Given all the animation jank, i’d say we hit this a long time ago. However, the Denon engineers are effing artists and scientists for what they’ve been able to pull out of these low-end SoCs! 
I so admire them and hope they are compensated very well.
5 Likes
Supporting multiple hardware platforms is doable. But it requires substantial efforts to keep the old devices up-to-date and tested(!) while delivering new, exclusive features for the more advanced and powerful hardware platforms.
I am building industrial control backends that are deployed on various (embedded) linux platforms. These include weak ARMv7 devices, comparable to the RK3288. The Rust/C++/C code itself is portable and the target platform doesn’t really matter during development.
5 Likes
Denon is using the RK3288-C (32-bit ARM quad-core Cortex-A17) for their units.
Pioneer cdj 3000 (probably the xdj az and opus quad also) uses A dual-core Arm® Cortex®-A57 (1.5GHZ) and a quad-core Arm® Cortex®-A53 (1.2GHZ).
You think Denon will jump from a17 to a53? Or even use 2 mpus? What does the flagship Prime4+ have for mpu? Hopefully not a RK3288-D or E or F. What ever the prime4+ is using will end up in the new single decks.
None of these processors would be enough to meet the specific expectations of DJs regarding the next generation of hardware, such as on-the-fly stem processing/separation (since this is probably the most extreme feature requiring the most computing power, let’s use it as a baseline).
At least a chipset equivalent to the processing power of an Apple M1 processor would be needed.
In other words, we would need to target at least a processor from the Dimensity 9000 series from Mediatek, or a processor from the Series 8 Gen 3 from Qualcomm. So at least 1 x Cortex X 4, 3 x cortex A720 and 4 x Cortex A520. The hardware would also need to be equipped with 16 GB of RAM
I don’t think there is even any equivalent from Rockchip at this time.
Any new architecture that would be presented and that would be inferior to the performance of the previously mentioned models, would probably still have a lot of difficulty in processing the separation of stems in real time, in an acceptable duration while maintaining a very high quality of separation without pre-calculation.
If you have a chipset capable of doing this you have a chipset capable of doing everything.
But these processors have a higher cost, which mechanically would lead to a substantial increase in the final price of the units. On the other hand, betting on the choice of a high-end processor will make the units even more futureproof (the rockchip rk3288 was already a very low-end processor when it was released in 2014)
So it’s a gamble, equipping the new units with a more expensive processor from the start but also more high-end and more powerful would allow Denon Dj/InMusic to be the first to offer this new functionality but this time, on the fly.
This would allow them to further widen the gap with the competition by striking with a big blow. It would also be a hell of a PR stunt that would attract attention and probably make Pioneer sweat.
Because even if there was a price increase of 20 to 30% of the final product, it would still allow Denon DJ to offer products lower than the current price of Pioneer units. But if Pioneer had to respond by releasing something as powerful, they would be forced to question their business model so as not to need to practice crazy prices (Pioneer prices are already crazy for just a device equipped with a Cortex A57 and a Cortex A53).
This would also allow them to change strategy and promote the development of a range of products over the very long term with a truly futureproof platform. See what the guys at inMusic were able to get out of a low-end Rk3288 that is 10 years old, then just imagine what they could do with a high-end chipset. Pioneer could release 3 generations of equipment before Denon Dj’s is even equaled or surpassed.
To give you an idea, look at the prices at which Poco (Xiaomi) Smartphones are sold, you can find a new Poco F6 with 12 GB of Ram and a Snapdragon 8s Gen 3 (which will tickle 1,500,000 points on Antutu) for less than $300.
Would you be willing to pay $300 more for a much more powerful product that would not be outdated in the next 10 years? Personally, yes.
4 Likes
I’d pay $300 more for a product that didn’t have stems and had extremely fast response times. Just sayin.
3 Likes
And again I probably overestimate the price increase because a Poco F6 integrates a screen, a chassis, speakers, cameras, a wifi/BT chip etc…
So many elements that should not be included in the final price.
If we just consider the processor and the increase in RAM and, assuming that there is a commercial negotiation between InMusic and the chipset manufacturer for a reduced price in the context of the purchase in large quantities, I think we should probably be on an increase between $150 and $200
For real-time stems you need at minimum an Apple a13 chip (which is less than a M1). These chip are made specifically for AI processing.
An IPhone 2022se (A15 chip) = $650cad LC6000 = $900cad
Combine these two and you get the your solution for real-time stems for $1550cad which is less than a SC6000 deck.
Maybe InMusic should just build decks that an iPad with an Engine app that can sit in lol 
1 Like
It depends on what we are talking about when we talk about real-time stems, there are real-time stems and real-time stems.
That is to say, which algorithm and with what level of separation quality.
If you are satisfied with the muddy and messy separation quality of the stems V1 algorithm used by the very first version of Djay Pro’s neural Mix, or the stems V1 algorithm of virtual DJ, the separation is indeed in real time but the separation quality is horrible. And in this case maybe an A13 can be enough
But we will leave that aside because objectively this is not what we are referring to. What most DJs expect is the cleanest possible separation quality as evidenced again by Phil Morse’s comparative video between the different platforms “Who sounds best”. So when we talk about real-time stems, we only mean the Stems V2 algorithm (also called Stems Pro).
Because to achieve this level of separation quality you will need much more than an A13, the Stems V2 algorithm is infinitely more demanding in terms of computing power than the Stems V1 algorithm.
Releasing a unit that can do real-time separation using stems V1 would be totally counterproductive. You would give the competition all the arguments to make fun of you, you would be ridiculed because you are constantly compared to what is done on software (this is already the case currently with the pre-calculated stems of engine which are at the bottom of the ranking and which nevertheless use Stems V2).
No, you need a more modern processor that will be much more efficient over time. If it is to save $50 or $100 per unit and be much more limited, it makes no sense. I think most DJs who are willing to spend $1500-$2000 are willing to spend an extra $100-$200 to get something much more durable with unmatched performance.
Let’s stop using already obsolete chipsets to make penny-pinching savings.
This is probably one of the ways to go.
1 Like
How do you determine this?
I have an A10 iPad, I have an A13 IPad Air, I have an A15 iPhone and I have a M1 Mac.
They all run the current version of Djay pro with the latest version of stems . They all sound the same on every device though my Rane one and Akai Amx. The only difference is that the latency activating stems on the A10 is high, the A13 the latency is less but noticeable, the A15 and the M1 is barely noticeable.
The only other software that I tested with stems is Serato and it wasn’t that bad but Algoriddim is better.
Do you even do any testing?
2 Likes
Economony of scale…
Just because a mass market device like a phone costs X does not necessarily mean a niche market product (DJ equipment) could be produced for just X more than the current price.
Stem algorithms work based on the processor’s processing power, that is to say how fast the processor is able to perform this separation.
So when you are a developer, you only have two axes on which you can work when you implement a real-time stem algorithm, the separation quality and the separation duration.
If you opt for a setting of your algorithm that will favor quality then depending on the processor on which you run it you will obtain a separation duration that will be more or less slow or fast.
This is the choice that Djay Pro seems to have made. Of course you get the same quality of separation at the end of the process but the processing time will be much slower on an A13 than on an M1. But where your M1 will perform your separation in less than 15 seconds without being saturated, your A13 will suffer by being at 100% of its workload and the separation process can take up to practically between 1 and 2 minutes depending on the tracks.
Of course you do not realize it because it is a processing that is done in the background. But by using 100% of the available resources between 1 and 2 minutes you expose yourself more to lags, blockages, or even crashes especially for live performance.
You can choose to choose a setting that will favor the speed of processing to the detriment of the quality of the analysis and separation. You get a much faster analysis that is less resource-intensive, but also much more superficial and less precise leading to the appearance of audio artifacts or frequency residues from other stems.
Or you can use a balanced setting, to find a kind of compromise between quality and speed of analysis in which neither the speed nor the quality of separation will be exceptional.
It is obvious that if Denon DJ were to release new equipment and wanted to keep its reputation for innovation, they will probably seek above all to favor quality because they know that they will be expected and compared on this point.
Either they arrive with muscular hardware and become a game changer that puts everyone in SLP, and which is designed to be futureproof with a large margin of performance.
Or they arrive with a limited entry-level processor, very fair or obsolete as they did with the rk3288 at the time to make some economies of scale of a few tens of dollars per unit.
And in this they risk finding themselves stuck with the next unanticipated technological advance (for example if Zplane releases a new Stems V3 or v4 algorithm that is even more demanding in terms of computing power).
So if you want to have both computing speed, very high quality and a power reserve to anticipate future advances you need a powerful chipset.
Also note that the most recent manufacturing chipsets have finer manufacturing processes. An Apple A13 uses a 7 Nm process while a snapdragon 8 Gen 3 uses a 4 Nm process. The finer the manufacturing process used, the less energy the chipset consumes. Which can significantly reduce your electricity bill or extend the autonomy of battery-powered units such as “prime go”
It is also important to point out that currently Stems engines do not use AI, it is an urban legend, finally it is mainly due to Algoriddim who made some marketing blushit with his Neural Mix story suggesting that it used AI calculations in one way or another. But in reality it is just marketing because AI is in the air, so it sells, so companies put the word AI on everything to look cutting edge. So NPU type units or AI calculation units are useless. A good GPU on the other hand, much faster for calculating complex algorithms is much more useful.
Stem engines are based only on algorithms, complex and demanding in computing power certainly, but they are nothing other than advanced algorithms.
1 Like
If the general public is willing to spend $1000 to $1300 on an iPhone that is certainly not worth its price (raise your hand if you have an iPhone here) while a Poco F6 or X7 pro gives you comparable performance for €700 to €800 less depending on the iPhone model you are comparing it to, then I think that these same people will have no problem spending an additional $100 or $200 to get a unit with a much more powerful chipset.
For my part, I would much rather settle for a Poco that will more than meet my needs and will only cost me $300 and put the savings into DJ equipment with a powerful processor. Because I will need this computing power more in my DJ equipment, rather than in my smartphone to browse the internet, go on social networks or play PUBG
On top of that, DJing and music in general is highly devalued. Most could not justify higher prices for dj equipment, which is why controllers an be so appealing.
It’s hilarious to me that when this art is devalued so much, you find a flood of new entrants into the market.
Why?
Likes and subs. Attention economy is strong!
Those are very nice predictions you have. I’ll wait for when your comments actually reflect reality.
I know for a fact that there will never be any perfect stem rendering. Governments around the world spends billions on their own version of stems for the imaging of the CMB and it still useless and has no reflection on physical reality.
Unless you know all the properties of the source material and can actually have control of the source materials, like what Native Instruments originally wanted to by trying to get artists to submit the original sources, you’ll always have ad hoc stems with artifacts no matter what processor or algorithm you use.
It will get to point where it will be pointless to go any further with it.
Stems can obviously be improved but right now it’s working well with me and for others. There are not many complainers on the Algoriddim forums about their version of stems btw.
You’re not disagreeing with me then, that smartphones sell in far greater quantities than DJ equipment? 
The most “close to perfect” stem creation will happen when artists decide to create a master and stems as product releases. In the post production film world, stems are created as part of the final mixing process.
2 Likes
Obviously I agree with you on this point, smartphones are a consumer good that is much more mainstream than DJ equipment, and in this respect I agree with you regarding the fact that the number of copies produced in the world is much greater than for DJ equipment. It therefore seems logical that the discounts on quantities concerning components are even more significant. This is a fair observation.
But this does not take away from the fact that sometimes (not to say often) DJ equipment manufacturers, in order to save a few bits of money, choose much more limited components, not to say already totally obsolete, while for a few dozen dollars more you could get much more efficient components.
When you buy DJ equipment like a prime 4+ which costs almost $2,500 or an SC6000 player which costs almost $1,700, you are no longer a few dozen dollars away from getting a smoother, faster, or more efficient experience.
Just look at the crazy number of people in our own niche that is DJing who are willing to spend $3,400 on an XDJ AZ or over $2,500 on a CDJ 3000.
If we take the example of mid-range Chinese smartphone brands like Xiaomi, we estimate that they take on average at best 20% margin per phone sold. And assuming that the chipset is one of the most expensive components of the product and that it represents about 1/3 of the manufacturing cost of the product alone, this would mean that Xiaomi negotiates it at around $80 to $100 per chip depending on the model.
Let’s take the most pessimistic range where Denon does not have Xiaomi’s volume negotiating power and would negotiate prices very poorly.
Let’s imagine that after price negotiation, they get a purchase price of $250 per chip and release a prime 4+ with a Snapdragon 8 Gen 3 and 16 Gb of Ram. Let’s even imagine that to compensate for the increase in costs related to these components they sell the unit at $3000 or that they sell a sc7000 at $2000 with these components, do you honestly think that people would still buy the products an AZ or a CDJ 3000 sold $500 more expensive with components only good enough to run a smartwatch?
I’m exaggerating on purpose because a snapdragon 8 Gen 3 will never cost $250 per chip even with the worst negotiated price.
So even assuming that Denon DJ doesn’t have the volume negotiating power of smartphone companies, it would be surprising if it cost them more than $200 per chip
And if Pioneer were to respond with equivalent units, they would probably reach $2900 for the CDJs and $4000 for the AIOs if they want to keep their pioneer tax/margin level.
In any case Denon DJ would still be cheaper than the competition
1 Like
This thread is about Namm 2025 not Namm 2032 right?
3 Likes
I’ve said this before
…I “get” that the whole point of Engine is it runs on stand alone devices without the need for a computer, but why not turn engine in to an iPad or iPhone app?
Phones & iPads have enough power for stem generation etc, so yes. …makes perfect sense to me. Just dock it to a pair of LC6000’s or similar, & off you go.
4 Likes
They’ve tried that before, and it didn’t do very well.
After they did the HS5500 with two layers, internal drive and a nice display (although mono), the next generation players went back to no internal storage, CD player style dot matrix display and an app on the iPad.
The DJ they partnered with at the time tried to do a demo set at BPM and just could not get the players paired with the iPad, so had to cancel and take his gear back to the van.