Nord User Forum

Does anybody know what sort of FlashROM is being used in the NP88? I'd like to replace it with larger capacity modules. Thanks.

Not possible. Even if you shoe-horned a new ROM in there, the OS wouldnt address it.

sad, we know.

So get Nord to patch the OS addressing problem, then.

It's simply not on that the memory is so limited in this day & age. I'm not yet a user of a Nord product, but this limitation seems unnecessarily restrictive.

When I see one piano sample taking almost half the available memory, that seriously worries me. It's great they have that kind of level of detail in the samples, but it does necessitate a memory upgrade.

You can't just swap in a new memory chip. You would need to at least format it with the right filesystem (which may be proprietary) and obtain a controller chip that reports the right size. If Nord's OS code is robust enough it would work at that point, but more often than not, embedded devices have NOR partition layouts burned into the OS kernel. I've actually ported open source firmware for wireless routers before and found myself having to patch the kernel to change the partition layout to allow for less kernel space and thus more room for user data. In embedded development, everything is done by direct addressing of subsystems. It's a pain.

Also, Nord uses NOR flash for their keyboards. NOR provides very fast random access suitable for use in sample recall (source), but it is much harder to get in large capacities than NAND since NOR's usual application is embedded devices that need to load up a small OS image. My question, of course, is whether NAND has gotten fast enough (SSDs?) to make it feasible to use NAND technology in keyboards.

To test this, I copied the Concert Grand 1 (Steinway model D) Large patch, which is 58.07MiB, to my SSD which is formatted as ext4 and hooked up on a 6Gb/s SATA bus on my desktop PC. I then told my desktop PC to drop any cached files from RAM ("echo 1 > /proc/sys/vm/drop_caches") to ensure a fair test, and then timed how long it took to simply read the contents of that file. 0.163s -- with the file completely gone from RAM, reading only from my SSD. Extrapolating a bit, I calculated an exact read rate of 356MiB/sec. The largest combination of piano patches that I have saved is the 196MiB XL bright grand and the 69MiB large BlueSwede upright -- these are 265MB total. That's 0.74 seconds to load those patches from a NAND based SSD into RAM, in the worst possible scenario.

Honestly, I would be okay with it taking 3/4 of a second to load my programs (plus probably another 0.2sec or so to load and process all of the settings) if it meant vastly more storage space in Nord's products. They could easily throw a 128GB NAND based SSD and 2GB of DDR3 RAM in there on top of an ARM Cortex A9 CPU and a good SATA controller, and the polyphony, processing power and modularity of the system would improve by orders of magnitude. In all honesty, I'm not sure how feasible it is to make an ARM based system that can handle a SATA 6Gbps bus, and perhaps someone with more experience in embedded development can contribute to this. Still... I'd love to get half an hour with Nord's hardware engineers. The technology they're using is rock solid but much of it is old and the technologies which did not meet their needs at design time very well might now.

metalfingers wrote:They could easily throw a 128GB NAND based SSD and 2GB of DDR3 RAM in there on top of an ARM Cortex A9 CPU and a good SATA controller, and the polyphony, processing power and modularity of the system would improve by orders of magnitude.

If you don't mind substituting that ARM with an Intel Atom, such a thing would exist. It's called a Kronos :-p

The Kronos did away with dedicated DSP processors (the Stage 2 has six Freescale DSPs) and replaced it with a single Atom (might be dual-core, don't know). It's pretty much a PC inside (including the loud fan) and they're using pretty standard RAM and SSDs. It runs Linux (the Yamaha Motif XS/XF do too). The 'magic' in the Kronos is in the software, the hardware is merely a platform to run the software on (a bit like a game console, really).

I assume you meant replacing the Freescale ColdFire MCU they have in the Nords with that ARM and keeping the DSPs? That ARM isn't powerful enough to get the sound quality of the Nords without some DSP help.

An important downside of DSPs is the fact you partially hardwire the functionality into your product. You see the same issue with the Access Virus TI - that is also a DSP-based instrument. They keep releasing new software with new features, but it's more like evolution than revolution. Because Korg chose to use fairly standard general purpose hardware for the Kronos (and Krome, or is it Chrome?) they could come out with completely new (and different) software for the Kronos if they wanted to.

Anyway, I think other workstation (synthesizer) companies will eventually follow the direction Korg has taken.

metalfingers wrote:Also, Nord uses NOR flash for their keyboards. NOR provides very fast random access suitable for use in sample recall (source), but it is much harder to get in large capacities than NAND since NOR's usual application is embedded devices that need to load up a small OS image. My question, of course, is whether NAND has gotten fast enough (SSDs?) to make it feasible to use NAND technology in keyboards.

As I understand it, the issue isn't merely one of NAND being "fast enough" -- it's that it can only be accessed as storage (like a hard drive), it cannot be accessed as memory (like RAM). So data located on NAND cannot be executed "in place" as data located on NOR can. Instead, it first must be copied into RAM. If the OS supports it, data can essentially be streamed in real time off storage (whether a fast hard drive or NAND), that's the way around it, and that's what the Kronos does (built on Linux). That's non-trivial which is why it probably makes more sense to use something like Linux rather than try to write such an OS yourself from scratch. But then, of course, you need to have essentially a Linux-capable computer inside your board, too.

anotherscott wrote:As I understand it, the issue isn't merely one of NAND being "fast enough" -- it's that it can only be accessed as storage (like a hard drive), it cannot be accessed as memory (like RAM). So data located on NAND cannot be executed "in place" as data located on NOR can. Instead, it first must be copied into RAM.

You could very well be right, but judging from the way the Nords have the flash connected to the DSP chips, it looks like they're 'streaming' the audio out of the flash chips. The DSPs have some on-board RAM, which should be enough for the simple programs they run and for mangling audio, etc.

anotherscott wrote:If the OS supports it, data can essentially be streamed in real time off storage (whether a fast hard drive or NAND), that's the way around it, and that's what the Kronos does (built on Linux). That's non-trivial which is why it probably makes more sense to use something like Linux rather than try to write such an OS yourself from scratch. But then, of course, you need to have essentially a Linux-capable computer inside your board, too.

DSP chips don't run a general-purpose Operating System (like Linux). Instead, they normally run very lean software stacks specifically meant for the task they need to perform. In the case of the Nords there is an additional microcontroller that takes care of general housekeeping (MIDI, updates, changing programs, etc.) and setting up the DSP chips. Some processors also have built-in DSPs - the ARM processor used in the Raspberry Pi is an example of this.

Some processors (like the Intel Atom used in the Kronos and Krome) also support several DSP instructions, but don't have a full-fledged DSP on board. Modern processors are fast enough to not really need a DSP anymore for most audio tasks and the DSP extensions make signal processing faster while still being able to develop software in a more general manner (CPU v.s. DSP).

We'll probably see fewer synths/keyboards using dedicated DSPs (like the Nords, Access Virus, Kurzweils) and more with a 'PC-like' setup like the Kronos and the Krome. The setup Korg uses for those makes a lot of (business) sense.

I'm not a hardware guy but I have 15 years in software development. Am I right in thinking that the hardware changes suggested here would mean a complete software rewrite? If that is the case it would be a huge undertaking. All experiences I know of show that building a new software architecture takes ten times longer than anyone could imagine. It might takes years to get it up to the quality level Nord has now.

Sorry for being pessimistic...