Calculate the TPS of the GB10 on local AI models

NVIDIA 128 GB LPDDR5X 273 GB/s October 2025

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

624 of 679 models it can run

Largest model it holds

Solar Open2 250B

250.3B · Q3_K_M · 6.9 tok/s

Fastest model

Gemma 3 QAT 1B

116 tok/s · 1B

What AI models can a GB10 run?

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.

624 models match

Calculating
Quantisation Fit
116 tok/s

98–139

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

98–139

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

69–185 · low confidence

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

69–185 · low confidence

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

69–185 · low confidence

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

69–185 · low confidence

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

64–171 · low confidence

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

63–168 · low confidence

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

63–168 · low confidence

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

63–168 · low confidence

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

63–168 · low confidence

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

58–154 · low confidence

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

58–154 · low confidence

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

58–154 · low confidence

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

58–154 · low confidence

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

80–113

Llama 3.2 1B 1.2B Sep 2024 2.2 GB 131k tokens Q8_0 Comfortable
92.8 tok/s

56–148 · low confidence

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

53–142 · low confidence

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

53–142 · low confidence

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

53–142 · low confidence

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

53–142 · low confidence

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

53–142 · low confidence

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

53–142 · low confidence

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

53–142 · low confidence

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

53–142 · 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

GB10 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
128 GB
Memory bandwidth
273 GB/s
Memory type
LPDDR5X
Memory bus width
256 bit
Memory clock
1.07 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
GB20B
Architecture
Blackwell 2.0
Generation
Server Blackwell(Bxx)
Foundry
TSMC
Process size
5 nm
Released
15 October 2025

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
1.67 GHz
Boost clock
2.42 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
6,144
Texture mapping units
384
Render output units
48
Streaming multiprocessors
48
Tensor cores
384
Ray tracing cores
48
L1 cache
250 KB
L2 cache
50 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)
29.7 TFLOPS
Single precision (FP32)
29.7 TFLOPS
Double precision (FP64)
464.3 GFLOPS
Pixel rate
116 GPixel/s
Texture rate
929 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)
140 W
Suggested power supply
300 W
Power connectors
None
Bus interface
PCIe 5.0 x16
Slot width
IGP
Dimensions
150 mm × 150 mm
Display outputs
1x HDMI

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
12.1
OpenCL
3.0

Listings

Where to buy a GB10

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

Capacity and bandwidth

Memory

128 GB

Bandwidth

273 GB/s

Largest model

Solar Open2 250B

With 128 GB of LPDDR5X, the GB10 is in the class of hardware that holds the largest open-weight models without splitting them across machines. Roughly 115.2 GB of that is reachable by an inference runtime once the driver takes its share.

At 273 GB/s across a 256-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 1.07 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 Solar Open2 250B (250.3B) at Q3_K_M compression, for about 6.9 tokens per second.

The chip and how it was built

The GB10 is built on the GB20B graphics processor, using NVIDIA's Blackwell 2.0 architecture, as part of the Server Blackwell(Bxx) generation.

The chip is manufactured by TSMC, on a 5 nm process. 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 October 2025. 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

29.7 TFLOPS

FP64

464.3 GFLOPS

Tensor cores

384

On paper the GB10 reaches 29.7 TFLOPS at half precision and 29.7 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 464.3 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.

The card carries 384 tensor cores across 48 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 1.67 GHz at base to 2.42 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 GB10 has 250 KB of L1 cache, backed by 50 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 6,144 shading units, 384 texture mapping units, and 48 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

140 W

The GB10 is rated at 140 W, with a 300 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 igp, measuring 150 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.

It connects over PCIe 5.0 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 a GB10 can run

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 Solar Open2 250B 250.3B · Q3_K_M · Jun 2026 6.9 tok/s
  2. 02 MiniMax-M2.7 229B · Q3_K_M · Mar 2026 1.4 tok/s
  3. 03 MiniMax-M2.5 229B · Q3_K_M · Feb 2026 1.4 tok/s
  4. 04 MiniMax-M2.1 229B · Q3_K_M · Dec 2025 1.4 tok/s
  5. 05 P1-235B-A22B 235B · Q3_K_M · Nov 2025 7.4 tok/s
  6. 06 Qwen3-235B-A22B-Thinking (Jul 2025) 235B · Q3_K_M · Jul 2025 7.4 tok/s
  7. 07 Qwen3-235B-A22B (Jul 2025) 235B · Q3_K_M · Jul 2025 7.4 tok/s
  8. 08 Qwen3-235B-A22B 235B · IQ4_XS · Apr 2025 6.7 tok/s
  9. 09 DeepSeek-V2.5 236B · Q3_K_M · Sep 2024 7.4 tok/s
  10. 10 DeepSeek-V2 (MoE-236B) 236B · Q3_K_M · May 2024 7.4 tok/s

The fastest AI models on a GB10

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 116 tok/s
  2. 02 Gemma 3 1B 1B · Q8_0 · 1.8 GB 116 tok/s
  3. 03 LLama 3..2 Typhoon 2 1B 1B · Q8_0 · 1.8 GB 116 tok/s
  4. 04 OLMo-1B 1B · Q8_0 · 1.8 GB 116 tok/s
  5. 05 HGRN 1B (WT 103) 1B · Q8_0 · 1.8 GB 116 tok/s
  6. 06 Pythia-1b 1B · Q8_0 · 1.8 GB 116 tok/s
  7. 07 OpenELM-1.1B 1.1B · Q8_0 · 1.9 GB 107 tok/s
  8. 08 TinyLlama-1.1B (1T token checkpoint) 1.1B · Q8_0 · 1.9 GB 105 tok/s
  9. 09 TinyLlama-1.1B (3T token checkpoint) 1.1B · Q8_0 · 1.9 GB 105 tok/s
  10. 10 DeciCoder-1B 1.1B · Q8_0 · 1.9 GB 105 tok/s

Step by step

How to work out the tokens per second of a GB10

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

    Start with the model, not the specification

    All 624 models the GB10 handles are already listed. The search box takes a name or a size such as 27b, which matches on parameter count.

  2. 02

    Match the context to your work

    Drag the slider to the conversation length you plan to work at. The cache grows with the conversation, and on 128 GB it is often what pushes a large model over the edge.

  3. 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.

  4. 04

    Look at the range, not just the number

    The figures are calculated, not measured. 116 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.

  5. 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 128 GB before settling on one.

  6. 06

    Open the model to compare cards

    Every model name in the table links to its own page, which runs the same calculation across every card we hold. That is where you see whether the GB10 is the right buy for it or merely a card that fits.

Answers

GB10 — common questions

01

How much memory does a GB10 have?

A GB10 has 128 GB of LPDDR5X memory. Around a tenth of that is reserved by the inference runtime and the driver, leaving roughly 115.2 GB available for a model and its conversation.

02

What is the memory bandwidth of a GB10?

The GB10 has 273 GB/s of memory bandwidth, across a 256-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.

03

What type of memory does a GB10 use?

It uses LPDDR5X clocked at 1.07 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.

04

Who makes the GB10?

The GB10 is a NVIDIA product, with the chip manufactured by TSMC, on a 5 nm process.

05

When was the GB10 released?

The GB10 was released in October 2025.

06

How much power does a GB10 use?

The GB10 has a rated board power of 140 W, and a 300 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.

07

How much cache does a GB10 have?

The GB10 has 250 KB of L1 cache, and 50 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.

08

What are the TFLOPS of a GB10?

The GB10 is rated at 29.7 TFLOPS at half precision and 29.7 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.

09

How many tensor cores does a GB10 have?

The GB10 has 384 tensor cores across 48 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.

10

Does the GB10 support CUDA?

Yes. The GB10 reports CUDA compute capability 12.1. 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.

11

What bus interface does the GB10 use?

It uses PCIe 5.0 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.

12

Is the GB10 good for running local AI models?

Its memory is large enough for models most desktop hardware cannot touch though its bandwidth means generation will feel slow on larger models. In total it runs 624 of the models we track. Whether that is enough depends entirely on which model you want — the table above answers that directly.

13

Can a GB10 run a model that does not fit in its memory?

Offloading past the card's 128 GB is possible and usually a false economy: the system-memory portion is slow enough to dominate the result.

14

Would two GB10 cards be twice as fast?

Pairing GB10 cards buys headroom rather than pace: 256 GB of combined memory, at roughly the same generation speed as one.

15

What AI models can a GB10 run?

624 of the 679 open-weight language models we track fit on a GB10 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.

16

What is the largest AI model a GB10 can run?

The largest model in our catalogue that fits on a GB10 is Solar Open2 250B at 250.3B parameters, compressed to Q3_K_M. It generates roughly 6.9 tokens per second and needs about 106.6 GB of the card's memory.

17

How many tokens per second does a GB10 produce?

It depends on the model. On a GB10 the fastest model we track is Gemma 3 QAT 1B at about 116 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.

18

Can a GB10 run a 7B model?

Yes. For example a GB10 runs Multi-Token Prediction 7B at Q8_0, using about 7.9 GB of memory and generating around 17.3 tokens per second.

19

Can a GB10 run a 13B model?

Yes. For example a GB10 runs DeepSeekMoE-16B at Q8_0, using about 17.5 GB of memory and generating around 40.2 tokens per second.

20

Can a GB10 run a 30B model?

Yes. For example a GB10 runs ERNIE-4.5-VL-28B-A3B at Q8_0, using about 29.2 GB of memory and generating around 23.0 tokens per second.

21

Can a GB10 run a 70B model?

Yes. For example a GB10 runs Qwen3-Coder-Next at Q8_0, using about 80.7 GB of memory and generating around 8.0 tokens per second.

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