They're not great at doing everything, just often better than doing it in software. The only thing a FPGA does better than an ASIC is take less time to develop, cost less in small quantities, and be reconfigurable.
@@saricubra2867 There will be an FPGA able to recreate the latest and greatest intel CPUs and faster. They are just not available yet. By then of course the newer CPU's will be faster. The point is FPGA are incredibly versatile and have lots of uses. One day all CPU's will be on FPGA type silicon. It's the future for sure.
One thing I wish he would have mentioned is how difficult FPGA's can be to program. Unless you have a very simple program, you'll have to break your program into at least a hundred different files to program it for an FPGA.
@@atif1538 they are called hardware description languages (hdl). The most common languages are verilog (or systemverilog) and vhdl. Most commercial companies use verilog whereas anything defense or government related uses vhdl.
@@hubertnnn No, ASIC is hardwired. A lot of toys are not hardwired but actually ESP/Arduino/Microcontroller running C code or something small like micro python. This is true to a lot of smart devices and toys that go beyond the cost of a simple PCB and IC.
@@hubertnnn No, ASICs (and FPGAs) are wired (and "programmed"/burned) to do some functions, microcontrollers runs software to do so, the function is implemented as software, it's a program. Processors and microcontrollers are special, large ASICs, wired to able to run machine code program. Cheap toys don't require full-hardware speed (implemented as ASIC or FPGA), a slow (compated to desktop processors) microcontroller can do the job by running program, which performs the function.
@@hubertnnn an ASIC is not running code, unlike a microprocessor. It is (like a microprocessor) a bundle of electronic circuits, but they are designed for a specific application (Application Specific Integrated Circuit). Building an ASIC is more difficult than programming an application on a microprocessor, but it has advantages (speed, efficiency, etc.).
@@kedharguhan Most modern FPGAs have a SERDES built in. so even 10 GB data will come in as a parallel stream, but much more slowly. Also, the data is likely coming in over multiple ports, and each port can afford to have its own dedicated logic (assuming that the FPGA is big enough). And the FPGA, like pretty much everything else, has gotten faster. But even hundreds of MHz should be enough to keep up with 10 GHz serial data.
@@harrkev Scalability was what I always loved about FPGAs (especially for logic like this which doesn't require much arbitration or interaction between design units) - you instantiate ten units within a FPGA, then if that's not enough you get a larger FPGA and instantiate a hundred of them! It really makes anything that can take advantage of parallel compute so much easier to deal with.
ASICs are fully optimized and dedicated logic. You get exactly what you want and not a gate more. It's not changeable but the fastest performance that can be achieved. (GPU's are ASICs.) FPGA are made of arrays of programmable blocks of logic, some of which you may not need for a specific function. They are slower and wasteful but for prototyping very nice. Many vendors have macro functions in their chips as well so you can just connect up to say a multiplier and not consume logic to make your own. Note that so much has been done on FPGAs that the libraries of completed designs are really valuable. This is why Intel bought Altera and AMD Xilinx. If you are doing System on a Chip - you just bought all anyone could ever ask for. When large ASICs are developed some teams take the course of buying or making an Asic Emulator. This is an array of FPGA's interconnected. You then take the ASIC's logic and partition it across the FPGA's. This gives you an accurate representation to run the design on to prove it functions before you spend millions building it. Note the emulator rarely runs at full speed but its a hell of a lot better than a running software simulation of a massive chip which can take a very long time. (Which is lost market opportunity.) You tend to find the tough bugs far faster. And of course the emulator can be used over and over again with fresh reloads of other unrelated efforts.
Yup most EDA is done on FPGAS. Billions of gates are emulated on large SOC and GPU designs. The debugging these emulators give saves manufacturers billions.
@@wtftwitter8225 - Yup - this has been part of my career. I recall one beast of an emulator we built from the ground up with 96 FPGAs found 60+ "corner case" really tough design flaws in less than a week. This would have taken 4 to 6 months with just the software based approach. The system was so successful the design was cost reduced (usually a version 2.0 post release phase) on the fly yielding substantial margin gains. I would not do a complex design any other way. This was for one of the largest ASICs TSMC had ever fabbed. Needless to say, I'm sold. And the fact that 8 months later, when that chip was shipping, that same emulator was pivoted to an entirely different effort unrelated to the original market was remarkable. Some happy engineers there.
One major benefit to using an FPGA that Linus didn't mention is that they consume FAR less power than CPUs/GPUs. Accelerating a task like a ML core on an FPGA can meet/outperform a GPU at 1/10th the power draw. Also, depending on your design, they can perform orders of magnitude faster than software equivalent algorithms. Heterogenous systems are the future!
An FPGA could be programmed as an advanced floating point unit. I found modern x86 floating point perfomance from AMD and Intel very weak (need a lot of cycles) , that's why Apple M1 lineup is so poweful despite running at 3.2GHz thanks to the powerful floating point, that's why it has been a huge sucess for music producers, sound designers, editors...
@@saricubra2867 The issue with floating point calculation is that it is very hard to do and almost never used (until it does) in a PC, so a CPU usually have just one floating point unit shared between multiple cores.
@@saricubra2867 Do you happen to know why x86 floating point performance is weak? Is it in applications using SSE or x87? If it's x87, you should be using SSE (the x87 stack is a huge performance killer, especially in modern OoO CPUs). Comparably speaking, an x86 CPU is far better at the actual FP uOPs than a FPGA. Most of the time, you want to avoid FP on a FPGA and instead used fixed point, because FP takes up a lot of hardware resources and cycles comparable speaking. The major benefit though, is that you can chain the execution graph together in a FPGA to be purely pipelined if you have a regular workload (which is the case for filters and synthesizers in sound - digital signal processing). I would imagine that most sound DSP applications use fixed point though.
Thats kind of obvious if they do one specific task it's possible to optimize hence waste less energy. Where as a general purpose computer can't optimize ahead similar to why consoles can perform better on games than PC do to optimizations ahead of time.
"You can think of them as Lego blocks, once you put them together they stay that way but you can always take them apart and put them back together and make something completely different" Linus never had these flat small lego blocks that were impossible to take apart after u stacked them.
That's the very point where Linus got it wrong though, FPGA's don't "stay that way", they forget their configuration every time they are powered down. They load their bitstream from a configuration ROM every time they are powered up.
As an FPGA Engineer it's incredibly exciting to hear FPGAs being talked about! Also, in some cases for ML applications FPGAs are being skipped over in favour of AI Accelerators (TPU, GroqChip, Cerebras, SambaNova, etc) in order to replace clusters of GPUs.
@@ghostmexi007 From my limited understanding of PLCs they are highly specialised devices and are not FPGAs, but are actually microprocessors that execute Ladder Logic programs. They can have large amount of dedicated I/O, ADCs, DACs, and different communication interfaces. Some PLCs may have some amount of switching fabric or some other kind of programmable glue logic, but I would not consider them to be true FPGAs.
@@ElToroLibre interesting thought they would be consider an FPGA seeing them as reprogramable cpus and the integrated logic circuit by the nature of the definition of an fpga well thanks for sharing this knowledge
What you do is really cool. I'm going on an FPGA/verilog rabbit hole after reading these comments from FPGA engineers. Starting to regret my BS in infotech... haha
@@RamyWarda This is still something you can do as a hobby! Depending on your experience you could start with SystemVerilog (as this is supposed to be closest to C) and a cheap development board, a number of them are well supported with lots of example projects to learn from. Some of them have displays, USB, buttons/switches, GPIO pins, VGA, HDMI or other audio/video output, etc. Even LTT have done videos of FPGAs used as retro game emulators! (Look for: "The Best Way To Play Retro Games?" by ShortCircuit). If you're feeling brave after learning a little bit about FPGAs you will have a great foundation to dive deeper into designing ASICs :D
I took an FPGA class in my school and we used Xilinx Vivado as the IDE. If that class taught me anything, it is to respect you FPGA engineers because oh man, I had such a hard time dealing with Vivado haha.
@@Schulzenberger Haha 100% The number of times where I literally made NO changes, and yet when I regenerate the bitstream and load the code to my FPGA and somehow it works is mind-boggling lol. I genuinely wonder how any hardware designed in Vivado works at all!
@@nashs.4206 I never used Vivado but I used Quartus Prime, and it's the same thing. The core program works rather well, but the simulator Modelsim is broken. It crashes randomly, which is kind of frustrating when editing code.
A good analogy would be, general CPUs are a toolbox, they can make many different products, and ASIC's are a production line, they make 1 product, but they make a lot of it , fast, and accurately.
And ASICs can go wide / parallel doing many things concurrently. Only multicore CPUs can process two threads in the same clock cycle. GPU's do many independent calculations in the same clock. This is what the CUDA cores are about - concurrency. This sort of loops back to LTT's recent analysis of the RTX3090 vs the CMP processor. They were related but not identical chips. If you were to look at the dataflow through each you would see custom lgic specifically added for graphics on one and AI on there other. Linus was spot on when he said the two markets had diverged AND could support independent chips due to the volume. The original NVIDIA Tesla's were really just relabelled GPU's with a manufacturing test mode plus API that flattened out that tremendous number crunching for general purpose vs graphics specifics. AI is getting really interesting in a similar way.
And building on that analogy, and FPGA is a 3D printer. You can make many different things with it and you get some degree of automation, but they're better at making certain classes of objects than others.
I've been an FPGA Engineer for 10 years. They really are great chips. You can pretty much do whatever you want in one, it just depends on how much time and money you want to spend 😉
Well FPGAs usually only have reconfigurable digital hardware, so everything analogue is in fixed parts like in a Microcontroller/SoC. Also chips for space or CERN need special considerations for radiation which make FPGAs unviable.
@@pascal7947 Radiation hardened FPGAs do exist from manufacturers such as Xilinx and Microsemi, and they have been used in applications related to space and particle physics. In regards to analog design, there is also the less popular FPAA (Field Programmable Analog Array), which I have only come across once about a decade ago.
Spot on. They are inefficient, slower, hotter and more costly. Its an artifact of their versatility and preset chip layout. Still, for where they fit that are really useful. If the final product can handle the cost premium its a fine go to market choice.
Currently making a generic cpu on an FPGA. Meaning it can be 16-, 32-, 64-, 128-, etc- bit. So I could theoretically have a 1 terabit or more cpu if the chip was big enough.
Well yes but actually no. FPGAs are simultaneously between ASICs and CPUs and also on the other side, past CPUs - yes, an FPGA might be used as simply a task-specific circuit (like an ASIC except reconfigurable to do something else), but the larger FPGAs are also complex enough to behave like a CPU, ANY CPU that fits within the resources they have; and in that sense, they are MORE general than a specific CPU. Not only can they run software like a CPU does, you get to choose WHICH CPU you want them to emulate.
Great video, but would've been awesome to mention the time or frequency domains that each of these are capable of working in because that's a big difference.
2 года назад+6
There is also new types of gates to calculate AI with analog circuit. There are bunch of resistors and ADC convertors inside ASIC.
3:30 the only thing I'd add is that FPGAs are not persistent, aka, they are volatile in that they need to be re-flashed if power is lost. Many FPGA boards will have a local non-volatile flash memory that automatically reprograms them on startup, or something more fancy like an MPSoC will use commands in the OS to reprogram
@@Liminal.Headspace I was so excited and tempted to buy one when they were released after seeing their booth at a tech show in NYC at the Javits Center (not sure why there hasn't been one since then and i'm talking long before the pandemic). Then the reviews came out and the lack of software that would make use of it just turned me off.
@@NightMotorcyclist It was an absolute waste of money almost 15 years ago. PhysX itself never went far. Ray tracing is where things are at, hopefully it'll get cheaper and better by the year. An actual dedicated ray tracing unit would be better than cramming it in an overtaxed GPU.
A better way to explain this: A CPU can do ANYTHING. But the down side is that it is always saying "What do I do next?" It does that step, and then says: "OK. What is my next step?" The whole instruction fetch and decode stuff takes up space and time (especially in the case of a cache miss). Also, a CPU does one thing at a time (only one ALU). Yes, I am ignoring superscaler. An ASIC or FPGA is pre-wired to know what to do, and it can also do multiple things at once. Need to multiply two lists of 100 numbers? A CPU will have to do that one at a time. You can put 100 multipliers on an ASIC or FPGA (yes, it will be big), and it can do them all at once. An ASIC is great of you are making thousands or millions. The NRE can be a million dollars or more. But they will run the fastest and be cheapest per unit. An FPGA will be more expensive, slower, and more power, but a hobbyist could easily make a design with just one of them, using free tools. Yes, I design ASICs for a living.
Datacentre ML and consumer retro consoles is a fairly limited selection of what FPGAs are used for (and probably a limited section of the market too). They are used a lot for moving around lots of data, like in LTE towers or in high end cameras/video switches etc. They also get used for glue logic and controllers because their timing is very controllable
3:35 With the caveat that they're volatile. If they lose power, they lose their programming. Generally they'll be paired with a ROM chip to program them on power-up. Since an FPGA doesn't emulate, but becomes the logic circuit you've built, you can turn an FPGA into a CPU if you want. (I've designed CPUs for FPGAs) But it's surreal because it's like, I have a CPU here! *unplugs it* and now that CPU is gone!
as a software programmer that acquired a couple of FPGA boards to learn to program them, the thing am seeing of note is that multiple FPGA digital circuits can be designed to run in parallel while based on the same clock signal. Depending on problem to be solved the FPGA manner of doing parallelism could be very powerful. An embedded CPU executing software instructions is very versitle and easy to program with lanfuages like C. But it's ability for parallelism is limited by how many cores the CPU has, which for embedded will typically be one, two, or four. Plus those cores need to really be programmed with some sort of OS kernel that facilitates managing different task for execution on the different cores. And those task need to be coordinated to work effectively and correctly, which can be the trickier aspect of multi-programming. Whereas the parallelism one designs into an FPGA is just inately what it does - it may require a means of feeding in input data and a means of collecting output data, but it doesn't have to be managed for parallel task execution in the way a multi-core CPU does. Of course, at the end of the day, my primary interest in FPGAs is to learn how to implement vintage computer CPUs :-)
@@saricubra2867 you still dont understand ,its not about the IPC, some games no matter what computer you use have issues with certain games games, especially certain graphical issues.thats what FGPA chips in these emulation consoles were used for. they are infact needed if you want true emulation, as for emulating it on another computer games run fine its just not always perfect.FPGA chips fix this. stuff with FPGA chips are also expensive and know people will pay the price for a true match for match emulation
@@karehaqt FPGA chips just ensure accuracy ,most games run accurate but some dont because of the software emulation.Its all common knowledge when it comes to emulation
FPGAs could be a neat way to learn about historical computer systems. You could emulate gaming consoles, obviously, but I'm guessing you could also emulate other types of computers from history, like 8-bit PCs, or even early computers like the ENIAC or UNIVAC systems.
One of the more impressive ASICs i know is a monitor screen Scaler. Time ago i was trying to repair a Monitor, i though maybe the fault was on the scaler chip circuit, so since i have some experience with microcontrollers i had the idea of replacing the scaler with one, but then after some research i realised it was impossible, the amount of information and the speed of processing was way beyond what a common microcontroler can do.
The first machine I ever built ran Windows 98 SE2 and an AMD K6-III with 256MB dual channel ram. I didn't even have an exhaust fan. I could edit videos, edit audio, browse online, be on IRC, DL from Newsgroups, and have emulation games (NESten) running at the same time. So I don't know what you're talking about saying that modern CPUs can do the things older CPUs couldn't? It's just a ridiculous statement. What you could say is software has evolved to the point that it requires a modern CPU to really multitask the way you did on the previous generation. That would have been honest and made sense.
Not only that, is the absolutely ridiculous IPC increase specially with Sandy Bridge, Haswell and Zen 1. Your K6 processor would struggle or be basically useless for music production, it would choke extremely hard, maybe the entire system would freeze. I run Android games on my 9 year old laptop with my Intel Core i7-4700MQ quad core processor, why? because it's waaaay faster than the Qualcomm octa core ARM chip from my current phone despite the difference on instruction sets. Massive difference on frequency and Haswell IPC is still good and AVX2 instructions are helping me. 64 bit instructions are huge, and 16GB of RAM is enough for lossless high definition video, unlike the K6 era (extremely crappy resolution compression and framerates). Things changed a lot since Core 2 Duo. The fact that you could do something back then is impressive, but would be extremely terrible and rough by modern standards.
This is like the difference between an all-in-one wireless routerswitchap, and a dedicated AP. I've got an all in one 2400Mbps, and a 1300Mbps AP, the 1300Mbps AP, with internal antennas, is about 8x faster than the all in one, but it can only do the one thing.
Some camera like from Blackmagic also use FPGA and can be upgraded with more function in the feature. Them problem with minner is they really one if sometimes minning is not good, they just resell the device easly. ASIC is not easy to resell.
what? no !!! As Electronics Engineer in Microelectronics I can clearly say that these information are wrong. To say, there is FPGA, ASICS and CPU is totally wrong information. A CPU is just a type of a device, such as GPU, or even 555. It is not a category of hardware as the other two. ASICS is the next step after prototyping with a FPGA, so you cannot change it. A CPU can be made with an FPGA, You can build a softcore in an FPGA, as well in an ASICS device. So, how the categorization in the video is made is very weird. I am really shocked !
@@jierenzheng7670 yes of course. It iw how we prototype things like that. In fact, you can find cores of old CPUs available on line on opencores and use directly. You can also find miners and other things.
ASICS isn't one I've heard much about, but it does sound interesting. FPGAs on the other hand have defiantly gotten my attention lately with all the retro gaming mods and FPGA clone consoles that add TONS of new features that would be seen only on new game consoles while also keep the original experience intact, what a world we live in.
The history of the naming can create confusion, but "microprocessor" just means CPU today (the micro part used to mean something when computers were the size of buildings, but not since everything scaled down to the chip level - like in the past 30 years). So really it's microcontroller vs process. A microcontroller is essentially a light weight CPU for embedded applications (toasters, dishwashers, etc.). Since they are for embedded applications, they're more like a system on chip with memory, "storage", and I/O on the same chip that is soldered to a board. One of their main staples is that they usually don't run a traditional OS, and only run one specialized application. They also are much slower but in exchange use less power and cost a lot less (you can get some microcontrollers for $0.20, that only use 0.01W, but runs at less than 1 MHz and has 128 bytes of memory).
Microprocessor: A CPU. Needs a bunch of stuff outside to be added (storage, memory, IO, etc.). Microcontroller: A CPU with almost everything that you need built-into it. You can easily have a "one chip" system where a microcontroller can have the code on-board to blink an LED or send and receive data over a serial port with practically no other parts. You find those in things like washing machines, microwaves, remote controls, etc. And because you have to build all of that stuff into one die, they are lower speed/capability and a separate CPU. But they are useful for a LOT of stuff.
I can't wait for AMD Ryzen to include FPGA ever since they successfully acquired Xilinx. Since FPGAs are very popular for seamless gaming emulation, maybe FPGA is the key to finally emulate the PS3, the hardest console to flawlessly emulate. With Ryzen and FPGA, development for RPCS3 will be done in a jiff.
There are cloud servers today with FPGA boards that can act as a specific CPU or a GPU, depending on the demand. That´s amazing. Soon AI will be programming FPGAs and new crazy ICs will be created or optimized.
Have you heard of that new adaptive transistor technology Linus? With this transistors can change the architecture by changing their direction electrically. That mean a zen 3 core can be changed to RDNA2 core or even next gen IP like Zen4. P cores can be changed to E cores. This things takes specialised chips one step further. Even ahead of modern CPU. This can literally save silicon waste.
0:51 Linus: Regular processors .... Also Linus : Shows us a Threadripper inserted in Zentih extreme MB ... That's the level of the game they're playing on :D
To summarize: CPUs run software telling them what to do and can be reprogrammed. FPGAs run software (bitstream) telling them what to be and they can be reprogrammed into another hardware configuration. ASICs are mask-programmed at the factory for a specific hardware config, but cannot be changed afterwards. On another note, FPGAs also have the advantage of dealing with handling data in true parallel fashion.
The Newtek chips were full custom ASICs. Incredible product, sadly the founder of the company passed away unexpectedly and they lost momentum. I think they are still around in some manner although now that everything is digital their original product is not needed. Comedian Dana Carvey's (Wayne's World) brother had a lot to do with that story.
@@LexFelron - A milestone product. It and the Amiga opened the home video market. Lucasfilm's Editdroid and Pixar brought Digital Post and CGI into the lexicon. Premier (and almost all desktop editors) are based on Editdroid's look and feel.
Something worth noting about FPGAs is that there are also fuse-based FPGAs which are only able to be programmed once in the field because fuses are used to permanently sever a connections that are not needed. These are good to have because unlike reprogrammable FPGAs that have to reset all of its gates every time it loses power, the fuse-based FPGAs are immediately ready when power is restored.
Are they still making those? I can't imagine the demand is too big, since you can't update them. I do remember using an Actel ProAsic years ago: an FPGA with flash built-in so that it is ready to go immediately without having to add a serial EEPROM. But flash and fuse-based FPGAs seem to be in the extreme minority.
Took me 35 hours of programing in Quartus (System Verilog HDL) to write a program that draws a line onto a DE1_SoC Development board (I'm studying Electrical Engineering in college). All of that work just to draw and clear a line onto a VGA output. Ugh that is the struggle with FPGA's they are a pain to code bc everything runs at the same time (unless you govern it by a clock, but still everything inside the always_ff @(posedge clk) still runs simultaneously.
If you want your code to run sequentially you would use a state machine, really just a case statement in a clocked process. You could also use a counter and conditional statements inside a clocked process to decide when lines of code run. But really you do want as many functions as possible to run in parallel without being blocked or having to wait, so pipelining and dataflow is very important.
i would say that you are wrong on this one guys. A cpu IS in fact, an asic. It has static non-reroutable logic gates as well and while they can do a lot of tasks, each task is done through an instruction set command that doesnt interfere with the logic gate connections, just like how the rest of the asic chips do. An FPGA chip though, can reroute its logic gates to simulate a custom processor or a completely different system, granted there are enough logic games.
He survived the Mentos with pepsi, but had it been Pop Rocks and Canada Dry... We wouldn't have Linus Gabriel Sebastian to kick around anymore. R.I.P. Mikey
I tend to skip through your sponsor segments, but I really like that you have chosen to work with a Ukrainian company like Grammarly right now. Watched it end-to-end.
Thats the difference between microprocessor and microcontroller. They are (mostly) made to handle one set of instructions that never change in the lifetime (or moderate upgrades thru firmware)
Maybe off topic , but might this explain why Google tensor didn't do as well in machine learning benchmarks versus apples bionic chip. My thought process here is, Google's tensor processor is partially an ASIC processor for specific machine learning tasks, for a specific Google features, which might explain the very large gap? Or is the bionic rly just that much better? Any computer science majors have any input???
ASICs are what YOU are using. If you have a cell phone, there are ASICs in there for the radio stuff (WiFi, Cell, Bluetooth, etc.). If you have a computer, there are ASICs in there for such things as power management, clock control, etc.. Really, ASIC just means any piece of silicon designed to do a specific job (IE not a CPU, and not discrete logic). FPGA is a chip that does nothing, but you can program it to do any job that an ASIC can do.
![Twenty One Pilots - The Line (from Arcane Season 2) [Official Audio]](http://i.ytimg.com/vi/XCz7WUotXs8/mqdefault.jpg)
FPGAs so great at doing everything, they dont do anything out of the box
That's why they AREN'T better than x86 based CPUs.
Clickbait title for the video
@@saricubra2867 That depends on what x86 cpu you are talking about. There is an x86 core that performs pretty well.
They're not great at doing everything, just often better than doing it in software. The only thing a FPGA does better than an ASIC is take less time to develop, cost less in small quantities, and be reconfigurable.
@@ZaPpaul Every x86 based equivalent and more powerful than Intel's Haswell.
@@saricubra2867 There will be an FPGA able to recreate the latest and greatest intel CPUs and faster. They are just not available yet. By then of course the newer CPU's will be faster. The point is FPGA are incredibly versatile and have lots of uses. One day all CPU's will be on FPGA type silicon. It's the future for sure.
As an FPGA engineer, this video was surprisingly good.
One thing I wish he would have mentioned is how difficult FPGA's can be to program. Unless you have a very simple program, you'll have to break your program into at least a hundred different files to program it for an FPGA.
@@larrimus4309 unless you get a myRIO and use labview
@@larrimus4309 what kind of programming languages do they use?
@@atif1538 they are called hardware description languages (hdl). The most common languages are verilog (or systemverilog) and vhdl.
Most commercial companies use verilog whereas anything defense or government related uses vhdl.
@@mwniels aight thanks
The chips inside cheap toys are usually tiny micro controllers that are burned with a program from the factory.
That is more or less the definition of asic
You mean flashed rather than burned perhaps?
@@hubertnnn No, ASIC is hardwired. A lot of toys are not hardwired but actually ESP/Arduino/Microcontroller running C code or something small like micro python. This is true to a lot of smart devices and toys that go beyond the cost of a simple PCB and IC.
@@hubertnnn No, ASICs (and FPGAs) are wired (and "programmed"/burned) to do some functions, microcontrollers runs software to do so, the function is implemented as software, it's a program. Processors and microcontrollers are special, large ASICs, wired to able to run machine code program. Cheap toys don't require full-hardware speed (implemented as ASIC or FPGA), a slow (compated to desktop processors) microcontroller can do the job by running program, which performs the function.
@@hubertnnn an ASIC is not running code, unlike a microprocessor. It is (like a microprocessor) a bundle of electronic circuits, but they are designed for a specific application (Application Specific Integrated Circuit). Building an ASIC is more difficult than programming an application on a microprocessor, but it has advantages (speed, efficiency, etc.).
I work in a cyber security company that uses FPGAs 😃, we use them to monitor network traffic at rates of hundreds of GB/s.
Which FPGAs are those ! I always thought fpgas don’t exceed MHz range
@@kedharguhan We mostly use Intel Stratix 10 FPGAs
@@kedharguhan Most modern FPGAs have a SERDES built in. so even 10 GB data will come in as a parallel stream, but much more slowly. Also, the data is likely coming in over multiple ports, and each port can afford to have its own dedicated logic (assuming that the FPGA is big enough).
And the FPGA, like pretty much everything else, has gotten faster. But even hundreds of MHz should be enough to keep up with 10 GHz serial data.
I meant serdes, my phone doesn't like that one
@@harrkev Scalability was what I always loved about FPGAs (especially for logic like this which doesn't require much arbitration or interaction between design units) - you instantiate ten units within a FPGA, then if that's not enough you get a larger FPGA and instantiate a hundred of them! It really makes anything that can take advantage of parallel compute so much easier to deal with.
ASICs are fully optimized and dedicated logic. You get exactly what you want and not a gate more. It's not changeable but the fastest performance that can be achieved. (GPU's are ASICs.) FPGA are made of arrays of programmable blocks of logic, some of which you may not need for a specific function. They are slower and wasteful but for prototyping very nice. Many vendors have macro functions in their chips as well so you can just connect up to say a multiplier and not consume logic to make your own. Note that so much has been done on FPGAs that the libraries of completed designs are really valuable. This is why Intel bought Altera and AMD Xilinx. If you are doing System on a Chip - you just bought all anyone could ever ask for.
When large ASICs are developed some teams take the course of buying or making an Asic Emulator. This is an array of FPGA's interconnected. You then take the ASIC's logic and partition it across the FPGA's. This gives you an accurate representation to run the design on to prove it functions before you spend millions building it. Note the emulator rarely runs at full speed but its a hell of a lot better than a running software simulation of a massive chip which can take a very long time. (Which is lost market opportunity.) You tend to find the tough bugs far faster. And of course the emulator can be used over and over again with fresh reloads of other unrelated efforts.
incorrect
@@fredbarfson2624 - Elaborate...
Yup most EDA is done on FPGAS. Billions of gates are emulated on large SOC and GPU designs. The debugging these emulators give saves manufacturers billions.
@@wtftwitter8225 - Yup - this has been part of my career. I recall one beast of an emulator we built from the ground up with 96 FPGAs found 60+ "corner case" really tough design flaws in less than a week. This would have taken 4 to 6 months with just the software based approach. The system was so successful the design was cost reduced (usually a version 2.0 post release phase) on the fly yielding substantial margin gains. I would not do a complex design any other way. This was for one of the largest ASICs TSMC had ever fabbed. Needless to say, I'm sold. And the fact that 8 months later, when that chip was shipping, that same emulator was pivoted to an entirely different effort unrelated to the original market was remarkable. Some happy engineers there.
@@dennisfahey2379 just a troll.
One major benefit to using an FPGA that Linus didn't mention is that they consume FAR less power than CPUs/GPUs. Accelerating a task like a ML core on an FPGA can meet/outperform a GPU at 1/10th the power draw. Also, depending on your design, they can perform orders of magnitude faster than software equivalent algorithms.
Heterogenous systems are the future!
An FPGA could be programmed as an advanced floating point unit. I found modern x86 floating point perfomance from AMD and Intel very weak (need a lot of cycles) , that's why Apple M1 lineup is so poweful despite running at 3.2GHz thanks to the powerful floating point, that's why it has been a huge sucess for music producers, sound designers, editors...
@@saricubra2867 The issue with floating point calculation is that it is very hard to do and almost never used (until it does) in a PC,
so a CPU usually have just one floating point unit shared between multiple cores.
@@saricubra2867 Do you happen to know why x86 floating point performance is weak? Is it in applications using SSE or x87? If it's x87, you should be using SSE (the x87 stack is a huge performance killer, especially in modern OoO CPUs).
Comparably speaking, an x86 CPU is far better at the actual FP uOPs than a FPGA. Most of the time, you want to avoid FP on a FPGA and instead used fixed point, because FP takes up a lot of hardware resources and cycles comparable speaking. The major benefit though, is that you can chain the execution graph together in a FPGA to be purely pipelined if you have a regular workload (which is the case for filters and synthesizers in sound - digital signal processing). I would imagine that most sound DSP applications use fixed point though.
@@saricubra2867 aww... The intel fanboy grew up into an apple fanboy!
Thats kind of obvious if they do one specific task it's possible to optimize hence waste less energy. Where as a general purpose computer can't optimize ahead similar to why consoles can perform better on games than PC do to optimizations ahead of time.
"You can think of them as Lego blocks, once you put them together they stay that way but you can always take them apart and put them back together and make something completely different"
Linus never had these flat small lego blocks that were impossible to take apart after u stacked them.
that's lenovo burned-to-never-work cpus
That's the very point where Linus got it wrong though, FPGA's don't "stay that way", they forget their configuration every time they are powered down. They load their bitstream from a configuration ROM every time they are powered up.
Somebody needs to purchase a "LEGO brick separator" tool. :)
@@robspiess those still dont do much for stubborn 1x2 stacked plates.
As an FPGA Engineer it's incredibly exciting to hear FPGAs being talked about! Also, in some cases for ML applications FPGAs are being skipped over in favour of AI Accelerators (TPU, GroqChip, Cerebras, SambaNova, etc) in order to replace clusters of GPUs.
Arent fpga chip used widely for plc aren't they?
@@ghostmexi007 From my limited understanding of PLCs they are highly specialised devices and are not FPGAs, but are actually microprocessors that execute Ladder Logic programs. They can have large amount of dedicated I/O, ADCs, DACs, and different communication interfaces. Some PLCs may have some amount of switching fabric or some other kind of programmable glue logic, but I would not consider them to be true FPGAs.
@@ElToroLibre interesting thought they would be consider an FPGA seeing them as reprogramable cpus and the integrated logic circuit by the nature of the definition of an fpga well thanks for sharing this knowledge
What you do is really cool. I'm going on an FPGA/verilog rabbit hole after reading these comments from FPGA engineers. Starting to regret my BS in infotech... haha
@@RamyWarda This is still something you can do as a hobby! Depending on your experience you could start with SystemVerilog (as this is supposed to be closest to C) and a cheap development board, a number of them are well supported with lots of example projects to learn from. Some of them have displays, USB, buttons/switches, GPIO pins, VGA, HDMI or other audio/video output, etc. Even LTT have done videos of FPGAs used as retro game emulators! (Look for: "The Best Way To Play Retro Games?" by ShortCircuit). If you're feeling brave after learning a little bit about FPGAs you will have a great foundation to dive deeper into designing ASICs :D
As an FPGA Engineer myself specializing in VHDL, I always find it soo cool to be a topic of discussion in your videos!
I took an FPGA class in my school and we used Xilinx Vivado as the IDE. If that class taught me anything, it is to respect you FPGA engineers because oh man, I had such a hard time dealing with Vivado haha.
@@nashs.4206 that’s what we work with 😉
@@nashs.4206 Vivado taught me what the word frustration really means
@@Schulzenberger Haha 100%
The number of times where I literally made NO changes, and yet when I regenerate the bitstream and load the code to my FPGA and somehow it works is mind-boggling lol. I genuinely wonder how any hardware designed in Vivado works at all!
@@nashs.4206 I never used Vivado but I used Quartus Prime, and it's the same thing. The core program works rather well, but the simulator Modelsim is broken. It crashes randomly, which is kind of frustrating when editing code.
A good analogy would be, general CPUs are a toolbox, they can make many different products, and ASIC's are a production line, they make 1 product, but they make a lot of it , fast, and accurately.
And ASICs can go wide / parallel doing many things concurrently. Only multicore CPUs can process two threads in the same clock cycle. GPU's do many independent calculations in the same clock. This is what the CUDA cores are about - concurrency. This sort of loops back to LTT's recent analysis of the RTX3090 vs the CMP processor. They were related but not identical chips. If you were to look at the dataflow through each you would see custom lgic specifically added for graphics on one and AI on there other. Linus was spot on when he said the two markets had diverged AND could support independent chips due to the volume. The original NVIDIA Tesla's were really just relabelled GPU's with a manufacturing test mode plus API that flattened out that tremendous number crunching for general purpose vs graphics specifics. AI is getting really interesting in a similar way.
And building on that analogy, and FPGA is a 3D printer. You can make many different things with it and you get some degree of automation, but they're better at making certain classes of objects than others.
I've been an FPGA Engineer for 10 years. They really are great chips. You can pretty much do whatever you want in one, it just depends on how much time and money you want to spend 😉
Well FPGAs usually only have reconfigurable digital hardware, so everything analogue is in fixed parts like in a Microcontroller/SoC.
Also chips for space or CERN need special considerations for radiation which make FPGAs unviable.
@@pascal7947 Radiation hardened FPGAs do exist from manufacturers such as Xilinx and Microsemi, and they have been used in applications related to space and particle physics. In regards to analog design, there is also the less popular FPAA (Field Programmable Analog Array), which I have only come across once about a decade ago.
Spot on. They are inefficient, slower, hotter and more costly. Its an artifact of their versatility and preset chip layout. Still, for where they fit that are really useful. If the final product can handle the cost premium its a fine go to market choice.
Currently making a generic cpu on an FPGA. Meaning it can be 16-, 32-, 64-, 128-, etc- bit. So I could theoretically have a 1 terabit or more cpu if the chip was big enough.
Well yes but actually no. FPGAs are simultaneously between ASICs and CPUs and also on the other side, past CPUs - yes, an FPGA might be used as simply a task-specific circuit (like an ASIC except reconfigurable to do something else), but the larger FPGAs are also complex enough to behave like a CPU, ANY CPU that fits within the resources they have; and in that sense, they are MORE general than a specific CPU. Not only can they run software like a CPU does, you get to choose WHICH CPU you want them to emulate.
Wow
Great video, but would've been awesome to mention the time or frequency domains that each of these are capable of working in because that's a big difference.
There is also new types of gates to calculate AI with analog circuit. There are bunch of resistors and ADC convertors inside ASIC.
Love it when professor Linus shows up. These are my favorites.
spasti
@@pandasdreamlygon Sowas sagt man nicht, Marcel.
As FPGA Engineer, I glad FPGA is getting more attention in tech media.
3:30 the only thing I'd add is that FPGAs are not persistent, aka, they are volatile in that they need to be re-flashed if power is lost. Many FPGA boards will have a local non-volatile flash memory that automatically reprograms them on startup, or something more fancy like an MPSoC will use commands in the OS to reprogram
Chips that do one thing really really well... brings back the promises of PhysX dedicated cards prior to nVidia's purchase.
Still have one laying around. Ageia was ahead of their time and it ruined them.
@@Liminal.Headspace I was so excited and tempted to buy one when they were released after seeing their booth at a tech show in NYC at the Javits Center (not sure why there hasn't been one since then and i'm talking long before the pandemic). Then the reviews came out and the lack of software that would make use of it just turned me off.
@@NightMotorcyclist It was an absolute waste of money almost 15 years ago. PhysX itself never went far. Ray tracing is where things are at, hopefully it'll get cheaper and better by the year.
An actual dedicated ray tracing unit would be better than cramming it in an overtaxed GPU.
A better way to explain this: A CPU can do ANYTHING. But the down side is that it is always saying "What do I do next?" It does that step, and then says: "OK. What is my next step?" The whole instruction fetch and decode stuff takes up space and time (especially in the case of a cache miss). Also, a CPU does one thing at a time (only one ALU). Yes, I am ignoring superscaler. An ASIC or FPGA is pre-wired to know what to do, and it can also do multiple things at once. Need to multiply two lists of 100 numbers? A CPU will have to do that one at a time. You can put 100 multipliers on an ASIC or FPGA (yes, it will be big), and it can do them all at once.
An ASIC is great of you are making thousands or millions. The NRE can be a million dollars or more. But they will run the fastest and be cheapest per unit. An FPGA will be more expensive, slower, and more power, but a hobbyist could easily make a design with just one of them, using free tools.
Yes, I design ASICs for a living.
Linus: These chips are better than CPUs.
Pringles: Of course! They're made of potatoes!
@Larma Kana why linus' cleanup didn't work?
Hummm. Tech potatoes. Yeah, maybe he is behind this article
Based on the semiconductor carbon as opposed to silicon... ;-)
0:41 I am impressed that RISC-V got a mention. Cool!
Datacentre ML and consumer retro consoles is a fairly limited selection of what FPGAs are used for (and probably a limited section of the market too). They are used a lot for moving around lots of data, like in LTE towers or in high end cameras/video switches etc. They also get used for glue logic and controllers because their timing is very controllable
Thank you for the video, but I also wanted to hear more industrial applications of fpga except game consoles)
3:35 With the caveat that they're volatile. If they lose power, they lose their programming. Generally they'll be paired with a ROM chip to program them on power-up. Since an FPGA doesn't emulate, but becomes the logic circuit you've built, you can turn an FPGA into a CPU if you want. (I've designed CPUs for FPGAs) But it's surreal because it's like, I have a CPU here! *unplugs it* and now that CPU is gone!
If you don't want to lose your program, then Flash-based FPGAs don't have this problem!
@@ElToroLibre Ooh! I didn't know about those! That's neat!
as a software programmer that acquired a couple of FPGA boards to learn to program them, the thing am seeing of note is that multiple FPGA digital circuits can be designed to run in parallel while based on the same clock signal. Depending on problem to be solved the FPGA manner of doing parallelism could be very powerful. An embedded CPU executing software instructions is very versitle and easy to program with lanfuages like C. But it's ability for parallelism is limited by how many cores the CPU has, which for embedded will typically be one, two, or four. Plus those cores need to really be programmed with some sort of OS kernel that facilitates managing different task for execution on the different cores. And those task need to be coordinated to work effectively and correctly, which can be the trickier aspect of multi-programming.
Whereas the parallelism one designs into an FPGA is just inately what it does - it may require a means of feeding in input data and a means of collecting output data, but it doesn't have to be managed for parallel task execution in the way a multi-core CPU does.
Of course, at the end of the day, my primary interest in FPGAs is to learn how to implement vintage computer CPUs :-)
its funny how he talked about emulation consoles, I was about to say they can be programmed to run identical to the retro systems architecture
Modern day CPUs have insane IPC that FPGAs aren't needed. Taking Nintendo aside, current consoles are all x86 based
@@saricubra2867 you still dont understand ,its not about the IPC, some games no matter what computer you use have issues with certain games games, especially certain graphical issues.thats what FGPA chips in these emulation consoles were used for. they are infact needed if you want true emulation, as for emulating it on another computer games run fine its just not always perfect.FPGA chips fix this. stuff with FPGA chips are also expensive and know people will pay the price for a true match for match emulation
@@kyles8524 Software emulation can be just as good, people should stop shitting on it just because it's software.
@@karehaqt FPGA chips just ensure accuracy ,most games run accurate but some dont because of the software emulation.Its all common knowledge when it comes to emulation
Super interesting to hear one of your favourite RUclipsrs talk about something you studie (embedded systems).
FPGAs could be a neat way to learn about historical computer systems. You could emulate gaming consoles, obviously, but I'm guessing you could also emulate other types of computers from history, like 8-bit PCs, or even early computers like the ENIAC or UNIVAC systems.
That's so strange, legit I am taking a course on FPGA's right now and I open techquickie to see this haha.
Ooo, the Lego analogy is a good way to explain what an FPGA is. Nice!
One of the more impressive ASICs i know is a monitor screen Scaler. Time ago i was trying to repair a Monitor, i though maybe the fault was on the scaler chip circuit, so since i have some experience with microcontrollers i had the idea of replacing the scaler with one, but then after some research i realised it was impossible, the amount of information and the speed of processing was way beyond what a common microcontroler can do.
0:05 Did Linus just burn himself, or...?
The first machine I ever built ran Windows 98 SE2 and an AMD K6-III with 256MB dual channel ram. I didn't even have an exhaust fan. I could edit videos, edit audio, browse online, be on IRC, DL from Newsgroups, and have emulation games (NESten) running at the same time. So I don't know what you're talking about saying that modern CPUs can do the things older CPUs couldn't? It's just a ridiculous statement. What you could say is software has evolved to the point that it requires a modern CPU to really multitask the way you did on the previous generation. That would have been honest and made sense.
Not only that, is the absolutely ridiculous IPC increase specially with Sandy Bridge, Haswell and Zen 1.
Your K6 processor would struggle or be basically useless for music production, it would choke extremely hard, maybe the entire system would freeze.
I run Android games on my 9 year old laptop with my Intel Core i7-4700MQ quad core processor, why? because it's waaaay faster than the Qualcomm octa core ARM chip from my current phone despite the difference on instruction sets. Massive difference on frequency and Haswell IPC is still good and AVX2 instructions are helping me.
64 bit instructions are huge, and 16GB of RAM is enough for lossless high definition video, unlike the K6 era (extremely crappy resolution compression and framerates).
Things changed a lot since Core 2 Duo.
The fact that you could do something back then is impressive, but would be extremely terrible and rough by modern standards.
FPGA Designer/Engineer here, thats a pretty spot on simplification of the whole topic
hey, if you don't mind me asking, how did you land your job? I am an Electronics Engineering student, and would love to work in this domain
@@allenqueen sure. Pretty simple: I applied for one and was accepted. Sidenote: pretty small company I currently work for, and I live in Germany
Your sponsor can help the people I get into arguments with on the internet.
This is like the difference between an all-in-one wireless routerswitchap, and a dedicated AP.
I've got an all in one 2400Mbps, and a 1300Mbps AP, the 1300Mbps AP, with internal antennas, is about 8x faster than the all in one, but it can only do the one thing.
Some camera like from Blackmagic also use FPGA and can be upgraded with more function in the feature.
Them problem with minner is they really one if sometimes minning is not good, they just resell the device easly. ASIC is not easy to resell.
Thanks for such quick explanation!
Love seeing more stuff about FPGAs..I got into them heavily because MiSTer a few years ago. I am learning various HDLs now and I love it..
I love how a ltt video gets a ad with Linus in it
Thanks for these helpful informations LMG
what? no !!!
As Electronics Engineer in Microelectronics I can clearly say that these information are wrong. To say, there is FPGA, ASICS and CPU is totally wrong information.
A CPU is just a type of a device, such as GPU, or even 555. It is not a category of hardware as the other two.
ASICS is the next step after prototyping with a FPGA, so you cannot change it. A CPU can be made with an FPGA, You can build a softcore in an FPGA, as well in an ASICS device. So, how the categorization in the video is made is very weird. I am really shocked !
Does that mean that we can make modular CPUs and GPUs with FPGA?
@@jierenzheng7670 yes of course. It iw how we prototype things like that. In fact, you can find cores of old CPUs available on line on opencores and use directly. You can also find miners and other things.
ASICS isn't one I've heard much about, but it does sound interesting. FPGAs on the other hand have defiantly gotten my attention lately with all the retro gaming mods and FPGA clone consoles that add TONS of new features that would be seen only on new game consoles while also keep the original experience intact, what a world we live in.
Switches are absolutely reliant on ASICs to run at wire speed.
FPGAs engineers... Know the big real engineers of ASICs, where the word "digital" is only a whisper.
FPGAs are also used to develop and test digital designs (like CPUs, MCUs, peripherals, etc.) before they are sent to the fab for printing on silicon.
Great script Jon!
Wonderful capability video, love it!
Much better than your previous attempts with this topic.
A TechQuickie video should have the word of its topic in the title
3:29 lol 🤣🤣😂 sounded like Linus has some electrical remanufacturing at that time stamp
Great! Thanks guys! 💜
This is the first time FPGAs have been explained in a way that I actually understand. Excellent video!
I think I want see Anthony get deep and dark with all the FPGA retro gaming now.
Do microcontroller vs microprocessor next please
The history of the naming can create confusion, but "microprocessor" just means CPU today (the micro part used to mean something when computers were the size of buildings, but not since everything scaled down to the chip level - like in the past 30 years).
So really it's microcontroller vs process.
A microcontroller is essentially a light weight CPU for embedded applications (toasters, dishwashers, etc.). Since they are for embedded applications, they're more like a system on chip with memory, "storage", and I/O on the same chip that is soldered to a board. One of their main staples is that they usually don't run a traditional OS, and only run one specialized application. They also are much slower but in exchange use less power and cost a lot less (you can get some microcontrollers for $0.20, that only use 0.01W, but runs at less than 1 MHz and has 128 bytes of memory).
Microprocessor: A CPU. Needs a bunch of stuff outside to be added (storage, memory, IO, etc.).
Microcontroller: A CPU with almost everything that you need built-into it. You can easily have a "one chip" system where a microcontroller can have the code on-board to blink an LED or send and receive data over a serial port with practically no other parts. You find those in things like washing machines, microwaves, remote controls, etc. And because you have to build all of that stuff into one die, they are lower speed/capability and a separate CPU. But they are useful for a LOT of stuff.
You're a good-looking chap, Linus. Don't put yourself down
Next video should be about hls (high level synthesis) or how to port c code to fpga
I' m just starting to learn Verilog!
This video made me think of what may be a strange question: whatever happened to math coprocessors?
I can't wait for AMD Ryzen to include FPGA ever since they successfully acquired Xilinx. Since FPGAs are very popular for seamless gaming emulation, maybe FPGA is the key to finally emulate the PS3, the hardest console to flawlessly emulate. With Ryzen and FPGA, development for RPCS3 will be done in a jiff.
There are cloud servers today with FPGA boards that can act as a specific CPU or a GPU, depending on the demand. That´s amazing.
Soon AI will be programming FPGAs and new crazy ICs will be created or optimized.
Have you heard of that new adaptive transistor technology Linus? With this transistors can change the architecture by changing their direction electrically. That mean a zen 3 core can be changed to RDNA2 core or even next gen IP like Zen4. P cores can be changed to E cores. This things takes specialised chips one step further. Even ahead of modern CPU. This can literally save silicon waste.
0:51
Linus: Regular processors ....
Also Linus : Shows us a Threadripper inserted in Zentih extreme MB ...
That's the level of the game they're playing on :D
To summarize: CPUs run software telling them what to do and can be reprogrammed. FPGAs run software (bitstream) telling them what to be and they can be reprogrammed into another hardware configuration. ASICs are mask-programmed at the factory for a specific hardware config, but cannot be changed afterwards. On another note, FPGAs also have the advantage of dealing with handling data in true parallel fashion.
What did Video Toasters use? Those were definitely ahead of their time and probably custom. ❤️❤️❤️
I am Jack's confusion...
What is that? Hotter, fatter, cooter? Drizzled in butter? IDK this language...
The Newtek chips were full custom ASICs. Incredible product, sadly the founder of the company passed away unexpectedly and they lost momentum. I think they are still around in some manner although now that everything is digital their original product is not needed. Comedian Dana Carvey's (Wayne's World) brother had a lot to do with that story.
@@dennisfahey2379 I was a big fan of it, in the late 90's for me, it was definitely the goto especially for live video feeds; Premier was for editing.
@@LexFelron - A milestone product. It and the Amiga opened the home video market. Lucasfilm's Editdroid and Pixar brought Digital Post and CGI into the lexicon. Premier (and almost all desktop editors) are based on Editdroid's look and feel.
Are the retro consoles having reviews on the main channel? it would be pretty cool
I wish we had dedicatated ASICS for raytracing.
Would you be kind enough to upload a video explaining current Nvidia and Radeon gpus naming / numbering convention?
After two years, I have come back to this video.
0:18 What the hell was THAT!?
Could Linus have done this video without his arms?
Something worth noting about FPGAs is that there are also fuse-based FPGAs which are only able to be programmed once in the field because fuses are used to permanently sever a connections that are not needed. These are good to have because unlike reprogrammable FPGAs that have to reset all of its gates every time it loses power, the fuse-based FPGAs are immediately ready when power is restored.
Are they still making those? I can't imagine the demand is too big, since you can't update them. I do remember using an Actel ProAsic years ago: an FPGA with flash built-in so that it is ready to go immediately without having to add a serial EEPROM. But flash and fuse-based FPGAs seem to be in the extreme minority.
Took me 35 hours of programing in Quartus (System Verilog HDL) to write a program that draws a line onto a DE1_SoC Development board (I'm studying Electrical Engineering in college). All of that work just to draw and clear a line onto a VGA output. Ugh that is the struggle with FPGA's they are a pain to code bc everything runs at the same time (unless you govern it by a clock, but still everything inside the always_ff @(posedge clk) still runs simultaneously.
If you want your code to run sequentially you would use a state machine, really just a case statement in a clocked process. You could also use a counter and conditional statements inside a clocked process to decide when lines of code run. But really you do want as many functions as possible to run in parallel without being blocked or having to wait, so pipelining and dataflow is very important.
I can't believe I lived to see an Innovus screenshot on an LMG video 😂
Well we've hit the end of times... Grammarly is sponsoring LTT videos...
What about a CPLD?
Interesting (to me) is that in the old days, I used to use a programming language called Asic. It was a free sub-set of Basic. Get it?
doritos still are the best chips
No blue lays are best 🤡
@@realhkv4082 blue?
@@spiddyman0079 i think he/she means ketchup lays
@@DeeezNuts but thats red
@@spiddyman0079 Im pretty sure ketchup lays is blue
i would say that you are wrong on this one guys. A cpu IS in fact, an asic. It has static non-reroutable logic gates as well and while they can do a lot of tasks, each task is done through an instruction set command that doesnt interfere with the logic gate connections, just like how the rest of the asic chips do. An FPGA chip though, can reroute its logic gates to simulate a custom processor or a completely different system, granted there are enough logic games.
This was quite informative.
I just bought my first FPGA development board
He survived the Mentos with pepsi, but had it been Pop Rocks and Canada Dry...
We wouldn't have Linus Gabriel Sebastian to kick around anymore.
R.I.P. Mikey
What's funny is the zilog z80 in the Sega Game Gear is called the 1 ASIC or 2 ASIC as the CPUs on them were very specialized
Good Information 👍🏻
I f'ing love FPGAs. Going to be receiving a vcu128 board soon
I tend to skip through your sponsor segments, but I really like that you have chosen to work with a Ukrainian company like Grammarly right now. Watched it end-to-end.
Comming in clutch to help me reviso for my exams xD NICE
Thats the difference between microprocessor and microcontroller. They are (mostly) made to handle one set of instructions that never change in the lifetime (or moderate upgrades thru firmware)
I need some cool spinning wheel that changes it from gaming to regular, that would be cool
Maybe off topic , but might this explain why Google tensor didn't do as well in machine learning benchmarks versus apples bionic chip. My thought process here is, Google's tensor processor is partially an ASIC processor for specific machine learning tasks, for a specific Google features, which might explain the very large gap? Or is the bionic rly just that much better? Any computer science majors have any input???
Really nobody else died at the kid with legos? That shit killed me
4:47 to CHOOOS
"Like that time I washed down a mentos with a bottle of pepsi." Loved it.
Really interesting stuff! 🤩
what if you make a mega cpu with both of the chips in the video
Asics are what the big cryoto mining companies are using.
ASICs are what YOU are using. If you have a cell phone, there are ASICs in there for the radio stuff (WiFi, Cell, Bluetooth, etc.). If you have a computer, there are ASICs in there for such things as power management, clock control, etc.. Really, ASIC just means any piece of silicon designed to do a specific job (IE not a CPU, and not discrete logic). FPGA is a chip that does nothing, but you can program it to do any job that an ASIC can do.
…Don't forget YLOD ( P h a t t ) PS3 (which can emulate PS2 using hardware accelerator), Apple Silicon, and Google Tensor
What are AG NG microchips and what do they do? Please and thank you
You could also call FPGAs EEPASIC
I thought this video was about the Florida professional Golf association
I wonder what kind of processor are in cameras. Cannon Sony Fuji etc they all mention to have powerful new processor that we have no specs to compare.