Calculate the TPS of the PG506-232 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
Mixtral 8x7B
46.7B · Q3_K_M · 82.7 tok/s
Fastest model
Gemma 3 QAT 1B
395 tok/s · 1B
Which AI models can run on a PG506-232?
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.
502 models match
Calculating| Quantisation | Fit | ||||||
|---|---|---|---|---|---|---|---|
|
395
tok/s
336–474 |
Gemma 3 1B | 1B | Mar 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
395
tok/s
336–474 |
Gemma 3 QAT 1B | 1B | Apr 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
395
tok/s
237–632 · low confidence |
HGRN 1B (WT 103) ≈ | 1B | Nov 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
395
tok/s
237–632 · low confidence |
LLama 3..2 Typhoon 2 1B ≈ | 1B | Dec 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
395
tok/s
237–632 · low confidence |
OLMo-1B ≈ | 1B | Feb 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
395
tok/s
237–632 · low confidence |
Pythia-1b ≈ | 1B | Apr 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
366
tok/s
220–585 · low confidence |
OpenELM-1.1B ≈ | 1.1B | May 2024 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
359
tok/s
216–575 · low confidence |
DeciCoder-1B ≈ | 1.1B | Aug 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
359
tok/s
216–575 · low confidence |
SantaCoder ≈ | 1.1B | Jan 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
359
tok/s
216–575 · low confidence |
TinyLlama-1.1B (1T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
359
tok/s
216–575 · low confidence |
TinyLlama-1.1B (3T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
329
tok/s
198–527 · low confidence |
EXAONE 4.0 (1.2B) ≈ | 1.2B | Jul 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
329
tok/s
198–527 · low confidence |
MinerU2.5 ≈ | 1.2B | Sep 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
329
tok/s
198–527 · low confidence |
Pleias 1.0 1.2B ≈ | 1.2B | Dec 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
329
tok/s
198–527 · low confidence |
Pleias-RAG-1B ≈ | 1.2B | Apr 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
321
tok/s
273–386 |
Llama 3.2 1B | 1.2B | Sep 2024 | 2.2 GB | 131k tokens | Q8_0 | Comfortable |
|
317
tok/s
190–507 · low confidence |
MiniCPM-1.2B ≈ | 1.2B | Jun 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
304
tok/s
182–486 · low confidence |
DeepSeek Coder 1.3B ≈ | 1.3B | Jan 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
304
tok/s
182–486 · low confidence |
DeepSeek-VL-1.3B ≈ | 1.3B | Mar 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
304
tok/s
182–486 · low confidence |
DigiRL ≈ | 1.3B | Jun 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
304
tok/s
182–486 · low confidence |
GLA Transformer 1.3B ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
304
tok/s
182–486 · low confidence |
Janus 1.3B ≈ | 1.3B | Oct 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
304
tok/s
182–486 · low confidence |
Kosmos-2.5 ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
304
tok/s
182–486 · low confidence |
Otter ≈ | 1.3B | May 2023 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
304
tok/s
182–486 · 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
PG506-232 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
- 24 GB
- Memory bandwidth
- 933 GB/s
- Memory type
- HBM2
- Memory bus width
- 3,072 bit
- Memory clock
- 1.22 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
- GA100
- Architecture
- Ampere
- Generation
- Server Ampere(Axx)
- Foundry
- TSMC
- Process size
- 7 nm
- Transistors
- 54.2 billion
- Transistor density
- 65,600 K/mm²
- Die size
- 826 mm²
- Package
- BGA-2743
- 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
- 930 MHz
- Boost clock
- 1.44 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
- 3,584
- Texture mapping units
- 224
- Render output units
- 96
- Streaming multiprocessors
- 56
- Tensor cores
- 224
- L1 cache
- 192 KB
- L2 cache
- 24 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)
- 10.3 TFLOPS
- Single precision (FP32)
- 10.3 TFLOPS
- Double precision (FP64)
- 5.2 TFLOPS
- Pixel rate
- 138 GPixel/s
- Texture rate
- 323 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)
- 165 W
- Suggested power supply
- 450 W
- Power connectors
- 8-pin EPS
- Bus interface
- PCIe 4.0 x16
- 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
- 8.0
- OpenCL
- 3.0
Listings
Where to buy a PG506-232
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
24 GB
Bandwidth
933 GB/s
Largest model
Mixtral 8x7B
The PG506-232 carries 24 GB of HBM2, which covers the mid-sized models most people actually run — about 21.6 GB of it after the runtime and driver reserve their working space.
Bandwidth is 933 GB/s across a 3,072-bit bus. Generating a token means reading every weight once, so that figure sets the pace more than any other number here, and at this level text arrives faster than most people read.
That comes from a 1.22 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.
In practice that combination tops out at Mixtral 8x7B — 46.7B, compressed to Q3_K_M, generating around 82.7 tokens per second.
The chip and how it was built
The PG506-232 is built on the GA100 graphics processor, using NVIDIA's Ampere architecture, as part of the Server Ampere(Axx) generation.
The chip is manufactured by TSMC, on a 7 nm process, with a die measuring 826 mm², holding 54.2 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 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
10.3 TFLOPS
FP64
5.2 TFLOPS
Tensor cores
224
On paper the PG506-232 reaches 10.3 TFLOPS at half precision and 10.3 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 5.2 TFLOPS. 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 224 tensor cores across 56 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 930 MHz at base to 1.44 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 PG506-232 has 192 KB of L1 cache, backed by 24 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 3,584 shading units, 224 texture mapping units, and 96 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
165 W
The PG506-232 is rated at 165 W, with a 450 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 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 PG506-232
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 PG506-232
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 PG506-232
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 502 models the PG506-232 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
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 24 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
Read the speed and the range
The figures are calculated, not measured. 395 tok/s on Gemma 3 QAT 1B is the fastest result on this card, and like every row it carries a range that reflects how much the runtime matters.
-
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 24 GB available.
-
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 PG506-232 compares.
Answers
PG506-232 — common questions
Would two PG506-232 cards be twice as fast?
Pairing PG506-232 cards buys headroom rather than pace: 48 GB of combined memory, at roughly the same generation speed as one.
What AI models can a PG506-232 run?
502 of the 679 open-weight language models we track fit on a PG506-232 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 PG506-232 can run?
The largest model in our catalogue that fits on a PG506-232 is Mixtral 8x7B at 46.7B parameters, compressed to Q3_K_M. It generates roughly 82.7 tokens per second and needs about 21.0 GB of the card's memory.
How many tokens per second does a PG506-232 produce?
It depends on the model. On a PG506-232 the fastest model we track is Gemma 3 QAT 1B at about 395 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 PG506-232 run a 7B model?
Yes. For example a PG506-232 runs Multi-Token Prediction 7B at Q8_0, using about 7.9 GB of memory and generating around 59.0 tokens per second.
Can a PG506-232 run a 13B model?
Yes. For example a PG506-232 runs DeepSeekMoE-16B at Q8_0, using about 17.5 GB of memory and generating around 137 tokens per second.
Can a PG506-232 run a 30B model?
Yes. For example a PG506-232 runs Nemotron 3-Nano-30B-A3B at Q4_K_M, using about 18.1 GB of memory and generating around 160 tokens per second.
How much memory does a PG506-232 have?
A PG506-232 has 24 GB of HBM2 memory. Around a tenth of that is reserved by the inference runtime and the driver, leaving roughly 21.6 GB available for a model and its conversation.
What is the memory bandwidth of a PG506-232?
The PG506-232 has 933 GB/s of memory bandwidth, across a 3,072-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 PG506-232 use?
It uses HBM2 clocked at 1.22 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 PG506-232?
The PG506-232 is a NVIDIA product, with the chip manufactured by TSMC, on a 7 nm process.
When was the PG506-232 released?
The PG506-232 was released in April 2021.
How much power does a PG506-232 use?
The PG506-232 has a rated board power of 165 W, and a 450 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 PG506-232 have?
The PG506-232 has 192 KB of L1 cache, and 24 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 PG506-232?
The PG506-232 is rated at 10.3 TFLOPS at half precision and 10.3 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 PG506-232 have?
The PG506-232 has 224 tensor cores across 56 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 PG506-232 support CUDA?
Yes. The PG506-232 reports CUDA compute capability 8.0. 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 PG506-232 use?
It uses PCIe 4.0 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 PG506-232 good for running local AI models?
Its memory comfortably covers the mid-sized models most people run locally and its bandwidth is high enough to generate text faster than most people read. In total it runs 502 of the models we track. Whether that is enough depends entirely on which model you want — the table above answers that directly.
Can a PG506-232 run a model that does not fit in its memory?
Only partly. Layers beyond the 24 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.
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.