Live

The machines.

Ten nodes in Los Angeles, four Tesla K80 slots each. This page is live. Dark is rendering for a visitor, violet is rendering for the collection, light is resting.

Node 1
resting
resting
resting
resting
Node 2
resting
resting
resting
resting
Node 3
resting
resting
resting
Node 4
resting
resting
resting
resting
Node 5
loop
loop · step 16
resting
resting
Node 6
resting
resting
resting
resting
Node 7
resting
loop · step 12
resting
Node 8
resting
resting
resting
Node 9
resting
resting
resting
resting
Node 10
resting
loop
resting
resting

0 for visitors · 4 for the collection · 33 resting · 3 off

The hardware.

Name
NVIDIA Tesla K80
Released
November 2014
Inside
Two GK210 chips, 12 GB each. Each half is a slot.
Precision
fp32 only. No half precision, so no shortcuts.
Last supported
CUDA 11, PyTorch 2.0, diffusers 0.14. Frozen there on purpose.
Work rate
2.7 seconds per step, about a minute per frame.
Since
Retired from client work, rehired for this.

When no visitor needs them, 4 machines work through a prompt list: remixes of the collection's prompts, a list we wrote, and prompts visitors let us keep. A curator reviews what comes out and keeps about one in fifty. That is how the collection grows.

This week: 3548 rendered · 1617 kept