EfficientNetV2-XL TPS calculator
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 we hold specifications for
Smallest card that fits
Tesla C1080
4 GB · Q8_0 · 177 tok/s
Fastest card
B200
16,290 tok/s · 180 GB
Which GPUs can run EfficientNetV2-XL?
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.
818 cards match
Calculating| Needs | Quantisation | Fit | |||||
|---|---|---|---|---|---|---|---|
|
16,290
tok/s
9,774–26,063 · low confidence |
B200 NVIDIA | 180 GB | 8,000 GB/s | Jan 2024 | 0.9 GB | Q8_0 | Comfortable |
|
16,290
tok/s
9,774–26,063 · low confidence |
B300 NVIDIA | 288 GB | 8,000 GB/s | Sep 2025 | 0.9 GB | Q8_0 | Comfortable |
|
13,008
tok/s
7,805–20,812 · low confidence |
Radeon Instinct MI350X AMD | 288 GB | 8,190 GB/s | Jan 2025 | 0.9 GB | Q8_0 | Comfortable |
|
13,008
tok/s
7,805–20,812 · low confidence |
Radeon Instinct MI355X AMD | 288 GB | 8,190 GB/s | Jan 2025 | 0.9 GB | Q8_0 | Comfortable |
|
10,403
tok/s
6,242–16,645 · low confidence |
Radeon Instinct MI300 AMD | 128 GB | 6,550 GB/s | Jan 2023 | 0.9 GB | Q8_0 | Comfortable |
|
9,957
tok/s
5,974–15,931 · low confidence |
H200 NVL NVIDIA | 141 GB | 4,890 GB/s | Nov 2024 | 0.9 GB | Q8_0 | Comfortable |
|
9,957
tok/s
5,974–15,931 · low confidence |
H200 SXM 141 GB NVIDIA | 141 GB | 4,890 GB/s | Nov 2024 | 0.9 GB | Q8_0 | Comfortable |
|
9,529
tok/s
5,718–15,247 · low confidence |
Radeon Instinct MI325X AMD | 256 GB | 6,000 GB/s | Oct 2024 | 0.9 GB | Q8_0 | Comfortable |
|
8,457
tok/s
5,074–13,532 · low confidence |
Radeon Instinct MI300A AMD | 128 GB | 5,325 GB/s | Dec 2023 | 0.9 GB | Q8_0 | Comfortable |
|
8,457
tok/s
5,074–13,532 · low confidence |
Radeon Instinct MI300X AMD | 192 GB | 5,325 GB/s | Dec 2023 | 0.9 GB | Q8_0 | Comfortable |
|
8,457
tok/s
5,074–13,532 · low confidence |
Radeon Instinct MI308X AMD | 192 GB | 5,325 GB/s | Dec 2023 | 0.9 GB | Q8_0 | Comfortable |
|
8,023
tok/s
4,814–12,836 · low confidence |
H100 NVL 94 GB NVIDIA | 94 GB | 3,940 GB/s | Mar 2023 | 0.9 GB | Q8_0 | Comfortable |
|
6,842
tok/s
4,105–10,947 · low confidence |
H100 PCIe 96 GB NVIDIA | 96 GB | 3,360 GB/s | Mar 2023 | 0.9 GB | Q8_0 | Comfortable |
|
6,842
tok/s
4,105–10,947 · low confidence |
H100 SXM5 80 GB NVIDIA | 80 GB | 3,360 GB/s | Oct 2022 | 0.9 GB | Q8_0 | Comfortable |
|
6,842
tok/s
4,105–10,947 · low confidence |
H100 SXM5 94 GB NVIDIA | 94 GB | 3,360 GB/s | Mar 2023 | 0.9 GB | Q8_0 | Comfortable |
|
6,842
tok/s
4,105–10,947 · low confidence |
H100 SXM5 96 GB NVIDIA | 96 GB | 3,360 GB/s | Mar 2023 | 0.9 GB | Q8_0 | Comfortable |
|
6,842
tok/s
4,105–10,947 · low confidence |
H800 SXM5 NVIDIA | 80 GB | 3,360 GB/s | Mar 2023 | 0.9 GB | Q8_0 | Comfortable |
|
5,209
tok/s
3,126–8,335 · low confidence |
Radeon Instinct MI250 AMD | 128 GB | 3,280 GB/s | Nov 2021 | 0.9 GB | Q8_0 | Comfortable |
|
5,209
tok/s
3,126–8,335 · low confidence |
Radeon Instinct MI250X AMD | 128 GB | 3,280 GB/s | Nov 2021 | 0.9 GB | Q8_0 | Comfortable |
|
4,341
tok/s
2,605–6,946 · low confidence |
Data Center GPU Max 1550 Intel | 128 GB | 3,280 GB/s | Jan 2023 | 0.9 GB | Q8_0 | Comfortable |
|
4,249
tok/s
2,549–6,798 · low confidence |
Data Center GPU Max Subsystem Intel | 128 GB | 3,210 GB/s | Jan 2023 | 0.9 GB | Q8_0 | Comfortable |
|
4,154
tok/s
2,492–6,646 · low confidence |
A100 SXM4 80 GB NVIDIA | 80 GB | 2,040 GB/s | Nov 2020 | 0.9 GB | Q8_0 | Comfortable |
|
4,154
tok/s
2,492–6,646 · low confidence |
A100X NVIDIA | 80 GB | 2,040 GB/s | Jun 2021 | 0.9 GB | Q8_0 | Comfortable |
|
4,154
tok/s
2,492–6,646 · low confidence |
A800 SXM4 80 GB NVIDIA | 80 GB | 2,040 GB/s | Aug 2022 | 0.9 GB | Q8_0 | Comfortable |
|
4,154
tok/s
2,492–6,646 · low confidence |
H100 CNX NVIDIA | 80 GB | 2,040 GB/s | Mar 2023 | 0.9 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
- Google,Google Brain
- Organisation type
- Industry,Industry
- Country
- United States of America
- Published
- 23 June 2021
- Authors
- Mingxing Tan, Quoc V. Le
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Vision
- Task
- Image classification, Neural Architecture Search - NAS
- Approach
- Supervised
- Numerical format
- FP16
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
- 208M
- Training data
- 14,180,000 tokens
- Epochs
- 30
- Batch size
- 4,096
208M for XL version (Table 7, page 7)
"ImageNet21k (Russakovsky et al., 2015) contains about 13M training images with 21,841 classes. The original ImageNet21k doesn’t have train/eval split, so we reserve randomly picked 100,000 images as validation set and use the remaining as training set... After pretrained on ImageNet21k, each model is finetuned on ILSVRC2012 for 15 epochs using cosine learning rate decay." 12.9M + 1.28M ~= 14,180,000
"Each model is trained for 350 epochs with total batch size 4096"
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
- 9.6 × 10¹⁹ FLOP
- How it was established
- Hardware
Table 7, page 7: 45 hours on 32 TPUv3 cores. "Each v3 TPU chip contains two TensorCores." TPU performance per chip = 123e12 FLOP/s 32 cores = 16 chips 123e12 FLOP/s per chip * (32 cores / 2 cores per chip) * 45 hours * 3600 seconds/hour * 0.30 utilization = 9.56e19 FLOP https://www.wolframalpha.com/input?i=123+terahertz+*+16+*+45+hours+*+0.3
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
- Google TPU v3
- Chips used
- 16
- Wall-clock time
- 45 hours
- Power draw
- 14.6 kW
- Compute cost
- $104
Table 7
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
- Open source
code and weights: https://github.com/google/automl/tree/master/efficientnetv2 Apache-2.0 license
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
- Highly cited,SOTA improvement
- Record confidence
- Confident
- Citations
- 4,324
"EfficientNetV2 achieves 87.3% top-1 accuracy on ImageNet ILSVRC2012, outperforming the recent ViT by 2.0% accuracy while training 5x-11x faster using the same computing resources."
Sources
Where this record came from and when it was last checked.
- Reference
- EfficientNetV2: Smaller Models and Faster Training
- Last updated
- 25 May 2026
The extremes
The ten fastest GPUs that run EfficientNetV2-XL
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.
- 01 B300 288 GB · 8,000 GB/s · Q8_0 16,290 tok/s
- 02 B200 180 GB · 8,000 GB/s · Q8_0 16,290 tok/s
- 03 Radeon Instinct MI350X 288 GB · 8,190 GB/s · Q8_0 13,008 tok/s
- 04 Radeon Instinct MI355X 288 GB · 8,190 GB/s · Q8_0 13,008 tok/s
- 05 Radeon Instinct MI300 128 GB · 6,550 GB/s · Q8_0 10,403 tok/s
- 06 H200 NVL 141 GB · 4,890 GB/s · Q8_0 9,957 tok/s
- 07 H200 SXM 141 GB 141 GB · 4,890 GB/s · Q8_0 9,957 tok/s
- 08 Radeon Instinct MI325X 256 GB · 6,000 GB/s · Q8_0 9,529 tok/s
- 09 Radeon Instinct MI300A 128 GB · 5,325 GB/s · Q8_0 8,457 tok/s
- 10 Radeon Instinct MI300X 192 GB · 5,325 GB/s · Q8_0 8,457 tok/s
The smallest GPUs that still run EfficientNetV2-XL
The cheapest route in, by memory capacity. A tight fit runs the model but leaves nothing spare for a longer conversation.
- 01 GeForce RTX 4010 4 GB · needs 0.9 GB · Q8_0 · comfortable 195 tok/s
- 02 RTX A400 4 GB · needs 0.9 GB · Q8_0 · comfortable 195 tok/s
- 03 RTX 500 Mobile Ada Generation 4 GB · needs 0.9 GB · Q8_0 · comfortable 261 tok/s
- 04 GeForce RTX 3050 A Mobile 4 GB · needs 0.9 GB · Q8_0 · comfortable 391 tok/s
- 05 Jetson Orin Nano 4 GB 4 GB · needs 0.9 GB · Q8_0 · comfortable 69.5 tok/s
- 06 Radeon RX 6450M 4 GB · needs 0.9 GB · Q8_0 · comfortable 203 tok/s
- 07 Radeon RX 6550M 4 GB · needs 0.9 GB · Q8_0 · comfortable 229 tok/s
- 08 Radeon RX 6550S 4 GB · needs 0.9 GB · Q8_0 · comfortable 203 tok/s
- 09 Arc A310 4 GB · needs 0.9 GB · Q8_0 · comfortable 164 tok/s
- 10 Arc Pro A30M 4 GB · needs 0.9 GB · Q8_0 · comfortable 169 tok/s
What the numbers mean
What it takes to run this model
Minimum card
Tesla C1080
Memory needed
0.9 GB
Fastest
16,290 tok/s
EfficientNetV2-XL is small enough at 208M 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 177 tokens per second.
The quickest result comes from a B200 at around 16,290 tokens per second — its 8,000 GB/s of bandwidth is what buys that.
What this model is
EfficientNetV2-XL was published by Google,Google Brain, in United States of America, in June 2021. It comes out of industry,Industry.
It works in Vision, and is recorded as doing image classification, Neural Architecture Search - NAS.
Published weights mean the model runs on your machine rather than someone else's, which is what makes the hardware question below answerable at all.
What decides the speed
The median result is around 457.4 tokens per second; 818 cards produce text faster than most people read it.
It is a dense model, so every parameter is read for every token produced. That makes speed track memory bandwidth almost exactly — a card with twice the bandwidth generates roughly twice as fast.
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
Training it took roughly 9.6 × 10¹⁹ FLOP of computation, on Google TPU v3 — a measure of what producing the model cost, not of how fast it answers.
Around 14,180,000 tokens went into training it.
It is tracked in the underlying dataset for one reason in particular: highly cited,SOTA improvement.
Step by step
How to choose a GPU for EfficientNetV2-XL
The table above has already assessed every card we hold specifications for against this model. Getting to your answer takes six steps.
-
01
Read the memory figure first
Every card here has been checked against EfficientNetV2-XL — around 0.9 GB at Q8_0. Capacity is the gate — a card either holds it or it does not.
-
02
Match the context to your actual use
Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for EfficientNetV2-XL.
-
03
Choose how far you will compress it
Each card runs the least-compressed copy it can hold — Q8_0 on the smallest card that fits. Setting a floor drops the cards that only manage EfficientNetV2-XL by squeezing it further than you would want.
-
04
Rank by throughput rather than spec sheet
Sort by speed to see how cards rank for EfficientNetV2-XL. It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 16,290 tok/s.
-
05
Read the fit column last
A tight fit runs EfficientNetV2-XL but leaves nothing spare for a longer conversation; comfortable has headroom. If you expect to grow the context, buy for comfortable.
-
06
Check the card from the other side
Following a card through to its own page shows every other model it can hold, which is the question that follows once EfficientNetV2-XL is settled.
Answers
EfficientNetV2-XL — common questions
What is EfficientNetV2-XL used for?
EfficientNetV2-XL works in Vision, and is recorded as handling image classification, Neural Architecture Search - NAS. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
Where can I download EfficientNetV2-XL?
The weights for EfficientNetV2-XL 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 EfficientNetV2-XL?
Around 9.6 × 10¹⁹ FLOP, on Google TPU v3. 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 EfficientNetV2-XL if it does not fit in my GPU?
Partly. Layers that do not fit sit in system memory and run at a fraction of the speed, so a mostly-offloaded EfficientNetV2-XL is rarely worth using. Every figure here assumes the whole model is on the card.
Would two GPUs run EfficientNetV2-XL faster?
Two cards buy memory rather than speed. That matters for EfficientNetV2-XL only if one card cannot hold it — 818 can, so a second adds little.
Why does the quantisation differ between cards for EfficientNetV2-XL?
A larger card holds a more accurate copy. Across the cards that run EfficientNetV2-XL, 1 compression levels are used; the floor control above pins it to one.
How accurate are these EfficientNetV2-XL speed estimates?
They are calculated from specifications rather than measured, and each carries a range — 9,774–26,063 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 EfficientNetV2-XL?
The smallest card in our catalogue that holds EfficientNetV2-XL is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 0.9 GB, and produces roughly 177 tokens per second. 818 cards in total can run it.
How fast is EfficientNetV2-XL on a GPU?
It depends on the card. The quickest we calculate is a B200 at about 16,290 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 818 of the cards that can run EfficientNetV2-XL clear that.
How much VRAM does EfficientNetV2-XL need?
About 0.9 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 EfficientNetV2-XL on a 8 GB GPU?
Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 0.9 GB and generating roughly 3,034 tokens per second — a comfortable fit.
Can I run EfficientNetV2-XL on a 12 GB GPU?
Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 0.9 GB and generating roughly 1,858 tokens per second — a comfortable fit.
Can I run EfficientNetV2-XL on a 16 GB GPU?
Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 0.9 GB and generating roughly 2,301 tokens per second — a comfortable fit.
Can I run EfficientNetV2-XL on a 24 GB GPU?
Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 0.9 GB and generating roughly 2,729 tokens per second — a comfortable fit.
Is EfficientNetV2-XL open source?
Its weights are published, so EfficientNetV2-XL 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 EfficientNetV2-XL have?
EfficientNetV2-XL has 208M parameters. 208M for XL version (Table 7, page 7). 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 EfficientNetV2-XL?
EfficientNetV2-XL was published by Google,Google Brain, based in United States of America, categorised as industry,Industry.
When was EfficientNetV2-XL released?
EfficientNetV2-XL was published in June 2021. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
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.