Referenced here. Baren Plateau - Thesis survey and brief validation by derwind<a href="https://blueqat.com/derwind/30f1495c-3a96-473e-b88e-a6b6cb6b7200 " rel="noopener noreferrer" target="_blank">https://blueqat.com/derwind/30f1495c-3a96-473e-b88e-a6b6cb6b7200 </a> Now, cuQuantum, NVIDIA&#39;s top-powered quantum computer simulator. It can perform the same calculations as a quantum computer on a GPU. Of course, it can be used as a hybrid or instead of a hybrid, as HPC-quantum hybrids are starting to be developed these days. This software is the spearhead of this trend. It is available free of charge. It was recently the fastest in the world. NVIDIA&#39;s world&#39;s fastest quantum circuit simulator! Running on the AIST ABCI supercomputer in Japan!<a href="https://blueqat.com/yuichiro_minato2/6816e9bf-a86b-430c-a48c-a4bacaa1fcac " rel="noopener noreferrer" target="_blank">https://blueqat.com/yuichiro_minato2/6816e9bf-a86b-430c-a48c-a4bacaa1fcac </a> I promote it simply because I am a fan of NVIDIA&#39;s cuQuantum. This time, the python version of the free jupyter environment called Google Colab has been updated to 3.8, so you can now use cuQuantum easily.I was able to use it because derwind has forked blueqat and written a gate set that supports cuQuantum. <pre class="ql-syntax" spellcheck="false">!python -V
</pre> Python 3.8.16. This works! The first things to look out for are. 1. change the runtime type to GPU. The first thing to notice is. This time we will use GPU, but I don&#39;t think it is set to GPU by default, so let&#39;s set it.Let&#39;s install it as soon as possible. All you need is cuquantum-python and blueqat&#39;s cuquantum-enabled backend. <pre class="ql-syntax" spellcheck="false">!pip install cuquantum-python git+https://github.com/derwind/Blueqat.git@cuquantum-0.1
</pre> Now restart the runtime once to reflect the installation. Once restarted, we can immediately perform the calculations. <pre class="ql-syntax" spellcheck="false">from blueqat import Circuit
Circuit(25).h[:].x[:].run(backend=&#34;cusv&#34;)
</pre> array([0.00017263+0.j, 0.00017263+0.j, 0.00017263+0.j, ... ,0.00017263+0.j, 0.00017263+0.j, 0.00017263+0.j], dtype=complex64) That&#39;s easy! The calculations were done using a state vector simulator, so all the results of this calculation are in state vectors. The state vector is limited by the amount of memory on the machine, so you cannot calculate or display more than the amount of memory on the GPU and CPU, so try to look at the limits and do the calculations. if you increase the GPU in clusters, you can calculate up to 40 or 41 qubits, but I think it is too hard with Google Colab. I think it&#39;s tough with Google Colab. I was able to calculate up to about 28 qubits. <pre class="ql-syntax" spellcheck="false">Circuit(28).h[:].x[:].run(backend=&#34;cusv&#34;)
</pre> This time I used the state vector cuStateVec. There is also another cuTensorNet included, so I will introduce that next time. That&#39;s all for this time.

Hearing that the GPU quantum computer simulator NVIDIA cuQuantum is now available on Google Colab (this time for state vector execution).

Yuichiro Minato