Console · Compute · ⌘K

5 solvers · 48 instruments

Ask it in words. Get the arithmetic.

Type a filter, a room, a coil, or a model and a graphics card. The console reads what you wrote, runs the exact function that renders the published page for it, and shows the answer — in your browser, in about a millisecond, with a link to the page and to the endpoint that returns the same numbers as JSON.

The same answers as an API

Worked, not illustrated

Every card below was solved when this page was built.

Not screenshots, not examples of what it might say. The build ran each phrase through the parser and the solver registry and rendered whatever came back — which is also why this page fails to build if one of them ever stops resolving.

will qwen3 30b fit on a 4090

On-device AI · VRAMquantized only

Qwen3 30B-A3B fits GeForce RTX 4090 at Q5_K_M, with 1.85 GiB spare.

Usable VRAM22.08GiB24 GiB nameplate

Weights20.23GiBmeasured · 5.69 bit/weight

Largest context20,168tokcapped by memory

Decode ceiling315.9tok/sbandwidth ÷ bytes read per token

Quants that fit3 of 6

Open the page →/api/v1/llm-vram-fit?gpu=rtx-4090&model=qwen3-30b-a3b

Weight bytes are the MEASURED size of a published GGUF file wherever one exists and params × bits/8 only where none does — the two differ by 33% on sub-1B models and by 89% on natively-quantized releases, so the basis travels with every figure.

lowpass at 1 kHz

DSP · biquadStable at every word length

Low-pass at 1 kHz, Q 0.7071 — -3.01 dB at the corner.

Corner1kHz

Gain at f₀-3.01dB

DC-0.00dB

Nyquist-334.74dB

Pole radius0.91160inside the unit circle

Group delay0.226ms

Open the page →/api/v1/biquad-design?freq=1000&type=lowpass

RBJ Audio EQ Cookbook coefficients solved in float64, with pole and zero radii from the quadratic roots (never from sqrt(|a2|)), the response curve, the real −3 dB crossings, and what 16/24/32-bit word lengths do to stability and in-band error.

12 x 16 x 8 room modes

Acoustics · room modes6 modes stack at 142.19 Hz

12 × 16 × 8 ft holds 175 modes, 18 of them axial.

length fundamental35.17Hz

width fundamental46.89Hz

height fundamental70.33Hz

Widest axial gap35.17Hzbelow 200 Hz

Ratio h : w : l1 : 1.50 : 2.00

Schroeder @ RT60 0.3s166.10HzRT60 assumed, not measured

Schroeder @ RT60 0.6s234.90HzRT60 assumed, not measured

Open the page →/api/v1/room-modes?height=8&length=16&width=12

Rectangular-room eigenmode solution f = (c/2)·sqrt((nx/Lx)² + (ny/Ly)² + (nz/Lz)²) at c = 343 m/s, with axial/tangential/oblique classification, pile-up clustering below 200 Hz, the Bolt-area proportion test (Bolt, 1946, JASA), and Schroeder crossovers at two ASSUMED RT60 values.

best proportions for 2400 cubic feet

Acoustics · room shapeInside the Bolt area

At 2,400 ft³, build 2.2467 : 1.2814 : 1 — 21.14 × 12.06 × 9.412 ft.

Length : width : height2.2467 : 1.2814 : 1

Dimensions21.14 × 12.06 × 9.412ft

Widest axial gap20.05Hzthe quantity being minimised

Penalty vs a full solve4.72%at exactly this volume

Band1,333–2,894ft³worst case 4.72% across the range

Open the page →/api/v1/room-shape-band?volume=2400

The inverse room problem: hold the volume, search the shape.

solenoid 25mm x 100mm 400 turns 2a

Magnetostatics · solenoidClosed form valid at this turn count

25 × 100 mm at 800 amp-turns gives 9.753 mT on axis.

Centre field9.753mT

Per amp-turn12.19µT/At

Field at the mouth51.1%of the centre value

Flat to 1%32.17mm32.2% of the winding

Infinite-solenoid error3.1%how far µ₀nI overstates it

Open the lane →/api/v1/solenoid-field?current=2&diameter=25&length=100&turns=400

Current-sheet solenoid solved in closed form, B(z) = (mu0*n*I/2)*[(L/2-z)/sqrt(R^2+(L/2-z)^2) + (L/2+z)/sqrt(R^2+(L/2+z)^2)], which at the centre reduces to mu0*N*I/sqrt(L^2+D^2).

peaking 2.5k q 1.41 +3db

DSP · biquadStable at every word length

Peaking EQ at 2.5 kHz, Q 1.41 — 3.00 dB at the corner.

Corner2.5kHz

Gain at f₀3.00dB

DC0.00dB

Nyquist0.00dB

Pole radius0.90828inside the unit circle

Group delay0.063ms

Open the lane →/api/v1/biquad-design?freq=2500&gain=3&q=1.41&type=peaking

llama 3.1 8b on an m4 max 32k context

On-device AI · VRAMcomfortable

Llama 3.1 8B Instruct fits Apple M4 Max (128GB) at FP16 / BF16, with 81.04 GiB spare.

Usable VRAM96.00GiB128 GiB nameplate

Weights14.96GiBmeasured · 16 bit/weight

Largest context131,072tokcapped by the model, not the card

Decode ceiling26.8tok/sbandwidth ÷ bytes read per token

Quants that fit6 of 6

Open the lane →/api/v1/llm-vram-fit?context=32768&gpu=apple-m4-max-128gb&model=llama-3-1-8b-instruct

How it works

One registry, six surfaces, no second implementation.

1 · Read

It parses, and shows you the parse

Units, dimension triples, filter names, model and accelerator names. Whatever it understood appears as chips above the answer, with anything it had to assume marked as assumed and anything you did not state marked as a default. A term that names two real devices is offered as two, never silently resolved to one.

2 · Solve

It runs the published function

Not a lookup and not a language model. The parse becomes the same query string/api/v1/* takes, and the same function that renders the published page produces the answer. An audit fails the build if a printed figure is anything other than a value the solver returned.

3 · Point somewhere

Every answer says where it lives

If your inputs land on a published page, the card links to that exact page. If they do not, it says so and links to the lane. Either way it hands you the endpoint that returns the identical numbers as JSON, with a stated method and no key.

The console is not a search box with arithmetic bolted on. It is the same solver registry behind the REST endpoints, the OpenAPI document, the MCP tool set and the agent documents, given a sixth surface — the one a person can type into. Press ⌘K anywhere on this site.