v4.0 add GPU memory as SWAP buffer & TCC_NN, Sycamore

As simulating an ever higher number of qubits with an ever increasing depth whilst adding SEPARATION to both QUNIT and QBDT becomes more commonplace optimisations for the use of large amounts of swap are implemented.

For this we use vramfs, bcache and mdadm. As the report on filesystem performance stated a COW filesystem such as ZFS carry too big of an overhead to be effective as the typical random IO workload that swap generates incurs too much latency. The caching system of a typical COW is superior to MDADM and therefore the need for a low overhead, low latency COW filesystem remains. In the mean time we use MDADM.

Also in this release is the output of both the TCC_NN and Sycamore 2022 runs that will be included in the paper that will be released at some point in the future.

v 3.5 Added tnn_d - High Qubit ranges

This halfway point release

• includes sycamore-qrack graph and can handle google data
• adds tnn_d performance benchmarks
• introduces high qubit ranges and benchmarks
• uses /qrack128 that can support up to 1024+ qbits
• some tipsy/bonsai graph improvements

3D QFT, 28qbits@14depth sycamore

Added:

• 3D QFT bonsai view as default
• minor parameter changes
• supremacy 28qbits@14depth

Ubuntu 20.04 - CUDAGL v11.4.1

Added:

• Bonsai precooked/realtime visualisation

Ubuntu 20.04 - CUDAGL v11.4.1

Added:

• Realtime conversion to Tipsy format
• Bonsai visualisation

ThereminQ Stack

OpenCL Stack with plugins and ES integration - Docker & K8s Orchestrators are supported - Images are autobuild on https://hub.docker.com/r/twobombs/thereminq. Delivered as-is, zero warranty, academic compliance not guaranteed :)

This work-in-progress is delivered on a best-effort basis.

Todo: AI integration ( see project https://github.com/users/twobombs/projects/2 for progress )