CPU vs FPGA

[linz]

teaching is an art form, visual is great for this kind of internal working

[jogara]

Very nice! From functional point of view, it would be also worth to show other things as well, like power consumption, clk, etc.

[dominicboyle]

It should also be noted that the FPGA architecture presented here can be further paralellized to reduce latency and permit higher data throughputs.

[magnuswootton]

cool visualization! i love machines!!!

[qrjftvx]

would like to see other comparison heck from network to ai chips

[Tigrou]

What is shown in this video (for the FPGA part) is also true for an ASIC.
The title "CPU vs ASIC" or "CPU custom silicon logic" would fit better.

[aivsdeveloper]

Which program did you use to create those animations?

[ClaymenSA]

What does a dedicated AI processor look like?

[davidthacher]

It is important to take note of a few things here. (All of these are mitigated with Algorithms and/or Computer System Architecture. There is so much to this area it can be bitter/sweet.)

The hardware size is larger with the FPGA and represents a higher level of specialization. This creates a performance increase that can sometimes be very significant, but may come at a larger cost. The memory bandwidth required by the CPU is fairly high here and this may not be actually reflective of the real world. As performance of the computation increases the memory bandwidth actually bottlenecks both equally. It should be noted that most computations end up being compute or IO bound. FPGAs are very good at accelerating compute bound applications. However IO bound applications are indifferent.

FPGA logic makes the most sense when it will have a high utilization. Most of the logic used must contribute to something and frequently. This is not true for CPUs and allows CPUs to model larger control circuits with little effort and cost. CPUs however eventually stall on memory performance which generally represents the limit of control complexity. Creating complexity this large in FPGAs is generally cost prohibitive. The parallel nature of FPGAs generally gives them an advantage in performance. In some state machines this advantage is trivial.

FPGAs may have memory contention stalls like CPUs. Architecture is more rigid with CPUs, but not all FPGA designs spring for the all the extra hardware.

FPGAs are direct and generally struggle to change. CPUs are indirect and can actually be too fluid. This actually serves as a larger consequence of optimization. Which is why optimizations should be measured carefully.

[shaikon]

Nice animation - but the bottom right matrix in the FPGA section should be "Memory" instead of "Control".

[pmcate]

Looks cool but doesn't help me understand what is going on. Any suggestions on textbooks?

[Rodrigo-jdwg]

I would love to know what I'm looking at

[straightupup]

That's one helluva visual - powerpoint?

[uccoskun]

it would be better if there is some explanations...

[gummadillinarasimha]

hey can anyone please help with creating such animations anyone any softwares... application or anything for like electronics related any block diagram related software for my presentation thesis

[newman]

what should i focus on this video to understand difference?

[Aziqfajar]

haha red dots go FHWOOOOMP FHWOOOOMP FHWOOOOMP

[alirezagumaryan]

plase make an animation to compare GPU and CPU

[ionescuandrei]

yes yes I do undertand what is going on

[tsunamio]

Well, you're basically playing redstone Minecraft at this point.