Rockchip RK3399 Sapphire Board is Sold for $75 (Promo)

Black Friday is coming soon, so we’ll start to see more an more interesting deals in the day ahead. Last summer, I wrote about what should be the cheapest RK3399 development board: Sapphire, Rockchip’s own evaluation board that sold for $75… on Taobao with limited availability. So not exactly the easiest board to get.

But today, I’ve learned that Varms, the company behind ROCK960 board, was selling Sapphire board for the same $75, plus shipping ($20 when I tried) with “many boards in stock”.

Click to Enlarge

Just a quick reminder about the specifications:

  • SoC – Rockchip RK3399 hexa-core big.LITTLE processor with two ARM Cortex A72 cores, four Cortex A53 cores, and an ARM Mali-T860 MP4 GPU
  • System Memory – 4 GB LPDDR3, dual channel
  • Storage – 8 GB eMMC flash, micro SD card
  • Video Output / Display Interfaces – 1x HDMI 2.0 up to 4K @ 60 Hz; MIPI DSI, eDP, LVDS via MXM connector
  • Connectivity – Gigabit Ethernet
  • USB – 1x USB 2.0 host port, 1x USB 3.0 port, 1x USB 3.0 type C port with DisplayPort over Type C support
  • Camera I/F – MIPI CSI signals
  • Expansion
    • 40-pin GPIO Raspberry Pi compatible header
    • 314-pin MXM connector with PCIe 2.0 x4 and other signals..
  • Misc – Power and reset keys
  • Power Supply – 12V/2A DC via power barrel; RK808-D PMIC

The board will ship with a 12V/2A power adapter. Some documentation for the board can be found in Rockchip Open Source website. The full development kit combining Sapphire, Excavator carrier board, and an eDP display is also sold, but with only minimal discount for a total of $349.

Support CNX Software - Donate via PayPal or become a Patron on Patreon

24
Leave a Reply

avatar
24 Comment threads
0 Thread replies
8 Followers
 
Most reacted comment
Hottest comment thread
12 Comment authors
wzyy2Richard krehbielwillytheguyuktkaiser Recent comment authors
  Subscribe  
newest oldest most voted
Notify of
vamrs
Guest

Wow, CNX is fast.

To clarify, 20$ shipping is for DHL.

Btw, rock960 is in PP now.

hex
Guest
hex

linuxable?

crashoverride
Guest
crashoverride

@hex
Currently, its not worth bothering with. You can look at the Firefly RK3399 forums to find out that pretty much nothing works and there is absolutely no support for anything.

The RK3399 was exciting “on paper”. None of the exciting features are supported in Linux and there is no expectation they ever will be:
https://github.com/rockchip-linux/libmali/issues/6

Until a RK3399 board vendor comes along that is serious about support and not just making more “orphan hardware”, consider alternatives.

blu
Guest
blu

@crashoverride
Is GLES3 functional at least?

fossxplorer
Guest
fossxplorer

And @willy wasn’t impressed with the perf of RK3399 neither.
I’m waiting for Orange Pi H6s, could be some good boards from Xunlong!

crashoverride
Guest
crashoverride

@blu
It is functional, but performance is abysmal due to RK’s X11 support. You can look at glmark2-es2 scores for any RK chip as verification. Off-screen performance is often cited instead due to this.

Member

@crashoverride
How many of these “exciting features” that don’t work involve the GPU?

Perhaps I’m in the minority, but I’m more interested in ARM SBC’s as mini servers, so Ethernet and SATA are much more important to me. I wonder how difficult it would be to make a base board for the Sapphire that takes the PCIe signals from the MXM connector and routes them to a mPCIe socket? Then SATA (and lots of other things) would be easily available.

blu
Guest
blu

@crashoverride
Ah, the ever-fashionable ‘we have a nice gpu but we can’t show frames worth shit’ issue.

SBC/SoC vendors never learn.

crashoverride
Guest
crashoverride

@Ian Tester
I can’t speak specifically for the Sapphire board, but the Firefly board had issues with GPU, USB 3.0, USB-C, WiFi, BT, and HDMI 2.0. There were also CPU scheduling issues (big.LITTLE). I did not test PCIe as ironically the mPCIe port does not actually have any PCIe lanes (only USB) and the M2 (B-key) port does not have SATA. I seem to recall the GB ethernet would not connect past 100MB, but I think that is a known issue with the PHY and certain cables (Cat 5e).

tkaiser
Guest
tkaiser

crashoverride :
and the M2 (B-key) port does not have SATA.

RK3399 has no SATA capabilities so either an USB or PCIe to SATA adapter is needed (Firefly folks sell a ‘PCIe M.2 to SATA3.0 Adapter Board’ which according to lspci is based on an ‘ASMedia Technology Inc. ASM1061 SATA IDE Controller (rev 02)’ — that could be an ASM1062 making use of two PCIe lanes so in case there are also two lanes routed to the connector the usual 400 MB/s barrier due to maxing out PCIe 2.0 x1 might not exist).

The XHCI controller is the same as in RK3328 which performs quite well but needs some patches RK recently provided. Anyway since I’ve not seen any PCIe/SATA attached storage benchmark results with any RK3399 device so far and the platform is quite expensive I’ll stay with the various others options for server/NAS use cases šŸ™‚

theguyuk
Guest
theguyuk

I keep suggesting it but to no avail. The open source hardware and software community should published a speciication of what is wanted and need. As well as a list of what at present fails and why.
The arm vendors are clueless, the customers need to stick what is needed in their face and the open source community need to tell them how it can be achieved, with what they have.

My own basic suggetion, drop the GPU from a TV soc, give it direct usb 3, true PCIe lanes that can run Nvidia or AMD GPU cards, and the SoC use DDR4 memory. Max memory 16 GB.

Sell it at a slight premium price, due to smaller market.

PCIe so you can plug in a card for what you need.

No it is not a top spec solution, but a market starter.

blu
Guest
blu

My own basic suggetion, drop the GPU from a TV soc, give it direct usb 3, true PCIe lanes that can run Nvidia or AMD GPU cards, and the SoC use DDR4 memory. Max memory 16 GB.
Sell it at a slight premium price, due to smaller market.

A slight premium with a dGPU? The BOM of said STB would quadruple!

willy
Guest
willy

Perhaps Iā€™m in the minority, but Iā€™m more interested in ARM SBCā€™s as mini servers, so Ethernet and SATA are much more important to me.

We’re two in this minority šŸ™‚

I wonder how difficult it would be to make a base board for the Sapphire that takes the PCIe signals from the MXM connector and routes them to a mPCIe socket? Then SATA (and lots of other things) would be easily available.

I don’t know. However if you want to play with some RK3399, the H96 Max could also be an option within then same price range depending on the time you’re willing to spend on it, and it comes with its heatsink and enclosure. I manage to boot Linux ot it (and have lost my notes, I don’t remember which image I used in the end). I still haven’t replaced the kernel however. The rockchip way of building and flashing images is a real pain. “It’s rock hard to program a rockchip”. I wish this situation changes and they stop shipping locked boot loaders, signed images and I don’t know what other crap and instead have an interruptible uboot and an editable extlinux.conf loading the kernel, initrd and dtb you place on the first partition. I gave up trying after remembering that it took me one year to linuxify my T034/CS008 and am not willing to spend so much effort on the H96 for a disappointingly slow CPU like the RK3399.

tkaiser
Guest
tkaiser

willy :
if you want to play with some RK3399, the H96 Max could also be an option

But AFAIK none of these TV boxes make one of the more interesting features of this SoC available: PCIe 2.1 x4. According to @vamrs their ROCK960 will make all 4 PCIe lanes and also GbE/RGMII available via an adapter board and if Pine folks decide to do RockPro there will be even a RK3399 board with GbE and a full PCIe x4 slot available next year.

Though no performance numbers available yet (at least I found none and based on experiences with other ‘PCIe on ARM SoC’ implementations I want to see this first — i.MX6 for example is not remotely able to saturate PCIe 2.x x1 bandwidth, our tests showed less than 50% of what Marvell PCIe lanes deliver).

theguyuk
Guest
theguyuk

@willy
I thought the push at present was for secure boot due to security issues?

Richard krehbiel
Guest
Richard krehbiel
Richard krehbiel
Guest
Richard krehbiel

@Richard krehbiel
Oh – and as if that wasn’t expensive enough, the Picocluster version.

https://www.picocluster.com/collections/marvell-macchiatobin

blu
Guest
blu

@Richard krehbiel
While I too would love for the price of the mcbinto to go down, you do realize the SBC subject of this thread and the mcbin are products of different classes, right?

Richard Krehbiel
Guest
Richard Krehbiel

@blu
I do indeed realize that. It was just a reality check (yeah, I hate those too) to commenters wishing for “server-grade” stuff at “STB-grade” prices.

Although… if you left out the 10GbE hardware…

blu
Guest
blu

@Richard Krehbiel
Gotcha. BTW, my impression of the price of the macchiato is that its price varies related to the prices of DDR4, but I don’t think it follows the same rate of change as DDR4 market price. Curiously enough, mcbin’s price did (almost) reach its preorder level not long ago, but that did not last long..

willy
Guest
willy

Richard Krehbiel :
It was just a reality check (yeah, I hate those too) to commenters wishing for ā€œserver-gradeā€ stuff at ā€œSTB-gradeā€ prices.

The thing is that we all know that most of the price of an STB comes from the SoC. Some SoCs are pretty powerful, have tons of I/O, and would be fairly capable as server chips if they were placed on suitable boards. But most often due to target use case, pins are not routed and very interesting features are simply lost. And that’s not counting with the porting work that often is done only by the community once the SoC is outdated.

I’m not really asking for a server board at the price of an STB, just for limited sacrifice of SoC features. When you buy a Clearfog, you don’t feel like you’re throwing your money through the sink. Some of us have played with the NSLU2 a decade ago. This device’s success came from all the mods that were made on it, trying to extract unused CPU signals from PCB traces and expanding its possibilities. Most of the STBs have a lot of wasted space where it would be possible to route unused pins to test points. It would serve no purpose for their target audience, but it could dope adoption. Just like NSLU2, WRT54G, WR703N.

Here we see a reasonable board corresponding to this full expandability and at a reasonable price. It just lacks an enclosure and a few connectors to become everyone’s board of choice, since it’s made to connect to a mainboard. But it’s a nice option already. I would immediately buy it if I didn’t have the H96. Now I’m less interested after having been disappointed by the RK3399’s performance.

wzyy2
Guest
wzyy2

@willy
rockchip opensource linux offer mainline u-boot and use editable extlinux.conf.
http://opensource.rock-chips.com/wiki_Main_Page

So it’s just firefly give you a shit. :- P

wzyy2
Guest
wzyy2

@crashoverride
See my comments in https://github.com/rockchip-linux/xserver/issues/6
If performance is abysmal in X11, then it’s not due to rockchip support, but X11 itself.
If you means low glamrk2-es2 scores, then it’s because of vsync, or you will see tearing when playing youtube. = =