Spec-Drafter TPS calculator

Open weights Peking University,Microsoft Research Asia 500M parameters October 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

818 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Tesla C1080

4 GB · Q8_0 · 73.7 tok/s

Fastest card

B200

6,776 tok/s · 180 GB

Which GPUs can run Spec-Drafter?

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

818 cards match

Calculating

					Needs	Quantisation	Fit
6,776 tok/s 4,066–10,842 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	1.2 GB	Q8_0	Comfortable
6,776 tok/s 4,066–10,842 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	1.2 GB	Q8_0	Comfortable
5,411 tok/s 3,247–8,658 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	1.2 GB	Q8_0	Comfortable
5,411 tok/s 3,247–8,658 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	1.2 GB	Q8_0	Comfortable
4,328 tok/s 2,597–6,924 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	1.2 GB	Q8_0	Comfortable
4,142 tok/s 2,485–6,627 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	1.2 GB	Q8_0	Comfortable
4,142 tok/s 2,485–6,627 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	1.2 GB	Q8_0	Comfortable
3,964 tok/s 2,379–6,343 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	1.2 GB	Q8_0	Comfortable
3,518 tok/s 2,111–5,629 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	1.2 GB	Q8_0	Comfortable
3,518 tok/s 2,111–5,629 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	1.2 GB	Q8_0	Comfortable
3,518 tok/s 2,111–5,629 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	1.2 GB	Q8_0	Comfortable
3,337 tok/s 2,002–5,340 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	1.2 GB	Q8_0	Comfortable
2,846 tok/s 1,708–4,554 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	1.2 GB	Q8_0	Comfortable
2,846 tok/s 1,708–4,554 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	1.2 GB	Q8_0	Comfortable
2,846 tok/s 1,708–4,554 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	1.2 GB	Q8_0	Comfortable
2,846 tok/s 1,708–4,554 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	1.2 GB	Q8_0	Comfortable
2,846 tok/s 1,708–4,554 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	1.2 GB	Q8_0	Comfortable
2,167 tok/s 1,300–3,467 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	1.2 GB	Q8_0	Comfortable
2,167 tok/s 1,300–3,467 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	1.2 GB	Q8_0	Comfortable
1,806 tok/s 1,084–2,889 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	1.2 GB	Q8_0	Comfortable
1,767 tok/s 1,060–2,828 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	1.2 GB	Q8_0	Comfortable
1,728 tok/s 1,037–2,765 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	1.2 GB	Q8_0	Comfortable
1,728 tok/s 1,037–2,765 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	1.2 GB	Q8_0	Comfortable
1,728 tok/s 1,037–2,765 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	1.2 GB	Q8_0	Comfortable
1,728 tok/s 1,037–2,765 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	1.2 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: Peking University,Microsoft Research Asia
Organisation type: Academia,Industry
Country: China
Published: 30 October 2023
Authors: Heming Xia, Tao Ge, Peiyi Wang, Si-Qing Chen, Furu Wei, Zhifang Sui

What it does

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

Domain: Language
Task: Translation

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: 500M
Training data: 39,321,600,000 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: 1.2 × 10²⁰ FLOP
How it was established: 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 V100
Chips used: 8
Power draw: 4.8 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: Open (non-commercial)

How it is classified

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

Record confidence: Confident

Sources

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

Reference: Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run Spec-Drafter

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 Spec-Drafter

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 it takes to run this model

Minimum card

Tesla C1080

Memory needed

1.2 GB

Fastest

6,776 tok/s

Spec-Drafter is small enough at 500M parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.

The smallest card that holds it is the Tesla C1080 with 4 GB, running it at Q8_0 and producing around 73.7 tokens per second.

At the other end, a B200 generates roughly 6,776 tokens per second on it, on the strength of 8,000 GB/s of memory bandwidth.

Background

Spec-Drafter was published by Peking University,Microsoft Research Asia, in China, in October 2023. The organisation is categorised as academia,Industry.

It works in Language, and is recorded as doing translation.

Because its weights were released, nothing about running it depends on a provider staying available — it is yours once downloaded.

Reading the throughput figures

The median result is around 190.3 tokens per second; 816 cards produce text faster than most people read it.

Being dense, it reads all of itself per token, which is why the ordering by speed below follows the ordering by memory bandwidth so closely.

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

Producing it required around 1.2 × 10²⁰ FLOP of arithmetic, on NVIDIA V100, which is a statement about the training budget rather than about inference.

The training set ran to roughly 39,321,600,000 tokens.

Step by step

How to choose a GPU for Spec-Drafter

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

The table lists every card that can hold Spec-Drafter — around 1.2 GB at Q8_0. That figure, not the card's headline performance, is what decides whether it runs.
02

Set the context length you will work at

Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for Spec-Drafter.
03

Set a quality floor

Compression is what makes Spec-Drafter fit smaller cards, at some cost in accuracy — Q8_0 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 Spec-Drafter follows memory bandwidth, not core counts, which is why the B200 tops it at 6,776 tok/s.
05

Read the fit column last

A tight fit runs Spec-Drafter but leaves nothing spare for a longer conversation; comfortable has headroom. If you expect to grow the context, buy for comfortable.
06

See what else that card runs

Following a card through to its own page shows every other model it can hold, which is the question that follows once Spec-Drafter is settled.

Answers

Spec-Drafter — common questions

How accurate are these Spec-Drafter speed estimates?

These are estimates with real error bars. The fastest result here, 4,066–10,842 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 Spec-Drafter?

The smallest card in our catalogue that holds Spec-Drafter is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 1.2 GB, and produces roughly 73.7 tokens per second. 818 cards in total can run it.

How fast is Spec-Drafter on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 6,776 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 816 of the cards that can run Spec-Drafter clear that.

How much VRAM does Spec-Drafter need?

About 1.2 GB at Q8_0 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 Spec-Drafter on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 1.2 GB and generating roughly 1,262 tokens per second — a comfortable fit.

Can I run Spec-Drafter on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 1.2 GB and generating roughly 773 tokens per second — a comfortable fit.

Can I run Spec-Drafter on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 1.2 GB and generating roughly 957 tokens per second — a comfortable fit.

Can I run Spec-Drafter on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 1.2 GB and generating roughly 1,135 tokens per second — a comfortable fit.

Is Spec-Drafter open source?

Its weights are published, so Spec-Drafter 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 Spec-Drafter have?

Spec-Drafter has 500M parameters. 0.5B 12-layer encoder + 2-layer decoder, d=512/2048. 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 Spec-Drafter?

Spec-Drafter was published by Peking University,Microsoft Research Asia, based in China, categorised as academia,Industry.

When was Spec-Drafter released?

Spec-Drafter was published in October 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 Spec-Drafter used for?

Spec-Drafter works in Language, and is recorded as handling translation. These are the areas it was designed around; they describe intent rather than a hard boundary.

Where can I download Spec-Drafter?

The weights for Spec-Drafter 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 Spec-Drafter?

Around 1.2 × 10²⁰ FLOP, on NVIDIA V100. 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 Spec-Drafter 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. Our figures for Spec-Drafter assume it is fully resident.

Would two GPUs run Spec-Drafter faster?

Capacity adds across cards; throughput does not. Since 818 of the cards we track already hold Spec-Drafter on their own, a second card is rarely the answer here.

Why does the quantisation differ between cards for Spec-Drafter?

Each card is shown running the least-compressed copy it can hold, and Spec-Drafter appears at 1 different compression levels across the cards that fit it. Bigger cards get the more accurate version.

Source

Original publication

Record last updated 28 November 2025

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