Rockchip RK3588 datasheet available, SBC’s coming soon

We had most Rockchip RK3588 specifications so far for the long-awaited Cortex-A76/Cortex-A55 processor, but at today’s Rockchip Developer Conference 2021, more information surfaces with impressive CPU and GPU benchmarks, and the Rockchip RK3588 datasheet has just dropped from the sky directly into my laptop, as such document usually does. At least two single board computers are expected to soon follow from Radxa and Pine64.

Rockchip RK3588 datasheet

Since we have the datasheet, we can confirm some of the specifications of RK3588, and get additional details:

  • CPU – 4x Cortex-A76 @ up to 2.4/2.6 GHz and 4x Cortex-A55 cores @ 1.8 GHz in dynamIQ configuration
  • GPU
    • Arm Mali-G610 MP4 “Odin” GPU with support for OpenGLES 1.1, 2.0, and 3.2, OpenCL up to 2.2 and Vulkan1.2
    • 2D graphics engine up to 8192×8192 source, 4096×4096 destination
  • AI Accelerator – 6 TOPS NPU 3.0 (Neural Processing Unit)
  • VPU
    • Video decoding
      • 8Kp60 H.265, VP9, AVS2, 8Kp30 H.264 AVC/MVC
      • 4Kp60 AV1
      • 1080p60 MPEG-2/-1, VC-1, VP8
    • Real-time 8Kp30 encoding with H.265/H.264; multi-channel encoding supported at lower resolutions
  • Memory I/F – LPDDR4/LPDDR4x/LPDDR5 up to 32GB
  • Storage – eMMC 5.1, SD/MMC, SATA 3.0 (multiplexed with PCIe 2.0), FSPI (Flexible SPI)
  • Video Output
    • Dual HDMI 2.1 / eDP 1.3 up to 8Kp60
    • Dual DisplayPort 1.4a up to 8Kp30 (multiplexed with USB 3.0)
    • Dual MIPI DSI output up to 4Kp60
    • Bt.1120 video output up to 1080p60
    • Optional dual LVDS up to 1080p60 via RK628 chip.
    • Up to four independent displays (up to 1x 8Kp60, 2x 4Kp60, 1x 1080p60)
  • Video Input/Camera
    • 48MP (2x 24MP) ISP with HDR and 3D NR support; multi-camera input
    • 2x MIPI DC (4-lane DPHY v2.0 or 3-lane CPHY V1.1)
    • 4x 2-lane MIPI CSI
    • DVP camera interface
    • HDMI Rx 2.0 interface up to 4Kp60 with HDCP 2.3 support
  • Audio
    • 2x 8-channel I2S, 2x 2-channel I2S
    • 2x SPDIF
    • 2x 8-channel PDM (for mic arrays)
    • 2-channel digital audio codec (16-bit DAC)
    • VAD engine
  • Networking – Dual Gigabit Ethernet
  • USB – 2x USB 3.1 Gen 1 up to 5 Gbps (multiplexed with DisplayPort), 1x USB 3.1 Gen 1 (multiplexed with Combo “PIPE PHY2”), 2x USB 2.0 OTG
  • PCIe – PCIe 3.0 up to 8 Gbps (1x 4-lane, or 2x 2-lane, or 4x 1-lane, or 1x 2-lane + 2x 1-lane
  • 3x Combo PIPE PHY interfaces
    • Combo PIPE PHY0/1 – SATA III or PCIe2.1 up to 5 Gbps
    • Combo PIPE PHY2  – SATA III, PCIe 2.1, or USB 3.0
  • Low-speed I/O – 5x SPI, 9x I2C, 10x UART, GPIOs, 12-bit ADC (SARADC)
  • Package – FCBGA1088L; 21.45 x 21.45mm
  • Manufacturing process – 8nm LP

There are additional details in the datasheet, including three Cortex-M0 MCUs for the PMU, NPU, and “PD_CENTER” with the latter related to the DDR memory controller. Rockchip RK3588 has several multiplex interfaces for SATA, PCIe 2.1, USB 3.0 in a way that looks similar to Rockchip RK3568.

I’m quite surprised they could use a Mali-G610 “sub-premium premium” GPU as it was announced together with Cortex-A510, Cortex-A710, Cortex-X2 Armv9 cores, but it also works in SoCs with older Armv8 cores so that’s good, and that’s why GPU performance is truly a big step, up to over 10 times faster, compared to Rockchip RK3399.

Pine64, Radxa Rockchip RK3588 SBC’s coming soon

The other good news is that Pine64 is working on an RK3588 SBC and Radxa ROCK5 is also under development, and there should be an announcement before the end of the year. It won’t mean the boards are ready to ship, but rather pre-orders for at least one of the boards will be opened, and early backers should be happy for the prices that are going to be offered. I’ve heard both boards may be showcased at RDC 2021 tomorrow, but this remains to be confirmed. I’ve also been told Rockchip RK3588 chip is really complex, and software support will take time.

Share this:

Support CNX Software! Donate via PayPal or cryptocurrencies, become a Patron on Patreon, or buy review samples

60 Replies to “Rockchip RK3588 datasheet available, SBC’s coming soon”

    1. It is estimated that mid-2022, fab wait times should be easing gone by 2023.
      i would guess that mass production would be after that time, exc. sampling bath, press kit, etc…..

  1. Now we’re talking, a proper multipurpose chip with decent processing power. The big question is, will Rockchip offer decent software support for it?

      1. What SoC vendor would make you say “it’s $X so definitely yes”? Not Allwinner, clearly. Amlogic?

        Anyway, the RK3588’s specs are exactly what I’m looking for, so I’ll take a leap of faith on the software and buy a SBC as soon as one is available.

    1. It will be interesting to see and with AMLogic seemingly blocking open source support for their S905X4 SOC it could open up great market opportunities for Rockchip to take the lead that they once had over ML.

      1. It’s just insane that it’s so hard to get a single company to offer a decent general purpose SoC with matching software support. There seems to be a huge demand, but zero interest in providing such a product.

    1. Boards are great and all but in 6 months or a year after this is released will it still have u-boot/kernel sources that build with current distros.. let alone a kernel with “well, there’s no merge conflicts” level grade LTS updates?

    2. Why is it none of you companies that makes these so called development boards, consider cooling? That board is going to be impossible to cool.

        1. You’re joking, right? This has an additional four cores, albeit low power ones, it has a much more powerful GPU, an NPU and much more PCIe connectivity, all of which adds heat. Yes, it’s built in a more advanced, but it’ll run hotter than the Pi 4.

          1. Odroid N2+ clocked to 2.4 ghz only does 6.2W when stressing the CPUs. A Raspberry Pi4B does 6-6.5W too. And consumes a bit more when overclocked, obviously.

            Now the Odroid N2+ has 2 fewer cores than 3588, but those A53s are clocked higher and obviously on a 12nm process rather than 8nm.

            5-6 watt when stressing the CPU is well within the realms of possibilities.

            Regarding the GPU, architectural improvements should have it much more efficient. Don’t know how much it will consume but I doubt it will be an order of a magnitude more than the OdroidN2, which has always been close to the Pi4 in power consumption.

            Really shouldn’t be anywhere near “impossible to cool”

          2. I never said the chip was impossible to cool, I said the Rock5 Model B is going to be impossible to cool. It has no sensible means of mounting a cooler, as there are no holes to use for mounting a cooler.

          3. A Pi4 doesn’t really have any sensible holes either. But even those hacky cases with a 20mm fan do their job well enough. As long as it doesn’t go north of 10W it shouldn’t be much of a problem.

          4. The Pi4 overheats without a heatsink. Had to stick one on, just for it not to fall over when playing videos. It’s terrible and yes, I loathe the RPi form factor.

          5. That’s fair.

            I’m fairly accepting of janky setups with those aftermarket cases.

            But I wish it’d be bigger too, like those bigger RK3399 board.

          6. It’s the largest SoC in rk35xx series (about double area than s922) and pcb quality and design is supportive to heat transfer and is declared for mobile internet purposes also.
            Average power demand is for lower cpu/gpu voltages (about <10% lower, but real values are TBD). Maximum current draw might be around 6-9 amps on SoC voltage level (depending on multimedia demand)?

          7. Mmm, I don’t think it will run hotter than a slightly overclocked RPi4. Mind that RPi4 had an ancient fab when it was released. Intel uses 22nm since 2012 and RPi4 is from 2019 and uses 28nm. Any board boasting a Cortex A76+A55 with better GPU with a 8nm fab will consume 8-10w tops, including overclocking

      1. Yep. What a pity! I’ve expect at least something like odroid N2{+}. It have decent heatsink. And you can add m.2 with that form factor.

      2. The 8nm process of the RK3588 vs. the 28nm process of the Broadcom on the RPi4 will definitely give it an advantage to power efficiency, low heat, and high performance.

  2. Its a great piece of hardware.
    If only had a good software support. Rockchip’s SoC always take some time to be really good for applications due to the community having to support it.

    1. That’s still better that for most chips where companies doesn’t work at all on kernel? Have you a better example than Rockchip for mainline support?

  3. Looks pretty nice overall. And dual-gigE, finally! This plus the 4 PCIe3 lanes offer nice possibilities… provided software support doesn’t take the usual 5 years to come, of course.
    To bad it got delayed by 2 years, it could likely have been even more impressive 2 years ago!

    Now let’s see what Tom comes up with and at what price. If it can make a decent dev armv8.2 board, I may get tempted.

    1. Dual gig just in time for when the market is moving to 2.5Gbps…
      Still a pretty decent chip overall though, assuming the software support is there.

      1. Yep I thought about it as well, but since the chip is 2yr late we couldn’t expect to see a new MAC IP block there I guess.

    2. I think the GPU was changed to this unreleased one during the 2 years of delays, so at least we got something out of it.

      1. Very likely it was considered as a showstopper for laptop usage due to being so late, and had to be changed or the product would die.

  4. Deja vu! It seems like this is exactly where we were at this time of year in 2019 – announcements.

    So lets just cut to the chase…
    How much does it cost? When will it be available (really available, not pre-order, not back-order)?

  5. how the fuck does arm survive with such perpetual pathetic shrivelled shit lowered expectations? this chip is not awful by arm standards but fuck man, the shittiest lowest crappiest x86 has >8x more io than this shrivelled peanut shit. what the fuck is wrong with arm. why are we happy or ok with such vastly vastly vaslty lowered expectations? power consumption isn’t all that, isn’t that stellar. cpu performance is droll; a bump but not superb. io is a ghostly shadow of what an intel atom would deliver, just vanishing & faint. this is somewhat cheaper than x86 but like, not a lot. alas of course amd no longer competes at the low end- that sucks. still, this is such a foggy shitty bare basics change, and i’m not mad that it’s going to make a big change- i’m just mad that there’s nothing the frell else out there to make any kind of difference. that the rk3588 is purely entirely totally competed with, in the <$100 system space. and many wont be that cheap.

      1. I don’t think that he actually said that ARM develops chips, but yes, you are right, there are ARM microprocessors for HPC, servers, etc – and companies do produce designs with the ability to move plenty of data to and from the chip. Some of these may well be available at a better price point than x86 chips, and if anyone knows of a desktop motherboard or SBC using one of them I at least would really like to know. I think that his complaint may be that for all processing power continues to get cheaper, there is a gap there is a gap in the market. It is very difficult for makers and even professional prototyping services who need something at a reasonable cost with more processing power, RAM, I/O and especially physical computing connectivity than you can get from the entry level Raspberry Pi and Arduino stuff – Unless you have the knowledge to design a custom PCB, or the physical space and budget in what you are making to use a still in production x86 motherboard and actually stick an expansion card(s) in. If anyone knows of a an SBC that has more oomph and is good for physical computing, I suspect a lot of people in the mechanical engineering side of things would like to hear more.

    1. Most mobile arm socs aren’t mean for servers, that’s why. Why to add dozen of pci lanes on a general purpose soc. What if you dont need them?? Then you are wasting them at a cost on price and power consumption. The guys who dev socs are far smarter than you, and that’s why.

    1. Cool. As everyone can see there’s room for a sufficiently large heat sink, so we can definitely expect cooling issues like with RK3399, as usual. Anyway, that’s normal, CPU manufacturers always have to find the sweet spot between performance and consumption.

  6. There’s also a Rockchip RK3588S chip. That must be why the GeekBench4 results leaked a few weeks ago were titled RK3588_S, and performance was a bit lower.

        1. No, it’s the same with less I/O. The rk3399S were just repurposed hw (aka rk3399s that aren’t good enough)

    1. Hello. Can you see the breakdown by sub-items of the test? And more. In Geekbench tests, 3588 is 1.5, 2 times faster than 922. And in Antutu, in 4. Why such a spread? Thank you.
      p.s. Test 922

  7. “I’m quite surprised they could use a Mali-G610 “sub-premium premium” GPU as it was announced together with Cortex-A510, Cortex-A710, Cortex-X2 Armv9 cores,
    but it also works in SoCs with older Armv8 cores so that’s good, and
    that’s why GPU performance is truly a big step, up to over 10 times
    faster, compared to Rockchip RK3399.”

    Mali G610 comes with closed-source BLOB and currently no support in open-source Panfrost drivers. A Mali Bifrost G76 with a generous number of EEs would have matched the CPU upgrade and have worked with open-source OpenGL and GL-ES drivers on launch.

    Was the G610 part of the original RK3588 design?

    A76+ CPUs and Bifrost GPUs can be found in now-affordable used phones. Units with cosmetic damage can be cheaper than the retail price of a N2+ or Vim3 Pro, and far below the ‘AI Development Boards’ being offered in the $500+ bracket.

    With some more mainline integration, an Exynos or Kirin phone could be attached via USB to a current SBC, run native Linux + Panfrost and render to offscreen buffer, streamed at low latency to the ‘desktop’ host device.

    While not a solution for all use-cases, this would be an alternative path for ARM linux users to doubled performance, while maintaining a portable fanless non-x86 lifestyle.

    I like having options.

      1. You just need really many of them, maybe you could get them manufactured for you by some Chinese second hand manufacturing company and hot glue them in place somewhere in the case of your industrial device.

    1. That’s pure expectation, but after what I read in dev blog, the jump from Mitgard T7xx and T8xx to Bitfrost G31, G52, G72, and few month later G78 that uses Valhall (start with G77), didn’t take so much months, after what I read, but GL3.1 support was only in september for G52. We could probably expect the same not too long progress for new generation GPU, hoping they will have one devboard soon enough. post GL 3.1 is still a WIP, most of features are here on 3.2 (but geometry shaders is not finished) and 3.3 and some functions on 4+ after mesa doc, and this is probably what needs more work with Vulkan driver PanVk, that just started around march 2021. Depending on the need. On most of the desktop, GL3.0, should be enough to work. Things like Blender (2.9 needs GL3.3, Blender 3.0 needs GL 4.3 to have real acceleration). It will still work for basic scenes else.

  8. Aside from anticipated performance, that I/O setup is simply amazing, from suprising additional HDMI input, to 2 GigE and large amount of serial, SPI, I2C, PCIe and audio interfaces. I can already so many more different devices that would utilize those capabilities, no matter the official software support.

  9. Board designers: please learn from the Odroid N2. Put the SoC on the bottom and the heat can trivially convect to the case. This yields a lot more case design flexibility.

    Thank you.

  10. Great, a Cortex A76+A55 being released on 2023. The only thing new here is the GPU.

    And of course, they will expect you to pay 100-120€ + shipping for a dev board.

    This thing will be interesting when the price of these is around 50€ with 8GB of RAM, and I don’t see why that wouldn’t happen.

Leave a Reply

Your email address will not be published.

Advertisement
Advertisement