What CAN this be used for?

[Blacklotus84]

Vortex - Intel Agilex FPGA PCIe Accelerator

https://coloradoengineering.com/products/vortex-intel-agilex-fpga-pcie-accelerator/

[Eigenvektor]

It has a field-programmable gate array (FPGA). As such I would assume it accelerates whatever you program it to accelerate. Naturally whatever software you're using that you want to accelerate would then have to be adapted to run that function on the card rather than using the CPU.

[Blacklotus84]

I had a theory that it was a sorta math co- processor silly me

[Eigenvektor]

2 minutes ago, Blacklotus84 said:

I had a theory that it was a sorta math co- processor silly me

In a sense it is. Let's say you have a use case where

 

- the CPU is too slow to handle it

- the GPU is unsuitable (i.e. your code has lots of branches and is not massively parallel)

- designing a custom hardware accelerator is cost prohibitive

 

Then you could buy such a card, program its FPGA for your use case then adapt your software to use it as an accelerator. You could probably also use it to design/validate a hardware accelerator before going into mass production of the actual IC.

[Blacklotus84]

So not as a general application accelerator for folding or games eh?

[Eigenvektor]

4 minutes ago, Blacklotus84 said:

So not as a general application accelerator for folding or games eh?

At least not without programming it as an accelerator for this use case and updating F@H or games to use said accelerator (and possibly writing some drivers too)

 

However both of these use cases already have custom silicon (the GPU) that is faster and less power hungry than an FPGA would be, provided it contained the same function blocks.

 

Graphics and folding are massively parallel, so the GPU is an ideal accelerator (same with physics and AI). This is more for edge cases where the GPU doesn't make sense. An FPGA programmed for a specific use case will be faster than the CPU, because it can then do something in a single step where the CPU would take multiple. But it will generally be slower and require more power than dedicated hardware. https://hardwarebee.com/fpga-vs-cpu-difference/

[Roswell]

21 minutes ago, Blacklotus84 said:

So not as a general application accelerator for folding or games eh?

FPGAs can do both, much... much more efficiently than a general purpose CPU or GPU. The thing is, you need to either find someone who has programmed a specific FPGA to do the task you want AND purchase/use the same FPGA that they did or you need to program it yourself.

 

4 minutes ago, Eigenvektor said:

However both of these use cases already have custom silicon (the GPU) that is faster and less power hungry than an FPGA would be, provided it contained the same function blocks.

FPGAs are vastly more power efficient than GPUs for a given task. There's a reason why they're so wildly popular in the mining world and people bend over backwards to protect the configuration and sell it for exorbitant amounts of money.

[Eigenvektor]

5 minutes ago, Roswell said:

FPGAs are vastly more power efficient than GPUs for a given task. 

What I meant was if you used an FPGA to build a generic purpose GPU it would use more power, because it would need the same flexibility (e.g. programmable shaders). If you programmed an FPGA to accelerate one particular game it should be more power efficient, but then it wouldn't work for anything else.

[Roswell]

8 minutes ago, Eigenvektor said:

What I meant was if you used an FPGA to build a generic purpose GPU it would use more power, because it would need the same flexibility (e.g. programmable shaders). If you programmed an FPGA to accelerate one particular game it should be more power efficient, but then it wouldn't work for anything else.

Ah ok, I misunderstood. That makes sense.