-------------------------------------------
High stakes for Trump as he departs for first Asia trip MSNBC - Duration: 6:34.
For more infomation >> High stakes for Trump as he departs for first Asia trip MSNBC - Duration: 6:34. -------------------------------------------
Facebook Messenger Hidden Secrets - Facebook Messenger Tricks For Girls 2017 In Urdu/Hindi - Duration: 3:50.
WELCOME TO MY CHANNEL
PLEASE SUBSCRIBE MY CHANNEL
THANK YOU
-------------------------------------------
02 LewTucker: The Brain as Model for parallel data structures - Duration: 10:40.
We were talking about the fastest supercomputers that are
extant at this point and you feel that there're still a lot smaller than
what you would want to see to really be able to yeah the if we take you know all
of the computers that are connected to the Internet and that you know doubling
probably every couple of years as it is now well pretty soon have the equivalent
computing power of one king brain really but that happens in a couple of years I
think - could be Wow I don't really I haven't
done the a very strict scientific analysis of that but I think that the
that that's not going to still be enough and that we still may there still may be
other ways to approach computing because that's always been the challenge and I
think that the number of devices that we have that are being connected to the
Internet is is growing even at a faster rate
so when you consider all of the smart phones and all of the IOT devices that
are that are coming about so we'll have computational power and now you have to
say how do you communicate because in the analogy the brain you have a neuron
that is connected to thousands of tens of thousands of other neurons so that
connectivity and most of the brain is actually in those connections so which
is why the connection making machine was particularly appealing because it's
recognizing that the connectivity is really at the heart of computing and
that the computing element the CPU or whatever can be very small so with the
connect at least the cm-1 it had the 12
dimensional good Leandre hypercube that richard fineman had suggested in that
that i ended up also using as a inspiration for the external form of the
machine did that hypercube structure survive into the CM - and was it in the
CM 5 yes but we we evolved in something what's known as a factory and the
factory is another kind of a cube like that the whole point is it's not a
strict hierarchy whereas you got higher and higher you go it's fewer and fewer
connections going up you needed to be able to do what's known as also the
cloths network today we talked about it whereby you have connectivity that that
is sort of constant at each level and therefore you have a lot of
communication bandwidth between the individual processors and even the data
centers today we're moving towards that model and the networking space because
we're finding that lo and behold inside of a data center today there's more
traffic going between servers than there is in and out of the data center i still
that's similar to a brain again there's so much more communication going between
the different processing elements so that it's single in obvious you know
example is sort of when you make a single query for google you know you
might type in you know find me you know picture of a cat and that will be farmed
out to hundreds of machines which will all be searching for that and their own
cross their own sort of segmented data stores and then aggregate the results
coming back so hundreds of machines are involved in answering a particular query
so the input and that but can be rather small compared to the amount of
communication amongst the processors themselves so for instance school really
does across of course yeah thousands of machines us in across many multiple data
centers as well and so every that you know every query that
processed in parallel and that's where a lot but the we're not using languages as
much as we're using this know today's sort of these microservices of being
able to have explicit communication between different services that are
returning results and analyzing the data so the the division of a problem into a
parallel structure is done at a micro service level and yeah and we had we had
to take shortcuts as well and one of the big shortcuts we took at that time was
what's known as a single instruction multiple data machine we were ready
running one program the same program on every machine and issuing those
instructions as a way to most efficiently perform this parallel
processing that model hasn't been continued today that model you do find
however in Nvidia and in video processing and so now chips we actually
have these Nvidia chips which are essentially small connection machines
that are operating in this kind of single instruction multiple data way
because they're processing an image and that's ideally suited for that kind of
processing power whereas general intelligence is more difficult to
achieve that way so there are other problems that Thinking Machines was
working on right at the beginning with things like modeling of weather and many
in many ways some of them in high-performance computing which I
haven't been very close to in the last several years but they are using much
the same algorithms for weather forecasting for you know QCD quantum
chromodynamics a lot of these physics problems and they are using much the
same algorithms and there you have being done a massively parallel supercomputers
today so those still exist and that was really where the connection machine led
to to using very very much and essentially tired
Center running a batch process running with a single problem at a time quantum
chromodynamics of course being the area exactly exactly right healthy so we're
highly influenced by behind Richards work and that's also where
high-performance and massively it was both Richard you know interest has only
been in combining a very different way of thinking about a problem within the
the technology you go about actually making it making it work so he was very
influential and not only what we worked on but how we thought about the problems
I can't stress enough the importance of what we did every day as a as a you know
in terms of our problem-solving was to try to think about the problem
differently you had to break out of the old mindset there were you know there
was a lot of bets being placed at that time if I remember that you couldn't
exceed by a factor like 10 a serial problem because you know which said that
you know no you can paralyze as much as you can but then the remaining part that
you can't paralyze will always slow you down and so there will be a limit to how
fast you can though go when M don't forgot was data parallelism and that was
know the data if you're matching if you're taking larger and larger and
larger amounts of data you can have full power was him across that data and
therefore we were consistently showing that I mean it was the difference
between a number and one of the other great things is that the people who came
to us with problems they were always most interesting problems I remember
working in something about space junk which was NASA that NASA had a problem
there was a lot of these remains of satellites and
nuts and bolts flying around whizzing round space and all we could detect was
a photon or two from each one of these things and so you get this like you know
scatter plot and in the next scatter plot know scatter plot and you had to
try to connect the dots to create the orbits of what these these particles
were we're spinning around and the best known serial algorithm showed you just
you need an enormous computing power you just couldn't do it instead we would
take the position that you can either map each small segment of the sky as a
cell and process all the cells at the same time you know and that way we can
immediately paralyze the problem and that's the typical kind of thing through
a course of a conversation you would you would arrive at a data parallel
algorithm which allowed you to really exploit the power of the machine and
even things such as sorting which you think of as being a completely serial
algorithm that you go through we could break that up into smaller segments sort
those merge sort merge and then using the power of the scan operation which
was implemented in hardware bring those results together can you explain this
scan this scan allowed you to to reduce it's part of the MapReduce algorithm
that people are using today that in parallel you perform some function
aggregating result out of a set some set and you can do those over many subsets
at the same time and so it's an it's the first part of sort of mapping of of data
to these different segments and then scanning is reducing those as a reduced
operation and since we were on a hypercube we could do that in parallel
and use the communications network where we had implemented the scan as this
reduction operation and so it's a combination with today and it's data
flow where we reproduce these things in a parallel sense quick quick quick
question moving from the boolean and cube to the
list of family the factory the factory if that happen at this cm 200 cm high
low that was the same 5 seemed to was still a hypercube and but the the
difference is very is very slight I mean a lot of it has to do with also the size
of the machine that we were building we were getting larger processors in a Sam
5 and more memory and therefore the machine was spread out over many
cabinets and we needed a way to be able to communicate now over distances and we
were still one of the fundamental problems with speed of life I mean being
able to have a signal that could go from one corner machine to another which is
why we came up with a physical design but you did which was trying to minimize
that by making it a cute and so that had to do it but constraints and and posed
by this actual speed of light Hey
-------------------------------------------
Ed On The Floor: EHX Glove Overdrive/Distortion Effects Pedal Demo For Guitar by DC Power - Duration: 16:56.
Alright alright today we have the Glove by Eelectro-Harmonix
Liiiiiiiikkkkke a glove, as our friend
Ace Ventura would say
Little bit'a my clean tone...
If you can hear my dog barking and nothing downstairs I apologize, but, umm..
He won't be joining us because he's decided that there's
something outside the door that's not there....
So, let's kick her on...
As usual on the clean channel of my ORANGE ROCKERVERB 50 MKIII
With the attenuator pretty much always all the way up because
this thing cranks as you know (already)
So!
12 o'clock
Um, as I understand it this thing is a OCD clone
As you know the OCD is one of the most popular pedals...
Pretty much, ever
Out of uh... Culver City California
their workshop is basically a fortress
Impenetrable, don't try knocking
Um, maybe email but just
look for them on Craigslist (or indeed) if you're
trying to, um...
get a soldering job or something of that nature
I believe their quote was "No rock stars"
But we're looking at EHX today Electro-Harmonix
So, enough about Fulltone
but Cali Represent
LA Represent
West LA Represent
Anyway,
So let's switch her on with the
Tone shift off at noon- see what happens
Actually, at uh...
At 12 o'clock it's not so bad..
let's do what we always pretty much do
turn the gain up
Let's turn that tone up...
Almost dimed...
probably the most OCD,
I've heard...
Let's turn that volume up...
Dimin' everything...
I feel like, um...
It gets a little muddy
So let's turn the gain down...
Let's turn the tone down
Uh, little bit more...
Alright, let's just put the volume back a little bit
gain up
It seems like the volume's affecting the tone a little bit...
Let's turn that back all the way up
Alright, just for Kicks,
we're gonna run it through the dirty app for a second
back to 12 o'clock, and then we're gonna check it out
With the shift knob, up...
So,
Dirty Channel
As usual forgive me for my horrible tapping,
Let's throw the gain up, we're gonna throw the tone up- we're gonna put the volume down...
(Inaudible expression of excitement)
DIME TIME!
Gain at 9 o'clock
(Inaudible expression of excitement)
Tone down... (11 o'clock)
Oh yeah, that's purdddy fun
Lets turn the volume down a little bit,
Turn the gain back up a bit...
Considering the volume's at about noon on my amp, on the dirty channel; that's pretty impressive...
Dirty channel off!
Clean channel back!
Let's see how much running it into a
Distorted channel or,
A distortion can do...
Let's turn the shift knob on
Or the shift switch on, tone up...
One sec go back to the other one
okay so it brings a little more shine into it...
Um, just everything at 12 o'clock real quick...
Oh definitely turned the sensitivity up
I'm gonna, roll my volume back a bit...
Gain all the way up...
Tone back...
Tone all the way upppppp....
let's put that gain back just a bit...
We can turn the gain up a little bit more, just because it's little more responsive...
Let's just dime it, why not?
Alright, dime the volume...
again excuse my awful tapping but for an overdrive that's pretty impressive so...
Let's bring everything back to 12 o'clock
Alright... Dirty channel
So, it's clearly introducing that 'sing-y' quality...
Fun with knobs...
DIMED!
DIME TIME!
(Ambiguous knob turning)
Okay
So,
General assessment The Glove is
Again in the words of Ace Ventura
Liiiiiiiikkkkke a glove,
It's very OCD-ish...
I think I might like it on the shift off a little bit better
Than the shift on
The shift on makes it more marshall-y it seems
The gain knob does more of that big muff-y thing where it...
kind of oscillates and and you know
tone wheels the gain so it gets a lot more
it gets a lot more sing to it
does that kind of mmmmmm thing if you turn the volume up real loud
I think as an overdrive it's pretty fantastic either way
I think it's a great...
I think it's a really great OCD clone
Obviously for less of a price but
There she is
However you feel about EHX
This one's a winner
Ed On The Floor
Peace \m/
Không có nhận xét nào:
Đăng nhận xét