Calculate the TPS of the Switch GPU 16nm on local AI models

NVIDIA 4 GB DDR4 26 GB/s August 2019

Every model in our catalogue assessed against this card at the context length and minimum quality you choose. Speed is an estimate for a single request, calculated from this card's memory bandwidth and the size of each model once compressed.

Calculated for this card

97 models it can run

679 models in our catalogue altogether

Largest model it holds

DeciLM 6B

5.7B · Q3_K_M · 4.4 tok/s

Fastest model

Gemma 3 QAT 1B

9.2 tok/s · 1B

Which AI models can run on a Switch GPU 16nm?

Set the inputs, read the answer

More context means more memory for the conversation cache. Speed is for a fresh conversation and does not change with this setting.

Hides models that would only fit by being compressed below this point.

97 models match

Calculating
Quantisation Fit
9.2 tok/s

3–18 · low confidence

Gemma 3 1B 1B Mar 2025 1.8 GB 33k tokens Q8_0 Comfortable
9.2 tok/s

3–18 · low confidence

Gemma 3 QAT 1B 1B Apr 2025 1.8 GB 33k tokens Q8_0 Comfortable
9.2 tok/s

3–18 · low confidence

HGRN 1B (WT 103) ≈ 1B Nov 2023 1.8 GB 131k tokens ? Q8_0 Comfortable
9.2 tok/s

3–18 · low confidence

LLama 3..2 Typhoon 2 1B ≈ 1B Dec 2024 1.8 GB 131k tokens ? Q8_0 Comfortable
9.2 tok/s

3–18 · low confidence

OLMo-1B ≈ 1B Feb 2024 1.8 GB 131k tokens ? Q8_0 Comfortable
9.2 tok/s

3–18 · low confidence

Pythia-1b ≈ 1B Apr 2023 1.8 GB 131k tokens ? Q8_0 Comfortable
8.5 tok/s

3–17 · low confidence

OpenELM-1.1B ≈ 1.1B May 2024 1.9 GB 131k tokens ? Q8_0 Comfortable
8.4 tok/s

3–17 · low confidence

DeciCoder-1B ≈ 1.1B Aug 2023 1.9 GB 131k tokens ? Q8_0 Comfortable
8.4 tok/s

3–17 · low confidence

SantaCoder ≈ 1.1B Jan 2023 1.9 GB 131k tokens ? Q8_0 Comfortable
8.4 tok/s

3–17 · low confidence

TinyLlama-1.1B (1T token checkpoint) ≈ 1.1B Oct 2023 1.9 GB 131k tokens ? Q8_0 Comfortable
8.4 tok/s

3–17 · low confidence

TinyLlama-1.1B (3T token checkpoint) ≈ 1.1B Oct 2023 1.9 GB 131k tokens ? Q8_0 Comfortable
7.7 tok/s

3–15 · low confidence

EXAONE 4.0 (1.2B) ≈ 1.2B Jul 2025 2.0 GB 131k tokens ? Q8_0 Comfortable
7.7 tok/s

3–15 · low confidence

MinerU2.5 ≈ 1.2B Sep 2025 2.0 GB 131k tokens ? Q8_0 Comfortable
7.7 tok/s

3–15 · low confidence

Pleias 1.0 1.2B ≈ 1.2B Dec 2024 2.0 GB 131k tokens ? Q8_0 Comfortable
7.7 tok/s

3–15 · low confidence

Pleias-RAG-1B ≈ 1.2B Apr 2025 2.0 GB 131k tokens ? Q8_0 Comfortable
7.5 tok/s

3–15 · low confidence

Llama 3.2 1B 1.2B Sep 2024 2.2 GB 54k tokens Q8_0 Comfortable
7.4 tok/s

3–15 · low confidence

MiniCPM-1.2B ≈ 1.2B Jun 2024 2.0 GB 131k tokens ? Q8_0 Comfortable
7.1 tok/s

2–14 · low confidence

DeepSeek Coder 1.3B ≈ 1.3B Jan 2024 2.1 GB 131k tokens ? Q8_0 Comfortable
7.1 tok/s

2–14 · low confidence

DeepSeek-VL-1.3B ≈ 1.3B Mar 2024 2.1 GB 131k tokens ? Q8_0 Comfortable
7.1 tok/s

2–14 · low confidence

DigiRL ≈ 1.3B Jun 2024 2.1 GB 131k tokens ? Q8_0 Comfortable
7.1 tok/s

2–14 · low confidence

GLA Transformer 1.3B ≈ 1.3B Aug 2024 2.1 GB 131k tokens ? Q8_0 Comfortable
7.1 tok/s

2–14 · low confidence

Janus 1.3B ≈ 1.3B Oct 2024 2.1 GB 131k tokens ? Q8_0 Comfortable
7.1 tok/s

2–14 · low confidence

Kosmos-2.5 ≈ 1.3B Aug 2024 2.1 GB 131k tokens ? Q8_0 Comfortable
7.1 tok/s

2–14 · low confidence

Otter ≈ 1.3B May 2023 2.1 GB 131k tokens ? Q8_0 Comfortable
7.1 tok/s

2–14 · low confidence

Phi-1 ≈ 1.3B Oct 2023 2.1 GB 131k tokens ? 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

Switch GPU 16nm full specification

Everything on record for this board, ordered by how much it bears on running a language model rather than by how a spec sheet would list it. Memory comes first because it decides the outcome; the rest is context.

Memory

The two specifications that decide what this card can run and how quickly. Capacity sets which models fit; bandwidth sets how many tokens per second they produce once they do.

Memory size
4 GB
Memory bandwidth
26 GB/s
Memory type
DDR4
Memory bus width
64 bit
Memory clock
1.6 GHz

The chip

Which processor is on the board and how it was manufactured. A smaller process size generally means more performance for the same power.

Graphics processor
GM20B
Architecture
Maxwell 2.0
Generation
Console GPU(Nintendo)
Foundry
TSMC
Process size
16 nm
Transistors
2 billion
Transistor density
20,000 K/mm²
Die size
100 mm²
Released
16 August 2019

Clock speeds

How fast the processor runs. Worth far less here than on a gaming benchmark: generating text is limited by memory bandwidth, so a higher clock barely moves the result.

Base clock
384 MHz
Boost clock
768 MHz

Processing units

What the chip contains. These drive graphics performance and matter mainly for processing a long prompt rather than for producing the answer.

Shading units
256
Texture mapping units
16
Render output units
16
Streaming multiprocessors
2

Theoretical performance

Peak arithmetic rates published for the board. These are ceilings that no real workload reaches, and generating text reaches a small fraction of them because it is limited by memory rather than arithmetic.

Half precision (FP16)
786.4 GFLOPS
Single precision (FP32)
393.2 GFLOPS
Double precision (FP64)
12.3 GFLOPS
Pixel rate
12 GPixel/s
Texture rate
12 GTexel/s

The board

What it takes to physically install and power the card — the practical constraints that decide whether it fits the machine you already own.

Power draw (TDP)
15 W
Dimensions
239 mm × 28 mm
Display outputs
1x USB Type-C

Software support

Which graphics and compute interfaces the card supports. CUDA compute capability is the one that bears on inference: below 7.0 there are no tensor cores, and modern inference software falls back to slower code paths.

CUDA compute capability
5.3
DirectX
12.1
OpenGL
4.6
Vulkan
1.4
OpenCL
1.2
Shader model
6.0

Listings

Where to buy a Switch GPU 16nm

No vendor is currently listing this card. Listings come from vendors who publish them here directly — browse the vendor directory to see who is selling what.

What the numbers mean

Why memory is the number that matters here

Memory

4 GB

Bandwidth

26 GB/s

Largest model

DeciLM 6B

At 4 GB of DDR4 the Switch GPU 16nm is limited to the smaller end of the catalogue. About 3.6 GB is actually available to a runtime, and a model has to fit entirely inside it before generating anything at all.

At 26 GB/s across a 64-bit bus, bandwidth is this card's real constraint. Every token requires reading the entire model out of memory, so a large model will feel slow here even when it fits.

The figure is the memory clock — 1.6 GHz here — multiplied by the bus width. It is why core counts predict generation speed so poorly.

In practice that combination tops out at DeciLM 6B — 5.7B, compressed to Q3_K_M, generating around 4.4 tokens per second.

The chip and how it was built

The Switch GPU 16nm is built on the GM20B graphics processor, using NVIDIA's Maxwell 2.0 architecture, as part of the Console GPU(Nintendo) generation.

The chip is manufactured by TSMC, on a 16 nm process, with a die measuring 100 mm², holding 2 billion transistors. A smaller process generally means more performance for the same power, though for language models it matters far less than the memory subsystem.

It was released in August 2019, roughly 6 years ago. Inference software support tends to follow hardware by a year or two, so a card of this age generally has mature, well-optimised code paths available to it.

Compute throughput, and why it matters less than it looks

FP16

786.4 GFLOPS

FP64

12.3 GFLOPS

On paper the Switch GPU 16nm reaches 786.4 GFLOPS at half precision and 393.2 GFLOPS at single precision. These are peak figures no real workload sustains, and generating text reaches only a small fraction of them — decoding is limited by memory rather than arithmetic, which is why a card can look enormously powerful here and still produce tokens at an ordinary rate.

Double-precision throughput is 12.3 GFLOPS. It has no bearing on running a language model — no inference runtime uses it — but it separates datacentre parts from consumer ones, since the latter deliberately restrict it.

Clocks run from 384 MHz at base to 768 MHz boosted. Worth far less here than on a gaming benchmark: raising the clock speeds up the arithmetic, and the arithmetic is not what generation is waiting on.

Cache and processing units

There are 256 shading units, 16 texture mapping units, and 16 render output units. These drive graphics workloads and contribute to prompt processing, but they sit idle for much of the time a model spends generating a reply.

Power, size and installation

Power draw

15 W

The Switch GPU 16nm is rated at 15 W. Running a language model keeps a card busy in bursts rather than continuously — it draws hard while generating and idles between requests — so sustained draw over a working day is usually well below the rated figure.

measuring 239 mm long. Worth checking against the case and power supply already in the machine, since the largest cards need considerably more of both than a typical desktop provides.

The extremes

The largest AI models that run on a Switch GPU 16nm

The biggest open-weight models that fit on this card, newest first. Each is shown at the best compression the card can hold.

  1. 01 Qwen3.5-4B 4B · Q5_K_M · Feb 2026 4.1 tok/s
  2. 02 Voxtral Mini 4.7B · Q4_K_M · Jul 2025 4.5 tok/s
  3. 03 Typhoon 2.1 Gemma 4B 4B · Q4_K_M · May 2025 5.3 tok/s
  4. 04 Qwen3-4B 4B · Q3_K_M · Apr 2025 6.2 tok/s
  5. 05 Gemma 3 QAT 4B 4B · Q4_K_M · Apr 2025 5.3 tok/s
  6. 06 Gemma 3 4B 4B · Q4_K_M · Mar 2025 5.3 tok/s
  7. 07 Phi-4-Multimodal 5.6B · Q3_K_M · Mar 2025 4.4 tok/s
  8. 08 Minitron 4B 4.2B · Q4_K_M · Nov 2024 5.1 tok/s
  9. 09 XVERSE-MoE-A4.2B 4.2B · Q4_K_M · Apr 2024 5.1 tok/s
  10. 10 DeciLM 6B 5.7B · Q3_K_M · Sep 2023 4.4 tok/s

The fastest AI models on a Switch GPU 16nm

Where this card produces tokens quickest. Smaller models dominate here, because generating each token means reading the whole model out of memory once.

  1. 01 Gemma 3 QAT 1B 1B · Q8_0 · 1.8 GB 9.2 tok/s
  2. 02 Gemma 3 1B 1B · Q8_0 · 1.8 GB 9.2 tok/s
  3. 03 LLama 3..2 Typhoon 2 1B 1B · Q8_0 · 1.8 GB 9.2 tok/s
  4. 04 OLMo-1B 1B · Q8_0 · 1.8 GB 9.2 tok/s
  5. 05 HGRN 1B (WT 103) 1B · Q8_0 · 1.8 GB 9.2 tok/s
  6. 06 Pythia-1b 1B · Q8_0 · 1.8 GB 9.2 tok/s
  7. 07 OpenELM-1.1B 1.1B · Q8_0 · 1.9 GB 8.5 tok/s
  8. 08 TinyLlama-1.1B (1T token checkpoint) 1.1B · Q8_0 · 1.9 GB 8.4 tok/s
  9. 09 TinyLlama-1.1B (3T token checkpoint) 1.1B · Q8_0 · 1.9 GB 8.4 tok/s
  10. 10 DeciCoder-1B 1.1B · Q8_0 · 1.9 GB 8.4 tok/s

Step by step

How to work out the tokens per second of a Switch GPU 16nm

You do not have to calculate anything by hand — the gputps.com calculator on this page has already worked it out for every model this card can hold. Reading off the answer takes six steps.

  1. 01

    Find the model in the table

    Every one of the 97 models this Switch GPU 16nm runs is in the table above. Search narrows it by name or by size.

  2. 02

    Set the context length you will actually use

    Longer conversations cost memory on top of the weights. With 4 GB to work in, that is frequently the difference between a model fitting and not.

  3. 03

    Choose how far you will compress

    By default the table picks the least-compressed copy that fits. Setting a floor removes models that only qualify through heavy compression.

  4. 04

    Look at the range, not just the number

    Each speed is an estimate for a single conversation, with a range beneath it — 9.2 tok/s on Gemma 3 QAT 1B at the top end here. The same card and model vary by thirty to fifty per cent between inference runtimes.

  5. 05

    Check the headroom before you decide

    A tight fit runs but leaves no room to raise the context later; comfortable has headroom. The memory column shows what each model needs against the 4 GB available.

  6. 06

    Cross-check against other hardware

    Each model page repeats this calculation for the whole catalogue. Worth a look before deciding: it shows what else runs the same model, and how the Switch GPU 16nm compares.

Answers

Switch GPU 16nm — common questions

01

What are the TFLOPS of a Switch GPU 16nm?

The Switch GPU 16nm is rated at 786.4 GFLOPS at half precision and 393.2 GFLOPS at single precision. These are peak arithmetic ceilings rather than achievable rates, and text generation reaches only a small fraction of them because it is limited by memory bandwidth instead.

02

Does the Switch GPU 16nm support CUDA?

Yes. The Switch GPU 16nm reports CUDA compute capability 5.3, which predates tensor cores. Capability 7.0 and above has tensor cores, which modern inference software uses; below that it falls back to slower code paths for quantised models.

03

Is the Switch GPU 16nm good for running local AI models?

Its memory limits it to smaller models though its bandwidth means generation will feel slow on larger models. In total it runs 97 of the models we track. Whether that is enough depends entirely on which model you want — the table above answers that directly.

04

Can a Switch GPU 16nm run a model that does not fit in its memory?

It can be split, with the overflow held in system memory — but that part drags the whole thing down, and none of the 4 GB figures on this page assume it.

05

Would two Switch GPU 16nm cards be twice as fast?

Pairing Switch GPU 16nm cards buys headroom rather than pace: 8 GB of combined memory, at roughly the same generation speed as one.

06

What AI models can a Switch GPU 16nm run?

97 of the 679 open-weight language models we track fit on a Switch GPU 16nm and can be run locally on it. The table on this page lists every one, with the memory it needs, the quantisation it runs at and an estimated generation speed.

07

What is the largest AI model a Switch GPU 16nm can run?

The largest model in our catalogue that fits on a Switch GPU 16nm is DeciLM 6B at 5.7B parameters, compressed to Q3_K_M. It generates roughly 4.4 tokens per second and needs about 3.5 GB of the card's memory.

08

How many tokens per second does a Switch GPU 16nm produce?

It depends on the model. On a Switch GPU 16nm the fastest model we track is Gemma 3 QAT 1B at about 9.2 tokens per second, while larger models run proportionally slower because each token requires reading the whole model out of memory once. Speeds are estimates for a single conversation at a time.

09

How much memory does a Switch GPU 16nm have?

A Switch GPU 16nm has 4 GB of DDR4 memory. Around a tenth of that is reserved by the inference runtime and the driver, leaving roughly 3.6 GB available for a model and its conversation.

10

What is the memory bandwidth of a Switch GPU 16nm?

The Switch GPU 16nm has 26 GB/s of memory bandwidth, across a 64-bit memory bus. This is the single best predictor of how fast it generates text, because producing each token means reading the entire model out of memory once.

11

What type of memory does a Switch GPU 16nm use?

It uses DDR4 clocked at 1.6 GHz. HBM types are found on datacentre accelerators and carry far more bandwidth than the GDDR used on desktop cards, which is why they generate tokens considerably faster at the same capacity.

12

Who makes the Switch GPU 16nm?

The Switch GPU 16nm is a NVIDIA product, with the chip manufactured by TSMC, on a 16 nm process.

13

When was the Switch GPU 16nm released?

The Switch GPU 16nm was released in August 2019.

14

How much power does a Switch GPU 16nm use?

The Switch GPU 16nm has a rated board power of 15 W. Generating text draws hard in bursts and idles between requests, so average consumption over a working session is normally well below the rated figure.

The other direction

Looking at it from the other side?

This page starts from the hardware. If you already know which model you want and need to know what it takes to run it, start from the model instead.

All GPUs