▸ MEASURED RIG REPORT$25PRO · 11 MODELS

1x NVIDIA L4

11 models measured22 GB VRAMllama.cpp b101562026-08-252.8 h on the machine

Before you spend $2,500 on this card, find out if it does what you need.

11 models loaded and timed on this machine, in one session, on one llama.cpp build. Read a whole one first if you like: the RTX 4090 and RTX 3090 reports are open in full.

One payment for this report, not a subscription. It stays yours, and it does not expire. You will be asked to sign in first, so the report stays attached to your account.

§ WHAT THIS REPORT SETTLES

Does Mistral Small 3.2 24B Instruct run on it?: It did, at Q4_K_M — 14.1 GB of 22 GB at peak
And how fast, model by model?: Decode, prefill and TTFT p50/p95 on 11 models
Which quantisation is worth running?: 2 models run across several rungs; Q4_K_M, Q3_K_M, Q5_K_M, Q6_K, Q8_0 measured on this card
And when a model is bigger than the card?: tok/s at each resident fraction, 8 models spilled into system RAM
How many people can use it at once?: 1 → 8 streams: aggregate tok/s, per-stream tok/s and TTFT p95
Does the quantisation cost answers?: Probed at temperature 0 on 11 models, scored per category
What does it cost to keep running?: Watts at the card, and dollars per million tokens, single stream and batched

§ PREVIEW · THE FASTEST 5MEASURED, NOT ESTIMATED

gpt-oss 20B · Q4_K_M92.0

Qwen3 4B Instruct 2507 · Q4_K_M83.3

Llama 3.1 8B Instruct · Q4_K_M50.3

Qwen3 8B · Q4_K_M48.9

Qwen3.5 9B · Q4_K_M43.5

DECODE TOK/S, SINGLE STREAM

Decode only, single stream, for the fastest 5. The remaining 6 models, and every other measurement, are listed below.

What is in this report

MODELS RUN ON THIS MACHINE11 MEASURED · 6 FAILED

· Llama 3.1 8B Instruct · Q4_K_M
· Qwen3 4B Instruct 2507 · Q4_K_M
· Qwen3 8B · Q4_K_M
· Qwen3.5 9B · Q4_K_M
· Gemma 3 12B Instruct · Q4_K_M
· Qwen3 14B · Q4_K_M
· gpt-oss 20B · Q4_K_M
· Mistral Small 3.2 24B Instruct · Q4_K_M
× Gemma 3 27B Instruct · Q4_K_M
× Qwen3 30B A3B Instruct 2507 · Q4_K_M
× GLM 4.7 Flash · Q4_K_M
× Qwen3 8B (Q3_K_M)
× Qwen3 8B (Q5_K_M)
× Qwen3 8B (Q6_K)
· Qwen3 8B (Q8_0)
· Gemma 3 12B Instruct (Q5_K_M)
· Gemma 3 12B Instruct (Q8_0)

Rows marked × failed to load or to run on this card. They are in the report with the stage they failed at, because a model that does not run is a result.

EVERY FIELD, MEASUREMENT BY MEASUREMENT

FOR EVERY MODEL

· Decode, single stream, tok/s
· Prefill, tok/s
· TTFT p50 and p95, ms
· Peak VRAM reported by the driver
· Weight size on disk, and load time

QUANTISATION

· Rungs measured: Q4_K_M, Q3_K_M, Q5_K_M, Q6_K, Q8_0
· 2 models run across several rungs on this card

OFFLOAD TO SYSTEM RAM

· 8 models run at several resident fractions
· tok/s at each fraction, against the fully resident rate

CONCURRENCY

· Streams: 1, 2, 4, 8
· Aggregate tok/s, per-stream tok/s and TTFT p95 at each level
· 11 models load-tested

ACCURACY AFTER QUANTISATION

· 11 models probed at temperature 0
· Score per category, and a separate looping check

POWER AND COST

· Watts sampled at the card for 11 models
· Cost per million tokens, single stream and batched

FITTED ACROSS THE RUN

· The card's decode constants: fixed ms per token, and implied bandwidth
· Fixed VRAM overhead before the weights — the intercept a fit calculator assumes is zero
· TTFT p95 against concurrent streams
· Each with its n and its correlation; a weak fit is quoted without a line

PROVENANCE

· GPU: NVIDIA L4
· llama.cpp build b10156
· Driver 570.195.03
· Run finished 2026-08-25
· 2.8 hours on the machine

WHAT YOU GET

· This page, in full, at this URL
· A PDF with the same tables and charts
· The underlying JSON, through the API, once the report is yours
· No expiry, and no subscription to keep reading it

Measured on a rented machine, not estimated and not supplied by a vendor. How the runs are taken, and what each figure does and does not cover, is on the methodology page.