Hardware design feedback

Re-post of a reply in another thread - I think it should be seen by the forum admins.

To all at nVidia:

We understand that we as the designers are responsible for our own products - this doesn’t stop other companies that offer modules like this from providing some guidance or suggestions based on their own experience in making their evaluation kits.

Altera/Intel has a PDN tool that will literally tell you if you have reached your target impedance in providing power to their SoC.

Other developers of SoM carrier modules will often put notes in the carrier design guide like:

This net includes a 0.1uF cap. Add the bulk capacitance necessary to meet the USB specification 
for your device application. This will generally be 1uF to 100uF and is specific to the mode of operation 
for the USB endpoint.


Switching supplies are recommended as the best design practice. These reduce wasted power and
result in a better product design. As temperature increases in a design, the cooling requirements
become more onerous. Higher temperatures also increase part failures and reduce the MTBF of a
design.

You’ve literally offered zero guidance at all - no specs on ripple, on module capacitance, even the sequencing and power-down information is nearly incoherent.

Then when folks come here and ask for help you just say ‘look at the eval board’ which by the way, ‘is not intended to be a product’.

So if the actual designers of the module can’t offer guidance or insight as to how to make a robust/rugged product with their own product, how are we supposed to have a better idea?

You are literally offering no support to basic hardware questions that any seasoned engineer would ask, and I myself asked and gave up when working directly with you.

These will be as successful as you are willing to support designers in using them, or people will give up and find a company that works harder to make designers have an easier time designing a very complex module into what is almost assuredly a more complex product.

Xilinx has like 10,000 pages of documentation on a single MPSoC family - you’ve got like 120 pages on designing hardware around this SoM. I hope you understand how shockingly light your support and documentation are.

I think NVIDIA is used to dealing with:

  1. Game developers, who use high-level drivers on major OS-es to optimize specific workloads.
    and
  2. Hardware OEMs, who start with a basic reference design, but then add their own marketable “twist,” and build hundreds of thousands of cards for the retail channel.

The Jetson line is a different beast, and I don’t think NVIDIA has (yet?) trained the katas and muscles needed to provide support in that area. OR they hope that they’ll snag a few big OEMs that they can funnel into their existing OEM support channel, which is more relationship based. (And also under NDA.)

I’m sorry, but I got the same feeling.
After trying to ask several questions, I’ve got answers like ‘we already answered this’.
Even if this is true, this is not a way to treat customers.

I’ve been supported National Semiconductors customers, in the PC area, and the motive was ‘Customer delight’ - do beyond customers expectation.

I’m doing contract design of a carrier board for a customer, who have selected Nvidia.
I don’t think that I’ll select Nvidia for next designs

WAKE UP - SUPPORT IS IMPORTANT !!!

Sure, reasonable explanation - but not an excuse to provide sub-standard support on a very complex module being designed into even more complex hosts.

Compare to the Raspberry Pi.

The Raspberry Pi support forums aren’t really any more detailed than the Jetson forums. In many ways, they provide less support and information, because there’s no remote debugger for the VideoCore, and no kernel build scripts, and no technical reference manual. What support exists for the Pi, often comes from the community and various blogs, rather than Broadcom.

It’s been my experience that, with complex leading-edge systems like these, you have to educate yourself, and go through the learning curve yourself. Because you’re at the leading edge, nobody has gone before you to beat a path you can follow.

Would I love it if NVIDIA support reps read the posts a little more carefully, and didn’t post with Python answers to C++ questions? Yes.

I think the support we get, is much better than you’d get for, say, an Intel system with EFI firmware, northbridge, CPU and integrated graphics. Intel is known for releasing those kinds of systems, then not providing enough information, and then killing the product after a year – has happened more than once!
If you have a real project that will make real money, NVIDIA will be happy to engage with you on a one-on-one basis and give you better contact with the product and engineering team on the inside.

I understand your point, but this is not the Raspberry Pi - it doesn’t cost $50, and I don’t expect the same level of support from a $50 hobbyist product that I do from a $700 OEM product.

No one is asking for them to do our jobs for us, simple guidance.

And if intel gives terrible support, that doesn’t become an excuse or reason to follow suit.

And to be clear, my company does have an NDA with them, this is a very real project with money behind it.

I am not dinking around with trying to see when my dog is in the room with 10 webcams. Legit application, serious project.

Sorry, as for the docs you are asking for, currently only the characters of capacitor is available to get from BOM file, in which the P/N and vendor are listed, customer can ask vendor for help accordingly. We are also collecting information from customers and partners to optimize our docs, thanks for your questions, i believe there will be related update once available.

The point is not looking up ESR/ESL values, the Intel PDN tool takes into account your board size, stackup, and your use of their SoC or FPGA, and then gives you a representative plot of the theoretical impedance of your power plane over frequency, and what your target should be given your use of their part. I’m sure this started as an internal document for them, and then was turned into a customer facing document because it was so useful.

https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/ug/ug_pdn_gen_device.pdf

External Media

This is super helpful for designers, and a great example of providing guidance based on tribal knowledge that your customers can’t be expected to have.

I don’t know why you are all acting like this is some novel idea - it’s really quite patronizing.

We have no idea how the internal support organization of NVIDIA works.

For all we know, the support engineers have a background in supporting end users trying to make games run on NVIDIA-based graphics cards? Or maybe they have supported Far East hardware manufacturers that clone the NVIDIA “development/reference” design and stamp out hundreds of thousands of cards to sell on Amazon and Newegg without further changes. In both those cases, they would have little experience with competitors tools for power impedance validation, and thus it would, actually, be a novel idea TO THEM.

Contrary to popular belief, companies are not monoliths, and what’s known in one part of the company may be an entire mystery to another part of the company. Couple this with the Jetson family being a totally new kind of business for NVIDIA, whereas Intel has literally been in the parts-for-OEMs business for half a century, and thus likely have built up better muscles for that kind of support.

I agree that the public face of NVIDIA Jetson support isn’t a great look, but I have a little more empathy with how they might have gotten themselves into this situation. I’m encouraged by this bit:

Now, here’s hoping they are able to turn these kinds of things around on the order of weeks-to-months, not on the order of “let’s make that a performance objective for the next business cycle planning session scheduled for 2021.”