NVIDIA Jetson Nano A FeaturePacked Arm Developer Kit For 99 USD Phoronix

One of the maximum interesting announcements out of NVIDIA's 2019 GTC convention is the advent of the Jetson Nano, NVIDIA's ultra-modern Arm developer board providing a Tegra SoC. This developer board is very exclusive from the past Jetson forums in that it's aiming for a completely cheap price point: just $99 USD.

image.php?id=nvidia-jetson-nano&image=jetson_nano_1_med

NVIDIA Jetson developer forums have traditionally been several hundred greenbacks or inside the case of the state-of-the-art excessive-performance presenting, the Jetson AGX Xavier commands a $1,299 USD fee. The Jetson Nano will retail for simply $99 USD even though glaringly the overall performance may not match that of the AGX Xavier. The Jetson Nano Developer Kit is passively cooled but there is a 4-pin fan header on the PCB and screw holes at the aluminum heatsink in case you want to mount a fan for higher cooling.

image.php?id=nvidia-jetson-nano&image=jetson_nano_2_med

With this low-value Jetson board, the Nano is using a Tegra chip just like what became observed in the Jetson TX1 some years lower back. This Tegra X1 SoC has a quad-core Cortex-A57 processor and 128-core NVIDIA Maxwell portraits... Not almost as thrilling as the X2 or AGX Xavier, but nevertheless not awful considering the SoCs commonly located in sub-$100 Arm developer boards.

image.php?id=nvidia-jetson-nano&image=jetson_nano_3_med

The Jetson Nano also gives 4GB of LPDDR4 reminiscence, Gigabit Ethernet, 12 MIPI lanes, four USB ports, and can drive up to 2 simultaneous shows. These functions and the Maxwell pictures easily positioned the Nano's abilities properly beforehand of maximum (or maybe all?) Arm developer boards hitting for the sub-$100 market. One of the benefits in the use of the older Tegra X1 design is that the open-source Linux kernel aid is in better form than the just-released SoCs and there is even the open-source Tegra Maxwell photos help within the Nouveau driver stack.

image.php?id=nvidia-jetson-nano&image=jetson_nano_4_med

Unlike the higher-stop Jetson boards featuring eMMC garage, the Jetson Nano is based upon a microSD card for storage. The connectivity at the developer package consists of four USB 3.0 Type-A ports, HDMI 2.0, DisplayPort 1.2, 40-pin header, MIPI CSI digicam connector, micro-SD slot, M.2 WiFi slot, and Gigabit Ethernet. That's one of the shortcuts in this board is there may be no integrated WiFi however does require an outside card in case you are interested by wireless connectivity.

image.php?id=nvidia-jetson-nano&image=jetson_nano_5_med

The Jetson Nano supports CUDA, TensorRT, and the alternative software program components of the higher-cease Jetson forums; the same JetPack software runs at the Nano. The "Linux 4 Tegra" at the Jetson Nano objectives Ubuntu 18.04 LTS even though we have visible other Linux distributions upload assist for other Jetson boards too.

Overall, the Jetson Nano is pretty a compelling product at $99 USD and we've got had it in our lab for a few days to deliver a few initial benchmark outcomes.

NVIDIA charges the Jetson Nano Developer Kit as being capable of gain 472 GFLOPs of FP16 compute electricity out of the Maxwell GPU, the usage of four Cortex-A57 cores is a great deal better than most sub-$one hundred Arm forums, the 4GB of LPDDR4 reminiscence is rated for 25.6 GB/s of memory bandwidth, and the video encode/decode is rated for 4K60 with up to 8 1080p30 HEVC decode streams of 4K30 4x 1080p30 HEVC encodes.

image.php?id=nvidia-jetson-nano&image=jetson_nano_6_med

I've simplest had a few days with the Jetson Nano Developer Kit up to now, but the enjoy up to now has been going first-rate and the overall performance is aggressive among the lower-priced Arm SBCs. The performance is far from the Jetson AGX Xavier, understandably, in addition to the Jetson TX2. Due to being short on time prior to the embargo raise and strolling into JetPack troubles at the TX1, there are not all of the GPU/compute exams on that older board. But really many of the likes of the Raspberry Pi and other low-fee forums, the Jetson Nano is competitive.

image.php?id=nvidia-jetson-nano&image=jetson_nano_7_med

The Arm SBC benchmark comparison for the Jetson Nano Developer Kit release day checking out protected the:

- Jetson TX1 Max-P
- Jetson TX2 Max-Q
- Jetson TX2 Max-P
- Jetson AGX Xavier
- Jetson Nano
- Raspberry Pi 3 Model B+
- ASUS Tinker Board
- ODROID XU-4

That changed into based totally upon the forums I had the time to freshly re-check given the quick turnaround time to release day as well as for the thrilling Arm boards in my ownership. Via the Phoronix Test Suite a wide range of benchmarks have been run while extra exams are imminent on Phoronix.

embed.php?i=1903186-HV-JETSONNAN05&sha=93a8228&c=62b87ae32421055d2f29bd48f484454c94961b85&p=1

The Jetson Nano GPU overall performance need to be more or less in step with the Jetson TX1 given the Maxwell GPU. Unfortunately TX1 JetPack troubles averted that assessment in time for launch day, however overall the overall performance of the Jetson Nano with CUDA isn't terrible while considering it's sub-$one hundred or even the Jetson TX2 still retails for $599 USD.

embed.php?i=1903186-HV-JETSONNAN05&sha=9ddf9b2&c=15b7d46c131638ab172f2382aff3c4e341e52d5d&p=1

For those thinking approximately OpenGL performance, the Jetson Nano can score almost 650 at 1080p with GLMark2. Meanwhile most low-price ARM SBCs do not even ship with running ARM (binary) graphics drivers by using default.

embed.php?i=1903186-HV-JETSONNAN05&sha=d85eaf0&c=7693cf4b748fad2c271bcb89164940e4b843357b&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=cef2417&c=cfb333ef67c7de9cc594bf6fbc84d2f67d1eab5c&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=4187158&c=ef5090e266cec7f3fda55978304f6aa4208d4517&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=f66a57e&c=3ff1dc362a1f52a48bd733e68bcfc87a809d8014&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=04b19aa&c=90a46d1b96a81665de452674363af9a8f7cf0dc6&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=0a76aee&c=80109c6799e9c85b803cf714ef07340430af75f4&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=de6171c&c=658a794bc6631ee1b360bf5e6f4a5fb1d1f3ffef&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=04e86e0&c=2cfc2939f7c4bade30a100a50dbaf1d208e53129&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=ba07b41&c=216609e8810730ce380373f32b77b9fdd256bca8&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=245c438&c=c7c3192ea0d426040784c80ca79fc0d0f2ade4ba&p=1

For the rate, the Jetson Nano TensorRT inference overall performance is looking superb.

embed.php?i=1903186-HV-JETSONNAN05&sha=24c9f8d&c=149afeae11c9f12778c0c839f9f2d57f0bf6aa65&p=1

The LCzero chess benchmark rankings around 15 nodes per 2d when simply the use of BLAS on the CPU cores....

embed.php?i=1903186-HV-JETSONNAN05&sha=5f4cd5a&c=94c1397628a33f5f2dd027de8fc2c5d1d3256d9d&p=1

But when walking this deep learning chess benchmark with CUDA, the Jetson Nano overall performance jumps to 140 nodes per 2nd.

embed.php?i=1903186-HV-JETSONNAN05&sha=be2817b&c=4ecdbd99750a0daa01188f4262a5be2dda8ef368&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=93b643e&c=ffba90bc6e56b93d1e0102003a1dd9e41a8e227c&p=1
embed.php?i=1903186-HV-JETSONNAN05&sha=a897ca1&c=32f6380f8c703d3d84bd215727a68137f54f31cb&p=1

Here's a examine the CPU performance of the Jetson Nano, which inside the case of the TTSIOD 3D Renderer is similar to the ODROID-XU4.

embed.php?i=1903186-HV-JETSONNAN05&sha=6ceb9bf&c=2fce56768ec8fbb681400a1706b2320d71ae5315&p=1

The Jetson Nano did come out a lot faster than the ODROID-XU4 for the multi-threaded Rust benchmarks.

embed.php?i=1903186-HV-JETSONNAN05&sha=432aff7&c=c784cc741b7806b94c28ea8a651f596a5c2f592b&p=1

For smooth analysis, here is a have a look at the geometric imply from a variety of CPU benchmarks that could run throughout all the examined SoCs. The Jetson Nano comes out slightly ahead of the ODROID-XU4, however keep in mind that is simplest with the Cortex CPU cores and not leveraging the Maxwell GPU.

Besides the CPU performance, the very succesful Maxwell GPU on the Jetson Nano is really something we are not used to seeing with those Arm SBCs specially with the NVIDIA Linux driver help whilst proprietary being plenty higher off than the standard proprietary Arm photographs motive force blobs we usually have to cope with. Beyond that, there's already Tegra Maxwell support within Nouveau should you need open-supply driver support albeit you then are losing out on CUDA and other GPU compute capabilities.

Also pleasant with the Jetson Nano Developer Kit is the first rate connectivity alternatives we additionally are not used to seeing on the low-charge Arm developer forums... There are four USB 3.0 ports, Gigabit Ethernet, both HDMI and DisplayPort that may be driven simultaneously, and powering over USB or a DC power connector.

Overall that is arguably the quality sub-$100 Arm developer board we have seen thus far depending upon your use-instances. The Jetson Nano will genuinely open up NVIDIA Tegra SoCs to appearing in extra low-cost DIY projects and other hobbyist use-instances in addition to commencing up GPU/CUDA acceleration that until now has surely no longer been feasible at the low value boards.

Stay tuned for extra benchmarks on the Jetson Nano Developer Kit as I've had greater time with the unit whilst look for this developer board to begin appearing at essential Internet retailers beginning tonight at $99 USD.

Let's block advertisements! (Why?)


//www.phoronix.com/vr.hypertext Preprocessor?View=27681
2019-03-18 23:04:14Z
52780243309969

0 Response to "NVIDIA Jetson Nano A FeaturePacked Arm Developer Kit For 99 USD Phoronix"

Post a Comment

Iklan Atas Artikel

Iklan Tengah Artikel 1

Iklan Tengah Artikel 2

Iklan Bawah Artikel