CPU - 10 Ph.D guys sitting in a room trying to solve one super hard problem. GPU - 1000 preschoolers drawing between lines.

One thing worth mentioning: GPUs actually do 4 dimensional matrix calculations rather than 3, because with 3 dimensions, Rotation and magnification require matrix multiplication while translation requires matrix addition. By adding an arbitrary dimension, the GPU is able to unify all three key transformations under a single multiplicative architecture.

"you've never seen a triangle that isn't flat." I just came from the non-euclidean geometry video.

Here’s a key acronym to remember about GPUs: “SIMD”. That‘s “Single-Instruction, Multiple-Data”. It has to do with the fact that a GPU can operate on a hundred or a thousand vertices or pixels at once in parallel, but it has to perform exactly the same calculation on all of them. Whereas a single CPU core can be described as “SISD” -- “Single-Instruction, Single-Data”. With multiple CPU cores, you get “MIMD” -- “Multiple-Instruction, Multiple-Data”, where each instruction sequence can be doing entirely different things to different data. Or in other words, multithreading. So even with all their massive parallelism, GPUs are still effectively single-threaded.

Hi missed one of the most important things. Why they are more efficient. Running GPU load in parallell don´t make it more power efficient. It also don´t use less die surface. Whats is done is that the GPU it just have one very narrow sett of instruction that is the same width and same length regardless. Typically it can be a 128 bit by 128 bit instruktion that can run on 2x64 bit, 4x32 bit or 8x16 bit. And it can only run a very limited number of instruction. Also, there is no branch, and no branch prediction. This cuts away a load of piples. In a GPU every pipeline does say 128 bit of instruction for every clock regardless of instruction. If you only need 16 by 16 bit of instruction, though luck, still have to fill up the whole 128 by 128 bit register. Also, a CPU got a FPU and a IPU and logic, and SIMD pipes. And every pipe have duplicates for bransch prediction. Some have 4 way branch prediction, so they have 4 pipelines for one instruction. So in a normal CPU you might have 4 pipes for integer, 4 for floating, 4 for SIMD and 4 for general logic (often combined with integer). Total of 16 pipes just to calculate one value. Why.. well to get the speed up. .. the clock speed that is and keep branch prediction errors down. A GPU just does one type of pipeline (in the older days there was two different, one for transform and one for rendering, but in most modern one they are integrated in the same pipeline). Also, there is one instruction for every 128 or 256 set of data. That is if you uses 32 bit on 128 bit pipe you get a 1:4 instruction reduction... because most data uses the same instruction. Typical 128 bit data can be a HDR rendering scen where two colors are mixed RGBA mixed with a other RGBA. Insteed of running first the R, then the G and then the B and so on. They just put the whole load in to the pipe and calculate all the data with the same instruction. Almost everything in graphics come in set of four. If there is a polygon, well triangle. It have 3 corners, and then a additional value for the polygon. So it still have 4 values. Some Graphics load use half precision. Then you can do two pixel or two polygon calculation at the same time in the same pipe. Actually, CPU have had that capability since Pentium 3. But its still just have one output per core. (even if its a 128bit simd output, most modern CPU even have 256 bit simd) The other part is the missing of branch .This remove a ton of problem for the chip designer. Firstly you don´t need any branch prediction. Secondly you don´t need and branching instructions and pipes. Removing the whole logic pipe. Thow you can still make branch like calculation multiplying with a matrix that gives a very set 0 or 1 value in the end matrix. But the GPU can never make any decision about it. This is also the main drawback with the GPU. it will continually calculate the same set of work orders on and on and it will only calculate set given work order. It can´t make work order. This makes the load very predictable. The GPU know what it will do several 10 000 clocks in for hand. This helps parallellism very much. But it can´t make a instruction. So the CPU still have to make a instruction for the GPU. But the load of the CPU can be very low. For example the CPU kan tell the GPU to render a trea of data. The trea in turn is a given list of objects, that have a given list of subobjekt, that have a given list of vertexes, that have a given list of textures and so on. This way the CPU can give the GPU very limited amount of information to make quite a lot of work. This have not always been the case. GPU prior to 2000 have to have specific list of textures and vertexes directly from the CPU, giving the CPU quite a lot of work load. The problem nowdays is that the game dev want the word to be "living" They there for want as little amount of data to be pre fixed to a set trea of object. There for the CPU can be running in feeding the GPU with 1000 and 1000 of objects for every frame. In DX11 this is a problem, because in DX11 just one CPU core can do the GPU feeding. Someone just never thought this would be a problem. This have been in DX since the first version. Finally in DX12 it will be updated.

woobly computerphile camera XDDD nice one hahaha.

This is one of the better simplified explanations of 3D graphics I've seen. Particularly the part about why triangles are used.

it's amazing that he didn't explain SIMD or Vector computing, because that's exactly what the difference is - and it's the real answer to why CPU and GPU are different. instead of looping over an expression of 256 array elements, you create huge registers that are a gang of 256 floating point elements. a CPU would have 256 threads progressing at different rates on each array element. a GPU would force all 256 items to be on the same instruction because it's working on these huge registers. this is why you can't run all CPU algorithms efficiently on a GPU. ie: for each i in [0,256]: c[i] = a[i] * b[i] ... is CPU thinking. Each thread progresses at its own rate. GPU thinking is: float32_times256 c,b,a; c = b * a; .... where c=b*a is one instruction, with three huge operands.

Thank you so much for the content and the people on the screen for their willingness to share knowledge! Bravo good sir, it takes a special kind of talent to explain something complicated this well and understandably. Exactly the kind of people you'd want to have tea/coffee/beer with!

2:45 - Nearly bringing up the 'w' coordinate and quick fix by baptizing it "transparency" :P

One of the features of OS X a few versions back was the ability to off-load repetitive, non-graphical tasks to the GPU rather than the CPU to operate faster.

We've learned all sorts of things about CPUs. Some more videos on GPUs would be really nice.

I've been always curious of this, thanks for a video to explain it!

An entire episode with no explanations on how stuff actually works. Well done.

Really nice simple explanation of a complicated area!

He is excellent at explaining these things. Please have him on more.

wow, so well explained, as simple as possible but no simpler. Its such a big area i cant believe he explained it so compactly and clearly. Hi is a real genus.

very well explained, thanks for the video!

good video ! i hope there will be more detailed videos about gpus und cpus !

Watching this video on one monitor, seeing my GTX970 busy folding away on the other monitor. How appropriate...

Nicely done. I use nVidia GPUs to do huge matrix operations in statistical modeling to speed up model estimation. Bayesian MCMC isn't really practical on large scales without the parallelism offered by GPU computing.

Such a pleasure to watch , wonderfull thanks !

You should have asked him exactly what role a CPU plays vs. a GPU in something like a game. What exactly does the GPU need from the CPU, etc.

Perfect, I just wished he had actually said that a GPU has lots of tiny processors (compute units) physically. Yes most people know this, but given how entry-level this explanation was (on purpose), the target audience might not know it. Very clear answer, I'm impressed :O

It is extremely easy to complain about a tiny glich in the game while sitting in front of a console without any clue how everything works. When u think of about the math involved in making it, it is mind blowing what a human mind is capable of.

I adore his crystal-clear step by step explanations. Great vid!

When you tell a joke to an engineer and he takes it seriously and fixes it instead. XDXD - 6:16

am I the only one who finds the way he says pixel weird, the stress seems to be in completely the wrong place

Can you guys eventually do a video on how a cpu works? thanks

Password cracking with a GPU is also faster than the CPU.

I didn’t understand half of what he was talking about but I could listen to him all day.

When you describe basically what the gpu is doing it really put into perspective how complicated modern graphics is.

*GPU:* Graphic objects can generally be broken into pieces and rendered in *parallel* at the machine level. *CPU:* Generally sequential by design to care of problems which have many *_dependencies_* which *cannot be solved in parallel* by their very nature. Ex. Seq1: *A = b + c;* Seq2: *D = A + E;* Seq3: *y = D - A;* Obviously, you have to do the thing in *sequence* in order to get the value of *y.* However, the *tasks* of *_relegating_* as to which problem (i.e. gaming or any other apps) during *execution time* is mainly a *function of the OS.* This is so as *Gaming Apps* is not *all* a *_graphic rendering_* task but also a combination of *logic, rules, and some AI.* Those are the reasons why you need *both.* Kinda like *_specialization_* of tasks.

Nice shout out to Bitcoin , but for clarification purposes Bitcoin mining on GPUs is now obsolete and most of it is done of Specialty ASICs. Mining is so competitive that If you attempt to mine with a GPU or even older generation ASIC you will lose money unless you have access to free electricity. The general point of mentioning mining on GPUs is important though as hashing double rounds of SHA256 is much more efficient on a GPU than CPU, although ASIC's are far superior to both. For Mining - ASIC > FPGA > GPU > CPU

thanks for adding your info so all humans can learn thanks again

I've been told that for computer modelling, GPU's aren't always a good solution. I've been told that for certain kinds of Finite Element Analysis, you can generally only calculate one step at a time, one node at a time, as the result of one node can affect the calculation required for all the other nodes it influences, and the same is true for those nodes at each time step. So these still need high speed CPU's to do the grunt work, as the task isn't easy to parallelize.

When it comes to game development 3D in unity do you need a powerful gpu?

It really isn't necessary to use floating point operations but it is done with floating point operations because there is hardware that does the floating point operations very well so they are making use of that. An example to clarify what I mean is doing square roots. You can do square roots without using floating point ops you can just use integer calculations then at the end you just place the decimal point in the correct position.

1:27 That's a suboptimal way of tessellating circles. All the triangles meeting in the center will increase the chance of visible artifacts occurring. (UKposts smartass answer mitigation disclaimer: I know it's not a circle and I saw that the triangles don't meet. Leaving a smaller copy of the same polygon for the center doesn't solve the problem though, does it?)

"wobbly Computerphile camera" (laughs) "can it fix my bad focussing?" ... my this is a self-deprecating episode isn't it? ;)

Most of the modern display devices provide a secondary display output for to use a second monitor. But how to display the content of the secondary linear framebuffer on the secondary monitor using a different resolution as the primary monitor for a selfmade bootmanager(no display driver loaded) with and without an UEFI-Bios?

thank you so much

I think a video on real time ray tracing would be interesting. There's a good chance that it'll be the future of graphics but it'll require different silicon to todays GPUs

Please start adding subtitle !

best explanation of gpu vs cpu I ever heard

Cool video.

can Computerphile do an episode on algorithmic art/music?

Hii sir.. can you help me...... I have amd A10 R7 7800 processor and amd radeon 2 GB graphic card model HD6450..... Soo can you suggest me they both work together and give best performance or graphic card is useless with this processor.....

Recently I made the mistake of trying to produce an entire 3d graphics engine from scratch in pjs, all starting with the following function: var raster = function(x, y, z) { return [x/z, y/z] }; This function allows me to take any point (x, y, z) and map it to the screen. This is the most vital, core component of the entire program. I eventually added cameraPos to the function so I could move anywhere. But there was one fatal problem. I could not rotate anything. I attempted to produce a function that took in 4 vectors (newx, newy, newz,) and (x, y, z) to produce a new imaginary grid that treats newx, newy, newz like x y z and transforms the point. I then tried to produce a function that generates 3 of the vectors the previous function calls for based on pitch, yaw, and roll, but I have yet to get it working.

I run very, very large AutoCAD files and they would take forever and a day to load. My supervisor has the exact same machine I do and he loads the same files in a fraction of the time. I asked him how and he said that he installed a graphics card. Just for displaying 2D images, the GPU makes all the difference. I got one and it is like magic.

I am currently doing "high throughput computing" :)

D-Wave made a quantum computer, get on it Computerphile. I am kind of super hyped right now.

I do 3D modelling recreationally and I hate working with triangles. I prefer quads because those types of polygons can be structured into bands. With those bands, a looping structure can select/manipulate multiple polygons at once. For example, imagine designing a human model. You can run bands of quads to form cylinder parts like arms, neck etc. From there, you can manipulate entire bands at once instead of going polygon after polygon.

is't the big difference architecture between those (kinda forgot but in school they teach something like this ) and was named by architects who made it , i know one was cheap to make because he could not take and give information at same time( that should be cpu ) and gpu could do it but it was much more expensive to produce harvard and von neumann i think. and slower one is von neumann expensive one is Harvard

that feeling when you're explaining to people that it's a $1000 card that just draws virtual triangles. I guess they're pretty good at adding and subtracting floats and doubles, but the triangle drawing is where the action is.

For a VP of Technology at "ARM" this was an extremely low tech explanation. But then all of my extra parallel processing is done on a NVIDIA card. A CPU today has eight or more cores. I hope the rest of this series has some more meat to the bone. We are not morons out here.

In what situations are objects modelled? Would it only be in CAD situations ?

can u please explain what is the difference between computer machine and computer device??

The spice must flow! (If you don't get the joke straight away, rewatch the video.)

you have just got a follower

Our professor on Star Clusters told us GPUs are far more effective in calculating clusters evolution. Since it is not possible to analytically derive the evolution for more than two bodies, it has to be done numerically. 'It's all just about data crunching', he said.

I think maybe GPU works like Analog TV which also can translate frequency into picture/ trinary data arrays. But CPU can do ALL of any of those and why CPU slow at the end? Maybe because of work load of crunching and sorting all of types of data arrays, that' why~! I just wondering, what would happened if GPU too used for processing tandem with CPU? will it bottle necked bc bridge/bus chaos or become awesome?

I thought that the triangles were flat only in euclidean space? as 2+2 is only 4 in decimal

Triangles can exist on curved surfaces. You can even have a triangle with 3 right angles when placed on the surface of a sphere.

“ That is coming”, prophetic words.

I'm using a Tesla GPU at my university for accelerating a certain physics simulation. Works quite well, I'm getting a speedup of around 10, compared to normal CPU processing.

And why do programs like C4D use the Processor for rendering a 3 dimensional space?

They also use GPU 's in Lattice QCD calculations.

You should tweak your audio and reduce the heavy bass more often, half of the videos on this channel want to blow my woofer.

Cuda; the best thing for the advancement of Machine Learning Research ever.

Gpu --> graphic rendering independently of cpu Cpu --> take user input and give it to gpu to draw the picture on the screen.. That's cool.. What is "vlar" and "foo" by the way?

"you've never seen a triangle that isn't flat." mmm, what about Spherical triangle

"Specialist" versus "General" computing? PS, after watching the video, the parallel computing is also a lot more important in a GPU. :)

The white pixel moving over the turned off monitor on the left is getting on my nerves... D:

Computerphile seem to be on a boat?

So I need a better gpu for gaming computer?

2:30 Games actually don't move the camera around. Everything else gets rotated around the camera.

I like this.

Would it be easier to say that the table top is a circle, rather than a bunch of triangles?

very nice video. I don't know a lot, and isn't the strongest in Math either, but I enjoyed this video very much

I'm ot sure what you mean by a triangle always being flat since the four sides of a pyramid are made of triangles and they aren't on planar surface? the bottom of course isn't a triangle

Nice

Graphics are just one possible application for the use of GPU's but remember that GPU's are designed as better number crunchers than CPU's. Possible uses are password cracking, calculating PI, Chaos theory, stock market prediction or anything that requires highly parallel computing. The guy being interviewed seemed to be only focused on graphics work which only the tip of the iceberg.

THE SPICE MUST FLOW!

Cool video everyone, but did you forget to pack a tripod? I feel like this was shot at sea. Keep up the good work otherwise.

What about hardware wise? What makes a CPU better at single computations than a GPU physically?

I like the fact that he says graphical processing unit, as in "we"

So a C.P.U is't capable of parallel computing? ( sorry if this is a stupid question, I just want to know the answer) .

This video has aged well. Can we have an update please.

but though gpu has a lot more processing cores named shader cores(vertex shader and fractal shader),their clock speed is much slower than cpu.So overall gpu isnt fast enough for running os ,programs and other shits

But the only thing you didn't say is how the gpu achieves that ability to do parallel computing. Does it just have 10.000 Cpus in it?

Anyone else notice this madman believes HE is a GPU? Great video!

I learned something

Great

I need help on upgrading my PC: GeForce GTX 950 AMD FX(tm)-4300 Quad-Core Processor 16GB RAM What should i upgrade next? CPU or GPU?

The other thing you can do extremely well on graphics hardware, is model a scene in 3d, with a large number of units, like sugar cubes. Astrophysicists want to model the universe after the big bang, or look at the growth of galaxies, or model dark matter. Meteorologists want to predict the weather 10 days out. Climate people want to know about global warming. Economists want to anticipate world trade next year. Boeing want a more efficient wing tip shape on the next airliner. The navy want a submarine with is 600 ft long and capable of travelling at over 30 mph while making almost no sound. The way a graphic processor treats a scene, as a series of data points which are pixels, works extremely well when you try to model the behaviour of other systems, like the ones I listed. This is what is behind CUDA and other efforts to use graphics hardware. It's very good at doing the stuff you need for large scale simulations.

A light field camera will help your focus problems. And a tripod.

Normalize your audio.

c ang g, done !

Nottingham in Malaysia! My alma meter!