Question about FPGA.

[Cidsor]

I am very little familiar with FPGA's, but I've read some articles on what they are and so on. If I understand it right, any GPU should be able to run as FPGA?

Now I know you can run 8 GPU's in a normal OS (with the exception of the few "quickfix" linux distro's), but could you potentially run e.g. 4x5970 on 4x PCI-ex x16 slots and 1x5850 as an FPGA on a x1 slot (or on any other slot for that matter)? And if it is possible, would the FPGA vs GPU mean a loss of computing power when mining or gaming?



 

[JoelKatz]

If I understand it right, any GPU should be able to run as FPGA?

It's possible you mean something by this that makes sense, but I can't figure out what it is. An FPGA is a physical device that is no very much like a GPU at all. In principle, a GPU can act like an FPGA just as a CPU can act like a GPU, but it won't be very efficient. If you want to mine on a GPU, we already know how to do that without any "pretend to be something else" inefficiency.

[Cidsor]

If I understand it right, any GPU should be able to run as FPGA?

It's possible you mean something by this that makes sense, but I can't figure out what it is. An FPGA is a physical device that is no very much like a GPU at all. In principle, a GPU can act like an FPGA just as a CPU can act like a GPU, but it won't be very efficient. If you want to mine on a GPU, we already know how to do that without any "pretend to be something else" inefficiency.

Was exactly what I was wondering about, thanks for the answer

[Minor]

The funny thing is that the truth is exactly the reverse: an FPGA can actually be made to act as a GPU (or any other digital electronic circuit for that matter - aren't FPGAs the coolest thing?).

[monological]

FPGAs are programmed with synthesized hdl code that actually represents the logic that eventually gets turned into ASICs like the GPUs in all these graphic cards.

[monological]

In other words, imagine you wanted to test out an AND gate. You can program an fpga on the fly to represent an AND gate. An ASIC is fabricated and can only do what it was designed to do. So in this simple example, it could only ever act as an AND gate, because the logic ant be changed. This is a very simplistic answer. Hope it doesnt confuse more than it helps.

[Gabi]

Of course an ASIC is much faster than a FPGA, but to produce them you need to invest some millions of $$ for masks/whatelse (after the investment their price is low of course)

[jackmaninov]

Maybe it would cost millions to develop an ASIC using the latest process technology, but it should be significantly cheaper to build a large, multi-core, low clock-speed ASIC for hashing. It's not like you're trying to put these things in laptops (though low power usage will be better in the long run for miners).

[Gabi]

The expense is for the masks used to make the chips. They cost millions and if you want to make a ASIC you need them

[monological]

Even for micrometer process nodes, it's still in the hundreds of thousands of dollars, and it'll take forever to recover that amount and it's super high risk.

[deepceleron]

An FPGA is a Field-Programmable Gate Array.

FPGA : logic :: EPROM : memory.
FPGA : IC :: DVD-RW : DVD

It is, at it's core, a mesh of single transistor equivalents that can be "burned" into logic gate truth tables and circuits to represent any arbitrary digital logic device, including a microprocessor. Without the FPGA, we would have to make such devices out of discrete logic components, or produce a complete masked integrated circuit design, only practical for large production runs. The actual burning of thousands of gates is largely simplified by modern design software which includes prebuilt logic cores.

A CPU is a microprocessor, which is designed to execute a large variety of instructions, a "jack-of-all-trades". GPUs, however, are designed to be very good at rendering graphics, which means they excel in certain types of parallel processed mathematics that are used in graphics rendering. Neither were designed with SHA256 hashing in mind, which is what mining needs, so a chip specifically designed to do only this can be even better. Designing and manufacturing a complete chip from scratch costs hundreds of thousands of dollars from application-specific integrated circuit fabs, so the FPGA is a compromise, in that it can be made into a specialized circuit optimized for one purpose in low production numbers, but the underlying device is not optimal for high speed efficient processing.