Innovate FPGA Design Contest

[blacksheeplogic]

Thinking about entering this years Innovate FPGA Design Contest. Out of the two Altera Development board I thought the C5P would be the easier option which arrived today. It's an open contest in case anyone is interested. Innvoate FPGA Design Contest This years theme is AI.

[hamster_nz]

I am too old-skool for this contest. I have yet to come to terms with the non-deterministic nature of most current AI.

I really should put some effort in and get to grips with it...

[OwO]

Too bad it's restricted to Altera FPGAs. The Xilinx 7 series which are the same process node as Cyclone V are much much faster and so all my DSP code are now optimized for Xilinx. My FFT core which achieves 450MHz Fmax on the slowest Kintex-7 only achieves 200MHz on the fastest speed grade Cyclone V.

I do have a few project ideas (mostly around DSP) but I think I'm too lazy to deal with contests.

[NorthGuy]

$5K for the most innovative FPGA design. Is this all it's worth?

[SiliconWizard]

Too bad it's restricted to Altera FPGAs.

Yeah, this is basically an Intel contest. I wouldn't call it an "FPGA contest".

[coppice]

I am too old-skool for this contest. I have yet to come to terms with the non-deterministic nature of most current AI.

I really should put some effort in and get to grips with it...

Did you grow up in an alternate universe where there is no noise? Engineering in this universe, I've never encountered anything which is truly deterministic.

[Boscoe]

I am too old-skool for this contest. I have yet to come to terms with the non-deterministic nature of most current AI.

I really should put some effort in and get to grips with it...

Did you grow up in an alternate universe where there is no noise? Engineering in this universe, I've never encountered anything which is truly deterministic.

How do you know the noise isn't ultimately deterministic? Just because you can't comprehend it doesn't mean it's random. I'm not say it isn't random, just thought I would point that out.

[Boscoe]

Too bad it's restricted to Altera FPGAs. The Xilinx 7 series which are the same process node as Cyclone V are much much faster and so all my DSP code are now optimized for Xilinx. My FFT core which achieves 450MHz Fmax on the slowest Kintex-7 only achieves 200MHz on the fastest speed grade Cyclone V.

I do have a few project ideas (mostly around DSP) but I think I'm too lazy to deal with contests.

Looks like the cheapest (on Digikey) Kintex 7 part is 4 times more expensive than that of the Cyclone 5. Not sure how they compare on size though.

[dmills]

Cannot really compare a kintex 7 with a cyclone 5, the comparable part would be a spartan 7 IMHO.

Regards, Dan.

 

[colorado.rob]

Cyclone 5 and Artix are about comparable, as are Cyclone 5 SOC and Artix-based Zynq.  But any direct comparisons will be far from perfect since the product lines and range of capabilities differs between the two vendors.  You'd have to get down to the product number to make a reasonable comparison.

[OwO]

I did a comparison between Zynq-7010 (XC7Z010-1) and Cyclone V SoC (5CSEMA5F31C6N).
The Fmax is 380MHz for the Zynq (should be identical to Artix) and 200MHz on the Cyclone V (fastest speed grade). In my several years of experience working with both Altera and Xilinx parts it was consistent that Altera parts are slower, but only recently did I directly compare and get some numbers.

[blacksheeplogic]

I did a comparison between Zynq-7010 (XC7Z010-1) and Cyclone V SoC (5CSEMA5F31C6N).
The Fmax is 380MHz for the Zynq (should be identical to Artix) and 200MHz on the Cyclone V (fastest speed grade). In my several years of experience working with both Altera and Xilinx parts it was consistent that Altera parts are slower, but only recently did I directly compare and get some numbers.

I'm a lot more comfortable with the Altrera tool set/workflow/IP these days making a switch to Xilinx difficult.

[Bassman59]

I did a comparison between Zynq-7010 (XC7Z010-1) and Cyclone V SoC (5CSEMA5F31C6N).
The Fmax is 380MHz for the Zynq (should be identical to Artix) and 200MHz on the Cyclone V (fastest speed grade). In my several years of experience working with both Altera and Xilinx parts it was consistent that Altera parts are slower, but only recently did I directly compare and get some numbers.

I'm a lot more comfortable with the Altrera tool set/workflow/IP these days making a switch to Xilinx difficult.

This is often a very good reason to choose a device -- you already know the tools, so the time to get the design completed should be faster. And if you're not pushing the performance envelope, then whether the devices are made by Brand X or Brand A (errr, I) or Brand L or brand M might not matter. All things being equal, of course.

If the design requires a feature that one vendor has and the others do not, you choose the part you need and if you have to learn the toolset, you do. But we're professionals, right? Learning a new toolset shouldn't take very long.

If your production quantities are such that per-part cost is significant, then you choose the cheaper part and learn its tools.

Of course, learning the tools is one thing. Switching to a new device family, even from the same vendor, will require learning the details of that family and how to exploit the features.

[colorado.rob]

I am too old-skool for this contest. I have yet to come to terms with the non-deterministic nature of most current AI.

I really should put some effort in and get to grips with it...

Humans are non-deterministic.  If your goal is to replace a human, whether for flipping burgers, driving a car, or flying a plane, all you need to do is show that the AI is more reliable than having a human do it.  I can't train two humans to do the same thing the same way.  And you want me to train a computer to be more deterministic than that?

There are self-driving cars that are more capable than your average teenage driver.  Which would you rather share the road with?

[hamster_nz]

There are self-driving cars that are more capable than your average teenage driver.  Which would you rather share the road with?

I would rather share the road with one inexperienced driver and thirty-nine experienced ones, than with forty inexperienced ones.

(Assuming it takes a year to drive competently, and people drive for forty years, and an even age distribution).

So if a self-driving cars are as good as an inexperienced driver, then I want the inexperienced driver thanks!

[NorthGuy]

Humans are non-deterministic.  If your goal is to replace a human, whether for flipping burgers, driving a car, or flying a plane, all you need to do is show that the AI is more reliable than having a human do it.  I can't train two humans to do the same thing the same way.  And you want me to train a computer to be more deterministic than that?

There are self-driving cars that are more capable than your average teenage driver.  Which would you rather share the road with?

AI is not up to par with humans so far. AI models neural networks which are found in human brain, as well as brains of other animals. Basically, AI is working on learning, developing  reflexes. Humans differ from animals in that they're capable of logical thinking while animals are not. If you relay this to AI, it's more like an animal than a human.

Of course, with time, AI may develop to the intelligence levels surpassing humans, but we will not know it. Just as a dog cannot comprehend human actions driven by logical thinking, a human will not comprehend the behaviour of superior AI. At some point it'll be too late. For the superior intelligence, the humans will be inferior beings, and it won't care much about humans, as we don't care about cows or sheep. Human dominance on the planet will be over and will never return. I think it's 20-30 years from now ...

[colorado.rob]

AI is not up to par with humans so far.

That's a bit imprecise.  Some AI is far better than humans at very task-specific functions.  And some are good a very broad topics that no one human can do.

Google Translate can do far more with it's AI translating human languages than any one human can.  Sure, some humans can do much better at specific languages.  I have yet to meet anyone that speaks as many human languages as the Google AI.  And I don't have access to a translator every time I click on a foreign website (or the money to pay for them).

What AI gives us is broad access to their abilities -- that either I do not have or want to learn.  My options are "Google Translate" or "go study Chinese for a year so I can sign up for a Taobao account."  Google Translate, please!