A new SBC from Radxa featuring the Qualcomm QCS6490 looks exciting.
radxa-dragon-640.webp (104.2 KB)
Source here https://mp.weixin.qq.com/s/SY96PzYzPdP3MISoz0n9nw
What time frame are we looking at for this devices release? Will Radxa make a SoM/CM version?
Is this official?
I would love to see a SoM/CM. But also a pico-ITX version like 5B/B+/T for better cooling management and richer/cleaner I/O.
I wrote an article about the Dragon Q6A. If you have any questions about the board I should cover in the full review let me know 
https://sbcwiki.com/news/articles/radxa-dragon-q6a-unboxing-and-first-look/
Did they implement USB 3.1 Type-C with DisplayPort output?
Can the board go to āsleep to ramā mode (like a tablet / laptop) out-of-the-box and if so, what is the current draw like?
Iām very interested in testing this board too. I am working on another vendorās board but is stuck on kernel 5.4.219ā¦ incredible.
Also very keen to explore the bootloader and see if thereās a chance of a clean Android / Linux dual boot solution!
This one I can sadly already answer with no. USB-C is just for power.
I wish they wouldāve used the native DP capability of the SoC instead of routing it through the RA620 for HDMI conversion and maybe use MIPI to HDMI for the port
Arenāt you concerned about the poor multi āperformanceā? Itās an octa-core SoC and multi is just three times the single score. Seriously?
When you check the individual results, you can see that thereās a huge discrepancy between results. I think it would be fair to expect at least 4x from single to multi, due to 4 big and 4 small cores, while the big ones combined deliver a total of 3.6 times the biggest oneās capacity. But some tests like file compression seem to be I/O bound since just about 1.5 times faster in multi-thread. Others are 3 or 4 times faster. We do not really know what is affecting performance this much. It could even be a low frequency DRAM which prevents cores from scaling.
Thereās no need for this since Geekbench 6 is known to not be a benchmark measuring āmultithreaded performanceā but a joke.
In this forum I multiple times already linked to @geerlingguyās testing of an 192-core Ampere ARM server that when āmeasuredā by GB6 is ā11.5 times fasterā multi vs. single, LMAO!
How to āimproveā multi performance? Just use the predecessor Geekbench 5 and now your machine āperformsā 84 times faster multi vs. single (while more realistic benchmarks like 7-zipās show a 155:1 ratio which resembles real-world performance more closely). Also just look at the consumption figures (279W GB6 vs. 586W with GB5)
TL;DR: GB6 (on anything other than x86) is a joke but people donāt care
Qcom documentation indicates that the QCS6490 (Snapdragon 782G) has only 2MB of L3 cache. This compares to the 4MB L3 of the competing QCS8250 (Snapdragon 865) and the 3MB L3 of the RK3588, which may be responsible for the lower scaling performance.
The thing that āreports performanceā (benchmark) is software and software contains bugs. In GB6ās case itās long known and pretty obvious that this software uses weird and highly uncommon memory access patterns that do not represent anything happening in the real world.
Why āperformā the four A76 on RPi 5 only two times faster than a single one? Since āmeasuredā by Geekbench 6. Want faster multi performance? Use GB 5 (or any other more realistic benchmark that doesnāt suck too much).
The Raspberry Pi Ltd. people even paid a contractor to hack together a āNUMA emulationā for their kernel which does one single thing: improve GB6 multi performance by a few percent, no other software affected.
Hey! Whatās fun with Radxa is that I really learn about new models when opening my mailbox after I get back from work 
Itās always a bit like Xmas before the date. I first thought it was the components I ordered from Ali that arrived early, until I unpacked everything and stumbled on this:
The specs are written on the back:
BTW itās strange that itās written ā6GBā on the model while valid ones are 2/4/8/16. Probably that itās 16 and the 1 wasnāt printed. Iām afraid Iāll have to wait a bit before testing it, Iām quite busy at the moment and next week is kernel recipes. Will probably check on the next week-end. Thanks to the team!
Nope since thatās inside:
RAM configuration:
description: TSOP 6392 MHz (0.2 ns)
physical id: 0
slot: Top - on board
size: 6GiB
width: 16 bits
clock: 2097MHz (0.5ns)
Hmm at first I was a bit confused by the ā16 bitā width mentioned above. I checked the schematic, and itās a dual-channel 16b (thus 32 total). I couldnāt find any reference for the marking on the board (CG439005200A0). I could imagine it would be a 6400 MT/s since itās what the SoC supports. But indeed, if yours detects 6GB, that should be it. Itās just surprising since not listed on the list of optional sizes on the box. Maybe thatās the size chosen for the early models.
Edit: the spec here mentions ā4GB / 6GB / 8GB / 12GB / 16GBā as the DRAM options, so itās just the printing on the boxās bottom thatās not up to date.
The 6GB and 12GB model was added later when we found that at this size it would be a price/volume sweet point. The dram is running at 5500MT/s at the moment.
Yes, with a newer version of the firmware, the actual memory frequency will be reported correctly in SMBIOS.
OK thanks for the clarification Jack, that makes sense!
Do you plan on increasing this in the next few weeks (to mark benchmark scores generated right now being preliminary)?
I actually care a lot about your insight. To answer your question if I am not worried about the performance results: At least not yet but I also only was doing only a first look. Iām planning to do a proper bench with your sbc-bench too as well as go more in depth in the full review. For why I am not worried is that the chip seems to be more power efficient or restricted than other Quad A78+A55 chips so my numbers were only early results with one benchmark that is only there to have a number at all.
I might not be able to squeeze every % point of performance out of it but I will try to provide a allround solid armbian experience while always being open for suggestions. (I still plan to incorporate your suggestions which were wrongfully dismissed for RK3588 someday too)