What do we want in the next Rock Pi?

CPU architecture does change from A72 to A76 that it would lead to higher IPC (hence in WiKi ppl claims that it goes up 2x DMIPS from A72@4.5 to A76@9.5), where a simple die shrink would only help in power consumption (which is also a good thing, of course) and a bit high core frequency (but you still need a deeper pipeline in general to go for higher clock speed). Even if the Wiki figures are a bit off, the 2 extra core should definitely bring the overall perf up 100%, which is actually very nice.

RK3588 is going for (Samsung) 8nm which is what nVidia RTX 30s are using, so I think given the GPU shortage (AMD CEO said we can’t expect worldwide fab capacity to turn better until end of year) it is safe to assume Radxa could only get their RK3588 Rock Pi sample / dev board by then, and release the final production board in Q1 or Q2 next year. A long wait for me indeed, sigh.

Yeah suppose so as really its an upgrade from the 75 > 73 > 72 but yeah the 8nm is still a backlog and likely a long one as there is a lot of high price silicon that has its foot in the door.
I have wanted a new GPU for over a year but its just crazy prices still, double MRP!

True.
I don’t think anyone have recieved any prototype yet.

For me anyway the RK3588 isn’t all that big of a loss anyway as the general purpose application area is sort of getting crowded with options from various vendors where Intel to AMD are also converging.
I am a big fan of reuse and my $85 NUC (i5-4250U) that had 8gb & 120gb SSD has been great buy and still on my todo list is to do something with the $25 coral mini pcie 4tops NPU.
Got a strip down, clean and new thermal paste and maybe the fan is noisier than new but comes in great for trying X86 only software.

So maybe not another general purpose application SBC, the Radxa Zero looks really interesting for multimedia as its video & audio capabilities look really great and fretting at the moment that uniqueness hasn’t been lost whilst opting for pads general purpose.

The RockPiS should of been a great audio SoC but dunno why the ADC is so noisy and the fixed PDM mic array via the POE hat sort loses all flexibility and cost effectiveness which has been a shame.
The A35 are great low power cpus but the way the memory was implemented it also ended up a bit slower than I expected.

The RK3568 PCIE3.0 1 ×2Lanes/2 ×1Lane@8Gbps is interesting seen as how the Sata enclosure picked up so much interest, the onboard 0.8t NPU is a bit limited but again with PCIe like the NUC you could add a Coral accelerator.

EdgeAI in terms of development and investment has overtaken CloudAI and maybe its about modules such as the Rockchip imageAi dsp chips or the Chipintelli voiceAi chips or NPU’s.

Adafruit look like they are going to add a 4 channel ADC to there line for use with a Pi.


Depends on the cost but thinking of just hooking up 2x ADC to a Radxa Zero, would be great if the noise on the RockPiS ADC was fixed though as that was the unique selling point of that SoC.

I didn’t see the release of the Radxa Zero so was a little late to the party, could be very interesting and waiting for a delivery.

With the potential that the RK3588 may never materialize, or that it may be high priced and in short supply if/when it does materialize, and with Amlogic A311D and S922X CPU based SBC’s clipping along at benchmarks nearly doubling the performance of RK3399 CPU based SBC’s, is it time for Radxa to consider an Amlogic CPU at the heart of the next Rock Pi (Rock Pi 5)?

RK3588 will be out in October as planned. And I assume we have the SBC out in Dec and shipping a batch before the Chinese New Year. ROCK Pi 5 will be RK3588 and I promise it will not be expensive. Remember, we build SBC for developers/geeks/hackers all over the world, we care developers who lives in less developed or developing countries.

s922x be in the consideration.

7 Likes

Yay, so just 4 more months until rk3588!

Its been the silicon delay that has been the problem but from specs the RK3588 seems to beat the s992x all round and after this length of time I don’t think a couple more months will be any harm.
The RK3588 has an extremely impressive set of specifications and guess why the delays have caused frustration.

S992x might be interesting as a cost effective step up from Radxa Zero maybe in a Pi3a+ sort of format of keeping it simple with a really good gpu/cpu price index.

This is indeed good news about the RK3588 and the Rock Pi 5. Will Rockchip deliver the RK3588 at the hoped for 8nm process?

There is one thing that might crop up with rk3588 and still slightly confused to which GPU but as far as I know only the Bifrost have MESA drivers which ends with the G52 being the highest with native linux drivers?

Question mark as am I right about that as don’t think Valhall has native driver support?
Panfrost lead developer Alyssa Rosenzweig commented today that reverse-engineering work has begun for Valhall with a focus on the Mali G78 and maybe will be faster than the time it took for Panfrost drivers or that is going to be some wait.

Would it comes with “reasonable cases” at release time?

I’m not familiar to Rock Pi but am looking forward to getting the new one based on RK3588. Checking the RockPi4 tutorial, it seems that it follows a [ 1st, 2nd, kernel ] boot sequence, right? would UEFI (based on ed2k) in addition to u-boot be a possibility as a 2nd stage bootloader?

If you will write uefi for it - then sure thing, it will be possible

Hey @jack

What’s cookin ?

https://twitter.com/theradxa/status/1423127586616446979?s=19

Is it?

1 Like

too bad that I am an enterprise architect now, not working with low level stuffs anymore so I can’t make it, though I know there are ppl doing so with Pi defintely, and as mentioned, based on edk2.

This is the one Debian relies on in making it bootable for the distribution.

IF the rock pi 5 is still in the design phase plz copy the layout of the ROCK 3A so the m2 nvme x2 slot is on the back and PLZ PUT THE SLOT IN THE RIGHT DIRECTION SO it dossent stick out at the wrong side and rais it 1 to 3 mm if you need clearence + makes the sbc a a lot more compact…the ribbon cable extenderboard makes no sence at all it only introduces more parts/points of failure and slower transfer since data needs to travel longer then direct mounting in the m2 slot…And if you think you gonna miss out most of us that buy a sbc have a set price in mind…if we have to spend 10 doller on a rather useless extension m2 card that means we will have to shave 10 doller of somewhere else…aka less ram…if you sell more boards with the higher ram on board you can get better bulk price for ur ram chips…

2280 (22mm x 80mm) dimensions of nvme
the socket itsel is 5mm so thats 85 mm(with side mounting)
dimension rock pi 85mm x 54mm
even if side mounting was a no go for radaxa surely adding 2 to 3 mm to the length of the board would not be a issue?) iit seems rather dumb cuz if you mount a board on top of it thats bigger then the board itself you might aswell add 3 mm to the board and keep other positions the same to keep compactebillety

@jack Did the possibility of a S922XJ companion to the Zero garner any further interest?

That would be the Zero Plus.

1 Like

That would be awesome, not sure about the name, but specs are pretty amazing. (Can not think of a better name though as it also gives version increment space, zero-one & zero-one-plus if there was every such a thing)
Would that be on a board something approx the size of a Pi3A+?
I deliberately put the J versions with the MP6 Mali as I think there is a lot of interest in a GPU biased model of relatively simple connectors.
Then again when it comes to reported specs and model numbers there seems to be a lot of confusion and its anyones guess but the extra power of the MP6 would likely be very saleable if cost effective.

I also keep thinking a higher density GPIO connector(s) might well be about time otherwise it could turn in pin mux revision musical chairs.
I keep wondering if higher density connector(s) (ribbon cable) have a range of offboard 40pin GPIO daughter boards so you can recreate common gpio and cope with different pin mux whilst also not losing pins.