Its not a silly dream, Apple ARM follow this approach & from what I see Qualcomm is moving in that direction. Its up to Radxa to select a CPUs which have a viable s/w stack. If its taking 2+ years to get RK3588 up streamed then as a business can’t you wait 2 years to bring a product to market. From experience I’ve dropped ARM & move backed to x86 to meeting client deliveries. For hobbyist it may be a different story.
Radxa 6 under development
2 Likes
At a guess yes if reading between the lines of the quad a78 also…
I don’t think Rickchip is too concerned there as they are looking to ship devices in qty, not just a few to hobbyists. The N100 doesn’t need a random heat sink as it comes with the correct cooling system as a ready device.
Likely a energy efficient quad A55 like Soc with 4 connectors of pcie x 2 lanes as seems embedded devices are often x2 lanes. It could be ausustor like with x4 nvme, Nas with Sata cards and additional highspeed 2.5gb Nic or USB or even AI cards.
Also combining a bigger big core count and GPU with a much higher core count, could provide for both the emulation and ML market with the ability to run LLMs via vulkan. The RK3588 without its GPU managed 4.64 Gflops/W
Both have specific focus but still cover a range of devices and by dropping unnecessary peripherals it keeps die size and cost down and eventual pcb size.
The nature of how things work is that Android BSP backed devices tend to come to the community with known IP from previous models and the SBC community does with it, what it can.
Mainline always takes time and some SoCs get very little attention.
The more interest and vendors behind a SoC, the faster things progress.
Mainlining sometimes gets bypassed by BSP backporting and can be a matter of stars aligning at a time…
Apple arm do WHAT? Man, apple arm doesn’t work in linux mainline, wake up. They quite literally develop their own OS for their own hardware as top 2 USA company.
As for Qualcomm here:
Release:
ThinkPad X13s - 2022Q1
Support:
2023Q3
So yeah, it is silly dream.
Well, they do ship it in quantity, you underestimate amount of hobbyist.
And this “correct cooling system” is loud or too hot.
Mekotronics Genio 1200, but the processor was out in End 2021 if I remember well.
(Not optimized)
And this “correct cooling system” is loud or too hot. without exaggeration price like seeed studio 120$ water cooler solution (just a bad clone of Michael Klements DIY fantastic projects) Rock Pi 6 could have 2x pins for top and bottom Ice tower cooler & Ice low profile cooler that reduce heat of the chip (it is very simple and cheap buying two good coolers if will get 4x X1 with 4x A78).
For that price, defenitely not gonna buy. That price range is very dangerous for ARM, because regardless of being powerful, you have much more powerful x86 mini PC’s with RAM, SSD, wifi and everything. And with recent CPU’s.
Not feasible, the board should always stay between 100-150€, no more.
3 Likes
Maybe Radxa should think of releasing 2 boards, something like R6 (basic functionality, CSI,DSI,M2,USB3, $100 ~ 150) and R6+ with all the features. ( > $250). It can please everyone…
I really like what is being done with the Zero boards and even the Rock S0 and the use of MX1.25 as IO headers.
There are quite a few boards where going along Rpi Pi terminology the B is the full board whilst the A is the cut down board that loses IO connectors (supposedly) but also maybe just shifted to a MX1.25 header.
I am actually the same about GPIO that maybe power pins stay 2.54mm but a higher density FPC goes to a daughter board that recreates common Pi 2.54 groupings.
Also as performance and cost increases dedicated soldered ram is a substantial part of board costs. I put in a voucher https://arace.tech/products/pre-order-milk-v-oasis-16-core-risc-v-desktop-coupon and I am prepared to spend around the $120 mark as the ram is standard DDR5 which because its quite a beast https://www.phoronix.com/news/SG2380-RISC-V-SoC-Upgrade I am prepared to go for 64-96gb because of my interest in LLMs, because at least if all fails I do have standard exchangeable dimms.
As boards increase in value the risk as an early adopter increases sharply as usually there is a balance in NVME, RAM and peripherals and as a board increases in value generally so do they.
Sorry about the name drop but when I saw the Sophgo @$120 and a voucher scheme it was almost instant that hell yeah I will risk a try on that. Its price range is the top end of what I want to risk as dubious from an unknown with an unknown of RiskV but what it has done is invest heavily with a focus to running LLMs and if this does then it will be amazing, but for other use it is prob not all that attractive.
$120 and add your own ram helped though as for LLMs really prob talking 64gb min.
Matching soldered ram on higher end boards is a big early adopter risk.
PS whilst I found the token link I noticed this.
https://arace.tech/products/milk-v-vega £80.00 for a 8 port 10gb managed hub is really cheap isn’t it?
2 Likes
It’s 2 port 10Gb, and unknown ASIC if there any
Ah not that cheap as I 1st thought but prob about %50 of mainstream brands at a guess, yeah the ASIC is RISK-V as that is what the Vega is a network switch from a brief glance.
If it had been 10g with 2.5g ports I would prob risked it 
1 Like
I would wait for interesting two parameters - power consumption and some benchmarks for those ports. There is nothing worse than slow and power hungry device, I think that it may be some issue here. I’m also looking for some network upgrade, I could get cisco 48 port 10G switch with 40G uplinks from work, specs were awesome and I had such space in my rack. But I found out it will eat about 640W on idle, which is ok for datacenter but not that much for home usage 
So I’m still on my super cheap d-links (48x1G + 2x10G), takes about 1.2W per active port
Dunno tdp is 4.5watt but likely would still prefer a https://www.broadbandbuyer.com/products/42569-qnap-qsw-2104-2t/ as likely have servers connected on 10gb with a 2.5g client network
It was purely something I saw when getting the voucher code for the Sophgo link as near all end devices I have are still 1gb, so not that big an interest as you only need 25 Mbps to stream 4k…
Currently because of content used and diversification of use, its not much of an issue at home.
It’s quite hard to respond with a number without sizing info, but the following points must be kept in mind:
- it must never be more expensive than the equivalent x86 device with the same features and performance ;
- a quad-core N100 at 3.4 GHz with 8G RAM and 256G SSD that consumes 6W is sold online between $147 and $169 with its enclosure, and has according to various tests online, roughly the same perf as the 8CX gen3 (some + and some -)
Thus I’d say that if Rock6 has nothing better than a $150 x86 box, it should stick in the <$200 range for the entry level otherwise it might be a failure. The success of ARM-based SBCs has always been to provide lower power draw and lower price for those who don’t seek x86-level performance. The Rock5B is the best example of this, it gives the feel of a small celeron at a similar price. But when approaching more decent x86 price, buyers will really think twice (stability, mainline support, availability of drivers, choice of OS, ease to repurpose the board etc).
There will always be some enthousiasts who will buy an ARM board because it’s ARM, because they want something different from x86, because they want a nice dev board etc. I guess a few of us here count as those who would buy one sample of such a board to give it a try and to have one ARMv9 at home. But I don’t think it’s realistic to count on blind fanboyism like RasPi does, so it’s very important to be careful on this point. E.g. if the board’s price starts high, you might be tempted to add some goodies on it to justify the price, and if it doesn’t see adoption it will be much harder to lower the price later.
I think that RAM size is a good justification to increase the price, but not too high, because users hesitate: if I buy a $300 device with 32G and finally don’t use it for whatever reason, I lost $300 and I cannot reuse its components. In the x86 world, you can often pick the DRAM stick for another board, or you can replace it with a larger one if you started with the small version.
A good way to keep the price low (and simplify routing) is to remove the storage: I think that nowadays users expect their storage to be on M.2 because it offers a lot of flexibility and performance with lots of choices of price/quality. So I’d just not put any eMMC on the board, just an SPI NOR to store a UEFI boot loader to make sure it boots out of the box on standard distros, and let users choose between the proposed SSD or SATA adapter or buy theirs. A 256G SSD is only $17, it no longer can be a justification for the price of a board.
2 Likes
Last time I checked X13S it was indeed “supported” in that it boots, but it showed much worse performance under linux than windows, most likely due to incorrectly tuned DDR controller. Not much surprised, we used to see this with RK3288 in the past with devices running with RAM at 200 MHz, often caused by u-boot only being built with the default training code. It could possibly be the same issue there.
1 Like
You’re right. At this point I’ll wait for Mongoose ARM-Samsung chip, that will get good perf and decent price.
Apple this year will have a new competitor like Qualcomm, with Nuvia chip, that has new competitors not only in mobile but in future also in desktop world, like MediaTek and Samsung (with Mongoose desktop chip).
just yesterday i was benchmarking my PC, and this is the value (yes, don’t write me nothing, because ATT isn’t optimize for windows and for desktop, but hey mobile chip nowadays can run games that decade ago was only for PC like total war, tomb raider, resident evil exc.)
1 Like
@willy sorry but this is wrong. A 6W CPU TDP means almost nothing. It won’t “consume 6W” at any point, unless you only count the CPU and no other components and disable turbo (which would make the system unusable). TDP only refers to CPU power at the base clock (800 MHz in case of N100).
I think many news outlets are complicit in perpetuating this misconception as they only copy and paste “6W CPU” with no explanation.
All TDP are wrong nowadays anyway, what matters to me is the amount of cooling required for a given task.
The Rock5B can just run fanless, just like the N5105 in my Odroid-H3 which is in the same price range (though a bit more expensive). The former delivers 1730 RSA2048 keys/s while the second delivers 2591, or just 50% faster. Compressing a large file is 15% faster on the x86 as well. All of these are elements that matter to me.
The N5105 is given for a TDP of 10W but it’s still fanless both on this machine and on the even smaller ones we have at work (MeLE quieter 3Q). If the CPU vendor uses the same measurement method for both CPUs, it’s likely that the 6W N100 is also suitable for being fanless. And if the vendor changed the measurement method, well, let’s stick to N5105 then
There is no mystery here. TDP refers to the base clock. No CPUs today run at base clock speeds. 10 W for N5105 also refers to the base clock (which is 2 GHz). So there you have it: 6W at 800 MHz vs 10W at 2 GHz. Which one is more power efficient? Not sure, but TDP won’t tell you. Because they usually turbo anyway.
btw, I agree that how much cooling you need matters most, but with these x86 chips you have a lot of flexibility with the power limit (you can lower it). But usually the faster ARM chips are more power efficient.
Even though its been an argument that the Intel isn’t as efficient, likely many buyers are not bothered about the small wattage difference.
The N100 especially if your a Pi5 owner you have to be enthuastic for the SBC if you merely want a desktop.
I guess Intel are dumping product low as the Intel mini PC market is very competitive to the SBC offerings we have.
I also think the cost of embedded ram starts to be a problem the more ‘powerfull’ a supposed SBC is.
The miniscule cost in the difference of running a N100 vs a Rk3588 isn’t a topic many bother about.
1 Like
Anyway, what I’m seeing is that N100 boards are fanless with a small heat sink and I can expect to place them into a tiny enclosure. You see, I ordered my Odroid-H3 for my server while I already had the Rock5B which would have been quite sufficient, and the reason for my choice was that the Rock5B was relying on a crappy kernel that I didn’t trust enough to put all my data on it. And I figured that the H3 was overall even more performant than the Rock5B. I was a bit disappointed not to have been able to use the Rock5B only for software reasons, but was actually glad to see that my choice ended up being more reliable, faster, fanless as well and not much more expensive. If I had to buy a fanless board today, I’d definitely go for a more modern N100.
2 Likes