FPGA based FFT - Could it work?

[drkirkby]

Originally Posted by kriesel

Try crawling before sprinting or flying. One TF kernel. Small ram requirements, and much less herculean than attempting FFT based PRP or P-1 implementation.
Or stop advocating someone else should do it.

I fail to see any reason not to suggest something is possibly worth looking at, that could form the basis of a student project. I am not saying anyone should do it, but it would not be inappropriate to look at for a degree level project.
The fact I don't wish to do it myself is largely irrelevant. I know where my skills lie, and it is not in digital logic design.

[kriesel]

Or do some online research as to what's already been done. At a glance, most student fpga fft projects shown are 8-32 point but there is a 1024-point mentioned in the first page of results for https://duckduckgo.com/?t=ffab&q=stu...ga+fft+project which took ~5 seconds to type a search string for and get results.
GIMPS wavefront size operands for LL, PRP, or P-1 require far larger fft sizes, ~6M words.

TF for the wavefront fits in tiny memory by comparison, with individual operands <12 bytes wide, although mfaktx show speed advantages with large sieve sizes ~2Gbits for fast gpus.

One of the issues with such projects for hobbyists is competition for time to complete it., both hardware and software to use it, versus the passage of time before the implementation is obsolete.

If FPGAs were advantageous, why wouldn't the leading GPU manufacturers use them? http://www.differencebetween.net/tec...-gpu-and-fpga/
A talk on FPGAs and GPUs from ~2 years ago, 39:13 run time https://www.youtube.com/watch?v=MO2Hxxxy6l4

[Uncwilly]

Similar threads:
https://www.mersenneforum.org/showth...highlight=FPGA
https://www.mersenneforum.org/showth...8281#post58281
https://www.mersenneforum.org/showth...highlight=FPGA
https://www.mersenneforum.org/showth...073#post505073

[drkirkby]

Originally Posted by GP2

Amazon AWS has just announced a new "F1" cloud instance type (FPGA). It is available in developer preview in the us-east-1 (N. Virginia) region.

More details here.

The specs say these are Xilinx UltraScale+ VU9P FPGAs

The last paragraph says

"As I mentioned earlier, we are launching in developer preview form today in the US East (N. Virginia) Region (we will expand to multiple regions when the instances become generally available in early 2017). "

which is 4-years ago.

It's an interesting concept that one can develop an FPGA application, then sell it on the Amazon marketplace. Plenty of people buy AWS credits for the compute engines. If an FPGA version was cheaper, someone could potentially sell that to GIMPS enthusiasts and potentially make money from it.

[retina]

Originally Posted by drkirkby

If an FPGA version was cheaper, someone could potentially sell that to GIMPS enthusiasts and potentially make money from it.

Go for it. More doing, less wondering.

[drkirkby]

Originally Posted by retina

Go for it. More doing, less wondering.

In case you did not notice, I did write above that I knew where my skills lied, and they were not in this area.

[retina]

Originally Posted by drkirkby

In case you did not notice, I did write above that I knew where my skills lied, and they were not in this area.

What is stopping you from learning new skills?

Or maybe you just trying to "encourage" someone else to work on your desires?

Also: https://www.britannica.com/story/lay...o-we-use-which

The past tense of lie (as in, to tell an untruth) is lied.
The past tense of lie is lay, but the past tense of lay is laid.

[kriesel]

Originally Posted by drkirkby

If an FPGA version was cheaper, someone could potentially sell that to GIMPS enthusiasts and potentially make money from it.

Big If there. And how much did you pay for CentOS, mprime, and gpuowl?

Originally Posted by drkirkby

In case you did not notice, I did write above that I knew where my skills lied, and they were not in this area.

Using "above" here to refer to an earlier post in an entirely different thread is not normal usage. Perhaps I'm way off base, but when I see "above" in that context, I look earlier in the SAME thread. That may even be too generous. Dictionary definitions of its use that way could be interpreted to require in the same POST.(See for example adv 5 or n 1 of https://www.thefreedictionary.com/above)
Necroposting here on a 3+year-dormant thread, and referencing a different post in a different thread https://mersenneforum.org/showpost.p...0&postcount=22 " I know where my skills lie, and it is not in digital logic design." does not fit these definitions of "above".

You ask questions, and then debate, with those who are well qualified to respond because their career history is relevant, despite your admitted lack of relevant qualification, experience, skill set, etc. Not good. Keep it up, and people will learn to ignore your questions, or add you to their ignore list to automate it.

[kriesel]

I've recently created a reference post on the FPGA for GIMPS use question. https://www.mersenneforum.org/showth...246#post577246


Would a mod please close this old thread or the other https://mersenneforum.org/showthread.php?t=26749?

[kriesel]

I've created a reference post re potential GIMPS use of FPGAs at https://www.mersenneforum.org/showpo...6&postcount=22.
Constructive comments are welcome in the reference post discussion thread https://www.mersenneforum.org/showthread.php?t=23383.

[chalsall]

Originally Posted by kriesel

I've created a reference post re potential GIMPS use of FPGAs at...

If I may please blow some sunshine...

This kind of collation and curation is underappreciated but very valuable.