Announcement

Collapse
No announcement yet.

NVIDIA Unveils $59 USD Raspberry Pi Competitor With Jetson Nano 2GB

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Originally posted by pkese View Post
    The problem with these specific Tegras is that A57 at 1.43 GHz only delivers about 50% performance of 1.5 GHz A72 found on RPi 4.

    So for many real life use cases, what you get is a reasonably performant GPU with an old CPU that's not capable of feeding it properly.
    Yup, I've been working with the 4gb model and just feeding 2 video streams (1280*960@75fps) to the GPU for encoding using Gstreamer uses 80% of the CPU, leaving very little room to do anything else.

    Comment


    • #32
      Originally posted by BesiegedAce View Post

      Yeah, while I'm no fan of nVidia, I feel like I've specifically read somewhere that Tegra has pretty good open source support, so I don't know why people are bringing that issue up for this board.
      To be fair, anyone interested in this is probably wanting to use it for CUDA and that's only supported in their proprietary stack.

      Comment


      • #33
        Originally posted by rene View Post
        vote with your wallet, and buy hardware with OPEN Source Linux drivers and preferably register level specification.
        The Tegra GPU kernel driver is open source. It isn't in kernel.org's tree because of the policy to only have one driver per device (nouveau is already there), but it is a GPL driver.

        Comment


        • #34
          Originally posted by pkese View Post
          The problem with these specific Tegras is that A57 at 1.43 GHz only delivers about 50% performance of 1.5 GHz A72 found on RPi 4.
          A72 is about 15% faster clock for clock than an A57. An A72 is (with some modest hand waving) an A57 with a better cache and load/store unit.

          Comment


          • #35
            Originally posted by jaxa View Post
            It's tough to care about a cut down version of a 2019 product (it feels like it has been longer than March 2019). Jetson Xavier NX could have been a successor to the Nano but it was priced at $400. At $200, I'm sure some people would buy that instead of a Pi4 8GB.

            If you don't need 4 GB of RAM for your IoT/ML/robotics project and have already been using the original Nano, it could be a good option. It's also a better option than Pi4 for console emulation due to the GPU, and 4 GB RAM isn't needed for that.

            Why did they remove the fourth USB3 port?
            Yes, I struggle to justify a tinker/dev board when it's >$200 and coming out of my pocket. Perhaps I just have expectations which are too great, but the ML datasets I've been tinkering with complain about 4GB of RAM, never mind 2GB.

            I'd love a Jetson AGX to play with, but $700?! When currency exchange and import costs are factored in, that's more than an RTX 3080 (when they finally become generally available...) and I can do more, faster with even a fairly old desktop with an RTX 3080 jammed in it.

            Originally posted by hax0r View Post
            M.2 slot please.
            I wish. The ability to use even the cheapest SATA M.2 device would be a massive step up over using microSD (and/or bodging root-from-USB) just from decent device firmware wear levelling would be a great step, even ignoring improved performance.

            Originally posted by WolfpackN64 View Post
            So it's more expensive then the Pi 4, has less memory at the same price point, has a weaker CPU and an old GPU architecture. Not much is known about the Pi 4's VideoCore VI, so I don't know if the old Maxwell really is more powerful.
            All the Pi's have really rubbish GPUs. The Pi4 is just to the point where 4K 30Hz is a feasible proposition, but any kind of 3D acceleration for a desktop is really outside of its ability.

            The Jetson Nano is the only SBC I've used which can push Cinnamon without it turning into a juddering slideshow.

            Originally posted by vegabook View Post
            Personally have run up against out-of-memory errors when running some Tensorflow benchmarks on my 4GB Jetson Nano, so this is a regression. Much better would have been a 119 USD 8GB version.
            Same. Although frankly if they're going to bother upgrading the RAM, I'd rather have a 16GB one. Sadly the only ARM-based SBC with 16GB+ of RAM and an nVidia GPU is $700 on Amazon.

            Originally posted by wizard69 View Post
            It is interesting that everybody compares the card to PI when there are so many similar boards. Odroid for examples has some very nice boards that would be competitive for some uses.
            I agree the ODROIDs are more capable than the RPi(4) but in my experience they're also quite a bit more finnicky. C1s, C2s, XU4s... they're all great bits of kit when they work, but when something goes a bit wonky, they're much more troublesome. They also cost a good chunk more than the RPi4 if you want to make them perform well - the eMMC modules are a night-and-day difference to microSD, IMO...

            Originally posted by SyXbiT View Post
            Can we be done with the A57 yet? It's ancient.
            Guess that's what nVidia had a license for. Will probably change now they've bought ARM!

            Originally posted by ed31337 View Post
            2GB RAM? No thanks. Even at 4GB, my RPi4 really could use more RAM. I keep eyeballing the 8GB version whenever it goes in stock, but at $75 I keep hesitantly passing, hoping maybe they'll someday have a 16GB or some other big improvement to justify spending my money.

            The old A57 CPU is a minus too. I was hoping being so old, it would not have speculative execution and all the security bugs that brings, but nope, A57 is on the vulnerable list. So no advantages, just slowness. Boo.
            I got as far as popping an 8GB RPi4 in my basket yesterday evening, then realised that I was basically going to drop $110 on something I didn't need and have little use for as it is functionally identical to something I already have (a 4GB RPi4).

            The A57 is actually not too bad in terms of speed, but something newer would be nice.

            Originally posted by BaumKuchen View Post
            Yeah... that price tag will go much higher once it arrives at your local store...

            And... you wanna tinker with a board, mess around with stuff, learn... and you pick stinky nvidia? I am wondering how far will you get with that.
            I bought a Jetson Nano because I was told it was impossible to get a program I use daily at work (and which requires CUDA) to work on anything other than x86, and I've always liked a challenge. Aside from having to rewrite a few x86 assembly hacks in pure C, it was fairly easy to do - the problem is limited RAM/VRAM.

            Originally posted by jaxa View Post
            2 GB is fine for retro gaming, LibreELEC, and some of the robotics/AI stuff it is ostensibly for.

            I don't expect to see a Pi5 or Pi6 with more than 8 GB. What would be nice is an ARM board that is around ODROID-H2 size, with two DDR4 SO-DIMM slots so you could put in 8, 16, 32, or even 64 GB of RAM. If it's on the market, it's probably not cheap.
            I'm sure I've seen an ARM64 SBC with SODIMM slots, but my Google-fu is failing me at the minute. You are right, though; it certainly wasn't cheap. Thinking about it, it might have been a Kickstarter or similar that never hit the target.

            Comment


            • #36
              Originally posted by Paradigm Shifter View Post
              I got as far as popping an 8GB RPi4 in my basket yesterday evening, then realised that I was basically going to drop $110 on something I didn't need and have little use for as it is functionally identical to something I already have (a 4GB RPi4).

              I'm sure I've seen an ARM64 SBC with SODIMM slots, but my Google-fu is failing me at the minute. You are right, though; it certainly wasn't cheap. Thinking about it, it might have been a Kickstarter or similar that never hit the target.
              Staggering the launch was a problem, since many people got 4 GB models when they would have paid for an 8 GB. I'm thinking of getting an 8 GB and giving away my 4 GB.

              I'm optimistic that the SODIMM situation will change, because SoC memory limits are rising right around now. The Broadcom BCM2711 in Raspberry Pi 4 can address up to 16 GB of RAM, way up from the 1 GB limit of the BCM2837, while competing SoCs like the Rockchip RK3399 can only do 4 GB. If something like the Rockchip RK3588 raises that to 16 or 32 GB, then there might be a reason to offer 1-2 SODIMM slots.

              Comment


              • #37
                Originally posted by Paradigm Shifter View Post
                I'm sure I've seen an ARM64 SBC with SODIMM slots
                I've seen 3:

                https://www.phoronix.com/scan.php?pa...-LX2K-Sep-Perf
                https://www.phoronix.com/scan.php?pa...core-developer
                32- and 64-bit ARM Open Platform Specifications. For software developers. For the maker community. For embedded OEMs. 64-bit ARM for $129.

                Comment


                • #38
                  Originally posted by jaxa View Post

                  Staggering the launch was a problem, since many people got 4 GB models when they would have paid for an 8 GB. I'm thinking of getting an 8 GB and giving away my 4 GB.

                  I'm optimistic that the SODIMM situation will change, because SoC memory limits are rising right around now. The Broadcom BCM2711 in Raspberry Pi 4 can address up to 16 GB of RAM, way up from the 1 GB limit of the BCM2837, while competing SoCs like the Rockchip RK3399 can only do 4 GB. If something like the Rockchip RK3588 raises that to 16 or 32 GB, then there might be a reason to offer 1-2 SODIMM slots.
                  The 4GB ones came quite late to Japan. Or rather, general availability did. The 2GB ones could be had from pretty much everywhere, but the 4GB ones were quite hard to find long after they became easy to get in the UK/US. If I'd known an 8GB one was coming, I'd have certainly waited as I still had an RPi3 and various ODROIDs to play with.

                  I'd like the Raspberry Pi Foundation to release more options for the compute modules, too. I know that the original idea for the RPi was something cheap 'n' cheerful to help teach about computers/programming/electronics, but its obvious that there is much more of a market out there.

                  Originally posted by andreano View Post
                  I've seen 3:
                  Thanks!

                  Comment


                  • #39
                    Originally posted by Paradigm Shifter View Post
                    The 2GB ones could be had from pretty much everywhere, but the 4GB ones were quite hard to find long after they became easy to get in the UK/US.
                    Lol, it seemed everyone except the Raspberry Pi Foundation expected the 4GB model to be torn away on launch day. Didn't they produce extra many 2GB models? I bought my 4GB in the evening that day, from the last webshop in the world that was shipping to Norway, and it was literally in Greek.
                    Last edited by andreano; 08 October 2020, 05:42 PM.

                    Comment


                    • #40
                      At least they didn't overproduce the 1GB models, eh?

                      Comment

                      Working...
                      X