RTX Spark
An RTX Spark runs 34 of the 34 open models tracked here at Q4_K_M with 8k context, computed on 16 October 2026. The largest is Mistral Small 4 119B (MoE), needing 72.5 GB and generating about 56 tokens per second.
- Memory
- 128 GB unified
- For the GPU
- 102 GB
- Bandwidth
- 300 GB/s
- Released
- 2026
- Availability
- Sold new
Open the RTX Spark in the LLM checker
What an RTX Spark runs, model by model
RTX Spark (128 GB) addresses 102 GB of its 128 GB of unified memory, worked out from what NVIDIA documents (see the sources below), at 300 GB/s. Every model below is at Q4_K_M with 8k of context, the setting most people actually use. Once the 5% safety margin comes off, the card has 96.9 GB to spend: memory decides what loads, bandwidth how fast it answers. Both are estimates from stated formulas, and the VRAM calculator shows the working.
On this card a model fits at or under 96.9 GB, and above 86.7 GB it is tight: it loads, but there is little left for the conversation, and a desktop on the same 128 GB can push it over. Tokens per second are single-stream estimates from memory bandwidth; ≤ marks a ceiling real runtimes fall well below.
Computed 16 October 2026The model to actually run on it
Nemotron 3.5 Bolt 74B-A9B (MoE) is the best use of this card: 45.9 GB of the 102 GB available, an estimated 40 tokens per second, comfortable for chat, and 131,072 tokens of context still available. It is a mixture-of-experts model, so all 74.9B parameters sit in memory but only 9B are read per token, which is why it is quick for its size.
How this is chosen, since no benchmark is quoted: models are compared by size, a mixture-of-experts model counting at the geometric mean of its total and active parameters (a rule of thumb, not a measurement). The pick is the biggest class that fits with memory to spare and answers at 25 tokens per second or better, and within that class the most recent general-purpose release; coding and reasoning specialists are listed above but not recommended by default.
Mistral Small 4 119B (MoE) is the largest model the card will hold, at 72.5 GB, with room for 131,072 tokens of context. It is a mixture-of-experts model reading only 6.5B parameters per token, so it is the faster of the two at an estimated 56 tokens per second; what it gives up is the depth of a dense model that reads all of its weights for every token.
If quality matters more than parameter count, Llama 3.3 70B fits at Q8_0 in about 81 GB. Q8 is near-lossless; a smaller model at Q8 often beats a larger one squeezed into Q3.
How much context actually fits
Model size is the question people ask; context is the one that bites. The KV cache grows linearly with the conversation, so a model that loads comfortably can still run out of memory halfway through a long document. On 102 GB, the context you get is whatever the weights leave behind, which is why two cards that run the same model can feel completely different in use. These are the largest contexts this card holds, at Q4_K_M weights.
| Model | Max context, FP16 KV | With Q8 KV cache |
|---|---|---|
| Nemotron 3.5 Bolt 74B-A9B (MoE) | 131,072 tokens | 131,072 tokens |
| Llama 3.3 70B | 131,072 tokens | 131,072 tokens |
| Qwen3 32B | 40,960 tokens · model limit | 40,960 tokens · model limit |
| Gemma 4 31B | 131,072 tokens | 131,072 tokens |
Every model in this table already reaches its limit at FP16, so quantising the KV cache buys nothing here. Capped at 128k tokens, or at the model's own native window where that is smaller, marked "model limit": free memory beyond that point buys nothing. Rows that hold far more context than their size suggests are hybrid, sliding-window or latent-attention models, which cache only a fraction of what a standard transformer does; each model page shows the working.
Where this card stops
Nothing in this list is out of reach. Every model tracked here fits at Q4_K_M with 8k of context, so the ceiling you will meet is context length and batch size, not parameter count.
Buy or rent
Buy one as a CUDA laptop that also holds large models, not for speed: a desktop RTX 3090 generates faster on anything that fits in 24 GB. Check the memory on the exact model, since the 128 GB configuration is the one that matters here. If the large models are an occasional need, renting a big card for those hours costs less than a machine that runs them slowly every day.
Retail prices drift, and a figure written today would be wrong within a month, so this page does not carry one. Put the figure you are actually looking at, with your electricity rate and your real monthly hours, into the build vs rent calculator. It returns the month owning becomes cheaper, or tells you it never does. Hours per month decides it far more often than the price does.
The honest take
NVIDIA's chip for Windows laptops and mini PCs: up to 128 GB shared with the GPU, read at 300 GB/s, with CUDA.
The DGX Spark idea in a Windows laptop: a Blackwell GPU and an Arm CPU sharing up to 128 GB, with CUDA running natively. NVIDIA's developer guide gives the memory 300 GB/s on a 256-bit bus, a little above the DGX Spark's 273 GB/s and about half an M5 Max's 614 GB/s, so large models fit and generate slowly; mixture-of-experts models, which read only a few billion parameters per token, are what it suits. How much of the 128 GB the GPU gets depends on a reserved block NVIDIA does not size, so this site plans around the 102 GB its rule guarantees.
What it is good at
- Up to 128 GB shared with the GPU in a laptop, and CUDA runs natively, so the NVIDIA software stack works as it does on a desktop card.
- 300 GB/s, a little above the DGX Spark's 273 GB/s on the same 256-bit width.
- Blackwell generation, so FP4 weight formats run natively.
What it is not
- 300 GB/s is about half an M5 Max and under a third of an RTX 4090: dense models above 30B generate slowly however well they fit.
- Windows 11 on Arm: x86 applications run under emulation, and NVIDIA's guide says native Arm builds perform better.
- Only the top chip reaches 128 GB; the version with the smaller GPU stops at 64 GB, and many laptops ship with less.
The thing people get wrong: Windows counts only the GPU's reserved block as dedicated memory, and NVIDIA's guide says NVML's memory query reports only that block, not the shared region the GPU also allocates from. A 128 GB machine can then look as if its GPU has far less memory than a model can actually use.
Compared with the alternatives
- DGX Spark: 126 GB · 273 GB/s · best fit Nemotron 3.5 Bolt 74B-A9B (MoE) at ~37 tok/s. NVIDIA's small Linux box with 128 GB the GPU can share: large models fit, but they are read at about a quarter of an RTX 4090's speed.
- MacBook Pro M5 Max: 96 GB · 614 GB/s · best fit Nemotron 3.5 Bolt 74B-A9B (MoE) at ~82 tok/s. Apple's 128 GB laptop: large models fit and are read at 614 GB/s, if you buy the 40-core GPU.
- Ryzen AI Max+ 395: 96 GB · 256 GB/s · best fit Nemotron 3.5 Bolt 74B-A9B (MoE) at ~34 tok/s. AMD's 128 GB laptop and mini-PC chip: up to 96 GB for the GPU on Windows, and an x86 route to large mixture-of-experts models.
How it stacks up against the other personal AI computers, from the Ghost Core to an RTX PRO 6000 workstation: personal AI computers compared.
All 39 cards are compared side by side on the GPU index, and the same numbers from the model's point of view are on each model fit table. To check one specific pairing rather than read a table, use the can-I-run-it checker.
Questions
- Can an RTX Spark run a 70B model?
- Yes. Llama 3.3 70B at Q4_K_M with 8k of context needs about 45.8 GB, inside the 102 GB this card can address, and it generates an estimated 5 tokens per second, which is slow; fine for batch jobs.
- What is the best model to run on an RTX Spark?
- Nemotron 3.5 Bolt 74B-A9B (MoE). At Q4_K_M and 8k context it needs about 45.9 GB of the 102 GB available and generates an estimated 40 tokens per second, which is comfortable for chat. If quality matters more than size, Llama 3.3 70B fits at Q8 in about 81 GB.
- How many tokens per second does an RTX Spark generate?
- It depends entirely on the size of the model, because generating a token means reading every active weight out of memory. At 300 GB/s and 70% efficiency, this card produces an estimated 45 tokens per second on an 8B model at Q4, and about 56 on the largest model it holds, Mistral Small 4 119B (MoE). Reading speed is roughly 5 to 8 words per second, so anything above 25 feels immediate. Treat 45 as a ceiling on 300 GB/s rather than a measurement; the higher a figure, the further real runtimes fall below it.
- Is 102 GB enough for running LLMs locally?
- It runs 34 of the 34 models tracked here at Q4_K_M with 8k of context, up to 119B parameters. The honest test is not the model list but the context: Nemotron 3.5 Bolt 74B-A9B (MoE) on this card holds about 131,072 tokens before memory runs out.
- RTX Spark or DGX Spark for local models?
- The DGX Spark holds more: 126 GB against 102 GB, but both run all 34 of these models at Q4_K_M with 8k of context, so the extra memory goes to longer context and higher precision. On speed, this card is faster: 300 GB/s against 273 GB/s, and bandwidth is what sets chat speed.
- What is the RTX Spark's memory bandwidth?
- 300 GB/s, from NVIDIA's RTX Spark porting guide: a 256-bit memory interface shared by the CPU and the GPU. NVIDIA's product pages print no bandwidth, and the larger figure in its launch blog describes the link between the CPU and the GPU, not the memory.
- RTX Spark or DGX Spark for local LLMs?
- Both share up to 128 GB with a Blackwell GPU and run CUDA. The RTX Spark reads memory a little faster, 300 GB/s against 273 GB/s, runs Windows 11 and comes in laptops. The DGX Spark runs Linux and gives its GPU more of the pool: 126 GB against the 102 GB this site plans for the RTX Spark. For a portable CUDA machine, the RTX Spark; for the most room for one model, the DGX Spark.
- Memory: weights (parameters × bytes per parameter) + KV cache (2 × layers × KV heads × head dim × context × bytes for a standard transformer; sliding-window, hybrid and latent-attention models counted as they actually cache) + 0.5 GB runtime + 4% of weights. Full derivation in the VRAM calculator.
- Speed: 70% × bandwidth ÷ active weight bytes, single stream, no batching, prompt processing excluded. A ceiling, not a measurement: the 7 of 34 figures marked ≤ are where real runtimes fall furthest below the number, because so few weights are read per token. See the tokens-per-second estimator.
- Card specification (manufacturer figures): 128 GB, 300 GB/s; see NVIDIA's own page for this machine. The bandwidth figure: NVIDIA's RTX Spark porting guide. Unified memory: NVIDIA reserves a block of memory for the GPU and lets it share 50 to 80% of the rest under Windows, but does not print the size of the reserved block. Whatever that size, the GPU can address at least 102 GB of a 128 GB machine, and figures here use 102 GB. Source: NVIDIA's RTX Spark unified-memory guide.
- Model architecture values come from each model's published config; the self-hosted LLM guide explains what each one changes.