ERNIE-4.5-300B-A47B TPS calculator

Open weights Baidu 300B parameters June 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

7 of 818 cards that can run it

Smallest card that fits

B200

180 GB · IQ4_XS · 154 tok/s

Fastest card

B200

154 tok/s · 180 GB

Which GPUs can run ERNIE-4.5-300B-A47B?

Set the inputs, read the answer

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

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

7 cards match

Calculating
Needs Quantisation Fit
154 tok/s

92–247 · low confidence

B200 NVIDIA 180 GB 8,000 GB/s Jan 2024 153.0 GB IQ4_XS Tight
91.2 tok/s

55–146 · low confidence

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 240.3 GB Q6_K Tight
75.2 tok/s

45–120 · low confidence

Radeon Instinct MI300X AMD 192 GB 5,325 GB/s Dec 2023 170.5 GB Q4_K_M Tight
75.2 tok/s

45–120 · low confidence

Radeon Instinct MI308X AMD 192 GB 5,325 GB/s Dec 2023 170.5 GB Q4_K_M Tight
72.8 tok/s

44–116 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 240.3 GB Q6_K Tight
72.8 tok/s

44–116 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 240.3 GB Q6_K Tight
65.6 tok/s

39–105 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 205.4 GB Q5_K_M Tight

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
Baidu
Organisation type
Industry
Country
China
Published
29 June 2025

What it does

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

Domain
Language
Task
Language modeling/generation, Quantitative reasoning, Code generation, Translation, Question answering
Numerical format
FP8

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
300B

MoE total parameters: 300B active parameters: 47B

Training data
tokens

"We commence with large-scale pre-training on trillions of pure-text tokens sourced from diverse domains."

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
2.8 × 10²⁴ FLOP

They say the model was trained on "trillions of tokens" Speculatively assuming 10T tokens: 6 FLOP / token / parameter * 47 * 10^9 active parameters * 10 * 10^12 assumed training tokens = 2.82e+24 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 H800 SXM5
Hardware utilisation
MFU 47.0%

Abstract says: "We achieve 47% Model FLOPs Utilization (MFU) during the pre-training of our largest ERNIE 4.5 language model." Page 4: "We efficiently pre-train ERNIE 4.5 using a heterogeneous hybrid parallelism approach and a hierarchical load balancing solution tailored to multimodal large models. Through our extreme optimizations, including efficient intra-node expert parallelism, FP8 mixed-precision training, and fine-grained recomputation methods, we achieve 47% Model FLOPs Utilization (MF…

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)
Training code
Unreleased

Apache 2.0 https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-Base-PT Apache 2.0 for inference code https://github.com/PaddlePaddle/ERNIE

Hugging Face
baidu

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
Record confidence
Speculative

Sources

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

Reference
ERNIE 4.5 Technical Report
Last updated
28 November 2025

What the numbers mean

The hardware side

Minimum card

B200

Memory needed

153.0 GB

Fastest

154 tok/s

At 300B parameters, ERNIE-4.5-300B-A47B is beyond what any single graphics card holds. Running it means either splitting it across several cards or renting hardware built for the job — 7 of the cards we track can hold it on their own, and all of them are datacentre parts.

At the low end, a B200 handles it — 180 GB, at IQ4_XS, for about 154 tokens per second.

A B200 is the fastest we calculate for it: about 154 tokens per second, from 8,000 GB/s of memory bandwidth.

What this model is

ERNIE-4.5-300B-A47B was published by Baidu, in China, in June 2025. It comes out of industry.

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

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. It is published under the baidu organisation on Hugging Face.

What decides the speed

The median result is around 75.2 tokens per second; 7 cards produce text faster than most people read it.

Mixture-of-experts routing is why the speeds here look high for the parameter count. Only a fraction is read per token, but the whole thing has to be loaded.

Without the attention layout on record, the memory column is an approximation. It is close enough to choose hardware by, and least reliable at long context.

How it was trained

Producing it required around 2.8 × 10²⁴ FLOP of arithmetic, on NVIDIA H800 SXM5, which is a statement about the training budget rather than about inference.

Step by step

How to choose a GPU for ERNIE-4.5-300B-A47B

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

  1. 01

    Check what it needs before anything else

    Look at what ERNIE-4.5-300B-A47B actually needs — around 153.0 GB at IQ4_XS. No amount of processing power compensates for a card that cannot hold it.

  2. 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: at long context ERNIE-4.5-300B-A47B can slip off a card that handles short questions easily.

  3. 03

    Decide how much compression you will accept

    The quantisation column varies by card, because a bigger card holds a more accurate copy of ERNIE-4.5-300B-A47B — IQ4_XS on the smallest card that fits. Set a floor to hold the comparison at one level.

  4. 04

    Sort by speed

    The speed ordering for ERNIE-4.5-300B-A47B is effectively an ordering by memory bandwidth, which is why the B200 tops it at 154 tok/s.

  5. 05

    Check the fit verdict before buying

    A tight fit runs ERNIE-4.5-300B-A47B but leaves nothing spare for a longer conversation; comfortable has headroom. If you expect to grow the context, buy for comfortable.

  6. 06

    See what else that card runs

    Every card name links to its own page, which runs the same calculation across the whole model catalogue. Worth a look before buying for ERNIE-4.5-300B-A47B alone — a card is usually bought for more than one model.

Answers

ERNIE-4.5-300B-A47B — common questions

01

When was ERNIE-4.5-300B-A47B released?

ERNIE-4.5-300B-A47B was published in June 2025.

02

What is ERNIE-4.5-300B-A47B used for?

ERNIE-4.5-300B-A47B works in Language, and is recorded as handling language modeling/generation, Quantitative reasoning, Code generation, Translation, Question answering. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

03

Where can I download ERNIE-4.5-300B-A47B?

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

04

How much compute was used to train ERNIE-4.5-300B-A47B?

Around 2.8 × 10²⁴ FLOP, on NVIDIA H800 SXM5. 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.

05

Can I run ERNIE-4.5-300B-A47B 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 43.6 GB. Our figures for ERNIE-4.5-300B-A47B assume it is fully resident.

06

Would two GPUs run ERNIE-4.5-300B-A47B faster?

Capacity adds across cards; throughput does not. Since 7 of the cards we track already hold ERNIE-4.5-300B-A47B on their own, a second card is rarely the answer here.

07

Why does the quantisation differ between cards for ERNIE-4.5-300B-A47B?

Each card is shown running the least-compressed copy it can hold, and ERNIE-4.5-300B-A47B appears at 4 different compression levels across the cards that fit it. Bigger cards get the more accurate version.

08

How accurate are these ERNIE-4.5-300B-A47B speed estimates?

These are estimates with real error bars. The fastest result here, 92–247 tok/s on the B200, could reasonably land anywhere in its published range depending on which runtime you use.

09

What GPU do I need to run ERNIE-4.5-300B-A47B?

The smallest card in our catalogue that holds ERNIE-4.5-300B-A47B is the B200, with 180 GB of memory. It runs the model at IQ4_XS using about 153.0 GB, and produces roughly 154 tokens per second. 7 cards in total can run it.

10

How fast is ERNIE-4.5-300B-A47B on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 154 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 7 of the cards that can run ERNIE-4.5-300B-A47B clear that.

11

How much VRAM does ERNIE-4.5-300B-A47B need?

About 153.0 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.

12

Is ERNIE-4.5-300B-A47B open source?

Its weights are published, so ERNIE-4.5-300B-A47B 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.

13

How many parameters does ERNIE-4.5-300B-A47B have?

ERNIE-4.5-300B-A47B has 300B parameters. MoE total parameters: 300B active parameters: 47B. 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.

14

Who created ERNIE-4.5-300B-A47B?

ERNIE-4.5-300B-A47B was published by Baidu, based in China, categorised as industry.

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.