Calculate the TPS of the GeForce RTX 5090 D 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
721 models in our catalogue altogether
Largest model it holds
Phi-3.5-MoE
60.8B · Q3_K_M · 187 tok/s
Fastest model
Gemma 3 QAT 1B
758 tok/s · 1B
Which AI models can run on a GeForce RTX 5090 D?
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.
543 models match
Calculating| Quantisation | Fit | ||||||
|---|---|---|---|---|---|---|---|
|
758
tok/s
644–910 |
Gemma 3 1B | 1B | Mar 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
758
tok/s
644–910 |
Gemma 3 QAT 1B | 1B | Apr 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
758
tok/s
455–1,213 · low confidence |
HGRN 1B (WT 103) ≈ | 1B | Nov 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
758
tok/s
455–1,213 · low confidence |
LLama 3..2 Typhoon 2 1B ≈ | 1B | Dec 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
758
tok/s
455–1,213 · low confidence |
OLMo-1B ≈ | 1B | Feb 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
758
tok/s
455–1,213 · low confidence |
Pythia-1b ≈ | 1B | Apr 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
702
tok/s
421–1,123 · low confidence |
OpenELM-1.1B ≈ | 1.1B | May 2024 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
689
tok/s
414–1,103 · low confidence |
DeciCoder-1B ≈ | 1.1B | Aug 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
689
tok/s
414–1,103 · low confidence |
SantaCoder ≈ | 1.1B | Jan 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
689
tok/s
414–1,103 · low confidence |
TinyLlama-1.1B (1T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
689
tok/s
414–1,103 · low confidence |
TinyLlama-1.1B (3T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
632
tok/s
379–1,011 · low confidence |
EXAONE 4.0 (1.2B) ≈ | 1.2B | Jul 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
632
tok/s
379–1,011 · low confidence |
LFM2-1.2B ≈ | 1.2B | Jul 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
632
tok/s
379–1,011 · low confidence |
MinerU2.5 ≈ | 1.2B | Sep 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
632
tok/s
379–1,011 · low confidence |
Pleias 1.0 1.2B ≈ | 1.2B | Dec 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
632
tok/s
379–1,011 · low confidence |
Pleias-RAG-1B ≈ | 1.2B | Apr 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
616
tok/s
524–740 |
Llama 3.2 1B | 1.2B | Sep 2024 | 2.2 GB | 131k tokens | Q8_0 | Comfortable |
|
608
tok/s
365–972 · low confidence |
MiniCPM-1.2B ≈ | 1.2B | Jun 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
583
tok/s
350–933 · low confidence |
DeepSeek Coder 1.3B ≈ | 1.3B | Jan 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
583
tok/s
350–933 · low confidence |
DeepSeek-VL-1.3B ≈ | 1.3B | Mar 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
583
tok/s
350–933 · low confidence |
DigiRL ≈ | 1.3B | Jun 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
583
tok/s
350–933 · low confidence |
GLA Transformer 1.3B ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
583
tok/s
350–933 · low confidence |
Janus 1.3B ≈ | 1.3B | Oct 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
583
tok/s
350–933 · low confidence |
Kosmos-2.5 ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
583
tok/s
350–933 · 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
GeForce RTX 5090 D 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
- 32 GB
- Memory bandwidth
- 1,790 GB/s
- Memory type
- GDDR7
- Memory bus width
- 512 bit
- Memory clock
- 1.75 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
- GB202
- Architecture
- Blackwell 2.0
- Generation
- GeForce 50
- Foundry
- TSMC
- Process size
- 5 nm
- Transistors
- 92.2 billion
- Transistor density
- 122,900 K/mm²
- Die size
- 750 mm²
- Released
- 30 January 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
- 2.02 GHz
- Boost clock
- 2.41 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
- 21,760
- Texture mapping units
- 680
- Render output units
- 176
- Streaming multiprocessors
- 170
- Tensor cores
- 680
- Ray tracing cores
- 170
- L1 cache
- 128 KB
- L2 cache
- 96 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)
- 104.8 TFLOPS
- Single precision (FP32)
- 104.8 TFLOPS
- Double precision (FP64)
- 1.6 TFLOPS
- Pixel rate
- 424 GPixel/s
- Texture rate
- 1,637 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)
- 575 W
- Suggested power supply
- 950 W
- Power connectors
- 1x 16-pin
- Bus interface
- PCIe 5.0 x16
- Slot width
- Dual-slot
- Dimensions
- 304 mm × 48 mm
- Display outputs
- 1x HDMI 2.1b, 3x DisplayPort 2.1b
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
- 12.0
- DirectX
- 12.2
- OpenGL
- 4.6
- Vulkan
- 1.4
- OpenCL
- 3.0
- Shader model
- 6.8
Listings
Where to buy a GeForce RTX 5090 D
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
32 GB
Bandwidth
1,790 GB/s
Largest model
Phi-3.5-MoE
GeForce RTX 5090 D carries 32 GB of GDDR7. That covers the mid-sized models most people actually run. Once the runtime and driver reserve their working space, roughly this much is left: 28.8 GB.
Memory bandwidth reaches 1,790 GB/s across a bus of 512 bits. 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.
The figure is the bus width multiplied by a memory clock of 1.75 GHz. 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 practical ceiling is Phi-3.5-MoE, 60.8B, compressed to Q3_K_M and generating around 187 tokens per second.
The chip and how it was built
GeForce RTX 5090 D is built on the graphics processor GB202, using the architecture Blackwell 2.0 from NVIDIA, as part of the generation GeForce 50.
The chip is manufactured by TSMC, on a process of 5 nm, with a die measuring 750 mm², holding 92.2 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 January 2025, roughly 1.62072445423 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
104.8 TFLOPS
FP64
1.6 TFLOPS
Tensor cores
680
On paper GeForce RTX 5090 D reaches 104.8 TFLOPS at half precision, and 104.8 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 reaches 1.6 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 680 tensor cores across 170 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 a base of 2.02 GHz to a boost of 2.41 GHz. 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
GeForce RTX 5090 D has an L1 cache of 128 KB, backed by an L2 cache of 96 MB. 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 21,760 shading units, 680 texture mapping units, and 176 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
575 W
GeForce RTX 5090 D is rated at 575 W, and the suggested system power supply is 950 W. 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 dual-slot, measuring 304 mm long, and needs 1x 16-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 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 that run on a GeForce RTX 5090 D
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 GeForce RTX 5090 D
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 GeForce RTX 5090 D
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
The table lists 543 models this card can run. Search by name, or by size — typing 27b matches on the parameter count even when the name never states it.
-
02
Match the context to your work
Set the context to your real working length. Short questions cost almost nothing, but a long document can consume a large share of 32 GB that is frequently the difference between a model fitting and not.
-
03
Set a minimum quality if you need one
By default the table picks the least-compressed copy that fits. Setting a floor removes models that only qualify through heavy compression.
-
04
Take the range as the answer
Speeds come with error bars for a reason. The best case here is 758 tok/s on Gemma 3 QAT 1B. The same card and model vary by thirty to fifty per cent between inference runtimes.
-
05
Read the fit verdict last
Tight means it works today; comfortable means it still works when the conversation grows. Compare what each model needs against an available 32 GB.
-
06
Check the same model from the other side
Following a model through to its own page lists all the hardware that can run it, so you can see how it compares against GeForce RTX 5090 D.
Answers
GeForce RTX 5090 D — common questions
GeForce RTX 5090 D— can it run 13B models?
Yes. For example it runs DeepSeekMoE-16B at Q8_0, using about 17.5 GB of memory and generating around 263 tokens per second.
GeForce RTX 5090 D— can it run 30B models?
Yes. For example it runs ERNIE-4.5-VL-28B-A3B at Q6_K, using about 22.7 GB of memory and generating around 219 tokens per second.
GeForce RTX 5090 D— how much memory does it have?
This card has 32 GB of GDDR7. Around a tenth is reserved by the inference runtime and the driver, leaving roughly 28.8 GB available for a model and its conversation.
GeForce RTX 5090 D— what is its memory bandwidth?
Memory bandwidth reaches 1,790 GB/s across a bus of 512 bits. 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.
GeForce RTX 5090 D— what type of memory does it use?
It uses GDDR7 clocked at 1.75 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.
GeForce RTX 5090 D— who makes it?
This is a product of NVIDIA, with the chip manufactured by TSMC, on a process of 5 nm.
GeForce RTX 5090 D— when was it released?
It was released in January 2025.
GeForce RTX 5090 D— how much power does it use?
Rated board power is 575 W, and the suggested system power supply is 950 W. Generating text draws hard in bursts and idles between requests, so average consumption over a working session is normally well below the rated figure.
GeForce RTX 5090 D— how much cache does it have?
The L1 cache is 128 KB, and the L2 cache is 96 MB. 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.
GeForce RTX 5090 D— what are its TFLOPS?
It is rated at 104.8 TFLOPS at half precision and 104.8 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.
GeForce RTX 5090 D— how many tensor cores does it have?
It has 680 tensor cores across 170 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.
GeForce RTX 5090 D— does it support CUDA?
Yes. It reports CUDA compute capability 12.0. 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.
GeForce RTX 5090 D— what bus interface does it 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.
GeForce RTX 5090 D— is it 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 543 of the models we track. Whether that is enough depends entirely on which model you want — the table above answers that directly.
GeForce RTX 5090 D— can it run a model that does not fit in its memory?
Only partly. Layers beyond the card's 32 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 GeForce RTX 5090 D cards be twice as fast?
No. A second card doubles the memory to 64 GB of combined memory, at roughly the same generation speed as one.
GeForce RTX 5090 D— which AI models can it run?
543 of the 721 open-weight language models we track fit on this card and can be run locally. The table on this page lists every one, with the memory it needs, the quantisation it runs at and an estimated generation speed.
GeForce RTX 5090 D— what is the largest AI model it can run?
The largest model in our catalogue that fits is Phi-3.5-MoE at 60.8B parameters, compressed to Q3_K_M. It generates roughly 187 tokens per second and needs about 27.2 GB of the card's memory.
GeForce RTX 5090 D— how many tokens per second does it produce?
It depends on the model. The fastest model we track here is Gemma 3 QAT 1B at about 758 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.
GeForce RTX 5090 D— can it run 7B models?
Yes. For example it runs Gemma 4 E4B at Q8_0, using about 9.8 GB of memory and generating around 168 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.