CogVLM-17B TPS calculator

Open weights Tsinghua University,Z.ai (Zhipu AI),Beihang University 17B parameters November 2023

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

295 cards that can run it

818 cards we hold specifications for

Smallest card that fits

GeForce GTX 1080 Ti

11 GB · Q3_K_M · 27.7 tok/s

Fastest card

B200

199 tok/s · 180 GB

Which GPUs can run CogVLM-17B?

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

295 cards match

Calculating

					Needs	Quantisation	Fit
199 tok/s 120–319 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	18.9 GB	Q8_0	Comfortable
199 tok/s 120–319 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	18.9 GB	Q8_0	Comfortable
159 tok/s 95–255 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	18.9 GB	Q8_0	Comfortable
159 tok/s 95–255 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	18.9 GB	Q8_0	Comfortable
127 tok/s 76–204 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	18.9 GB	Q8_0	Comfortable
122 tok/s 73–195 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	18.9 GB	Q8_0	Comfortable
122 tok/s 73–195 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	18.9 GB	Q8_0	Comfortable
117 tok/s 70–187 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	18.9 GB	Q8_0	Comfortable
103 tok/s 62–166 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	18.9 GB	Q8_0	Comfortable
103 tok/s 62–166 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	18.9 GB	Q8_0	Comfortable
103 tok/s 62–166 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	18.9 GB	Q8_0	Comfortable
98.2 tok/s 59–157 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	18.9 GB	Q8_0	Comfortable
83.7 tok/s 50–134 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	18.9 GB	Q8_0	Comfortable
83.7 tok/s 50–134 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	18.9 GB	Q8_0	Comfortable
83.7 tok/s 50–134 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	18.9 GB	Q8_0	Comfortable
83.7 tok/s 50–134 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	18.9 GB	Q8_0	Comfortable
83.7 tok/s 50–134 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	18.9 GB	Q8_0	Comfortable
63.7 tok/s 38–102 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	18.9 GB	Q8_0	Comfortable
63.7 tok/s 38–102 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	18.9 GB	Q8_0	Comfortable
55.8 tok/s 33–89 · low confidence	GeForce RTX 3080 12 GB NVIDIA	12 GB	912 GB/s	Jan 2022	10.0 GB	IQ4_XS	Tight
55.8 tok/s 33–89 · low confidence	GeForce RTX 3080 Ti NVIDIA	12 GB	912 GB/s	May 2021	10.0 GB	IQ4_XS	Tight
53.1 tok/s 32–85 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	18.9 GB	Q8_0	Comfortable
52.0 tok/s 31–83 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	18.9 GB	Q8_0	Comfortable
50.8 tok/s 30–81 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	18.9 GB	Q8_0	Comfortable
50.8 tok/s 30–81 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	18.9 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: Tsinghua University,Z.ai (Zhipu AI),Beihang University
Organisation type: Academia,Industry,Academia
Country: China
Published: 6 November 2023
Authors: Weihan Wang, Qingsong Lv, Wenmeng Yu, Wenyi Hong, Ji Qi, Yan Wang, Junhui Ji, Zhuoyi Yang, Lei Zhao, Xixuan Song, Jiazheng Xu, Bin Xu, Juanzi Li, Yuxiao Dong, Ming Ding, Jie Tang

What it does

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

Domain: Multimodal, Vision, Language
Task: Image captioning, Visual question answering, Chat
Approach: Supervised
Base model: Vicuna-7B v0
Numerical format: BF16

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: 17B
Training data: 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: 6.3 × 10²² FLOP
How it was established: Reported
Fine-tuning compute: 2 × 10²² FLOP

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 (restricted use)
Training code: Unreleased

How it is classified

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

Foundation model: Yes
Why it is tracked: SOTA improvement
Record confidence: Confident
Citations: 794

Sources

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

Reference: CogVLM: Visual Expert for Pretrained Language Models
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run CogVLM-17B

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 CogVLM-17B

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

What you need to run it

Minimum card

GeForce GTX 1080 Ti

Memory needed

9.0 GB

Fastest

199 tok/s

CogVLM-17B is small enough at 17B parameters that hardware is rarely the obstacle — 295 of the cards we track can run it, including cards several years old.

The least hardware that works is a GeForce GTX 1080 Ti. Its 11 GB is enough at Q3_K_M compression, giving roughly 27.7 tokens per second.

The quickest result comes from a B200 at around 199 tokens per second — its 8,000 GB/s of bandwidth is what buys that.

Background

CogVLM-17B was published by Tsinghua University,Z.ai (Zhipu AI),Beihang University, in China, in November 2023. The organisation is categorised as academia,Industry,Academia.

It works in Multimodal, Vision, Language, and is recorded as doing image captioning, Visual question answering, Chat.

It is derived from Vicuna-7B v0 rather than trained from scratch, which is the usual way a specialised model is produced.

The weights being open is what puts this page in the calculator rather than only in the catalogue: it is a model you can actually hold.

Reading the throughput figures

Half the cards that hold it manage more than 20.6 tokens per second, and 251 exceed reading speed outright.

Every weight participates in every token here, so bandwidth is the whole story: the ranking below is effectively a ranking of memory throughput.

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.

What went into building it

The training run consumed about 6.3 × 10²² FLOP. That figure describes the cost of creating it and has no bearing on how quickly it generates text.

Its inclusion criterion is sOTA improvement.

Step by step

How to choose a GPU for CogVLM-17B

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

01

Start from the memory column

The table lists every card that can hold CogVLM-17B — around 9.0 GB at Q3_K_M. That figure, not the card's headline performance, is what decides whether it runs.
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 CogVLM-17B can slip off a card that handles short questions easily.
03

Choose how far you will compress it

Each card runs the least-compressed copy it can hold — Q3_K_M on the smallest card that fits. Setting a floor drops the cards that only manage CogVLM-17B by squeezing it further than you would want.
04

Compare tokens per second, not specifications

Sort by speed to see how cards rank for CogVLM-17B. It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 199 tok/s.
05

Read the fit column last

Tight means CogVLM-17B 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 CogVLM-17B.

Answers

CogVLM-17B — common questions

Why does the quantisation differ between cards for CogVLM-17B?

A larger card holds a more accurate copy. Across the cards that run CogVLM-17B, 5 compression levels are used; the floor control above pins it to one.

How accurate are these CogVLM-17B speed estimates?

These are estimates with real error bars. The fastest result here, 120–319 tok/s on the B200, could reasonably land anywhere in its published range depending on which runtime you use.

What GPU do I need to run CogVLM-17B?

The smallest card in our catalogue that holds CogVLM-17B is the GeForce GTX 1080 Ti, with 11 GB of memory. It runs the model at Q3_K_M using about 9.0 GB, and produces roughly 27.7 tokens per second. 295 cards in total can run it.

How fast is CogVLM-17B on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 199 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 251 of the cards that can run CogVLM-17B clear that.

How much VRAM does CogVLM-17B need?

About 9.0 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 CogVLM-17B on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at IQ4_XS, using about 10.0 GB and generating roughly 55.8 tokens per second — a tight fit.

Can I run CogVLM-17B on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q5_K_M, using about 13.0 GB and generating roughly 50.3 tokens per second — a tight fit.

Can I run CogVLM-17B on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 18.9 GB and generating roughly 33.4 tokens per second — a tight fit.

Is CogVLM-17B open source?

Its weights are published, so CogVLM-17B 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 CogVLM-17B have?

CogVLM-17B has 17B parameters. CogVLM-17B has 10 billion vision parameters and 7 billion language parameters. However, "the total number of trainable parameters is 6.5B". "CogVLM model comprises four fundamental components: a vision transformer (ViT) encoder, an MLP adapter, a pretrained large language model (GPT), and a visual expert module." ViT: EVA2-CLIP-E, last layer removed (5B params with last layer, non-trainable) MLP adapter: 2 layers, parameter count unavailable GPT: Vicuna1.5-7B (7B params) Visual expert module: parameter count unclear. 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 CogVLM-17B?

CogVLM-17B was published by Tsinghua University,Z.ai (Zhipu AI),Beihang University, based in China, categorised as academia,Industry,Academia.

When was CogVLM-17B released?

CogVLM-17B was published in November 2023. 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 CogVLM-17B used for?

CogVLM-17B works in Multimodal, Vision, Language, and is recorded as handling image captioning, Visual question answering, Chat. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.

Where can I download CogVLM-17B?

The weights for CogVLM-17B 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 CogVLM-17B?

Around 6.3 × 10²² FLOP. 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 CogVLM-17B if it does not fit in my GPU?

It can be split between the card and system memory, but CogVLM-17B generates painfully slowly that way — the nearest miss we calculate is short by 2.0 GB. Nothing on this page assumes offloading.

Would two GPUs run CogVLM-17B faster?

Two cards buy memory rather than speed. That matters for CogVLM-17B only if one card cannot hold it — 295 can, so a second adds little.

Source

Original publication

Record last updated 25 May 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