiling
LFM2.5 24B-A2B24B-A2B	Q8_0	26.73 GB	+5.3 GB	≤724Ceiling
Devstral Small 1.1 24B24B	Q6_K	20.91 GB	+11.1 GB	57–75Derived
Magistral Small 1.2 24B24B	Q6_K	20.91 GB	+11.1 GB	57–75Derived
Codestral 22B22B	Q4_K_M	15.03 GB	+17 GB	82–108Derived
ERNIE 4.5 21B-A3B21B-A3B	Q8_0	24.44 GB	+7.6 GB	≤592Ceiling
GPT-OSS 20B21B MoE	MXFP4	12.76 GB	+19.2 GB	≤905Ceiling

14B · 19

RTX 5090 — what LLMs can it run? — 14B
Model	Quant	Est. VRAM	Headroom	tok/s
InternLM2 20B Chat20B	Q5_K_M	15.59 GB	+16.4 GB	78–103Derived
DeepSeek-Coder-V2-Lite Instruct16B-A2.4B	Q8_0	17.56 GB	+14.4 GB	≤740Ceiling
DeepSeek-V2-Lite Chat16B-A2.4B	Q4_K_M	10.08 GB	+21.9 GB	≤1297Ceiling
StarCoder2 15B15B	Q4_K_M	10.19 GB	+21.8 GB	121–160Derived
Qwen3 14B Instruct14B	Q5_K_M	11.65 GB	+20.4 GB	105–139Derived
Qwen2.5 14B Instruct14B	Q5_K_M	11.73 GB	+20.3 GB	105–139Derived
DeepSeek-R1-Distill-Qwen-14B14B	EXL2 4.65bpw	9.75 GB	+22.3 GB	≤221Ceiling
Phi-4 14B14B	Q5_K_M	11.77 GB	+20.2 GB	105–139Derived
Phi-3 Medium 14B Instruct14B	Q6_K	12.86 GB	+19.1 GB	96–127Derived
Mistral Nemo 12B Instruct12B	Q6_K	11.14 GB	+20.9 GB	110–145Derived
Gemma 3 12B IT12B	Q5_K_M	9.84 GB	+22.2 GB	127–168Derived
Mellum2.1 12B-A2.5B Thinking12B-A2.5B	Q8_0	13.6 GB	+18.4 GB	≤711Ceiling
Stable LM 2 12B Chat12B	Q4_K_M	8.35 GB	+23.7 GB	150–198Derived
Gemma 4 12B IT12B	Q8_0	13.82 GB	+18.2 GB	87–115Derived
Llama 3.2 11B Vision Instruct11B	Q8_0	12.9 GB	+19.1 GB	94–124Derived
Solar 10.7B Instruct11B	Q4_K_M	7.6 GB	+24.4 GB	169–224Derived
Falcon 3 10B Instruct10B	Q4_K_M	7.28 GB	+24.7 GB	174–231Derived
Gemma 2 9B Instruct9B	Q8_0	11.7 GB	+20.3 GB	112–148Derived
Qwen3.5 9B9B	Q8_0	10.37 GB	+21.6 GB	112–148Derived

7B · 25

RTX 5090 — what LLMs can it run? — 7B
Model	Quant	Est. VRAM	Headroom	tok/s
GLM-4-9B-Chat9B	Q8_0	10.16 GB	+21.8 GB	115–152Derived
Qwen3-VL 8B Instruct8B	Q8_0	10.39 GB	+21.6 GB	118–155Derived
Granite 4.2 8B8B	Q8_0	10.45 GB	+21.6 GB	118–156Derived
Ministral 3 8B Instruct8B	Q8_0	10.02 GB	+22 GB	122–161Derived
Qwen2-VL 7B Instruct7B	Q4_K_M	5.49 GB	+26.5 GB	219–289Derived
Granite 3.1 8B Instruct8B	Q4_K_M	5.88 GB	+26.1 GB	221–293Derived
Qwen3 8B Instruct8B	Q6_K	7.64 GB	+24.4 GB	164–216Derived
Llama 3.1 8B Instruct8B	Q8_0	9.47 GB	+22.5 GB	129–170Derived
Nous Hermes 3 Llama 3.1 8B8B	EXL2 4.65bpw	5.43 GB	+26.6 GB	≤405Ceiling
Aya 23 8B8B	Q4_K_M	5.64 GB	+26.4 GB	226–299Derived
OpenChat 3.6 8B8B	EXL2 4.65bpw	5.43 GB	+26.6 GB	≤405Ceiling
DeepSeek-R1-Distill-Llama-8B8B	Q5_K_M	6.51 GB	+25.5 GB	193–255Derived
InternLM2 7B Chat7B	Q4_K_M	5.45 GB	+26.6 GB	234–310Derived
Qwen2.5 7B Instruct7B	Q6_K	6.77 GB	+25.2 GB	176–232Derived
Qwen2.5-Coder 7B Instruct7B	EXL2 4.65bpw	4.87 GB	+27.1 GB	≤426Ceiling
DeepSeek-R1-Distill-Qwen-7B7B	EXL2 4.65bpw	4.87 GB	+27.1 GB	≤426Ceiling
Gemma 4 E4B ITE4B	Q8_0	8.46 GB	+23.5 GB	137–182Derived
OLMo 2 7B Instruct7B	Q8_0	10.3 GB	+21.7 GB	142–187Derived
OLMo 3 7B Instruct7B	Q8_0	10.3 GB	+21.7 GB	142–187Derived
WizardLM-2 7B7B	Q4_K_M	5.14 GB	+26.9 GB	250–331Derived
Mistral 7B Instruct v0.37B	Q6_K	6.75 GB	+25.3 GB	185–245Derived
Zephyr 7B Beta7B	Q6_K	6.75 GB	+25.3 GB	185–245Derived
Gemma 4 E2B ITE2B	Q8_0	5.2 GB	+26.8 GB	≤382Ceiling
Qwen3.5 4B4B	Q8_0	4.97 GB	+27 GB	≤408Ceiling
Gemma 3 4B IT4B	Q8_0	5.05 GB	+27 GB	≤413Ceiling

≤3B · 14

RTX 5090 — what LLMs can it run? — ≤3B
Model	Quant	Est. VRAM	Headroom	tok/s
Qwen3 4B Instruct4B	Q6_K	4.05 GB	+28 GB	≤574Ceiling
Phi-4 Mini Instruct3.8B	Q8_0	4.81 GB	+27.2 GB	≤462Ceiling
Phi-3.5 Mini Instruct3.8B	Q8_0	5.89 GB	+26.1 GB	≤464Ceiling
Granite 4.2 3B3B	Q8_0	4.41 GB	+27.6 GB	≤486Ceiling
Llama 3.2 3B Instruct3B	Q8_0	4.05 GB	+28 GB	≤553Ceiling
Qwen2.5 3B Instruct3B	Q8_0	3.59 GB	+28.4 GB	≤574Ceiling
SmolLM3 3B3B	Q8_0	3.73 GB	+28.3 GB	≤576Ceiling
Gemma 2 2B Instruct2B	Q8_0	3.34 GB	+28.7 GB	≤681Ceiling
Qwen3.5 2B2B	Q8_0	2.23 GB	+29.8 GB	≤905Ceiling
Qwen3 1.7B Instruct1.7B	Q8_0	2.39 GB	+29.6 GB	≤1030Ceiling
Qwen2.5 1.5B Instruct1.5B	Q8_0	1.83 GB	+30.2 GB	≤1156Ceiling
Llama 3.2 1B Instruct1B	Q8_0	1.51 GB	+30.5 GB	≤1434Ceiling
Qwen3.5 0.8B0.8B	Q8_0	0.94 GB	+31.1 GB	≤2212Ceiling
Qwen2.5 0.5B Instruct0.5B	Q8_0	0.6 GB	+31.4 GB	≤3657Ceiling

How this list is built

Each row is the lowest-perplexity-loss quant of that model whose estimated total — weights plus KV cache at 4K context plus activation buffer — uses at most 88% of the card. That is the calculator's "green" threshold, so every row here has real headroom rather than only just fitting. Raise the context length and the list shortens; the calculator lets you check any combination directly. The speed column is this card: a measured run, a range scaled from this card's own Llama 2 7B benchmark, or a bandwidth ceiling. A figure that only exists for the RTX 4090 is labelled as that card.

A further 1 models load but with no headroom to spare (up to 105% of VRAM) — the Quant Hub’s GPU chips count those too, which is why its number is higher.

Open this card in the VRAM calculator Feel this speed Browse all 112 models

Measured on this card

No benchmark runs in this index were recorded on a RTX 5090. Every figure on this page is calculated from the model architecture and the quant level — treat them as estimates, not measurements.

Cards with the same budget

The same 32GB in system memory is a different machine — everything on this page fits there, at a speed set by your DIMMs rather than a graphics bus: 32 GB RAM (CPU)

Setup guides for this card

Step-by-step guides written for the RTX 5090's runtime or memory size, with the commands and the checks that prove the model is running on the card.

RTX 5090: What 32GB Actually Buys for Local LLMs

Common questions

What is the best local LLM for an RTX 5090?

For everyday use, GLM-4.7-Flash at Q4_K_M — about 19.2 GB of the card's 32 GB at 4K context, leaving 12.8 GB for a longer window. If you want the largest thing that will load, that is Nemotron 3.5 Bolt 74B-A9B at IQ2_XXS (23.1 GB, 9.0 GB spare). "Best" here means best fit for the memory budget — this index does not run task benchmarks, so it cannot tell you which model is smarter.

Can an RTX 5090 run Llama 3.2 90B Vision Instruct?

No. Its smallest build here, Q3_K_M, needs about 46.2 GB and this card has 32.0 GB usable — short by 14.2 GB before any context beyond 4K. The largest model this card does clear is Nemotron 3.5 Bolt 74B-A9B.

How many tokens per second does an RTX 5090 do on an 8B model at Q4?

This index has no measured run on this card, so it does not publish a figure. What can be stated from specifications: generating a token requires reading every weight once, Llama 3.1 8B Instruct at Q4_K_M is 4.6 GB of weights, and this card moves 1,792 GB/s — a ceiling near 388 tok/s. Batch size, context length, the runtime and how much of the model sits in cache all take you below it.

RTX 5090 or RTX 4090 for local LLMs?

They do not hold the same set: RTX 5090 fits 90 and RTX 4090 86 of 112 at 4K. Speed is the other difference — 1,792 GB/s against 1,008 GB/s, a 1.8× gap in how fast the weights can be read, which is what token generation is bound by. The RTX 5090 generates faster on any model both can hold.

This page's figures change when the model or the runtime does.Last updated 2026-10-20 RSS → /feed.xml