Calculate the TPS of the Radeon Pro Duo 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
DeciLM 6B
5.7B · Q3_K_M · 80.1 tok/s
Fastest model
Gemma 3 QAT 1B
169 tok/s · 1B
Which AI models can run on a Radeon Pro Duo?
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 | ||||||
|---|---|---|---|---|---|---|---|
|
169
tok/s
101–271 · low confidence |
Gemma 3 1B | 1B | Mar 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
169
tok/s
101–271 · low confidence |
Gemma 3 QAT 1B | 1B | Apr 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
169
tok/s
101–271 · low confidence |
HGRN 1B (WT 103) ≈ | 1B | Nov 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
169
tok/s
101–271 · low confidence |
LLama 3..2 Typhoon 2 1B ≈ | 1B | Dec 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
169
tok/s
101–271 · low confidence |
OLMo-1B ≈ | 1B | Feb 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
169
tok/s
101–271 · low confidence |
Pythia-1b ≈ | 1B | Apr 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
157
tok/s
94–251 · low confidence |
OpenELM-1.1B ≈ | 1.1B | May 2024 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
154
tok/s
92–246 · low confidence |
DeciCoder-1B ≈ | 1.1B | Aug 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
154
tok/s
92–246 · low confidence |
SantaCoder ≈ | 1.1B | Jan 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
154
tok/s
92–246 · low confidence |
TinyLlama-1.1B (1T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
154
tok/s
92–246 · low confidence |
TinyLlama-1.1B (3T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
141
tok/s
85–226 · low confidence |
EXAONE 4.0 (1.2B) ≈ | 1.2B | Jul 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
141
tok/s
85–226 · low confidence |
MinerU2.5 ≈ | 1.2B | Sep 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
141
tok/s
85–226 · low confidence |
Pleias 1.0 1.2B ≈ | 1.2B | Dec 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
141
tok/s
85–226 · low confidence |
Pleias-RAG-1B ≈ | 1.2B | Apr 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
138
tok/s
83–220 · low confidence |
Llama 3.2 1B | 1.2B | Sep 2024 | 2.2 GB | 54k tokens | Q8_0 | Comfortable |
|
136
tok/s
81–217 · low confidence |
MiniCPM-1.2B ≈ | 1.2B | Jun 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
130
tok/s
78–208 · low confidence |
DeepSeek Coder 1.3B ≈ | 1.3B | Jan 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
130
tok/s
78–208 · low confidence |
DeepSeek-VL-1.3B ≈ | 1.3B | Mar 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
130
tok/s
78–208 · low confidence |
DigiRL ≈ | 1.3B | Jun 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
130
tok/s
78–208 · low confidence |
GLA Transformer 1.3B ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
130
tok/s
78–208 · low confidence |
Janus 1.3B ≈ | 1.3B | Oct 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
130
tok/s
78–208 · low confidence |
Kosmos-2.5 ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
130
tok/s
78–208 · low confidence |
Otter ≈ | 1.3B | May 2023 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
130
tok/s
78–208 · 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
Radeon Pro Duo 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
- 512 GB/s
- Memory type
- HBM
- Memory bus width
- 4,096 bit
- Memory clock
- 500 MHz
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
- Capsaicin
- Architecture
- GCN 3.0
- Generation
- Radeon Pro GCN
- Foundry
- TSMC
- Process size
- 28 nm
- Transistors
- 8.9 billion
- Transistor density
- 14,900 K/mm²
- Die size
- 596 mm²
- Released
- 26 April 2016
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 GHz
- Boost clock
- 1 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
- 4,096
- Texture mapping units
- 256
- Render output units
- 64
- L1 cache
- 16 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)
- 8.2 TFLOPS
- Single precision (FP32)
- 8.2 TFLOPS
- Double precision (FP64)
- 512 GFLOPS
- Pixel rate
- 64 GPixel/s
- Texture rate
- 256 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)
- 350 W
- Suggested power supply
- 750 W
- Power connectors
- 3x 8-pin
- Bus interface
- PCIe 3.0 x16
- Slot width
- Dual-slot
- Dimensions
- 277 mm
- Display outputs
- 1x HDMI 1.4a, 3x DisplayPort 1.2
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.
- DirectX
- 12.0
- OpenGL
- 4.6
- Vulkan
- 1.2
- OpenCL
- 2.1
- Shader model
- 6.5
Listings
Where to buy a Radeon Pro Duo
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
4 GB
Bandwidth
512 GB/s
Largest model
DeciLM 6B
At 4 GB of HBM the Radeon Pro Duo 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.
The memory bus moves 512 GB/s across a 4,096-bit bus. That is the number that governs generation speed — arithmetic per byte read is small enough that the bus, not the cores, is what everything waits on.
The figure is the memory clock — 500 MHz here — multiplied by the bus width. It is why core counts predict generation speed so poorly.
Put together, the largest model that fits is DeciLM 6B at 5.7B, running Q3_K_M and producing around 80.1 tokens per second.
The chip and how it was built
The Radeon Pro Duo is built on the Capsaicin graphics processor, using AMD's GCN 3.0 architecture, as part of the Radeon Pro GCN generation.
The chip is manufactured by TSMC, on a 28 nm process, with a die measuring 596 mm², holding 8.9 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 2016, roughly 10 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
8.2 TFLOPS
FP64
512 GFLOPS
On paper the Radeon Pro Duo reaches 8.2 TFLOPS at half precision and 8.2 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 512 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 1 GHz at base to 1 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 Radeon Pro Duo has 16 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 4,096 shading units, 256 texture mapping units, and 64 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
350 W
The Radeon Pro Duo is rated at 350 W, with a 750 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 277 mm long, and needs 3x 8-pin. 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 3.0 x16. 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 Radeon Pro Duo
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 Radeon Pro Duo
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 Radeon Pro Duo
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
Find the model in the table
All 97 models the Radeon Pro Duo handles are already listed. The search box takes a name or a size such as 27b, which matches on parameter count.
-
02
Match the context to your work
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.
-
03
Pin the comparison to one quality level
Each model is shown at the best compression this card can hold. A minimum quality hides the ones that only fit by being squeezed further than you would accept.
-
04
Take the range as the answer
Speeds come with error bars for a reason. The best case here is 169 tok/s on Gemma 3 QAT 1B, and which inference software you use moves that by thirty to fifty per cent.
-
05
Check the headroom before you decide
The fit column separates models that just fit from those with room to spare — worth checking against the card's 4 GB before settling on one.
-
06
Check the same model from the other side
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 Radeon Pro Duo compares.
Answers
Radeon Pro Duo — common questions
Is the Radeon Pro Duo good for running local AI models?
Its memory limits it to smaller models and its bandwidth gives usable, if unspectacular, generation speeds. 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.
Can a Radeon Pro Duo run a model that does not fit in its memory?
Only partly. Layers beyond the 4 GB sit in system memory and run at a fraction of the speed, so a mostly-offloaded model is rarely worth using. Every figure here assumes it is fully resident on the card.
Would two Radeon Pro Duo cards be twice as fast?
Capacity adds, throughput does not. Two of them give you 8 GB to work with rather than twice the tokens per second — every figure here is for a single Radeon Pro Duo.
What AI models can a Radeon Pro Duo run?
97 of the 679 open-weight language models we track fit on a Radeon Pro Duo 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 Radeon Pro Duo can run?
The largest model in our catalogue that fits on a Radeon Pro Duo is DeciLM 6B at 5.7B parameters, compressed to Q3_K_M. It generates roughly 80.1 tokens per second and needs about 3.5 GB of the card's memory.
How many tokens per second does a Radeon Pro Duo produce?
It depends on the model. On a Radeon Pro Duo the fastest model we track is Gemma 3 QAT 1B at about 169 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.
How much memory does a Radeon Pro Duo have?
A Radeon Pro Duo has 4 GB of HBM 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.
What is the memory bandwidth of a Radeon Pro Duo?
The Radeon Pro Duo has 512 GB/s of memory bandwidth, across a 4,096-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 Radeon Pro Duo use?
It uses HBM clocked at 500 MHz. 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 Radeon Pro Duo?
The Radeon Pro Duo is a AMD product, with the chip manufactured by TSMC, on a 28 nm process.
When was the Radeon Pro Duo released?
The Radeon Pro Duo was released in April 2016.
How much power does a Radeon Pro Duo use?
The Radeon Pro Duo has a rated board power of 350 W, and a 750 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 Radeon Pro Duo have?
The Radeon Pro Duo has 16 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 Radeon Pro Duo?
The Radeon Pro Duo is rated at 8.2 TFLOPS at half precision and 8.2 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.
Does the Radeon Pro Duo support CUDA?
No. CUDA is NVIDIA-only, and the Radeon Pro Duo is a AMD card. It runs language models through ROCm, Vulkan or Metal depending on the software, which are less mature than the CUDA path — our estimates apply a penalty for that.
What bus interface does the Radeon Pro Duo use?
It uses PCIe 3.0 x16. 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.