LLMBOTTLENECK.COM
Meta / llama4_text / file from the unsloth mirror

Llama-4-Maverick-17B-128E-Instruct

401.6 billion parameters, routing 1 of 128 experts per token.

Open in the calculator →

Architecturepublished data
Quick answer
Llama-4-Maverick-17B-128E-Instruct needs about 248.9 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. None of the 18 common devices listed below holds it entirely at this setting. No hardware for it? Rent a GPU that runs it, priced live.

Sized at 8,192 tokens of context with the whole model in device memory, across 18 common devices. Speeds are estimates, not benchmarks.

Configuration read from a mirror of a restricted repository

The official repository is meta-llama/Llama-4-Maverick-17B-128E-Instruct (licence: other). Meta's repository is licence-gated. This configuration is read from Unsloth's ungated copy; it is not byte-compared with Meta's file.

Weight-file size by format

FormatSizeRangeBasis
FP16803.5 GB562.2 GB – 811.2 GBsize range
Q8_0427.0 GB298.7 GB – 432.9 GBsize range
Q6_K329.8 GB230.6 GB – 335.2 GBsize range
Q5_K_M286.9 GB193.3 GB – 335.2 GBsize range
Q5_0280.3 GB193.3 GB – 335.2 GBsize range
Q4_K_M246.5 GB158.1 GB – 335.2 GBsize range
Q4_0233.7 GB158.1 GB – 335.2 GBsize range
Q3_K_M200.8 GB120.8 GB – 335.2 GBsize range
Q2_K159.0 GB92.2 GB – 335.2 GBsize range

A file size is not the memory a run needs: the KV cache and the runtime reserve come on top, and the calculator adds both.

What the labels mean
published data
Read from a published source — a file's byte count, a model's configuration or a manufacturer's specification — or exact arithmetic on such values. Not a measurement on a machine.
reconstructed size
Weight size reconstructed from the pinned architecture, because no published file exists.
size range
Only a lower and an upper bound are claimed for this weight size.

Hardware ladder

Q4_K_M at 8,192 tokens · 0 of 18 devices hold it entirely
DeviceMemoryVerdictDecodeCalculator
NVIDIA GeForce RTX 5070248.9 GB of 12 GB47 layers on system RAM~9.3 tok/s with offloadOpen →
AMD Radeon™ RX 9070 XT248.9 GB of 16 GB46 layers on system RAM~9.5 tok/s with offloadOpen →
NVIDIA GeForce RTX 4080248.9 GB of 16 GB46 layers on system RAM~9.5 tok/s with offloadOpen →
NVIDIA GeForce RTX 5080248.9 GB of 16 GB46 layers on system RAM~9.6 tok/s with offloadOpen →
NVIDIA GeForce RTX 5070 Ti248.9 GB of 16 GB46 layers on system RAM~9.5 tok/s with offloadOpen →
NVIDIA GeForce RTX 5060 Ti 16GB248.9 GB of 16 GB46 layers on system RAM~9.3 tok/s with offloadOpen →
NVIDIA GeForce RTX 4090248.9 GB of 24 GB44 layers on system RAM~10.0 tok/s with offloadOpen →
AMD Radeon™ RX 7900 XTX248.9 GB of 24 GB44 layers on system RAM~10.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 3090248.9 GB of 24 GB44 layers on system RAM~9.9 tok/s with offloadOpen →
NVIDIA GeForce RTX 5090248.9 GB of 32 GB43 layers on system RAM~10 tok/s with offloadOpen →
NVIDIA RTX 6000 Ada Generation248.9 GB of 48 GB40 layers on system RAM~11 tok/s with offloadOpen →
Apple M4 Pro248.9 GB of 64 GBdoes not fitnot calibratedOpen →
Apple M5 Pro248.9 GB of 64 GBdoes not fitnot calibratedOpen →
AMD Ryzen AI Max+ 395 with Radeon 8060S248.9 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M3 Max248.9 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M4 Max248.9 GB of 128 GBdoes not fitnot calibratedOpen →
NVIDIA DGX Spark248.9 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M2 Ultra248.9 GB of 192 GBdoes not fitnot calibratedOpen →

Decode is a calibrated estimate from the published calibration; capacity is the manufacturer's published ceiling, not guaranteed free memory. Fitting in memory is not the same as loading: whether the runtime and version you have supports this architecture and format on that machine has not been tested here. Offloaded rows assume there is enough system RAM for the overflow — the calculator checks that against the RAM you declare. Each “Open” link keeps this model, format and context.

Run it in the cloud

Rent a GPU that runs Llama-4-Maverick-17B-128E-Instruct

Every rentable machine that holds the whole model, from Vast.ai and RunPod, with this site's speed estimate for each and the provider's own price, read live. Cheapest first; sort by cost per token or speed instead, or switch the format.

Prices · 18:30 UTC, 22 Sept
7 configurations in stock · 9 out of stock
MachineSpeedPer hourPer M tokensWhereRent
8× L40S384 GBrecommendedcheapestbest value~588 tok/sFaster than you read$3.73$1.76Vast.aiMarketplace · 99.9% reliableRent on Vast.ai
8× RTX PRO 5000384 GB~915 tok/sFaster than you read$6.81$2.06Vast.aiMarketplace · 98.3% reliableRent on Vast.ai
2× H200 NVL282 GB~817 tok/sFaster than you read$7.20$2.45Vast.aiMarketplace · 98.7% reliableRent on Vast.ai
1× B300270 GB~655 tok/sFaster than you read$7.89$3.34RunPodSecure CloudRent on RunPod
4× H100 SXM320 GBfastest~1141 tok/sFaster than you read$9.07$2.21Vast.aiMarketplace · 99.9% reliableRunPod $13.96/h Rent on Vast.ai
2× H200282 GB~817 tok/sFaster than you read$9.18$3.12RunPodSecure CloudVast.ai $10.00/h Rent on RunPod
4× H100 PCIe320 GB~681 tok/sFaster than you read$9.59$3.91Vast.aiMarketplace · 98.3% reliableRent on Vast.ai
No GPU at all

Use Llama-4-Maverick-17B-128E-Instruct by the token

For one person chatting, that is about 2.7× cheaper than the best-value rented card above ($1.76 per million tokens). A rented GPU pays off when you keep it busy — many requests batched together, long agent runs — or when the data must stay on a machine you control, or the exact file you want is not served anywhere.

First time renting a GPU? How it works, in four steps
  1. Create an account and add credit. Both providers are prepaid: $10 is enough to try any card on this page for hours. A new RunPod account that arrives through a referral link gets a one-time $5 credit when it first adds $10.
  2. Pick the card and the count shown here, and a template: llama.cpp or vLLM for an OpenAI-compatible API, or one with a web chat if you only want to talk to the model. “How to launch it” below gives the exact image and command.
  3. Wait for the download. The weights are fetched on the machine; tens of gigabytes take a few minutes on a data-centre connection. You pay by the second from the moment it starts.
  4. Stop it when you are done. A running machine bills even when idle. A stopped one costs nothing per hour, though its disk is usually still billed until you delete it.
How to launch it

No Q4_K_M file of Llama-4-Maverick-17B-128E-Instruct is catalogued here, so there is no exact command to vouch for. Start the machine from the llama.cpp server image ghcr.io/ggml-org/llama.cpp:server-cuda and point -hf at a Q4_K_M build from the publisher or a quantizer you trust — search Hugging Face for one. Check its file size against the memory figure on this page before you rent.

Every machine holds the whole model at Q4_K_M and 8,192 tokens of context, no offload. Speeds are this site’s single-stream decode estimates; prices are what each provider’s own API quoted, on-demand, for the whole machine. Vast hosts below 98% measured reliability are left out.

Referral links Vast.ai, RunPod and Novita pay us a share of what you spend if you sign up through these buttons. It costs you nothing, and it never decides an order or a recommendation: both are computed from the live price and the speed, and options that pay us nothing are listed and recommended on the same terms. How we rank

Buying a card instead, or paying by the token? Compare a year of Llama-4-Maverick-17B-128E-Instruct three ways →

Under the hood

Architecture, read from the publisher’s file

The numbers every figure above is computed from, with the file they came from.

Architecture

✓ Architecture read from the mirror's config.json

Retrieved 2026-09-20 at pinned commit 86c5ecb6f2fb; parameter count from the safetensors index at the same revision.

Architecture
llama4_text
Layers
48
Hidden size
5,120
Attention heads
40
KV heads
8
Head dimension
128
Feed-forward width
8,192
Vocabulary
202,048
Context ceiling
1,048,576
RoPE theta
500,000
Experts
128
Experts / token
1

Where the memory goes

tokenembedding× 48 decoder blocksGrouped-query attention40 query · 8 KV headsfull contextRouted experts1 of 128 per tokenall residentoutputprojectiongrows with contextcapacity ≠ traffic

Grouped-query attention shares each key/value head across 5 query heads, so the KV cache is 20% of what multi-head attention would need at the same context.

This is a multimodal checkpoint (vision); the diagram and memory sizing above cover the text decoder. Encoder/audio/image activations are not included in the KV or weight figures.

Open the exact config.json this page was read from (the mirror) ↗

More from Meta

Llama-3.2-1B-InstructLlama-3.2-3B-InstructLlama-3.3-70B-InstructLlama-4-Scout-17B-16E-InstructMeta-Llama-3.1-70B-Instruct

Use this answer in your own product

As JSON, from the API
curl -s https://llmbottleneck.com/v1/analyze \
  -H "Authorization: Bearer $LLMB_KEY" -H "content-type: application/json" \
  -d '{"model":"unsloth-llama-4-maverick-17b-128e-instruct","quantization":"Q4_K_M","context":8192,"hardware":"nvidia-geforce-rtx-4090"}'

Same engine, same evidence, every field sourced. Free key in one step →

On your page, as a widget
<script src="https://llmbottleneck.com/widget.js"
  data-model="unsloth-llama-4-maverick-17b-128e-instruct" data-quantization="Q4_K_M"
  data-hardware="nvidia-geforce-rtx-4090" data-context="8192"></script>

No key needed. Unbranded, with your own buy button, on Pro and Business →

Citing this page

LLM Bottleneck. “Llama-4-Maverick-17B-128E-Instruct VRAM and hardware requirements.” Architecture from unsloth/Llama-4-Maverick-17B-128E-Instruct at revision 86c5ecb6f2fb, retrieved 2026-09-20. https://llmbottleneck.com/models/unsloth-llama-4-maverick-17b-128e-instruct

Every figure above is either the published value or a reconstruction whose measured error is on the accuracy page.

llmbottleneck
catalogue 2026-10-03models 327devices 135