LLaVA-NeXT-34B (LLaVA-1.6) TPS calculator

Open weights University of Wisconsin Madison,ByteDance,Nanyang Technological University,University of California (UC) Berkeley 34.8B parameters January 2024

Each card below is assessed against this model at the context length and minimum quality you choose. Speed is an estimate for a single request, calculated from the card's memory bandwidth and the size of the model once compressed.

Calculated for this model

132 of 818 cards that can run it

Smallest card that fits

RTX A4500

20 GB · Q3_K_M · 21.1 tok/s

Fastest card

B200

97.5 tok/s · 180 GB

Which GPUs can run LLaVA-NeXT-34B (LLaVA-1.6)?

Set the inputs, read the answer

Context length tokens

A longer conversation needs more memory, which can push this model off smaller cards.

Minimum quality

Hides cards that would only fit the model by compressing it below this point.

Search cards

Card memory

Also show cards that cannot run it

132 cards match

Calculating

					Needs	Quantisation	Fit
97.5 tok/s 59–156 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	37.9 GB	Q8_0	Comfortable
97.5 tok/s 59–156 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	37.9 GB	Q8_0	Comfortable
77.9 tok/s 47–125 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	37.9 GB	Q8_0	Comfortable
77.9 tok/s 47–125 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	37.9 GB	Q8_0	Comfortable
62.3 tok/s 37–100 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	37.9 GB	Q8_0	Comfortable
59.6 tok/s 36–95 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	37.9 GB	Q8_0	Comfortable
59.6 tok/s 36–95 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	37.9 GB	Q8_0	Comfortable
57.0 tok/s 34–91 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	37.9 GB	Q8_0	Comfortable
50.6 tok/s 30–81 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	37.9 GB	Q8_0	Comfortable
50.6 tok/s 30–81 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	37.9 GB	Q8_0	Comfortable
50.6 tok/s 30–81 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	37.9 GB	Q8_0	Comfortable
48.0 tok/s 29–77 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	37.9 GB	Q8_0	Comfortable
41.0 tok/s 25–66 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	37.9 GB	Q8_0	Comfortable
41.0 tok/s 25–66 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	37.9 GB	Q8_0	Comfortable
41.0 tok/s 25–66 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	37.9 GB	Q8_0	Comfortable
41.0 tok/s 25–66 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	37.9 GB	Q8_0	Comfortable
41.0 tok/s 25–66 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	37.9 GB	Q8_0	Comfortable
40.7 tok/s 24–65 · low confidence	DRIVE A100 PROD NVIDIA	32 GB	1,870 GB/s	May 2020	25.8 GB	Q5_K_M	Tight
40.7 tok/s 24–65 · low confidence	GRID A100A NVIDIA	32 GB	1,870 GB/s	May 2020	25.8 GB	Q5_K_M	Tight
40.1 tok/s 24–64 · low confidence	GeForce RTX 5090 D V2 NVIDIA	24 GB	1,340 GB/s	Aug 2025	19.7 GB	IQ4_XS	Tight
39.0 tok/s 23–62 · low confidence	GeForce RTX 5090 NVIDIA	32 GB	1,790 GB/s	Jan 2025	25.8 GB	Q5_K_M	Tight
39.0 tok/s 23–62 · low confidence	GeForce RTX 5090 D NVIDIA	32 GB	1,790 GB/s	Jan 2025	25.8 GB	Q5_K_M	Tight
36.5 tok/s 22–58 · low confidence	A30X NVIDIA	24 GB	1,220 GB/s	Apr 2021	19.7 GB	IQ4_XS	Tight
31.2 tok/s 19–50 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	37.9 GB	Q8_0	Comfortable
31.2 tok/s 19–50 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	37.9 GB	Q8_0	Comfortable

Speeds are estimates for a single request — one conversation at a time — calculated from memory bandwidth, model size and quantisation. Real throughput varies with the inference runtime and its version. Figures published by hardware vendors measure many simultaneous requests and are much higher.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation: University of Wisconsin Madison,ByteDance,Nanyang Technological University,University of California (UC) Berkeley
Organisation type: Academia,Industry,Academia,Academia
Country: United States of America, China, Singapore
Published: 30 January 2024
Authors: Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, Yong Jae Lee

What it does

The problem areas the model was built for. A model can carry several of each.

Domain: Multimodal, Language, Vision
Task: Visual question answering, Chat, Question answering

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters: 34.8B
Training data: 89,338,000 tokens

Training compute

The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.

Training compute: 2.6 × 10²⁰ FLOP
How it was established: Hardware

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Training hardware: NVIDIA A100
Chips used: 32
Chip-hours: 768
Wall-clock time: 24 hours
Power draw: 25.3 kW

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights: Open — downloadable
Model access: Open weights (unrestricted)

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Record confidence: Likely

Sources

Where this record came from and when it was last checked.

Reference: LLaVA-NeXT: Improved reasoning, OCR, and world knowledge
Last updated: 11 February 2026

The extremes

The ten fastest GPUs for LLaVA-NeXT-34B (LLaVA-1.6)

Ranked by estimated tokens per second, newest card first where speeds tie. Because generation is bound by memory bandwidth, this ordering follows bandwidth rather than any gaming benchmark.

The smallest GPUs that still run LLaVA-NeXT-34B (LLaVA-1.6)

The cheapest route in, by memory capacity. A tight fit runs the model but leaves nothing spare for a longer conversation.

What the numbers mean

Hardware requirements in practice

Minimum card

RTX A4500

Memory needed

17.7 GB

Fastest

97.5 tok/s

With 34.8B parameters, LLaVA-NeXT-34B (LLaVA-1.6) lands in the range a serious desktop card can handle once the weights are compressed. 132 of the cards we track can run it.

The entry point is the RTX A4500: 20 GB of memory, Q3_K_M compression, roughly 21.1 tokens per second.

The quickest result comes from a B200 at around 97.5 tokens per second — its 8,000 GB/s of bandwidth is what buys that.

What this model is

LLaVA-NeXT-34B (LLaVA-1.6) was published by University of Wisconsin Madison,ByteDance,Nanyang Technological University,University of California (UC) Berkeley, in United States of America, in January 2024. academia,Industry,Academia,Academia is the category the publisher falls under.

It works in Multimodal, Language, Vision, and is recorded as doing visual question answering, Chat, Question answering.

The weights are published, so it can be downloaded and run on your own hardware indefinitely, offline, with no account attached.

What decides the speed

Across every card that can run it, the middle of the range is about 20.7 tokens per second, and 103 of them clear the ten tokens per second that roughly matches reading speed.

It is a dense model, so every parameter is read for every token produced. That makes speed track memory bandwidth almost exactly — a card with twice the bandwidth generates roughly twice as fast.

Memory here is estimated from size rather than computed from the architecture, which is not recorded for this model — the numbers are indicative rather than exact.

Training and provenance

Training it took roughly 2.6 × 10²⁰ FLOP of computation, on NVIDIA A100 — a measure of what producing the model cost, not of how fast it answers.

It was trained on about 89,338,000 tokens of text.

Step by step

How to choose a GPU for LLaVA-NeXT-34B (LLaVA-1.6)

The table above has already assessed every card we hold specifications for against this model. Getting to your answer takes six steps.

01

Read the memory figure first

Look at what LLaVA-NeXT-34B (LLaVA-1.6) actually needs — around 17.7 GB at Q3_K_M. No amount of processing power compensates for a card that cannot hold it.
02

Set the context length you will work at

The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect: at long context LLaVA-NeXT-34B (LLaVA-1.6) can slip off a card that handles short questions easily.
03

Choose how far you will compress it

The quantisation column varies by card, because a bigger card holds a more accurate copy of LLaVA-NeXT-34B (LLaVA-1.6) — Q3_K_M on the smallest card that fits. Set a floor to hold the comparison at one level.
04

Compare tokens per second, not specifications

Sort by speed to see how cards rank for LLaVA-NeXT-34B (LLaVA-1.6). It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 97.5 tok/s.
05

Check the fit verdict before buying

Tight means LLaVA-NeXT-34B (LLaVA-1.6) loads and works, with no room to raise the context later. Comfortable means you can. The difference matters more than a few tokens per second.
06

Open the card you have settled on

Each card page repeats this sweep for every model we hold. It answers what else the hardware is good for, beyond LLaVA-NeXT-34B (LLaVA-1.6).

Answers

LLaVA-NeXT-34B (LLaVA-1.6) — common questions

How accurate are these LLaVA-NeXT-34B (LLaVA-1.6) speed estimates?

They are calculated from specifications rather than measured, and each carries a range — 59–156 tok/s on the B200, for instance. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

What GPU do I need to run LLaVA-NeXT-34B (LLaVA-1.6)?

The smallest card in our catalogue that holds LLaVA-NeXT-34B (LLaVA-1.6) is the RTX A4500, with 20 GB of memory. It runs the model at Q3_K_M using about 17.7 GB, and produces roughly 21.1 tokens per second. 132 cards in total can run it.

How fast is LLaVA-NeXT-34B (LLaVA-1.6) on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 97.5 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 103 of the cards that can run LLaVA-NeXT-34B (LLaVA-1.6) clear that.

How much VRAM does LLaVA-NeXT-34B (LLaVA-1.6) need?

About 17.7 GB at Q3_K_M compression, which is what the smallest card that runs it uses. Less compression needs more: the figures in the memory column above are recalculated for each card, because each one holds the least-compressed version it can.

Can I run LLaVA-NeXT-34B (LLaVA-1.6) on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at IQ4_XS, using about 19.7 GB and generating roughly 40.1 tokens per second — a tight fit.

Is LLaVA-NeXT-34B (LLaVA-1.6) open source?

Its weights are published, so LLaVA-NeXT-34B (LLaVA-1.6) can be downloaded and run on your own hardware. Note that open weights is not the same as open source in the full sense — it says nothing about the training data, the training code, or the commercial terms attached.

How many parameters does LLaVA-NeXT-34B (LLaVA-1.6) have?

LLaVA-NeXT-34B (LLaVA-1.6) has 34.8B parameters. 34.75B. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

Who created LLaVA-NeXT-34B (LLaVA-1.6)?

LLaVA-NeXT-34B (LLaVA-1.6) was published by University of Wisconsin Madison,ByteDance,Nanyang Technological University,University of California (UC) Berkeley, based in United States of America, categorised as academia,Industry,Academia,Academia.

When was LLaVA-NeXT-34B (LLaVA-1.6) released?

LLaVA-NeXT-34B (LLaVA-1.6) was published in January 2024. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

What is LLaVA-NeXT-34B (LLaVA-1.6) used for?

LLaVA-NeXT-34B (LLaVA-1.6) works in Multimodal, Language, Vision, and is recorded as handling visual question answering, Chat, Question answering. These are the areas it was designed around; they describe intent rather than a hard boundary.

Where can I download LLaVA-NeXT-34B (LLaVA-1.6)?

The weights for LLaVA-NeXT-34B (LLaVA-1.6) are published, though we do not hold a repository link for it. This site calculates hardware requirements rather than hosting model files.

How much compute was used to train LLaVA-NeXT-34B (LLaVA-1.6)?

Around 2.6 × 10²⁰ FLOP, on NVIDIA A100. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.

Can I run LLaVA-NeXT-34B (LLaVA-1.6) if it does not fit in my GPU?

It can be split between the card and system memory, but LLaVA-NeXT-34B (LLaVA-1.6) generates painfully slowly that way — the nearest miss we calculate is short by 7.3 GB. Nothing on this page assumes offloading.

Would two GPUs run LLaVA-NeXT-34B (LLaVA-1.6) faster?

A second card roughly doubles the memory available but not the generation rate. With 132 cards already able to run LLaVA-NeXT-34B (LLaVA-1.6) alone, the case for pairing is weak.

Why does the quantisation differ between cards for LLaVA-NeXT-34B (LLaVA-1.6)?

Because capacity varies, so does how hard LLaVA-NeXT-34B (LLaVA-1.6) has to be squeezed — 6 distinct levels appear in the table above. Set a minimum quality to compare at one.

Source

Original publication

Record last updated 11 February 2026

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.

All AI models