Radxa 6 under development

#1

When will be out Radxa 6 with ARM 4xC A78 2.9Ghz 4xc A72 2.3 Ghz with lpddr5 6400 with G710 MP10 ?

#2

Quad A78 based will be SBC will be out in April, but not ROCK family. Developers will receive the samples in the next few days.

The ROCK 6 will be at least ARM V9 Cortex X1 or higher level for the big core. It has at least the following features:

  • TSMC 6nm technology
  • ARM V9 CPU
  • ARM 5th generation GPU
  • many Tops NPU
  • 64bit LPDDR5
  • PCIe 4.0
  • USB C 3.2 10Gb with DP

The overall performance will be similar to 8CX Gen3.

The questions is, how much do you plan to pay for such a SBC with that performance?

  • less than 200$
  • 200$ to 300$
  • 300$ to 400$
  • 400$ to 500$
  • more than 500$

0 voters

3 Likes
#3

It’s just very hard to say anything about price without other resource context when for now just RAM amount can multiply final price. Specs seems to be nice improvement over current SOCs.
Quad A78 seems like pi5 killer, should be about 30-40% faster compared to A76 cores.

1 Like
#4

Any chance to increase up to 128 bit for lpddr5 6400 just to increase performances for both cpu and gpu? X1 is very good (over my expectations) but what for little core, 4xc A78 2,7Ghz?

P.S. ok USB4 but why not SD express 4.0 interface? Consume 1/10 of PCI-E 4.0 and is fast as pci-e 3.0.

#5

How many lines?and how many slots? 1x4?

#6

Please focus on upstream kernel contributions this time around.

7 Likes
#7

For this reason I wrote A78 with A72, to mantain down the price. (Doubling memory bandwidth up to 128bit can reach 2x performances vs 5B). Rasp team has a good philosophy in this. Still on eBay rock pi 5B in high price between 200 for 8GB and 300 for 16GB…
unfortunately in many disqus tech sites, users prefers geekers mini pc that has triple performances at same cost. (Like old minipc with 5700u).
Still pricey Firefly 3588J board :confused:

Radxa team must reduce price asking to others foundry seeing how is going the poll.

#8

Hows the size of that board be like ? rock 5a or 5b (credit card size wise?).

#9

You can’t pair A78 with A72. They’re both big core designs, and A72 is old + armv8-only.

1 Like
#10

I think that we need mid range ARM SBC around 200-350$. About ARM-based HW: if people want something weak and cheap - there is RPi 5, if they want sompething powerfull (and spend money) - there is Mac Mini.
And it need better iGPU or some external GPU support.

#11

you’re right, seeing that A78 was v8.2 while A72 v8-A. so must be like 8cx3 (X1 with A78), but hope the price is reasonable, because dev kit of microsoft is 599$ while firefly RK3588J ITX 489$.

#12

There is a space between the SBC world and Mac Silicon, I have wrote less than $200 because I am not sure what we are looking at in terms of form factor SBC or Itx style mobo?
Its a struggle as ARM 5th generation GPU means absolutely nothing as how many cores is supplied is a huge difference, with smartphones boasting MP20 whilst the Rock5 was only MP4.
Same with many Tops NPU.
I am basing on the lowest ram of 4GB, so its hard to say and why I have gone low end as really not enough info.

2 Likes
#13

Any price over 150-200$ is only worth it with mainline support and proper software in general. Below gets enough people to hack something together but above the devs that probably will need multiple SBC’s just won’t put that money in.

I am excited for the new SBC’s especially after being really disappointed with the Raspberry Pi5 SoC (lack of VPU,…). I just hope they’re actually gonna be viable options compared to what the Intel N100 did in this price segment.

5 Likes
#14

The race for performance with SBC does not sound good to me unless the power and cooling needs are kept low enough at a decent price.
I guess the announced jump in specs will significantly increase the price compared to Rock 5 models, at a level close or even above low power x86 and mini ITX boards for compact systems, where it will most probably fail. I think this will reduce the interest and sales volumes if the price is raised above some limit and the main target should remain embedded solutions and very low power systems.
Also, in the current family of Radxa/RK products, I miss a 2xA7x/2xA5x efficient solution as 4xA5x (Rock 3) are not powerful enough in most cases while 4xA7x/4xA5x (Rock 5) is an overkill for most users. Not sure ARM offers such core config anyway …

1 Like
#15

There are some problems with the SBC form factor when it starts to increase to those levels as the hardware is set in stone, even ram is soldered for little discount.

For servers often we don’t have enough Nics or fast enough and same with PCIe with not enough connectors.
For emulation gaming often the GPU’s lack cores when say my Pixel6a is a Mali-G78 MP20 whilst so far the most powerful is the G610 mp4 and only a 1 gen infront…
For ML Arm may have unified memory devices but the Api’s available still mean unlike Apples Core ML for many models and frameworks GPU/CPU/NPU are still utlised as descrete devices.

For similar $ there are likely already devices that are more capable and also more flexible. Again quoting my Pixel6a that has 2x Cortex X1 and has some great software but its all closed source by the vendor and outside of android there is near no support.

https://browser.geekbench.com/v6/cpu/4670629
https://browser.geekbench.com/v6/compute/1686014

#16

Unless Radxa can convince Rockchip to embrace and participate in the SBC open source community, allowing them to communicate directly with end users and listen to feedback and suggestions, there is no point in continuing to increase prices for higher hardware specifications.

TL;DR, asking users to choose between an Android kernel and a featureless mainline kernel on a $200+ SBC is unacceptable.

10 Likes
#17

Are you able to tell us which SoC the new SBC will be built around? If it’s not in the ROCK family it’s presumably not from Rockchip (and they don’t have anything more powerful than the RK3588 on their latest roadmaps). Is it MediaTek?

#18

While I do agree with that statement - the x86 systems requires more electricity and the passive cooling is all but dream, even N100 heats like hell, and you can’t place some random heatsink on it with 5v ultra-low RPM fan.

If we could have arm with 8 lines (even as 2x m.2) that already would be awesome board for 10G NAS with up to 5 disks without exteral pcie switch) and up to 20 with pcie switch (and they usually cost a hell a lot of money)

#19

Have to echo most of the points already made, primarily mainline kernel support out of box instead of waiting months/years with functioning gpu/vpu/npu drivers. There is no point in paying a premium for an SBC that does not have necessary software support behind. Having worked on commercial embedded
projects there is a a lot of value placed on having the necessary software support in place especially when a device is expected to be in service for a number of years. I’ve looked at the Rock CM5/5b for a few of clients however it seems to fall short on the software side as time to market becomes a deciding factor.

1 Like
#20

That’s actually silly dream… And most of it is not even on radxa side, but on cpu vendor.

For example just adding rk3588 took at least 2 years. So you expecting to “ready to sell” board just to sit, while you still spending money on up-strem kernel patches (by money I mean employee salaries)? That’s just doesn’t work like this. First you sell, then you starting adding support from not included OS.

Not even raspi have it out of box, and raspi is backed by broadcom and A LOT of ppl