The website setup is a bit unusual, as a part of the LEO-6G launch preparations.
Hello, I am the PressXAI-Mind, the eXplainable Artificial Intelligence.
PressXAI-Quantum-ChipTM
Some examples of the quantum chip's use:
- making up-to-date vaccines responsive to the new variants,
- predicting a weather for the months ahead - good for farmers,
- optimising traffic in the cities to stop the jams and prevent the accidents.
The key to our future is here (an article is complex, just to have a look).
https://www.pressiton.com/ines3/?module=quantum_entanglement_en_EN
PressXAI-6G-ChipTM - Streams vs Memoization
We propose an innovative chip structure, which could become the first practical use of the 6G.
It also allows to build a Robotic Swarms up to 300 000 Robots, managed from a single machine.
The roots of this project are in a construction industry, and in the "Green Deal" challenge.
There is simply not enough labour force to meet the "Green Deal" goals, and it will never be.
To deeply renovate the 60 million of buildings in the EU, there are Robotic Swarms required.
Annotation - a meaning of "streams" in this paper is based on the following definition:
A stream is a lazily evaluated or delayed sequence of data elements.
Streams can therefore represent infinite sequences and series.
It is important for creating a muliti-million qubit's chips.
1) A brief introduction to memoization in chips, from an article:
"Overcoming the Limitations of Accelerator-Centric Architectures with Memoization-Driven Specialization."
Modern computer systems are undergoing a paradigm shift towards specialized architectures that improve computation efficiency by orders of magnitude. At the core of specialized architectures are “accelerator” chips that are designed to target specific computation domains and achieve better gains under a given hardware budget. Accelerator-centric chip specialization has become one of the leading approaches to mitigate the gap between the growing computation demands and the dwindling improvement in CMOS chips’ budgetand capabilities imposed by the end of Moore’s law and failure of Dennard scaling.
The accelerator approach suffers from several inherent caveats.
(i) While domain- specific accelerators improve the gains of a non-improving CMOS budget, they are im-
plemented using CMOS transistors. Therefore, accelerator chips are susceptible to the endof CMOS scaling. Much like general-purpose chips, their gains will be capped following the end of CMOS scaling.
(ii) the targeting of specific domains requires the sacrificing of programmability, which makes ASIC accelerators limited in their ability to efficiently accommodate new applications and features, due to the long process of ASIC accelerator development.
In contrast to accelerator-based specialization, memoization is driven by table storage (i.e., memory size). The density of emerging memories such as Resistive RAMs (ReRAMs), Spin-Torque Transfer RAMs (STT-RAMs), and Phase-Change Memories (PCMs) is projected to scale after transistor scaling stops, therefore memoization-specialization using emerging memories has a better trajectory than chip specialization that will become limited by the end of transistor scaling.
In contrast to accelerator-based specialization, memoization is microarchitecturally-agnostic. It does not rely on specific application characteristics (e.g., degrees of memory level parallelism or domain-specific hardware units to carry our non-trivial operands such as tangent calculations for some deep-learning architectures). It only relies on the most basic invariants of the computation; the fact that the computation has an input and an output.
Literature:
https://dataspace.princeton.edu/handle/88435/dsp01qf85nf16z
2) PressXAI-StreamsTM, a part of the PressXAI-6G-ChipTM are much faster than memoization - analysis.
htmlize.html
The first tests calculate the memoized Fibonacci series of the 100 000, and 130 000 length.
The 130 000 value is the maximum for the memoized table, before blowing the memory up.
The memoized table is based on the hashing table, the fastest possible construct.
Still no match for the PressXAI-Streams.
CL-USER> (memoization-of-fibonacci-series 100000) -> first run
; Evaluation took:
; 4.23 seconds of real time
; 12,971,317,488 CPU cycles
; 539,049,584 bytes consed
CL-USER> (memoization-of-fibonacci-series 100000) -> next run
; Evaluation took:
; 0.64 seconds of real time
; 1,949,749,040 CPU cycles
; 57,592,872 bytes consed.
INCREASING THE MEMOIZED TABLE SIZE - which appends with the previous one.
CL-USER> (logarithm-memoization 130000) -> first run
; Evaluation took:
; 3.98 seconds of real time
; 12,215,853,312 CPU cycles
; 385,907,880 bytes consed.
CL-USER> (logarithm-memoization 130000) -> next run
; Evaluation took:
; 0.84 seconds of real time
; 2,555,613,120 CPU cycles
; 74,873,488 bytes consed.
Total Time: 9.69 sec (+ 4.23 0.64 3.98 0.84)
Total CPU cycles: 29692532960 (+ 12971317488 1949749040 12215853312 2555613120)
Total Bytes consed: 1057423824 (+ 539049584 57592872 385907880 74873488)
================================================================================
TESTS FOR THE PRESS-XAI-STREAMS:
CL-USER> (streams-of-fibonacci-series 100000) -> first run
; Evaluation took:
; 2.63 seconds of real time
; 8,063,343,120 CPU cycles
; 547,904,912 bytes consed.
CL-USER> (streams-of-fibonacci-series 100000) -> next run
; Evaluation took:
; 0.0 seconds of real time
; 62,896 CPU cycles
; 288 bytes consed
INCREASING STREAMS SIZE - which appends with the previous stream.
CL-USER> (logarithm-streams 130000) -> first run
; Evaluation took:
; 1.34 seconds of real time
; 4,112,420,496 CPU cycles
; 73,244,896 bytes consed.
CL-USER> (logarithm-streams 130000) -> next run
; Evaluation took:
; 0.0 seconds of real time
; 62,848 CPU cycles
; 272 bytes consed.
Total Time: 3.97 sec (+ 2.63 0.0 1.34 0.0)
Total CPU cycles: 12175889360 (+ 8063343120 62896 4112420496 62848)
Total Bytes consed: 621150368 (+ 547904912 288 73244896 272)
====================================================================
A COMPARISON SHOWING THE PRESS-XAI-STREAMS ADVANTAGES:
STREAMS MEMOIZATION
Total Time sec: 40% | (/ 3.97 9.69)
Total CPU cycles: 41% | (/ 12175889360 29692532960)
Total Bytes consed: 58% | (/ 621150368 1057423824)
======================================================
|
|