Gpu/cpu virtualization idea

[septic slayer]

so i had this really dumb thought. Is it possible to use a gpu like an NVIDIA tesla card and trick you hypervisor into think it is a cpu of some sort and then use that "cpu" (modified gpu) as a normal Intel or amd cpu in a vm along side a gpu that has been passed thru? 

if it is possible and someone has done it could you please add a link .

thanks in advance! 

[Radium_Angel]

no.

 

[septic slayer]

1 hour ago, Radium_Angel said:

no.

 

thanks 

[Radium_Angel]

Just now, septic slayer said:

thanks 

It's a neat idea, but sadly, no go, a GPU and CPU operate on an entirely different principle when it comes to capabilities.

 

[septic slayer]

1 hour ago, Radium_Angel said:

It's a neat idea, but sadly, no go, a GPU and CPU operate on an entirely different principle when it comes to capabilities.

 

yes i was doing research on it and found nothing so i came here jsut incase some body some where had the sam crazy idea and made it possible but sadly not

[Radium_Angel]

Just now, septic slayer said:

yes i was doing research on it and found nothing so i came here jsut incase some body some where had the sam crazy idea and made it possible but sadly not

Of course, you could be the 1st to whip up a program to make this happen!

You'll be richer than Bezos after that, and give everyone else reason to hate you as you extend the GPU shortage, but hey, money! 

[the gamer that is bad]

5 hours ago, septic slayer said:

yes i was doing research on it and found nothing so i came here jsut incase some body some where had the sam crazy idea and made it possible but sadly not

I have had the same question and I want to do it but I struggled to make a Minecraft server so that is way out of my skill set but if you get it working please let me know

[EscapementDog]

There are three kinds of processor architectures in this world that we use commonly.  CISC, RISC and Vector.  CPUs you put into desktops are CISC, ARMs used in phones are RISC, and GPUs are Vector. The three different processor types are strong with different kinds of software workloads. You can look up those architectures in wiki for more info and I think its really neat.  However, I'll give a brief explanation of what each is good at.

 

RISC - Risc is good at doing the most amount of computing with the least amount of circuits. Because it is less complex it can do things like run at low power and is generally just a lot cheaper to manufacture. Its limit is that it can't do nearly as much computing as the other two processors. Excitement about the Apple ARM aside, you'll find that per sq. mm, it can't do as much computation as the other two cpu architectures. By the way, this is a controversial statement STILL but you can see in practice RISC is more bang for the buck (whether the coinage is cost or power) but not the most BANG you can jam into a box.  

 

Vector - Vector processors compute MANY functions in parallel so they only good for computer problems that can be arranged in a way where you need to compute millions of the same function with just different parameters. Need to compute lighting for 3D model or hashes for crypto currency, its a great great solution because those workloads are basically the same math with just different numbers going in. Vectors are also kind of RISC like, only instead of reducing circuitry and power use they just do more computing.  It's kind of like taking 1000s of cheap phones and mashing them together on one chip only each phone can only run the SAME program, but just a tiny piece of a bigger thing. Like each one draws a single pixel that you put together to get a whole screen frame of minecraft.

 

CISC - It's the jack of all trades.  Inside a CISC processors are RISC and VECTOR processors. Most of the circuits in a CISC chip are about trying to transform many different functions with no relationship to each into workloads that are RISC-ish and Vector-ish looking. It looks ahead to see what's coming in the program, it will take code that could compute in 10 different ways and run all ways simultaneously and throw out the the 9 that weren't needed. It will try to take problems that are serial by design, and try to run some of them in parallel. It's a magic beast that tries to compute ANYTHING it gets, now matter how sloppy, as quickly as possible. 

 

So no, you can't run general computing on a GPU but that is okay.  We are in a world where our computing has gotten complex and these three processor types have evolved into their own paths from the old common ancestors.