Calculate the TPS of the A16 PCIe on local AI models
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
679 models in our catalogue altogether
Largest model it holds
Nemotron 3-Nano-30B-A3B
31.6B · Q3_K_M · 40.2 tok/s
Fastest model
Gemma 3 QAT 1B
84.8 tok/s · 1B
Which AI models can run on a A16 PCIe?
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.
432 models match
Calculating| Quantisation | Fit | ||||||
|---|---|---|---|---|---|---|---|
|
84.8
tok/s
72–102 |
Gemma 3 1B | 1B | Mar 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
84.8
tok/s
72–102 |
Gemma 3 QAT 1B | 1B | Apr 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
84.8
tok/s
51–136 · low confidence |
HGRN 1B (WT 103) ≈ | 1B | Nov 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
84.8
tok/s
51–136 · low confidence |
LLama 3..2 Typhoon 2 1B ≈ | 1B | Dec 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
84.8
tok/s
51–136 · low confidence |
OLMo-1B ≈ | 1B | Feb 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
84.8
tok/s
51–136 · low confidence |
Pythia-1b ≈ | 1B | Apr 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
78.5
tok/s
47–126 · low confidence |
OpenELM-1.1B ≈ | 1.1B | May 2024 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
77.0
tok/s
46–123 · low confidence |
DeciCoder-1B ≈ | 1.1B | Aug 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
77.0
tok/s
46–123 · low confidence |
SantaCoder ≈ | 1.1B | Jan 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
77.0
tok/s
46–123 · low confidence |
TinyLlama-1.1B (1T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
77.0
tok/s
46–123 · low confidence |
TinyLlama-1.1B (3T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
70.6
tok/s
42–113 · low confidence |
EXAONE 4.0 (1.2B) ≈ | 1.2B | Jul 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
70.6
tok/s
42–113 · low confidence |
MinerU2.5 ≈ | 1.2B | Sep 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
70.6
tok/s
42–113 · low confidence |
Pleias 1.0 1.2B ≈ | 1.2B | Dec 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
70.6
tok/s
42–113 · low confidence |
Pleias-RAG-1B ≈ | 1.2B | Apr 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
68.9
tok/s
59–83 |
Llama 3.2 1B | 1.2B | Sep 2024 | 2.2 GB | 131k tokens | Q8_0 | Comfortable |
|
67.9
tok/s
41–109 · low confidence |
MiniCPM-1.2B ≈ | 1.2B | Jun 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
65.2
tok/s
39–104 · low confidence |
DeepSeek Coder 1.3B ≈ | 1.3B | Jan 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
65.2
tok/s
39–104 · low confidence |
DeepSeek-VL-1.3B ≈ | 1.3B | Mar 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
65.2
tok/s
39–104 · low confidence |
DigiRL ≈ | 1.3B | Jun 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
65.2
tok/s
39–104 · low confidence |
GLA Transformer 1.3B ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
65.2
tok/s
39–104 · low confidence |
Janus 1.3B ≈ | 1.3B | Oct 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
65.2
tok/s
39–104 · low confidence |
Kosmos-2.5 ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
65.2
tok/s
39–104 · low confidence |
Otter ≈ | 1.3B | May 2023 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
65.2
tok/s
39–104 · 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
A16 PCIe 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
- 16 GB
- Memory bandwidth
- 200 GB/s
- Memory type
- GDDR6
- Memory bus width
- 128 bit
- Memory clock
- 1.56 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
- GA107
- Architecture
- Ampere
- Generation
- Server Ampere(Axx)
- Foundry
- Samsung
- Process size
- 8 nm
- Transistors
- 8.7 billion
- Transistor density
- 43,500 K/mm²
- Die size
- 200 mm²
- Package
- FCBGA-1358
- Released
- 12 April 2021
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
- 1.31 GHz
- Boost clock
- 1.76 GHz
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
- 1,280
- Texture mapping units
- 40
- Render output units
- 32
- Streaming multiprocessors
- 10
- Tensor cores
- 40
- Ray tracing cores
- 10
- L1 cache
- 128 KB
- L2 cache
- 2 MB
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)
- 4.5 TFLOPS
- Single precision (FP32)
- 4.5 TFLOPS
- Double precision (FP64)
- 140.4 GFLOPS
- Pixel rate
- 56 GPixel/s
- Texture rate
- 70 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)
- 250 W
- Suggested power supply
- 600 W
- Power connectors
- 8-pin EPS
- Bus interface
- PCIe 4.0 x8
- Slot width
- Dual-slot
- Dimensions
- 267 mm
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
- 8.6
- DirectX
- 12.2
- OpenGL
- 4.6
- Vulkan
- 1.4
- OpenCL
- 3.0
- Shader model
- 6.8
Listings
Where to buy a A16 PCIe
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
What the memory subsystem means for AI
Memory
16 GB
Bandwidth
200 GB/s
Largest model
Nemotron 3-Nano-30B-A3B
16 GB of GDDR6 puts the A16 PCIe comfortably into small and mid-sized models, with roughly 14.4 GB usable once the driver overhead is taken out. The largest models are out of reach without splitting them.
At 200 GB/s across a 128-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.
That comes from a 1.56 GHz memory clock across the bus width above. Widening the bus and raising the clock are the two levers a manufacturer has, which is why a card with unremarkable cores can still generate quickly.
The practical ceiling is Nemotron 3-Nano-30B-A3B at 31.6B, held at Q3_K_M and running at roughly 40.2 tokens per second.
The chip and how it was built
The A16 PCIe is built on the GA107 graphics processor, using NVIDIA's Ampere architecture, as part of the Server Ampere(Axx) generation.
The chip is manufactured by Samsung, on a 8 nm process, with a die measuring 200 mm², holding 8.7 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 April 2021, roughly 5 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
4.5 TFLOPS
FP64
140.4 GFLOPS
Tensor cores
40
On paper the A16 PCIe reaches 4.5 TFLOPS at half precision and 4.5 TFLOPS 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 140.4 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.
The card carries 40 tensor cores across 10 streaming multiprocessors. These accelerate the matrix arithmetic at the heart of a transformer, and they are what make prompt processing — reading a long document before answering — dramatically faster than it would otherwise be.
Clocks run from 1.31 GHz at base to 1.76 GHz 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
The A16 PCIe has 128 KB of L1 cache, backed by 2 MB of L2. Cache absorbs a share of the memory traffic that would otherwise hit the main bus, which is the one place on this page where a number other than bandwidth quietly affects generation speed — a large L2 lets more of the working set stay close to the cores.
There are 1,280 shading units, 40 texture mapping units, and 32 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
250 W
The A16 PCIe is rated at 250 W, with a 600 W power supply suggested for the whole system. 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.
The board occupies a dual-slot, measuring 267 mm long, and needs 8-pin EPS. 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.
It connects over PCIe 4.0 x8. The interface governs how quickly a model is loaded from disk into the card, not how fast it runs once there, so a narrower link costs a few seconds at startup and nothing thereafter.
The extremes
The largest AI models that run on a A16 PCIe
The biggest open-weight models that fit on this card, newest first. Each is shown at the best compression the card can hold.
The fastest AI models on a A16 PCIe
Where this card produces tokens quickest. Smaller models dominate here, because generating each token means reading the whole model out of memory once.
Step by step
How to work out the tokens per second of a A16 PCIe
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.
-
01
Start with the model, not the specification
Every one of the 432 models this A16 PCIe runs is in the table above. Search narrows it by name or by size.
-
02
Match the context to your work
Set the context to your real working length. Short questions cost almost nothing; a long document can consume a large share of the card's 16 GB.
-
03
Choose how far you will compress
Compression is what lets bigger models fit. The quality control drops any model that needs more of it than you are willing to give.
-
04
Take the range as the answer
Speeds come with error bars for a reason. The best case here is 84.8 tok/s on Gemma 3 QAT 1B, and which inference software you use moves that by thirty to fifty per cent.
-
05
Read the fit verdict last
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 16 GB available.
-
06
Open the model to compare cards
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 A16 PCIe compares.
Answers
A16 PCIe — common questions
Is the A16 PCIe good for running local AI models?
Its memory covers small and mid-sized models, though the largest are out of reach though its bandwidth means generation will feel slow on larger models. In total it runs 432 of the models we track. Whether that is enough depends entirely on which model you want — the table above answers that directly.
Can a A16 PCIe 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 16 GB figures on this page assume it.
Would two A16 PCIe cards be twice as fast?
Capacity adds, throughput does not. Two of them give you 32 GB to work with rather than twice the tokens per second — every figure here is for a single A16 PCIe.
What AI models can a A16 PCIe run?
432 of the 679 open-weight language models we track fit on a A16 PCIe 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.
What is the largest AI model a A16 PCIe can run?
The largest model in our catalogue that fits on a A16 PCIe is Nemotron 3-Nano-30B-A3B at 31.6B parameters, compressed to Q3_K_M. It generates roughly 40.2 tokens per second and needs about 14.4 GB of the card's memory.
How many tokens per second does a A16 PCIe produce?
It depends on the model. On a A16 PCIe the fastest model we track is Gemma 3 QAT 1B at about 84.8 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.
Can a A16 PCIe run a 7B model?
Yes. For example a A16 PCIe runs Multi-Token Prediction 7B at Q8_0, using about 7.9 GB of memory and generating around 12.7 tokens per second.
Can a A16 PCIe run a 13B model?
Yes. For example a A16 PCIe runs DeepSeekMoE-16B at Q6_K, using about 13.7 GB of memory and generating around 42.8 tokens per second.
Can a A16 PCIe run a 30B model?
Yes. For example a A16 PCIe runs ERNIE-4.5-VL-28B-A3B at Q3_K_M, using about 12.9 GB of memory and generating around 45.4 tokens per second.
How much memory does a A16 PCIe have?
A A16 PCIe has 16 GB of GDDR6 memory. Around a tenth of that is reserved by the inference runtime and the driver, leaving roughly 14.4 GB available for a model and its conversation.
What is the memory bandwidth of a A16 PCIe?
The A16 PCIe has 200 GB/s of memory bandwidth, across a 128-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.
What type of memory does a A16 PCIe use?
It uses GDDR6 clocked at 1.56 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.
Who makes the A16 PCIe?
The A16 PCIe is a NVIDIA product, with the chip manufactured by Samsung, on a 8 nm process.
When was the A16 PCIe released?
The A16 PCIe was released in April 2021.
How much power does a A16 PCIe use?
The A16 PCIe has a rated board power of 250 W, and a 600 W system power supply is suggested. Generating text draws hard in bursts and idles between requests, so average consumption over a working session is normally well below the rated figure.
How much cache does a A16 PCIe have?
The A16 PCIe has 128 KB of L1 cache, and 2 MB of L2 cache. Cache absorbs part of the memory traffic that would otherwise reach the main bus, so a larger L2 gives a modest lift to generation speed beyond what bandwidth alone predicts.
What are the TFLOPS of a A16 PCIe?
The A16 PCIe is rated at 4.5 TFLOPS at half precision and 4.5 TFLOPS 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.
How many tensor cores does a A16 PCIe have?
The A16 PCIe has 40 tensor cores across 10 streaming multiprocessors. They accelerate the matrix arithmetic a transformer is built from, which mainly speeds up processing a long prompt rather than producing the reply.
Does the A16 PCIe support CUDA?
Yes. The A16 PCIe reports CUDA compute capability 8.6. 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.
What bus interface does the A16 PCIe use?
It uses PCIe 4.0 x8. This governs how fast a model is loaded onto the card rather than how fast it runs once loaded, so it costs a few seconds at startup and nothing during generation.
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.