next »

[gimymblert]

OKAY
this one is technically more a though experiment, just asking some people who might no a bit more  

Would it be possible to compile a prog directly to electronic schematic rather than machine code for a specific "generic" processor? I understand the irony as language tend to be based on idea about computer architecture, and code are made with those in mind, but let's ignore that point and think about the implication of this idea ... what would that mean in practice?

[Boreal]

You're basically describing an ASIC (application-specific integrated circuit) or FPGA (field-programmable [logic] gate array) here.

http://en.wikipedia.org/wiki/Application-specific_integrated_circuit
http://en.wikipedia.org/wiki/Field-programmable_gate_array

First you would use an FPGA to implement the architecture at a small/slow scale and then use that to build up your own ASIC.

[gimymblert]

THANKS That's interesting! I assume it could exist.

I looking at things like the rasPI and how cheap it is, so much it might be profitable to sell game directly in a boxed pi for a price similar to commercial game (60€). So with 3D printing and the maker culture taking off in hobbyist circle, I was wondering what would be the next step? What if creating chip became cheap enough for hobbyist?
https://www.adafruit.com/blog/2014/08/07/parallax-propeller-1-p8x32a-is-open-source-oshw-parallaxinc/
I found this today but I haven't look at it yet

Imagine compiling a game directly to circuit what would be the implication?

[jgrams]

Imagine compiling a game directly to circuit what would be the implication?

Like this? http://arcade.gadgetfactory.net/

The implications are mostly about performance. So in the early days (Spacewar, Pong, etc.) you generally built games in custom hardware. I have a bunch of electronics books from the 80s which have projects like that.

But as hardware has gotten more powerful, those things tend to fall by the wayside and people just build things on general purpose hardware. There's an interesting "intro to rendering" talk by John Carmack from a few years back where he says that several times in answer to people's questions about tricky rendering techniques.

So...I don't see a particular application, except for fun/learning.

Oh, and maybe custom controllers and such. But that's much more about building the mechanical stuff than the electronics.

[gimymblert]

I have look to inform me on the matter, and yes it seems it's rather specializing in prototyping how when the environment is too unstable (whatever that mean) but performance is not important, microsoft have adopt it for its server tech (

).

How we might also be on the verge of a revolution too apparently, it has the convenience of gpu (parallelism) and cpu (sequence) through reconfiguration "scaling" . Most importantly advance in optimization of compiling code to core partition was key. But the real future is in hybrid, we are already seeing gpu-cpu hybrid, fgpa hybrid also appeared, there might be a convergence.

I can see in game a re configurable part akin to the cell in the ps3 a generic multicore for common operation and fgpa for making game specific architecture, because a game like minecraft don't exactly have the same need than a game like call of duty. FGPA could reconfigure itself to handle massive array of data (parallelism) or complex ai (branching logic) or even visual processing (kinect, pseye). Am I correct? I'm no engineer!

[Boreal]

I'd say it's possible but I doubt that kind of approach will beat a good old AMD HSA chip, at least for the time being.  Especially now that GCC 5 is about to come out with a whole new parallel processing capability.

[gimymblert]

I don't know the real difference with fgpa but isn't the use of microcode a way to have programmable hardware?

The n64's RSP was programmable through microcode, so was the ps2 VU0 and VU1, and it helpt performance immensely over the life course of the console (doubling resolution, framerate, 5x the tri throughput for the n64, at least factor5's games) at the cost of engineer knowledge.

HSA seem to work with a better memory architecture to remove bottleneck (by sharing the same space) it does seems to give the actual advantage of FGPA type of architecture (which HPRC architecture embraced) , ie you still have to design algorithm around the hardware rather the opposite. Some people are advocating for more micro code to allow more efficient high level language (

around 9:00).

But yeah, as far I understand, hsa favor tradition and the work built on it, while FGPA might break that and have less knowledge base built around it.

At least that's what I understand from reading/viewing a few things on internet XD ...

[Quarry]

Practically an advanced Game & Watch?

[gimymblert]

http://danstrother.com/fpga-nes/

[gimymblert]

FOR FUN (I don't plan anything and I close my curiosity now):

It seems 2013 was a turning point where price fell through the roof culminating to a series of hobbyist FPGA board.

https://www.sparkfun.com/news/1203
go to the comments!

http://www.edn.com/electronics-products/electronic-product-reviews/fpga-boards-under--100--hands-on-reviews-collection/4431752/FPGA-boards-under--100--Introduction
http://www.xess.com/static/media/appnotes/FpgasNowWhatBook.pdf
http://www.joelw.id.au/FPGA/CheapFPGADevelopmentBoards

[gimymblert]

About Cpu + gpu + FPGA
http://www.vision-systems.com/articles/print/volume-14/issue-2/technology-trends/image-processing-multiprocessing-system-supports-fpgas-gpus-and-cpus.html

http://www.theinquirer.net/inquirer/news/2255611/nvidia-says-fpgas-choke-video-processing-performance

http://stackoverflow.com/questions/317731/cuda-vs-fpga

http://forums.xilinx.com/t5/Embedded-Development-Tools/GPUs-vs-FPGAs/td-p/60112

http://www.anandtech.com/show/7334/a-look-at-alteras-opencl-sdk-for-fpgas

Some date a bit and concern expressed have lowered

THIS TIME I'M DONE promise!
http://fpgablog.com/

[gimymblert]

[Quarry]

ARM?

[gimymblert]

I'm sorry
https://www.crowdsupply.com/kosagi/novena-open-laptop
But that's cool