Calculate the TPS of the FirePro S9050 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
ERNIE-4.5-21B-A3B
21B · Q3_K_M · 62.3 tok/s
Fastest model
Gemma 3 QAT 1B
87.2 tok/s · 1B
Which AI models can run on a FirePro S9050?
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.
396 models match
Calculating| Quantisation | Fit | ||||||
|---|---|---|---|---|---|---|---|
|
87.2
tok/s
52–140 · low confidence |
Gemma 3 1B | 1B | Mar 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
87.2
tok/s
52–140 · low confidence |
Gemma 3 QAT 1B | 1B | Apr 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
87.2
tok/s
52–140 · low confidence |
HGRN 1B (WT 103) ≈ | 1B | Nov 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
87.2
tok/s
52–140 · low confidence |
LLama 3..2 Typhoon 2 1B ≈ | 1B | Dec 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
87.2
tok/s
52–140 · low confidence |
OLMo-1B ≈ | 1B | Feb 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
87.2
tok/s
52–140 · low confidence |
Pythia-1b ≈ | 1B | Apr 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
80.8
tok/s
48–129 · low confidence |
OpenELM-1.1B ≈ | 1.1B | May 2024 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
79.3
tok/s
48–127 · low confidence |
DeciCoder-1B ≈ | 1.1B | Aug 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
79.3
tok/s
48–127 · low confidence |
SantaCoder ≈ | 1.1B | Jan 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
79.3
tok/s
48–127 · low confidence |
TinyLlama-1.1B (1T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
79.3
tok/s
48–127 · low confidence |
TinyLlama-1.1B (3T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
72.7
tok/s
44–116 · low confidence |
EXAONE 4.0 (1.2B) ≈ | 1.2B | Jul 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
72.7
tok/s
44–116 · low confidence |
MinerU2.5 ≈ | 1.2B | Sep 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
72.7
tok/s
44–116 · low confidence |
Pleias 1.0 1.2B ≈ | 1.2B | Dec 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
72.7
tok/s
44–116 · low confidence |
Pleias-RAG-1B ≈ | 1.2B | Apr 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
70.9
tok/s
43–113 · low confidence |
Llama 3.2 1B | 1.2B | Sep 2024 | 2.2 GB | 131k tokens | Q8_0 | Comfortable |
|
69.9
tok/s
42–112 · low confidence |
DeepSeekMoE-16B | 16B | Jan 2024 | 10.0 GB | 4k tokens | Q4_K_M | Tight |
|
69.9
tok/s
42–112 · low confidence |
MiniCPM-1.2B ≈ | 1.2B | Jun 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
67.1
tok/s
40–107 · low confidence |
DeepSeek Coder 1.3B ≈ | 1.3B | Jan 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
67.1
tok/s
40–107 · low confidence |
DeepSeek-VL-1.3B ≈ | 1.3B | Mar 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
67.1
tok/s
40–107 · low confidence |
DigiRL ≈ | 1.3B | Jun 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
67.1
tok/s
40–107 · low confidence |
GLA Transformer 1.3B ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
67.1
tok/s
40–107 · low confidence |
Janus 1.3B ≈ | 1.3B | Oct 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
67.1
tok/s
40–107 · low confidence |
Kosmos-2.5 ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
67.1
tok/s
40–107 · low confidence |
Otter ≈ | 1.3B | May 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
FirePro S9050 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
- 12 GB
- Memory bandwidth
- 264 GB/s
- Memory type
- GDDR5
- Memory bus width
- 384 bit
- Memory clock
- 1.38 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
- Tahiti
- Architecture
- GCN 1.0
- Generation
- FirePro Server(Sx000)
- Foundry
- TSMC
- Process size
- 28 nm
- Transistors
- 4.3 billion
- Transistor density
- 12,300 K/mm²
- Die size
- 352 mm²
- Released
- 7 August 2014
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
- 900 MHz
- Boost clock
- 900 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
- 1,792
- Texture mapping units
- 112
- Render output units
- 32
- L1 cache
- 16 KB
- L2 cache
- 0.75 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.
- Single precision (FP32)
- 3.2 TFLOPS
- Double precision (FP64)
- 806.4 GFLOPS
- Pixel rate
- 29 GPixel/s
- Texture rate
- 101 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)
- 225 W
- Suggested power supply
- 550 W
- Power connectors
- 1x 8-pin
- Bus interface
- PCIe 3.0 x16
- Slot width
- Dual-slot
- Dimensions
- 254 mm
- Display outputs
- 1x 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
- 11.1
- OpenGL
- 4.6
- Vulkan
- 1.2
- OpenCL
- 1.2
- Shader model
- 5.1
Listings
Where to buy a FirePro S9050
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
Memory: the specification that decides everything
Memory
12 GB
Bandwidth
264 GB/s
Largest model
ERNIE-4.5-21B-A3B
12 GB of GDDR5 puts the FirePro S9050 comfortably into small and mid-sized models, with roughly 10.8 GB usable once the driver overhead is taken out. The largest models are out of reach without splitting them.
At 264 GB/s across a 384-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 1.38 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 biggest thing it holds is ERNIE-4.5-21B-A3B (21B) at Q3_K_M compression, for about 62.3 tokens per second.
The chip and how it was built
The FirePro S9050 is built on the Tahiti graphics processor, using AMD's GCN 1.0 architecture, as part of the FirePro Server(Sx000) generation.
The chip is manufactured by TSMC, on a 28 nm process, with a die measuring 352 mm², holding 4.3 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 2014, roughly 11 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
FP64
806.4 GFLOPS
Double-precision throughput is 806.4 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.
Clocks run from 900 MHz at base to 900 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
The FirePro S9050 has 16 KB of L1 cache, backed by 0.75 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 1,792 shading units, 112 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
225 W
The FirePro S9050 is rated at 225 W, with a 550 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 254 mm long, and needs 1x 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 FirePro S9050
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 FirePro S9050
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 FirePro S9050
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
Search for the model you want
All 396 models the FirePro S9050 handles are already listed. The search box takes a name or a size such as 27b, which matches on parameter count.
-
02
Decide how long your conversations run
Drag the slider to the conversation length you plan to work at. The cache grows with the conversation, and on 12 GB it is often what pushes a large model over the edge.
-
03
Choose how far you will compress
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
Look at the range, not just the number
Each speed is an estimate for a single conversation, with a range beneath it — 87.2 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.
-
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 12 GB available.
-
06
Cross-check against other hardware
Following a model through to its own page lists all the hardware that can run it, so you can see where the FirePro S9050 sits against the alternatives.
Answers
FirePro S9050 — common questions
Can a FirePro S9050 run a model that does not fit in its memory?
Offloading past the card's 12 GB is possible and usually a false economy: the system-memory portion is slow enough to dominate the result.
Would two FirePro S9050 cards be twice as fast?
Capacity adds, throughput does not. Two of them give you 24 GB to work with rather than twice the tokens per second — every figure here is for a single FirePro S9050.
What AI models can a FirePro S9050 run?
396 of the 679 open-weight language models we track fit on a FirePro S9050 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 FirePro S9050 can run?
The largest model in our catalogue that fits on a FirePro S9050 is ERNIE-4.5-21B-A3B at 21B parameters, compressed to Q3_K_M. It generates roughly 62.3 tokens per second and needs about 10.1 GB of the card's memory.
How many tokens per second does a FirePro S9050 produce?
It depends on the model. On a FirePro S9050 the fastest model we track is Gemma 3 QAT 1B at about 87.2 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.
Can a FirePro S9050 run a 7B model?
Yes. For example a FirePro S9050 runs DeepSeek Coder 6.7B at Q6_K, using about 9.9 GB of memory and generating around 18.9 tokens per second.
Can a FirePro S9050 run a 13B model?
Yes. For example a FirePro S9050 runs DeepSeekMoE-16B at Q4_K_M, using about 10.0 GB of memory and generating around 69.9 tokens per second.
How much memory does a FirePro S9050 have?
A FirePro S9050 has 12 GB of GDDR5 memory. Around a tenth of that is reserved by the inference runtime and the driver, leaving roughly 10.8 GB available for a model and its conversation.
What is the memory bandwidth of a FirePro S9050?
The FirePro S9050 has 264 GB/s of memory bandwidth, across a 384-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 FirePro S9050 use?
It uses GDDR5 clocked at 1.38 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 FirePro S9050?
The FirePro S9050 is a AMD product, with the chip manufactured by TSMC, on a 28 nm process.
When was the FirePro S9050 released?
The FirePro S9050 was released in August 2014.
How much power does a FirePro S9050 use?
The FirePro S9050 has a rated board power of 225 W, and a 550 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 FirePro S9050 have?
The FirePro S9050 has 16 KB of L1 cache, and 0.75 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.
Does the FirePro S9050 support CUDA?
No. CUDA is NVIDIA-only, and the FirePro S9050 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 FirePro S9050 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.
Is the FirePro S9050 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 396 of the models we track. Whether that is enough depends entirely on which model you want — the table above answers that directly.
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.