Then you could have programs dynamically create hardware depending on their needs.
Name:
Anonymous2016-01-30 18:22
There's supposed to be some Xeons coming out that will take a hybrid approach to this.
Name:
Anonymous2016-01-31 1:10
FPGAs are always slower, take more power/generate more heat, take more silicon real estate, and are far more expensive than static chips.
Name:
Anonymous2016-01-31 7:26
>>3 Those things will improve as the market grows.
Name:
Anonymous2016-01-31 10:10
>>4 Improve but never reach the efficiency of static chips.
Name:
Anonymous2016-01-31 11:00
>>5 If you pour enough money into something, you can make it as efficient as you like. Intel proved that.
Name:
Anonymous2016-01-31 12:09
>>6 A bunch of money was poured into Lisp and AI but the result had terrible efficiency. Your argument is invalid.
Name:
Anonymous2016-01-31 12:09
>>6 No you can't. FPGAs by definition have to flip more variable electrical state to perform the same work as static chips. They will always have to travel longer traces.
This is like comparing interpreters to compilers and saying that interpreters will catch up. You're just stupid.
Name:
Anonymous2016-01-31 12:43
>>7 A bunch of money was poured into Lisp and AI. The result was hella fast compared to everything else, but the notion of symbolic AI was weak and didn't go anywhere.
>>12 What's the point of that? Fortran does't even have an amb operator.
Name:
Anonymous2016-02-01 1:16
Forth is fun.
Name:
Anonymous2016-02-01 1:36
I want everything to be sandwiches. Unlimited sandwiches works.
Name:
Anonymous2016-02-04 2:14
>>1 All high performance computing will be slow if it ran on FPGA computers.
Name:
Anonymous2016-02-04 2:59
>>16 Software running on the best high performance general purpose hardware is almost always beaten by even crappy FPGAs.
Name:
Anonymous2016-02-04 5:52
>>17 Citation needed. Let's say I am a video producer and I edit videos and audio all day long. I sincerely doubt that there is a FPGA computer working today that performs this kind of task more efficiently* than the standard general purpose desktop computers today.
* Pick any metric you want and I still think that FPGA computers will be slower
Name:
Anonymous2016-02-05 1:09
>>18 Look at the speed for bitcoin miners that were running on the highest end graphics cards compared to the FPGA implementation. It's similar to the efficiency gains of OpenGL 2.0's shaders over OpenGL 1.0's fixed pipeline.
Name:
Anonymous2016-02-05 4:49
Not everything is amenable to being partitioned into a fixed array of gates. If you're doing something that involves a lot of data transformations (sorry, the speed of light is finite so mutable data will be with us forever) you can do much much worse than random access memory.
Maybe one day we'll have hardware that is plastic enough that you will be able to transform it on microsecond time scales, but we aren't there yet.
Name:
Anonymous2016-02-05 18:01
>>20 It's all gates at the end of things, and FPGAs can be given DMA to main memory anyway.
Name:
Anonymous2016-02-06 5:55
FPGA JIT
Name:
Anonymous2016-02-07 3:21
>>21 FPGA gates are not arranged in a fashion that's amenable to running at high speeds or with low power consumption. If you're going to be limited by main memory bandwidth anyway you might as well use a programmable ASIC and save some watts.