It seems Arace is accepting pre-orders.
Upcoming ARM v9 product with 45 TOPS NPU
Specwise feels nice. Cortex-A720 has MTE, MPAM, SVE, SVE2 and many other cpu features I would like to play with.
But boardwise? No traces of SoC in mainline kernel nor in EDK2-platforms. “@cixtech.com” is present in THREE kernel commits.
This is Arm. Mainline support is not granted.
I wanted Arm desktop 10 years ago. Gave up on getting such a few years ago. And then board appears without trace of support.
How many months/years it will take to get it fully supported in mainline?
Yes but prohibitive shipping costs to Europe unfortunately (dhl/fedex) 
mmmm i was thinking, if the main PCI-E slot support bifurcation maybe Radxa could make an adapter board to 4 M.2 slots @ x2 4.0? It would be great for NAS.
There are already plenty of boards doing that, and some even using a PCIe bridge, without requiring bifurcation. I don’t see much value for Radxa in creating yet another one, it’s like making one’s own RAID or SATA card.
1 Like
Because the pcie is x8 electrical, the adapters in the market that relies on bifurcation would be only be able to do 2 M.2 at x4. So it is only possible with adapters that include a bridge, and those will likely be super expensive or sacrifice perf.
If instead you build one adapter you should be able to do 4 M.2 slots each one x2 4.0 PCIE, if it is wired correctly for this case, no bridge and nothing fancy requiered, just the M.2 slots and some supporting electronics to handle power delivery.
8 M.2 @ x1 4.0 might be possible too, it all depends on the bifurcation capabilities of the SoC.
2 Likes
There’s still some choice, in 5 minutes of searching I could find:
- PCIe x1 to 4x NVME via a bridge for 35 EUR
- PCIe x4 to 4x NVME via bifurcation for 84 EUR
- PCIe x8 to 4x NVME via a bridge for 137 EUR
I don’t find these overly excessive. Also the benefit of the bridge is that any device can run at full speed, even though I agree that 2xGen4 is already great for a single device. Bifurcation requires support from the PCIe controller and I don’t know if all support many combinations. The fact that I can’t seem to find a passive x8->4*x2 seems to indicate that support even on PCs is not granted :-/
Based on the pricing of the items you found I would suspect that the PCIe switches used there are Gen3 or even Gen2, aren’t they?
One thing people await now for years is something like a QNAP QM2-4P-484: x8 to 4 x x2 but with a Gen4 switch (the older QM2-4P-384 and QM2-4P-284 are based on Gen3/Gen2 switches)
Interesting point regarding PCIe speeds. The x4 one is shown as being usable with many serial ports and doesn’t mention any speed, so it could very well be a gen2 or so.
The other one (e.g. that one) explicitly says PCIe 4.0, and it gets an x8 connector. It can be found from 137 to 155 EUR on the same site depending on vendors, it’s just hard to search (aliexpress search engine…).
For the 84 EUR one, I just can’t find it anymore at the moment (search engine again) so I can’t check.
And that’s misleading or outright wrong since the specs mention the PCIe switch being an ASM2824 which is PCIe Gen3 only. So it works in PCIe Gen4 and even Gen5 slots but the negotiated speed with both host and devices will always be Gen3.
Since that one mentioned bifurcation there’s no PCIe switch involved which could downgrade the speed so if I understand correctly ‘all’ you have to deal within a passive Gen4 setup is signal integrity which then might downgrade the negotiated speed to Gen3 or even Gen2 based on results of link training 
All that is fairly plausible, that’s the magic of aliexpress. I too tend to search for chips to verify the specs, but here I didn’t see it :-/
I also noticed that,
73€ DHL and 83€ fedex, both of them will charge yet another 15€ for import duties + tax,
$80 for shipping to USA,
32GB RAM option with case and including all costs will be about 500€ (and roughly the same in $)
Also I see that all packages are now about 2-3x times more expensive (even for smaller things like X4). I expect that this policy is caused by christmas delays, I’ve seen this in few shops, on the other hand it’s pre-order with postponed delivery, when prices should be normal and some cheaper options (like 4px) should be yet again available.
@Arace @Arace-Martin @Eric-Arace - any chance for a change?
2 Likes
Looks like an awesome machine!
Will there be a fanless metal case?
Is the sound card at least decent? With previous onboard sound on the Rock 5B I recall it had awful background noise and I had to use a USB sound card to not have that.
I am Crush from Arace shop.
Thank you for your interest in our product!
Regarding your feedback that the shipping cost is too high, we have taken measures to optimize the shipping cost. By adjusting the packaging to reduce the volume of the goods, we have successfully lowered the shipping cost.
Actually, we set the shipping cost based on the actual shipping cost table of our forwarder to ensure that you are paying the shipping cost based on the actual cost and we do not charge any additional fees.
3 Likes
Thanks @Arace for your response. I trust you to make the shipping price as low as possible with the shipping company, but nevertheless it remains at 167 EUR for me for 2 boards, which inflates the price by 25%, this is super high. The problem doesn’t come from you but from these companies (DHL and Fedex). Maybe you should have an option to use a regular postal service ? This is particularly valid for pre-orders where the consumer doesn’t expect super-fast delivery (which is the main reason why these companies charge this high for transporting your products).
Thank you for your support!
For some high-value products, although regular shipping may offer a more economical option, there is indeed a risk of lost packages without compensation, resulting in significant losses.
In order to ensure customers’ shopping experience, for high-value products, we choose to use DHL or FedEx logistics companies, as they provide more reliable transportation services and better guarantees in case of issues.
It actually depends on companies. For example in France you can have an insurance up to 1000 EUR for approx 1.25% of the declared value. In the example above it would have cost 9.40 EUR for the 665 EUR parcel (and it can probably be lowered further if you declare the product value, i.e. loss for the sender, instead of the customer’s price). I guess that most european postal services have similar options. These are way less expensive than DHL/Fedex where most of the cost comes from the guy paid to wait in front of your door while you’re at work…
Totally agree
DHL and Fedex has significant higher prices for import duties, they demand for order confirmation and screenshot to declare item value and add as much as possible for tax and their safety. Probably they will not include any early bird discount (like the one for orion) so base amount for all taxes will be higher.
So high price can be justified by really fast delivery, but that is not the case here 
For example Trading Shenzen sends 1000€ phones with 16€ shipping and I never hear about any lost package. Here by default all packages are insured up to 1k€, only once I needed to check that out and I got refund
1 Like
Not to mention that FedEX is either a complete and utter mess or a total scam when it comes to tax and import fee collection.
I had to pay the courier in cash once, because the fee amount was below the 150 euro threshold and they abused me with repeated invoices and payment reminders (the fees were paid for because I wouldn’t have been able to receive the package otherwise). They never responded to my messages either.
I’m glad that for cheaper stuff 4PX can be used. And it’s pretty reliable.
2 Likes
Do we have any info yet about the differences between the “Big” vs the “Medium” Cortex-A720 cores in this SoC? I’m assuming it’s more than just the clock speed difference: 2.8GHz vs 2.4GHz…
Perhaps the Mediums have been configured for the area optimized implementation (same area as the Cortex-A78, but according to Arm giving 10% more perf than that baseline plus v9.2 arch features)?
Looking at footnote 2 from the page I linked to above the area optimized implementation has L1i and L1d caches that are 32KB each and a 128KB L2, as opposed to 64KB L1i & L1d and a 512KB L2 for the “full” config.