We are very happy to announce the release of NxSDK 1.0.0. With this release we have upgraded the OS support to Ubuntu 20.04 and Python support to 3.8.x. Please create a new virtual environment before installing NxSDK 1.0.0 on INRC Cloud. This release is not backward compatible and won't work with 16.04. We are in process of upgrading our Cloud Systems to Ubuntu 20.04 in next 2 weeks so that users can use this release on the INRC Cloud. Users who have their own Ubuntu 20.04 VMs or Kapoho Bay can use this release. Some of the highlights of this release are :
New Features/Improvements
General
lakemont_driver has been re-named to nx
Timestamps for time and energy probes are collected on x86 cores between tStart and tEnd, as indicated in the probe condition, reducing the number of times the x86 and host must sync.
Boot-up times accelerated by initializing only used chips and powering down the unused cores.
Energy probes on Kapoho Bay now include start and end temperatures (all timesteps in between will return NaN to indicate no valid temerature reading)
Fixed a bug where the number of unique axonal delays per core was artificially limited by NxSDK
The location of N2Board has changed. It must now be imported from nxsdk.arch.n2a.n2board, or through the N2A API as nxsdk.api.n2a.NxBoard
The syntax board.n2Chips[].n2Cores[] is still supported for now, but will give a deprecation warning because the syntax is changing to board.nxChips[].nxCores[] to make changing between chip and core versions easier
NxSDK Modules
NxSlayer auto modules: Release of NxSlayer auto modules that enable seamless network creation and execution of SLAYER-Loihi trained models in Loihi hardware. The auto modules are available as nxSlayer.auto.{s2lDataset, Network}
NxSlayer benchmarking module: Utililty to create multiple copies of NxSlayer auto network and replicate it over multiple Loihi chips for accurate energy benchmarking. It is available as nxSlayer.benchmark.MultiChipNetwork