Llama 2-70B TPS calculator

Open weights Meta AI 70B parameters July 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

61 cards that can run it

818 cards we hold specifications for

Smallest card that fits

A100 PCIe 40 GB

40 GB · IQ4_XS · 23.2 tok/s

Fastest card

B200

48.4 tok/s · 180 GB

Which GPUs can run Llama 2-70B?

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

61 cards match

Calculating

					Needs	Quantisation	Fit
48.4 tok/s 41–58	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	72.5 GB	Q8_0	Comfortable
48.4 tok/s 41–58	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	72.5 GB	Q8_0	Comfortable
38.7 tok/s 23–62 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	72.5 GB	Q8_0	Comfortable
38.7 tok/s 23–62 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	72.5 GB	Q8_0	Comfortable
30.9 tok/s 19–49 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	72.5 GB	Q8_0	Comfortable
29.6 tok/s 25–36	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	72.5 GB	Q8_0	Comfortable
29.6 tok/s 25–36	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	72.5 GB	Q8_0	Comfortable
29.5 tok/s 25–35	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	56.2 GB	Q6_K	Comfortable
29.5 tok/s 25–35	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	56.2 GB	Q6_K	Comfortable
28.3 tok/s 17–45 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	72.5 GB	Q8_0	Comfortable
26.1 tok/s 22–31	GRID A100B NVIDIA	48 GB	1,870 GB/s	May 2020	39.9 GB	Q4_K_M	Tight
25.1 tok/s 15–40 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	72.5 GB	Q8_0	Comfortable
25.1 tok/s 15–40 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	72.5 GB	Q8_0	Comfortable
25.1 tok/s 15–40 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	72.5 GB	Q8_0	Comfortable
23.8 tok/s 20–29	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	72.5 GB	Q8_0	Tight
23.2 tok/s 20–28	A100 PCIe 40 GB NVIDIA	40 GB	1,560 GB/s	Jun 2020	35.8 GB	IQ4_XS	Tight
23.2 tok/s 20–28	A100 SXM4 40 GB NVIDIA	40 GB	1,560 GB/s	May 2020	35.8 GB	IQ4_XS	Tight
23.2 tok/s 20–28	A800 PCIe 40 GB NVIDIA	40 GB	1,560 GB/s	Nov 2022	35.8 GB	IQ4_XS	Tight
20.3 tok/s 17–24	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	72.5 GB	Q8_0	Tight
20.3 tok/s 17–24	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	72.5 GB	Q8_0	Tight
20.3 tok/s 17–24	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	72.5 GB	Q8_0	Tight
18.7 tok/s 16–22	RTX PRO 5000 Blackwell NVIDIA	48 GB	1,340 GB/s	Mar 2025	39.9 GB	Q4_K_M	Tight
17.9 tok/s 15–22	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	56.2 GB	Q6_K	Comfortable
17.9 tok/s 15–22	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	56.2 GB	Q6_K	Comfortable
17.9 tok/s 15–22	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	56.2 GB	Q6_K	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: Meta AI
Organisation type: Industry
Country: United States of America
Published: 18 July 2023
Authors: Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan,…

What it does

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

Domain: Language
Task: Language modeling, Language modeling/generation, Question answering
Approach: Supervised

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: 70B
Training data: 2,000,000,000,000 tokens
Epochs: 1
Batch size: 4,000,000

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: 8.1 × 10²³ FLOP
How it was established: Hardware,Operation counting

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 SXM4 80 GB
Chips used: 1,000
Chip-hours: 1,720,320
Wall-clock time: 1,728 hours (72 days)
Hardware utilisation: MFU 41.9%
Power draw: 795.6 kW
Compute cost: $1,102,561
Data centre: Meta’s Research Super Cluster

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
Hugging Face: meta-llama

How it is classified

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

Likely above 10²³ FLOP: Yes
Why it is tracked: Historical significance,Significant use,Highly cited,Training cost
Record confidence: Confident
Citations: 16,911

Sources

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

Reference: Llama 2: Open Foundation and Fine-Tuned Chat Models
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run Llama 2-70B

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 Llama 2-70B

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

The hardware side

Minimum card

A100 PCIe 40 GB

Memory needed

35.8 GB

Fastest

48.4 tok/s

Llama 2-70B sits at 70B parameters, which puts it above consumer hardware and into the range where a card is bought for this purpose rather than repurposed for it. 61 of the cards we track can hold it.

At the low end, a A100 PCIe 40 GB handles it — 40 GB, at IQ4_XS, for about 23.2 tokens per second.

Top of the range is the B200, at roughly 48.4 tokens per second thanks to 8,000 GB/s of bandwidth.

Background

Llama 2-70B was published by Meta AI, in United States of America, in July 2023. The organisation is categorised as industry.

It works in Language, and is recorded as doing language modeling, Language modeling/generation, Question answering.

Because its weights were released, nothing about running it depends on a provider staying available — it is yours once downloaded. It is published under the meta-llama organisation on Hugging Face.

Reading the throughput figures

Across every card that can run it, the middle of the range is about 17.1 tokens per second, and 49 of them clear the ten tokens per second that roughly matches reading speed.

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

Its attention layout is on file, so the memory figures are computed exactly rather than approximated.

What went into building it

Producing it required around 8.1 × 10²³ FLOP of arithmetic, on NVIDIA A100 SXM4 80 GB, which is a statement about the training budget rather than about inference.

It was trained on about 2,000,000,000,000 tokens of text.

It is tracked in the underlying dataset for one reason in particular: historical significance,Significant use,Highly cited,Training cost.

Step by step

How to choose a GPU for Llama 2-70B

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

Look at what Llama 2-70B actually needs — around 35.8 GB at IQ4_XS. No amount of processing power compensates for a card that cannot hold it.
02

Decide how long your conversations run

Set the context to what you will actually use. The cache grows with the conversation, and it is the usual reason Llama 2-70B stops fitting a card that seemed fine.
03

Set a quality floor

Compression is what makes Llama 2-70B fit smaller cards, at some cost in accuracy — IQ4_XS on the smallest card that fits. A minimum quality removes the ones that go too far.
04

Compare tokens per second, not specifications

Ranking by tokens per second for Llama 2-70B follows memory bandwidth, not core counts, which is why the B200 tops it at 48.4 tok/s.
05

Look at the headroom, not just the fit

The fit column separates cards that just manage Llama 2-70B from those with room to spare. Buy for the second if the context might grow.
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 Llama 2-70B.

Answers

Llama 2-70B — common questions

Would two GPUs run Llama 2-70B faster?

Capacity adds across cards; throughput does not. Since 61 of the cards we track already hold Llama 2-70B on their own, a second card is rarely the answer here.

Why does the quantisation differ between cards for Llama 2-70B?

Each card is shown running the least-compressed copy it can hold, and Llama 2-70B appears at 4 different compression levels across the cards that fit it. Bigger cards get the more accurate version.

How accurate are these Llama 2-70B speed estimates?

They are calculated from specifications rather than measured, and each carries a range — 41–58 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 Llama 2-70B?

The smallest card in our catalogue that holds Llama 2-70B is the A100 PCIe 40 GB, with 40 GB of memory. It runs the model at IQ4_XS using about 35.8 GB, and produces roughly 23.2 tokens per second. 61 cards in total can run it.

How fast is Llama 2-70B on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 48.4 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 49 of the cards that can run Llama 2-70B clear that.

How much VRAM does Llama 2-70B need?

About 35.8 GB at IQ4_XS 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.

Is Llama 2-70B open source?

Its weights are published, so Llama 2-70B 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 Llama 2-70B have?

Llama 2-70B has 70B parameters. Llama has been released in 7B, 13B, 34B, and 70B variants. 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 Llama 2-70B?

Llama 2-70B was published by Meta AI, based in United States of America, categorised as industry.

When was Llama 2-70B released?

Llama 2-70B was published in July 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 Llama 2-70B used for?

Llama 2-70B works in Language, and is recorded as handling language modeling, Language modeling/generation, Question answering. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

Where can I download Llama 2-70B?

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

How much compute was used to train Llama 2-70B?

Around 8.1 × 10²³ FLOP, on NVIDIA A100 SXM4 80 GB. 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 Llama 2-70B if it does not fit in my GPU?

Only by offloading, which is usually a false economy: the part in system memory drags the whole thing down — the nearest miss we calculate is short by 11.1 GB. Our figures for Llama 2-70B assume it is fully resident.

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