Eagle 2 TPS calculator

Open weights NVIDIA,Nanjing University,Tsinghua University,Hong Kong Polytechnic University,Johns Hopkins University,New York University (NYU) 8.9B parameters January 2025

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

582 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Quadro 6000

6 GB · Q3_K_M · 15.6 tok/s

Fastest card

B200

379 tok/s · 180 GB

Which GPUs can run Eagle 2?

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

582 cards match

Calculating

					Needs	Quantisation	Fit
379 tok/s 228–607 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	10.3 GB	Q8_0	Comfortable
379 tok/s 228–607 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	10.3 GB	Q8_0	Comfortable
303 tok/s 182–485 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	10.3 GB	Q8_0	Comfortable
303 tok/s 182–485 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	10.3 GB	Q8_0	Comfortable
242 tok/s 145–388 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	10.3 GB	Q8_0	Comfortable
232 tok/s 139–371 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	10.3 GB	Q8_0	Comfortable
232 tok/s 139–371 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	10.3 GB	Q8_0	Comfortable
222 tok/s 133–355 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	10.3 GB	Q8_0	Comfortable
197 tok/s 118–315 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	10.3 GB	Q8_0	Comfortable
197 tok/s 118–315 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	10.3 GB	Q8_0	Comfortable
197 tok/s 118–315 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	10.3 GB	Q8_0	Comfortable
187 tok/s 112–299 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	10.3 GB	Q8_0	Comfortable
159 tok/s 96–255 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	10.3 GB	Q8_0	Comfortable
159 tok/s 96–255 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	10.3 GB	Q8_0	Comfortable
159 tok/s 96–255 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	10.3 GB	Q8_0	Comfortable
159 tok/s 96–255 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	10.3 GB	Q8_0	Comfortable
159 tok/s 96–255 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	10.3 GB	Q8_0	Comfortable
126 tok/s 76–202 · low confidence	CMP 170HX 8 GB NVIDIA	8 GB	1,490 GB/s	Sep 2021	7.1 GB	Q5_K_M	Tight
121 tok/s 73–194 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	10.3 GB	Q8_0	Comfortable
121 tok/s 73–194 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	10.3 GB	Q8_0	Comfortable
108 tok/s 65–172 · low confidence	CMP 170HX 10 GB NVIDIA	10 GB	1,560 GB/s	Sep 2021	8.2 GB	Q6_K	Tight
101 tok/s 61–162 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	10.3 GB	Q8_0	Comfortable
99.0 tok/s 59–158 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	10.3 GB	Q8_0	Comfortable
96.8 tok/s 58–155 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	10.3 GB	Q8_0	Comfortable
96.8 tok/s 58–155 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	10.3 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: NVIDIA,Nanjing University,Tsinghua University,Hong Kong Polytechnic University,Johns Hopkins University,New York University (NYU)
Organisation type: Industry,Academia,Academia,Academia,Academia,Academia
Country: United States of America, China, Hong Kong
Published: 20 January 2025
Authors: Zhiqi Li, Guo Chen, Shilong Liu, Shihao Wang, Vibashan VS, Yishen Ji, Shiyi Lan, Hao Zhang, Yilin Zhao, Subhashree Radhakrishnan, Nadine Chang, Karan Sapra, Amala Sanjay Deshmukh, Tuomas Rintamaki, Matthieu Le, Ilia Karmanov, Lukas Voegtle, Philipp Fischer, De-An Huang, Timo Roman, Tong Lu, Jose M. Alvarez, Bryan Catanzaro, Jan Kautz, Andrew Tao, Guilin Liu, Zhiding Yu

What it does

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

Domain: Vision, Robotics, Language
Base model: Qwen2.5-7B

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: 8.9B
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: 4.7 × 10²² FLOP

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 H100 SXM5 80GB
Chips used: 256
Wall-clock time: 131 hours
Power draw: 352.1 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 (non-commercial)
Training code: Unreleased
Hugging Face: nvidia

How it is classified

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

Why it is tracked: SOTA improvement
Record confidence: Confident
Citations: 57

Sources

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

Reference: Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Models
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run Eagle 2

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 Eagle 2

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

Quadro 6000

Memory needed

5.1 GB

Fastest

379 tok/s

Eagle 2 reaches a parameter count of 8.9B. That is small enough that hardware is rarely the obstacle, including on cards several years old. The number of cards we track that can run it: 582.

The entry point is Quadro 6000, with a memory capacity of 6 GB, running it at a compression of Q3_K_M and producing around 15.6 tokens per second.

At the other end sits B200, generating roughly 379 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

What this model is

Eagle 2 was published by NVIDIA,Nanjing University,Tsinghua University,Hong Kong Polytechnic University,Johns Hopkins University,New York University (NYU), in the country recorded as United States of America, during January 2025. The category the publisher falls under is industry,Academia,Academia,Academia,Academia,Academia.

It works in the domain of Vision, Robotics, Language.

Its starting point was an existing base model, Qwen2.5-7B. Most models at this scale are adapted from an existing base rather than built from nothing.

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. On Hugging Face it is published under the organisation nvidia.

What decides the speed

Half the cards that hold it manage more than 21.3 tokens per second. Producing text faster than most people read it: 541 of them.

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

Its internal architecture is not on file, so memory is approximated from the parameter count and marked accordingly. Expect the real figure to differ, more so at long context.

How it was trained

Producing it required arithmetic totalling around 4.7 × 10²² FLOP, on hardware recorded as NVIDIA H100 SXM5 80GB. That figure measures what producing the model cost, and has no bearing on how fast it answers.

The reason it appears in this catalogue at all: sOTA improvement.

Step by step

How to choose a GPU for Eagle 2

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

01

Check what it needs before anything else

Every card here has been checked against Eagle 2, needing around 5.1 GB at a compression of Q3_K_M. Capacity is the gate — a card either holds it or it does not.
02

Match the context to your actual use

The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect, because at long context a card that handles short questions easily can be dropped by Eagle 2.
03

Decide how much compression you will accept

Compression is what makes a model fit smaller cards, at some cost in accuracy, reaching a compression of Q3_K_M on the smallest card that fits. Setting a minimum quality drops the cards that only manage it by squeezing further than you would want, and holds the comparison at one level.
04

Rank by throughput rather than spec sheet

Sort by speed to see how cards rank for Eagle 2. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 379 tok/s.
05

Look at the headroom, not just the fit

A tight fit runs, but leaves nothing spare for a longer conversation, in the case of Eagle 2. Comfortable means you can grow the context later. That difference matters more than a few tokens per second, so buy for comfortable if you expect to.
06

Open the card you have settled on

Each card page repeats this sweep for every model we hold, answering what else the hardware is good for beyond Eagle 2.

Answers

Eagle 2 — common questions

Eagle 2— can I run it on a GPU holding 8 GB?

Yes. The card CMP 170HX 8 GB, holding 8 GB, runs it at a compression of Q5_K_M, using about 7.1 GB and generating roughly 126 tokens per second. The fit is tight.

Eagle 2— can I run it on a GPU holding 12 GB?

Yes. The card GeForce RTX 3080 Ti, holding 12 GB, runs it at a compression of Q8_0, using about 10.3 GB and generating roughly 43.3 tokens per second. The fit is tight.

Eagle 2— can I run it on a GPU holding 16 GB?

Yes. The card Tesla V100 SXM2 16 GB, holding 16 GB, runs it at a compression of Q8_0, using about 10.3 GB and generating roughly 53.6 tokens per second. The fit is comfortable.

Eagle 2— can I run it on a GPU holding 24 GB?

Yes. The card GeForce RTX 5090 D V2, holding 24 GB, runs it at a compression of Q8_0, using about 10.3 GB and generating roughly 63.6 tokens per second. The fit is comfortable.

Eagle 2— is it open source?

Its weights are published, so it 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.

Eagle 2— how many parameters does it have?

It has a parameter count of 8.9B. Table 4 https://huggingface.co/nvidia/Eagle2-9B "8.93B params". 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.

Eagle 2— who created it?

It was published by NVIDIA,Nanjing University,Tsinghua University,Hong Kong Polytechnic University,Johns Hopkins University,New York University (NYU), based in United States of America, an organisation categorised as industry,Academia,Academia,Academia,Academia,Academia.

Eagle 2— when was it released?

It was published in January 2025.

Eagle 2— what is it used for?

It works in the domain of Vision, Robotics, Language. 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.

Eagle 2— where can I download it?

Its weights are published on Hugging Face, under the organisation nvidia. We do not host model files — this site calculates what hardware is needed to run them.

Eagle 2— how much compute was used to train it?

Training consumed around 4.7 × 10²² FLOP, on hardware recorded as NVIDIA H100 SXM5 80GB. 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.

Eagle 2— can I run it if it does not fit in my GPU?

Only by offloading, which is usually a false economy: the part held in system memory drags the whole thing down. The nearest miss we calculate falls short by 1.6 GB. Every figure here assumes the whole model is resident on the card.

Eagle 2— would two GPUs run it faster?

Two cards buy memory rather than speed, which matters only if one card cannot hold it. The number that can: 582. So a second card is rarely the answer here.

Eagle 2— why does the quantisation differ between cards?

Because capacity varies, so does how hard it has to be squeezed. The number of distinct levels in the table above: 4. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

Eagle 2— how accurate are these speed estimates?

These are estimates with real error bars, and any of them could reasonably land anywhere in its published range depending on which runtime you use. The fastest result here: 228–607 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

Eagle 2— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Quadro 6000, with a memory capacity of 6 GB. It runs the model at a compression of Q3_K_M using about 5.1 GB, and produces roughly 15.6 tokens per second. The number of cards able to run it in total: 582.

Eagle 2— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 379 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and the number of cards clearing that: 541.

Eagle 2— how much VRAM does it need?

It needs about 5.1 GB at a compression of Q3_K_M, 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.

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