Calculate the TPS of the H100 SXM5 96 GB 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
Largest model it holds
dots.llm1
142B · IQ4_XS · 24.6 tok/s
Fastest model
Gemma 3 QAT 1B
1,423 tok/s · 1B
What AI models can a H100 SXM5 96 GB 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.
609 models match
Calculating| Quantisation | Fit | ||||||
|---|---|---|---|---|---|---|---|
|
1,423
tok/s
1,210–1,708 |
Gemma 3 1B | 1B | Mar 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
1,423
tok/s
1,210–1,708 |
Gemma 3 QAT 1B | 1B | Apr 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
1,423
tok/s
854–2,277 · low confidence |
HGRN 1B (WT 103) ≈ | 1B | Nov 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,423
tok/s
854–2,277 · low confidence |
LLama 3..2 Typhoon 2 1B ≈ | 1B | Dec 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,423
tok/s
854–2,277 · low confidence |
OLMo-1B ≈ | 1B | Feb 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,423
tok/s
854–2,277 · low confidence |
Pythia-1b ≈ | 1B | Apr 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,318
tok/s
791–2,108 · low confidence |
OpenELM-1.1B ≈ | 1.1B | May 2024 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,294
tok/s
776–2,070 · low confidence |
DeciCoder-1B ≈ | 1.1B | Aug 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,294
tok/s
776–2,070 · low confidence |
SantaCoder ≈ | 1.1B | Jan 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,294
tok/s
776–2,070 · low confidence |
TinyLlama-1.1B (1T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,294
tok/s
776–2,070 · low confidence |
TinyLlama-1.1B (3T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,186
tok/s
712–1,897 · low confidence |
EXAONE 4.0 (1.2B) ≈ | 1.2B | Jul 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,186
tok/s
712–1,897 · low confidence |
MinerU2.5 ≈ | 1.2B | Sep 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,186
tok/s
712–1,897 · low confidence |
Pleias 1.0 1.2B ≈ | 1.2B | Dec 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,186
tok/s
712–1,897 · low confidence |
Pleias-RAG-1B ≈ | 1.2B | Apr 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,157
tok/s
983–1,388 |
Llama 3.2 1B | 1.2B | Sep 2024 | 2.2 GB | 131k tokens | Q8_0 | Comfortable |
|
1,141
tok/s
684–1,825 · low confidence |
MiniCPM-1.2B ≈ | 1.2B | Jun 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,095
tok/s
657–1,751 · low confidence |
DeepSeek Coder 1.3B ≈ | 1.3B | Jan 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,095
tok/s
657–1,751 · low confidence |
DeepSeek-VL-1.3B ≈ | 1.3B | Mar 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,095
tok/s
657–1,751 · low confidence |
DigiRL ≈ | 1.3B | Jun 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,095
tok/s
657–1,751 · low confidence |
GLA Transformer 1.3B ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,095
tok/s
657–1,751 · low confidence |
Janus 1.3B ≈ | 1.3B | Oct 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,095
tok/s
657–1,751 · low confidence |
Kosmos-2.5 ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,095
tok/s
657–1,751 · low confidence |
Otter ≈ | 1.3B | May 2023 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
1,095
tok/s
657–1,751 · 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
H100 SXM5 96 GB 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
- 96 GB
- Memory bandwidth
- 3,360 GB/s
- Memory type
- HBM3
- Memory bus width
- 5,120 bit
- Memory clock
- 1.31 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
- GH100
- Architecture
- Hopper
- Generation
- Server Hopper(Hxx)
- Foundry
- TSMC
- Process size
- 5 nm
- Transistors
- 80 billion
- Transistor density
- 98,300 K/mm²
- Die size
- 814 mm²
- Released
- 21 March 2023
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
- 1.35 GHz
- Boost clock
- 1.98 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
- 16,896
- Texture mapping units
- 528
- Render output units
- 24
- Streaming multiprocessors
- 132
- Tensor cores
- 528
- 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)
- 267.6 TFLOPS
- Single precision (FP32)
- 66.9 TFLOPS
- Double precision (FP64)
- 33.5 TFLOPS
- Pixel rate
- 48 GPixel/s
- Texture rate
- 1,045 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)
- 700 W
- Suggested power supply
- 1,100 W
- Power connectors
- 8-pin EPS
- Bus interface
- PCIe 5.0 x16
- Slot width
- SXM Module
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
- 9.0
- OpenCL
- 3.0
Listings
Where to buy a H100 SXM5 96 GB
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
96 GB
Bandwidth
3,360 GB/s
Largest model
dots.llm1
With 96 GB of HBM3, the H100 SXM5 96 GB is in the class of hardware that holds the largest open-weight models without splitting them across machines. Roughly 86.4 GB of that is reachable by an inference runtime once the driver takes its share.
Its 3,360 GB/s across a 5,120-bit bus is at the top of what exists. Since each token means reading the whole model out of memory once, that translates almost directly into generation speed — this card is bandwidth-rich enough that model size stops being the limiting factor long before the bus does.
That comes from a 1.31 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 dots.llm1 — 142B, compressed to IQ4_XS, generating around 24.6 tokens per second.
The chip and how it was built
The H100 SXM5 96 GB is built on the GH100 graphics processor, using NVIDIA's Hopper architecture, as part of the Server Hopper(Hxx) generation.
The chip is manufactured by TSMC, on a 5 nm process, with a die measuring 814 mm², holding 80 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 March 2023, roughly 3 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
267.6 TFLOPS
FP64
33.5 TFLOPS
Tensor cores
528
On paper the H100 SXM5 96 GB reaches 267.6 TFLOPS at half precision and 66.9 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 33.5 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 528 tensor cores across 132 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 1.35 GHz at base to 1.98 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 H100 SXM5 96 GB 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 16,896 shading units, 528 texture mapping units, and 24 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
700 W
The H100 SXM5 96 GB is rated at 700 W, with a 1,100 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 sxm module, 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 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 H100 SXM5 96 GB 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.
The fastest AI models on a H100 SXM5 96 GB
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 H100 SXM5 96 GB
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
All 609 models the H100 SXM5 96 GB handles are already listed. The search box takes a name or a size such as 27b, which matches on parameter count.
-
02
Set the context length you will actually use
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 96 GB.
-
03
Set a minimum quality if you need one
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
Each speed is an estimate for a single conversation, with a range beneath it — 1,423 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
Read the fit verdict last
The fit column separates models that just fit from those with room to spare — worth checking against the card's 96 GB before settling on one.
-
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 H100 SXM5 96 GB compares.
Answers
H100 SXM5 96 GB — common questions
Can a H100 SXM5 96 GB run a 70B model?
Yes. For example a H100 SXM5 96 GB runs Qwen3-Coder-Next at Q8_0, using about 80.7 GB of memory and generating around 98.8 tokens per second.
How much memory does a H100 SXM5 96 GB have?
A H100 SXM5 96 GB has 96 GB of HBM3 memory. Around a tenth of that is reserved by the inference runtime and the driver, leaving roughly 86.4 GB available for a model and its conversation.
What is the memory bandwidth of a H100 SXM5 96 GB?
The H100 SXM5 96 GB has 3,360 GB/s of memory bandwidth, across a 5,120-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 H100 SXM5 96 GB use?
It uses HBM3 clocked at 1.31 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 H100 SXM5 96 GB?
The H100 SXM5 96 GB is a NVIDIA product, with the chip manufactured by TSMC, on a 5 nm process.
When was the H100 SXM5 96 GB released?
The H100 SXM5 96 GB was released in March 2023.
How much power does a H100 SXM5 96 GB use?
The H100 SXM5 96 GB has a rated board power of 700 W, and a 1,100 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 H100 SXM5 96 GB have?
The H100 SXM5 96 GB 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.
What are the TFLOPS of a H100 SXM5 96 GB?
The H100 SXM5 96 GB is rated at 267.6 TFLOPS at half precision and 66.9 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 H100 SXM5 96 GB have?
The H100 SXM5 96 GB has 528 tensor cores across 132 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 H100 SXM5 96 GB support CUDA?
Yes. The H100 SXM5 96 GB reports CUDA compute capability 9.0. 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 H100 SXM5 96 GB 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.
Is the H100 SXM5 96 GB good for running local AI models?
Its memory is large enough for models most desktop hardware cannot touch and its bandwidth puts it among the fastest hardware available for generation. In total it runs 609 of the models we track. Whether that is enough depends entirely on which model you want — the table above answers that directly.
Can a H100 SXM5 96 GB 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 96 GB figures on this page assume it.
Would two H100 SXM5 96 GB cards be twice as fast?
No. A second H100 SXM5 96 GB doubles the memory to 192 GB, which lets you hold models neither could hold alone, but generation does not split that way. These figures describe one card.
What AI models can a H100 SXM5 96 GB run?
609 of the 679 open-weight language models we track fit on a H100 SXM5 96 GB 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 H100 SXM5 96 GB can run?
The largest model in our catalogue that fits on a H100 SXM5 96 GB is dots.llm1 at 142B parameters, compressed to IQ4_XS. It generates roughly 24.6 tokens per second and needs about 78.3 GB of the card's memory.
How many tokens per second does a H100 SXM5 96 GB produce?
It depends on the model. On a H100 SXM5 96 GB the fastest model we track is Gemma 3 QAT 1B at about 1,423 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 H100 SXM5 96 GB run a 7B model?
Yes. For example a H100 SXM5 96 GB runs Multi-Token Prediction 7B at Q8_0, using about 7.9 GB of memory and generating around 212 tokens per second.
Can a H100 SXM5 96 GB run a 13B model?
Yes. For example a H100 SXM5 96 GB runs DeepSeekMoE-16B at Q8_0, using about 17.5 GB of memory and generating around 494 tokens per second.
Can a H100 SXM5 96 GB run a 30B model?
Yes. For example a H100 SXM5 96 GB runs ERNIE-4.5-VL-28B-A3B at Q8_0, using about 29.2 GB of memory and generating around 282 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.