1 of 5

DIGITAL CELL TWIN

Week 8

of 11

Human Cell Digital Twin

CUDA installed, design locked in — and the old model's graphs fixed

2 of 5

RECAP

What We Worked On This Week

02

01

Scaled to 800K voxels

The new spatial model now runs on the P100 at roughly 800,000 voxels.

02

Installed CUDA

The GPU compute stack is in place — diffusion now runs in parallel.

03

Fixed the graphs

Decreased the time interval on the old model, resolving the artifacts.

3 of 5

NEW MODEL · SPATIAL DIFFUSION

Scaled Up and Onto CUDA

03

800,000 voxels on the P100

A large jump in spatial resolution, on real GPU hardware.

CUDA installed

Every voxel updates simultaneously ,the point of the spatial approach.

Design changes made

Model restructured around the GPU, first approach.

Env variables & distance defined

Units, spacing and boundaries are now concrete.

SPATIAL RESOLUTION

800,000

voxels, live on the P100

CUDA installed

Design restructured

Env vars + distance defined

4 of 5

OLD MODEL · COMPARTMENTAL

Smaller Time Steps, Cleaner Graphs

04

THE ISSUE

The graphs showed artifacts — the output didn't track the chemistry smoothly.

THE FIX

We decreased the time interval, stepping time forward in smaller increments.

WHY IT WORKS

A smaller Δ means each state is computed from a closer previous state, so the curve resolves instead of overshooting.

t(x) = t(x–1) + Δ smaller Δ → smoother curve

BEFORE → AFTER

large Δ — artifacts

small Δ — clean

Same chemistry. The only difference is the size of the time step.

5 of 5

LOOKING AHEAD

What We Plan to Do

05

Push past 800K voxels

New model — raise resolution further now that CUDA is running.

01

Move P100 → H100

New model — shift onto the H100 for full-scale parallel compute.

02

Add visualization

New model — render the diffusion so the spread is visible.

03

Validate the old model

Old model — confirm the corrected graphs, continue expert review.

04

Week 8 of 11 — the spatial model is real and running. Next is scale and visibility.