ALBERT-xxlarge 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 · 157 tok/s
Fastest card
B200
14,418 tok/s · 180 GB
Which GPUs can run ALBERT-xxlarge?
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 | |||||
|---|---|---|---|---|---|---|---|
|
14,418
tok/s
8,651–23,069 · low confidence |
B200 NVIDIA | 180 GB | 8,000 GB/s | Jan 2024 | 1.0 GB | Q8_0 | Comfortable |
|
14,418
tok/s
8,651–23,069 · low confidence |
B300 NVIDIA | 288 GB | 8,000 GB/s | Sep 2025 | 1.0 GB | Q8_0 | Comfortable |
|
11,513
tok/s
6,908–18,421 · low confidence |
Radeon Instinct MI350X AMD | 288 GB | 8,190 GB/s | Jan 2025 | 1.0 GB | Q8_0 | Comfortable |
|
11,513
tok/s
6,908–18,421 · low confidence |
Radeon Instinct MI355X AMD | 288 GB | 8,190 GB/s | Jan 2025 | 1.0 GB | Q8_0 | Comfortable |
|
9,208
tok/s
5,525–14,732 · low confidence |
Radeon Instinct MI300 AMD | 128 GB | 6,550 GB/s | Jan 2023 | 1.0 GB | Q8_0 | Comfortable |
|
8,813
tok/s
5,288–14,101 · low confidence |
H200 NVL NVIDIA | 141 GB | 4,890 GB/s | Nov 2024 | 1.0 GB | Q8_0 | Comfortable |
|
8,813
tok/s
5,288–14,101 · low confidence |
H200 SXM 141 GB NVIDIA | 141 GB | 4,890 GB/s | Nov 2024 | 1.0 GB | Q8_0 | Comfortable |
|
8,435
tok/s
5,061–13,495 · low confidence |
Radeon Instinct MI325X AMD | 256 GB | 6,000 GB/s | Oct 2024 | 1.0 GB | Q8_0 | Comfortable |
|
7,486
tok/s
4,491–11,977 · low confidence |
Radeon Instinct MI300A AMD | 128 GB | 5,325 GB/s | Dec 2023 | 1.0 GB | Q8_0 | Comfortable |
|
7,486
tok/s
4,491–11,977 · low confidence |
Radeon Instinct MI300X AMD | 192 GB | 5,325 GB/s | Dec 2023 | 1.0 GB | Q8_0 | Comfortable |
|
7,486
tok/s
4,491–11,977 · low confidence |
Radeon Instinct MI308X AMD | 192 GB | 5,325 GB/s | Dec 2023 | 1.0 GB | Q8_0 | Comfortable |
|
7,101
tok/s
4,261–11,361 · low confidence |
H100 NVL 94 GB NVIDIA | 94 GB | 3,940 GB/s | Mar 2023 | 1.0 GB | Q8_0 | Comfortable |
|
6,056
tok/s
3,633–9,689 · low confidence |
H100 PCIe 96 GB NVIDIA | 96 GB | 3,360 GB/s | Mar 2023 | 1.0 GB | Q8_0 | Comfortable |
|
6,056
tok/s
3,633–9,689 · low confidence |
H100 SXM5 80 GB NVIDIA | 80 GB | 3,360 GB/s | Oct 2022 | 1.0 GB | Q8_0 | Comfortable |
|
6,056
tok/s
3,633–9,689 · low confidence |
H100 SXM5 94 GB NVIDIA | 94 GB | 3,360 GB/s | Mar 2023 | 1.0 GB | Q8_0 | Comfortable |
|
6,056
tok/s
3,633–9,689 · low confidence |
H100 SXM5 96 GB NVIDIA | 96 GB | 3,360 GB/s | Mar 2023 | 1.0 GB | Q8_0 | Comfortable |
|
6,056
tok/s
3,633–9,689 · low confidence |
H800 SXM5 NVIDIA | 80 GB | 3,360 GB/s | Mar 2023 | 1.0 GB | Q8_0 | Comfortable |
|
4,611
tok/s
2,767–7,377 · low confidence |
Radeon Instinct MI250 AMD | 128 GB | 3,280 GB/s | Nov 2021 | 1.0 GB | Q8_0 | Comfortable |
|
4,611
tok/s
2,767–7,377 · low confidence |
Radeon Instinct MI250X AMD | 128 GB | 3,280 GB/s | Nov 2021 | 1.0 GB | Q8_0 | Comfortable |
|
3,842
tok/s
2,305–6,148 · low confidence |
Data Center GPU Max 1550 Intel | 128 GB | 3,280 GB/s | Jan 2023 | 1.0 GB | Q8_0 | Comfortable |
|
3,760
tok/s
2,256–6,017 · low confidence |
Data Center GPU Max Subsystem Intel | 128 GB | 3,210 GB/s | Jan 2023 | 1.0 GB | Q8_0 | Comfortable |
|
3,677
tok/s
2,206–5,883 · low confidence |
A100 SXM4 80 GB NVIDIA | 80 GB | 2,040 GB/s | Nov 2020 | 1.0 GB | Q8_0 | Comfortable |
|
3,677
tok/s
2,206–5,883 · low confidence |
A100X NVIDIA | 80 GB | 2,040 GB/s | Jun 2021 | 1.0 GB | Q8_0 | Comfortable |
|
3,677
tok/s
2,206–5,883 · low confidence |
A800 SXM4 80 GB NVIDIA | 80 GB | 2,040 GB/s | Aug 2022 | 1.0 GB | Q8_0 | Comfortable |
|
3,677
tok/s
2,206–5,883 · low confidence |
H100 CNX NVIDIA | 80 GB | 2,040 GB/s | Mar 2023 | 1.0 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
- Toyota Technological Institute at Chicago,Google
- Organisation type
- Academia,Industry
- Country
- United States of America
- Published
- 9 February 2020
- Authors
- Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Language
- Task
- Language modeling/generation, Question answering
- Approach
- Self-supervised learning
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
- 235M
- Training data
- 3,300,000,000 tokens
- Epochs
- 79.4
- Batch size
- 2,097,152
Pretraining same as for BERT - Wikipedia and BookCorpus "For the pre-training corpus we use the BooksCorpus (800M words) (Zhu et al., 2015) and English Wikipedia (2,500M words)"
Sequences are capped at 512 tokens; 10% of the time they'll use an input less than 512 long. Batches are over 4096 sequences. Tokens per batch: 2,097,152
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
- 2.4 × 10²¹ FLOP
- How it was established
- Hardware,Third-party estimation,Operation counting
32 hours of training 512 TPU V3s 0.33 utilization rate 123000000000000 FLOP / chip / sec * 512 TPUs * 32 hours * 3600 sec / hour * 0.33 [assumed utilization] = 2.3940956e+21 FLOP "We train all models for 125,000 steps unless otherwise specified" "All the model updates use a batch size of 4096 " "We always limit the maximum input length to 512, and randomly generate input sequences shorter than 512 with a probability of 10%." 6 FLOP / parameter / token * 235000000 parameters * 512 tokens per s…
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
- 512
- Wall-clock time
- 32 hours
- Power draw
- 471.2 kW
- Compute cost
- $4,440
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
Apache 2.0 code/weights. repo includes training code: https://github.com/google-research/ALBERT
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
- Record confidence
- Speculative
- Citations
- 7,447
Sources
Where this record came from and when it was last checked.
- Reference
- ALBERT: A Lite BERT for Self-supervised Learning of Language Representations.
- Last updated
- 25 May 2026
The extremes
The ten fastest GPUs that run ALBERT-xxlarge
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 14,418 tok/s
- 02 B200 180 GB · 8,000 GB/s · Q8_0 14,418 tok/s
- 03 Radeon Instinct MI350X 288 GB · 8,190 GB/s · Q8_0 11,513 tok/s
- 04 Radeon Instinct MI355X 288 GB · 8,190 GB/s · Q8_0 11,513 tok/s
- 05 Radeon Instinct MI300 128 GB · 6,550 GB/s · Q8_0 9,208 tok/s
- 06 H200 NVL 141 GB · 4,890 GB/s · Q8_0 8,813 tok/s
- 07 H200 SXM 141 GB 141 GB · 4,890 GB/s · Q8_0 8,813 tok/s
- 08 Radeon Instinct MI325X 256 GB · 6,000 GB/s · Q8_0 8,435 tok/s
- 09 Radeon Instinct MI300A 128 GB · 5,325 GB/s · Q8_0 7,486 tok/s
- 10 Radeon Instinct MI300X 192 GB · 5,325 GB/s · Q8_0 7,486 tok/s
The smallest GPUs that still run ALBERT-xxlarge
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 1.0 GB · Q8_0 · comfortable 173 tok/s
- 02 RTX A400 4 GB · needs 1.0 GB · Q8_0 · comfortable 173 tok/s
- 03 RTX 500 Mobile Ada Generation 4 GB · needs 1.0 GB · Q8_0 · comfortable 231 tok/s
- 04 GeForce RTX 3050 A Mobile 4 GB · needs 1.0 GB · Q8_0 · comfortable 346 tok/s
- 05 Jetson Orin Nano 4 GB 4 GB · needs 1.0 GB · Q8_0 · comfortable 61.5 tok/s
- 06 Radeon RX 6450M 4 GB · needs 1.0 GB · Q8_0 · comfortable 180 tok/s
- 07 Radeon RX 6550M 4 GB · needs 1.0 GB · Q8_0 · comfortable 202 tok/s
- 08 Radeon RX 6550S 4 GB · needs 1.0 GB · Q8_0 · comfortable 180 tok/s
- 09 Arc A310 4 GB · needs 1.0 GB · Q8_0 · comfortable 145 tok/s
- 10 Arc Pro A30M 4 GB · needs 1.0 GB · Q8_0 · comfortable 150 tok/s
What the numbers mean
What you need to run it
Minimum card
Tesla C1080
Memory needed
1.0 GB
Fastest
14,418 tok/s
ALBERT-xxlarge is small enough at 235M parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.
The least hardware that works is a Tesla C1080. Its 4 GB is enough at Q8_0 compression, giving roughly 157 tokens per second.
A B200 is the fastest we calculate for it: about 14,418 tokens per second, from 8,000 GB/s of memory bandwidth.
About this model
ALBERT-xxlarge was published by Toyota Technological Institute at Chicago,Google, in United States of America, in February 2020. academia,Industry is the category the publisher falls under.
It works in Language, and is recorded as doing language modeling/generation, Question answering.
The weights are published, so it can be downloaded and run on your own hardware indefinitely, offline, with no account attached.
How fast it runs, and why
Across every card that can run it, the middle of the range is about 404.9 tokens per second, and 818 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.
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.
How it was trained
Producing it required around 2.4 × 10²¹ FLOP of arithmetic, on Google TPU v3, which is a statement about the training budget rather than about inference.
Around 3,300,000,000 tokens went into training it.
Its inclusion criterion is highly cited.
Step by step
How to choose a GPU for ALBERT-xxlarge
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 ALBERT-xxlarge actually needs — around 1.0 GB at Q8_0. No amount of processing power compensates for a card that cannot hold it.
-
02
Match the context to your actual use
Set the context to what you will actually use. The cache grows with the conversation, and it is the usual reason ALBERT-xxlarge stops fitting a card that seemed fine.
-
03
Decide how much compression you will accept
Compression is what makes ALBERT-xxlarge 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
Sort by speed
Ranking by tokens per second for ALBERT-xxlarge follows memory bandwidth, not core counts, which is why the B200 tops it at 14,418 tok/s.
-
05
Look at the headroom, not just the fit
Tight means ALBERT-xxlarge loads and works, with no room to raise the context later. Comfortable means you can. The difference matters more than a few tokens per second.
-
06
Check the card from the other side
Each card page repeats this sweep for every model we hold. It answers what else the hardware is good for, beyond ALBERT-xxlarge.
Answers
ALBERT-xxlarge — common questions
Who created ALBERT-xxlarge?
ALBERT-xxlarge was published by Toyota Technological Institute at Chicago,Google, based in United States of America, categorised as academia,Industry.
When was ALBERT-xxlarge released?
ALBERT-xxlarge was published in February 2020. 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 ALBERT-xxlarge used for?
ALBERT-xxlarge works in Language, and is recorded as handling language modeling/generation, Question answering. 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.
Where can I download ALBERT-xxlarge?
The weights for ALBERT-xxlarge 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 ALBERT-xxlarge?
Around 2.4 × 10²¹ 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 ALBERT-xxlarge 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 ALBERT-xxlarge assume it is fully resident.
Would two GPUs run ALBERT-xxlarge faster?
Capacity adds across cards; throughput does not. Since 818 of the cards we track already hold ALBERT-xxlarge on their own, a second card is rarely the answer here.
Why does the quantisation differ between cards for ALBERT-xxlarge?
Because capacity varies, so does how hard ALBERT-xxlarge has to be squeezed — 1 distinct levels appear in the table above. Set a minimum quality to compare at one.
How accurate are these ALBERT-xxlarge speed estimates?
These are estimates with real error bars. The fastest result here, 8,651–23,069 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 ALBERT-xxlarge?
The smallest card in our catalogue that holds ALBERT-xxlarge is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 1.0 GB, and produces roughly 157 tokens per second. 818 cards in total can run it.
How fast is ALBERT-xxlarge on a GPU?
It depends on the card. The quickest we calculate is a B200 at about 14,418 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 ALBERT-xxlarge clear that.
How much VRAM does ALBERT-xxlarge need?
About 1.0 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 ALBERT-xxlarge on a 8 GB GPU?
Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 1.0 GB and generating roughly 2,685 tokens per second — a comfortable fit.
Can I run ALBERT-xxlarge on a 12 GB GPU?
Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 1.0 GB and generating roughly 1,644 tokens per second — a comfortable fit.
Can I run ALBERT-xxlarge on a 16 GB GPU?
Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 1.0 GB and generating roughly 2,037 tokens per second — a comfortable fit.
Can I run ALBERT-xxlarge on a 24 GB GPU?
Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 1.0 GB and generating roughly 2,415 tokens per second — a comfortable fit.
Is ALBERT-xxlarge open source?
Its weights are published, so ALBERT-xxlarge 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 ALBERT-xxlarge have?
ALBERT-xxlarge has 235M parameters. 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.
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.