I was reading a LI article (it sounds like a desperate start of the year, I agree) by someone from Natixis, the French bank (please, don't make fun of me),
The author explains:

nous nous basons énormément sur les méthodes de simulations de Monte Carlo. Nous avons une infrastructure qui permet de simuler plus
de 100 000 000 000 de nombres aléatoires à la seconde pour valoriser les portefeuilles et produire les sensibilités qui sont essentielles pour le pilotage des risques.

In short, Natixis infrastructure generates some 100Bln random numbers per seconds for them the compute derivatives sensitivities and assess risk.
My question (non-banker, here) : Is that the usual order of magnitude? what is the usual computational power for that at banks?
I am asking this because 100Bln RV per second seem huge to me (maybe biased by technology available in my home country, proud France).
Apologies if my question is not relevant at all.
 

Ah, 100,000,000 syndrome?

https://forum.wilmott.com/viewtopic.php?f=34&t=102135

BTW where will you store them? Supply and demand?

Like in the old days, Cray computers could crunch data so fast for the piddly VAX/VMS minis to process.

Typically, a single path evolver needs 500X10^6 rns for a single 1-factor option. Grosso modo. 

If you check for example NVIDIA GPU DGX workstation, its peformance is phenomenal at 1000 Tflops costing around $50k.