Visions of what will be.

Join Date
Apr 2002
Location
No income tax, no capital gains tax. Freedom!
Posts
8,351
nVidia has bought Arm Holdings. Many of the more powerful PLCs already use Arm CPUs.


OK, so what?


PLCs execute ladder in sequence but it is a simply trying to imitate relay circuits like the ones I was trained on back in the 1970s are probably didn't change much in the previous 70 years before then.


Things are going to change. As you know nVidia makes graphics cards and specifically GPUs called CUDA cores. These allow computations in parallel and even do AI applications. The CUDA cores could effectively process a relay ladder circuit in parallel so the scan times are insignificant. This will affect motion controllers too. imagine being able to process many axes of motion in parallel. The only draw backs now are software and the heat generated by the GPUs.


You may think that the speed is not necessary but wait. People will find a use for it and the money will go to the ones who find a way.



nVidia is moving our cheese.
https://en.wikipedia.org/wiki/Who_Moved_My_Cheese?

I can think of how this will change the motion control industry. Distributed motion control will change. The need for smart remote devices will drop. There will still be a need for the distributed I/O through a network.



BTW, I bought a GTX 1080 with 2560 CUDA cores for playing Stockfish ,a chess program, on my computer. Human players can't beat Stockfish or Lela Chess 0.
 
Things are going to change. As you know nVidia makes graphics cards and specifically GPUs called CUDA cores. These allow computations in parallel and even do AI applications. The CUDA cores could effectively process a relay ladder circuit in parallel so the scan times are insignificant. This will affect motion controllers too. imagine being able to process many axes of motion in parallel. The only draw backs now are software and the heat generated by the GPUs.

You may think that the speed is not necessary but wait. People will find a use for it and the money will go to the ones who find a way.

People have wondered for years why PLCs are as slow as they are in comparison to PC based platforms, and that any day now things would be revolutionized. I'm not saying its a bad idea, it just seems like one of those things where surely if it had a benefit someone would have done it by now. I somehow can't quite picture a secret cabal of automation manufacturers actually getting Rockwell and Siemens to agree on ANYTHING.


I think running motion axes in parallel seems like the first good revolutionary (and idiot-proof) usage of multi-threading that I've heard of in an automation context. We've seen communications moved to a separate thread in Logix, and from what I hear that mostly seems to have just complicated things (asynchronous scan). On the Siemens side, the 1518MFP lets you run C/C++ code in parallel in a separate runtime in a separate core, but that is super niche. I feel like the average PLC guy would have so many race conditions if his code started getting multi-threaded.



Peter,
I feel like you would have more knowledge than most of what goes on behind the scenes with the product design. Is there a reason this hasn't been done so far? Obviously, confidential info stays confidential, but is it something like this acquisition suddenly bring the cost structure into a place where it makes sense, whereas it didn't before? Is keeping the chips reliably cool enough to maintain the expected lifespan the big limitation?
 
Is there a reason this hasn't been done so far?

I can tell you why in some of the industries I played in for many years.

Approvals, insurance, hardware ratings (temperature/power/surge withstand/MTBF/etc.), product life cycle, hardware track record and cost.

An approval cycle, TUV, ETL, regardless of the tester; testing to a standard, you choose which one, is a large cost. Altering that now approved system, depending on the change, can trigger a new round of partial or full testing.

If the hardware temperature operating range is too narrow, it will not be used.

If the hardware manufacture will not guarantee a many year production, it will not make the cut.

Lots of reasons. My2c.

 
Last edited:
I guess we can re visit this topic in a few years but my prediction is that there won't be much noticeable change for vast majority of us. the reason is still the same, for most industrial application, the speed is just not needed.

That doesn't mean there could be other, yet unknown development that could happen. I don't know... real time AI learning with minimal programming due to large and powerful software framework and such...
 
PLC's are alot about backwards compatability, if they want to keep the legacy market the suppliers cant go with parallell computing because old code wont behave the same.
I bet most suppliers use a "virtual machine" as an abstraction layer on the hardware that will translate the logic to bytecode for the real CPU.
 
The big thing is market inertia.
Harry is right about about a lot of applications don't require speed or they don't see the need for speed. For instance in servo hydraulic control there is no need for speed in most applications because the MDT rods don't update that fast. However, there are testing applications where high frequencies like 400-500Hz are desired. Right now the limitations are mechanical.

The power requirements are much greater. This requires bigger power supplies and fans just like for your PC. Our second generation motion controllers used only a little more than one watt. Our latest one is much more power but it is 7 watts and has a big aluminum heat sink.

Getting the UL, CSA, CE etc approvals is still a paperwork shuffle because the voltages are below 24v. It is extortion. The costs get past on to the customer. There was a thread a long time ago by Mike Granby of Red Lion complaining about certifications.

However, there is a GTX1080 graphics card in my PC NOW. The technology exists NOW and has been around for a few years. 10+ years ago I was at a wood panel show in Atlanta. I saw a scanner that used 4 or 5 video cards to process the scan data for imaging a log. That guy was ahead of his time.
https://deltamotion.com/peter/Videos/Daqota Scanner-desktop.mp4
Now this can probably done with one video card.

It could be that PLCs stay the same but the peripherals get the new technology.
 
The power requirements are much greater. This requires bigger power supplies and fans just like for your PC. Our second generation motion controllers used only a little more than one watt. Our latest one is much more power but it is 7 watts and has a big aluminum heat sink.


However, there is a GTX1080 graphics card in my PC NOW. The technology exists NOW and has been around for a few years.
Sure, but it's 180W, literally 25x more power.


I guess one of the things I wasn't thinking about before is that we desire passive cooling in industry, not the typical active cooling (2 fans built in) common in PC parts. A PC can get by with a "small" heat sink, because it has a bunch of fans. A PLC typically doesn't, although I have seen them in automation components before.
 
Backward compatibility as mentioned is one significant factor. Most of us are still using 4-20mA signal.

The one area where we did see some major progress is HMI where there are clear separation.
 
A PLC typically doesn't, although I have seen them in automation components before.

As someone that had custom made controllers and IO blocks that required a dedicated cooling fan (and had alarms for when it stopped) to maintain, I really don't want to go back to that. It sucked having to schedule some time to power down the IO block or controller and replace the little fan or risk having the part die on you sooner as well as the annoying fan alarm.
 
BTW, I bought a GTX 1080 with 2560 CUDA cores for playing Stockfish ,a chess program, on my computer. Human players can't beat Stockfish or Lela Chess 0.

If you can't beat Stockfish, was the purchase of the GTX 1080 justifiable ? Asking for a friend ....

But if they've got mega fast processors, or if your PCs have multiple cores, you could make them work for a living - take a look at FoldingAtHome (Don't be put off by the HUGE garing font on the landing page....). I've been folding with my various PCs for over 10 years, mainly for Cancer research, but obviously recent events have shifted the perspective ...

EDIT : FAH just pinches the CPU power when it is sitting in it's "idle" loop, waiting for something else to happen. Same technology as used on SETI, I believe.
 
Last edited:
If you can't beat Stockfish, was the purchase of the GTX 1080 justifiable ? Asking for a friend ....
Yes, Stockfish analyzes board positions faster and solves chess problems faster.
If I get motivated I can even download the CUDA development system and do some simulations or write some extremely fast system identification code.
CUDA cores can be used for neural nets too.
 

Similar Topics

Good day is there somewhere i can see the situation and compatibility regarding different firmware revisions. I have a 2711-K5A5, series H and...
Replies
4
Views
186
Today I was trying to install firmware update to new out of the box CompactLogix processor. CompactFLASH dialog box did not show any revisions...
Replies
10
Views
224
I am working for a client and currently installed on their computer studio 5000. they need to give me a backup file on their plc with rslogix 5000...
Replies
1
Views
405
I am working for a client and currently installed on their computer studio 5000. they need to give me a backup file on their plc with rslogix 5000...
Replies
10
Views
753
Hello, I have rslogix processors operating at anywhere from rev 16.0 to 30.0. I have the full option of FBD routine creation in the higher...
Replies
9
Views
2,788
Back
Top Bottom