Nord Piano 5 incoming

[Fleer]

As per this last post: https://www.norduserforum.com/nord-piano-grand-forum-f10/any-news-on-nord-piano-5-t19182-20.html?sid=738e0c48f4db027d36c4737f696055f6

[stoken6]

 

This would need to include some surprises to interest me.

 

Cheers, Mike.

[ElmerJFudd]

And $500 shaved off its street price.

[jeffincltnc]

So... let's figure 2GB of piano sample library storage instead of 1GB. (matches the spec of Nord Grand and Nord Stage 3).

 

Anything else?

[Delaware Dave]

I guess they couldn't get it right the first 4 times.

[dje31]

I know I've pined for a Nord piano multiple times...but since Yamaha is now giving the ability to update sounds like Nord, at a lower price point, that interest has faded.

 

Now a Stage 3 is another matter...

[Nathanael_I]

Nord - the company that never gives you exactly what you want - everything is always crippled in some nonsensical way. I do have a Nord Grand that I enjoy thoroughly - and NO FATAR ACTION! Which is a huge plus in my book. But the MIDI implementation is skinny. Why can't it be a great master controller, have some wheels? Because it is a piano? Silly. It is not a piano. It is digital and it runs software. It can be whatever it is programmed to be. I don't regret it, but Nord's skinny piano samples mean little in a world of spectacular software pianos. I just wish they would make the best instrument they are capable of, not endlessly narrow "positioned" things.

[ElmerJFudd]

Nord - the company that never gives you exactly what you want - everything is always crippled in some nonsensical way. I do have a Nord Grand that I enjoy thoroughly - and NO FATAR ACTION! Which is a huge plus in my book. But the MIDI implementation is skinny. Why can't it be a great master controller, have some wheels? Because it is a piano? Silly. It is not a piano. It is digital and it runs software. It can be whatever it is programmed to be. I don't regret it, but Nord's skinny piano samples mean little in a world of spectacular software pianos. I just wish they would make the best instrument they are capable of, not endlessly narrow "positioned" things.

 

They all do this to an extent on boards that are less than $4k+, and even then. But Nord, they make the game as obvious as Apple. An endless revamping of the same product with minor tweaks. Every once in a while there"s a tech advancement that makes a significant advancement in sound but we"ve been on sample playback for a long time. More detailed samples, longer samples, more samples. Lower latency, better fidelity, better quality action. And sometimes an intuitive user interface change. But otherwise, long gaps between anything revolutionary. So to keep product movement. Minor updates. Korg plays the color game, Nord plays the cheap out on memory game. Yamaha, it does or doesn"t have strong resonance. Etc etc.

[DannyKeys2]

So... let's figure 2GB of piano sample library storage instead of 1GB. (matches the spec of Nord Grand and Nord Stage 3).

 

I never understand why modern digital pianos have such relatively paltry storage space. Flash memory is ridiculously cheap. What would the cost difference between 1gb and 2gb be? $1?

[ElmerJFudd]

Apparently big enough to cut into their profit margin as a viable business. And leaves them nothing to upgrade each year.

[AnotherScott]

I never understand why modern digital pianos have such relatively paltry storage space. Flash memory is ridiculously cheap. What would the cost difference between 1gb and 2gb be? $1?

The kind of flash in the Nord (coincidentally called NOR flash) is relatively pricey. See: https://www.embedded.com/flash-101-nand-flash-vs-nor-flash/

[analogika]

So... let's figure 2GB of piano sample library storage instead of 1GB. (matches the spec of Nord Grand and Nord Stage 3).

 

I never understand why modern digital pianos have such relatively paltry storage space. Flash memory is ridiculously cheap. What would the cost difference between 1gb and 2gb be? $1?

As mentioned, this is NOR flash, which is an order of magnitude or two more expensive than NAND flash.

 

The decision here is that NOR flash writes very slowly but reads at speeds comparable to RAM, so that samples can be played directly from storage.

 

If you opt for cheaper solid-state storage, you need to add RAM and boot an operating system that loads samples into RAM at boot or on program change. That"s a fundamentally different design decision and what Korg has done with the Kronos.

[AnotherScott]

Info from the norduser forum indicates that changes likely include these:

 

* ability to layer/detune (and presumably split) two pianos, as on the Nord Stage 3

* ability to layer (and presumably split) two of the sample section sounds, as on the Nord Stage 3

* split point crossfade, as on the Nord Stage 3 and Electro 6

* reverbs from the Wave 2

* addition of a 73-key version

 

So to at least some extent, you could kind of look at it essentially as a two-panel version of the Nord Piano 4, or as a version of the two-panel Nord Stage 3 without the synth and external MIDI zone functions.

[Adan]

73-key version might peak my interest as a product, though I don't really need it. Less interesting if it's Fatar TP100 action.

[jeffincltnc]

Not sure the source of those changes, but interesting that the layer/detune and crossfade would launch in Piano 5 and not an update to Nord Grand.

[AnotherScott]

Not sure the source of those changes, but interesting that the layer/detune and crossfade would launch in Piano 5 and not an update to Nord Grand.

Source was supposedly a picture that appeared on--and then disappeared from--instagram.

 

It's also not impossible that the Nord Grand might be getting a refresh along with the Nord Piano.

[bfields]

As mentioned, this is NOR flash, which is an order of magnitude or two more expensive than NAND flash.

 

The decision here is that NOR flash writes very slowly but reads at speeds comparable to RAM, so that samples can be played directly from storage.

 

If you opt for cheaper solid-state storage, you need to add RAM and boot an operating system that loads samples into RAM at boot or on program change. That"s a fundamentally different design decision and what Korg has done with the Kronos.

 

I don't quite get this. It seems these days SSDs can manage thousands of random IOPs without average latencies under a millisecond, and maximum latencies still only a few milliseconds. That should be good enough to stream samples off disk at pretty high polyphony, shouldn't it?

 

I mean, just babbling here, my day job's storage-adjacent but I'm no expert on the performance requirements and economic constraints at play here.

[stoken6]

73-key version might peak my interest as a product, though I don't really need it. Less interesting if it's Fatar TP100 action.

+1 and +1. Wasn't there an earlier Nord Piano with a 7x-key TP100?

 

Cheers, Mike.

[AnotherScott]

I don't quite get this. It seems these days SSDs can manage thousands of random IOPs without average latencies under a millisecond, and maximum latencies still only a few milliseconds. That should be good enough to stream samples off disk at pretty high polyphony, shouldn't it?

It's not just about speed, it's also about functionality. The processor can see NOR flash as if it were RAM. The data on an SSD or NAND flash must be copied into RAM to be used, it cannot be used directly. Real-time streaming access can still happen quickly with some amount of pre-loading (which is why streaming a VST from disk can work at all), but that requires a different architecture, additional RAM, and an operating system that can manage it. This is what Kronos does, by actually incorporating a full linux computer inside. (Related, NAND/SSD isn't actually capable of high speed truly random access, which is also why caching and copying into RAM is part of the process.)

 

There are some other possible solutions. Some boards store everything on low-cost NAND flash, but have to copy everything into RAM at startup, so you have besides the cost of the flash, the cost of the same amount of RAM, plus you slow the startup as all the data must be copied into RAM at each startup. I believe you find this approach in Dexibells and some Korgs (e.g. Krome).

 

In the years since Kronos, Yamaha and Kurzweil have come up with their own (patented) ways of making use of at least some amount of lower cost flash without requiring the software/hardware overhead of the Kronos approach. Kurzweil has what they call flashplay. Yamaha's approach with their SWP70 chip (and contrasted with how it differs from SSD streaming) is well explained at http://sandsoftwaresound.net/new-workstation-namm-2016/

 

Nord's approach is pricey... but it allowed them to do something that was pretty revolutionary in 2005 (when they released the first Nord Stage, which was the first model that allowed you to re-write the board's "permanent" memory). It's an interesting question as to how much more mileage they can get out of that approach, as newer technologies may become increasingly cheaper to implement, but would require an entire redesign. Lucklily, NOR flash, relatively pricey as it may be, is still a whole lot cheaper than it used to be (which has allowed Nord to eventually expand the Stage series from 128 mb of piano memory to 2 GB). There may also be other tradeoffs, e.g. the pricier NOR flash is still more robust and may be less prone to data loss in the long term. Yamaha did in fact shift from the NOR flash model used in the Motif XF/MOXF/Tyros ($300 for a 1 GB expansion card, IIRC) to the aforementioned SWP70+NAND approach in the Montage/MODX/Genos/newer PSRs. But Yamaha is a huge company, and they patented what they're doing, so it's not obvious that a boutique company could make the same kind of transition.

[AnotherScott]

Wasn't there an earlier Nord Piano with a 7x-key TP100?

Yes, the Nord Piano 2. My guess is that they didn't do one for the 3 or 4 because, unlike when the 2 came out, there were Electro models that basically did the same thing (with the addition of organ). Back in the NP2 days, the Electro was still monotimbral, so the fact that the Piano could split/layer further differentiated it from the Electro. But now it looks like they may be introducing some more differentiating features.

[bfields]

Related, NAND/SSD isn't actually capable of high speed truly random access

 

See, this is the part I'm not convinced of. I mean, look at the 4k random access chart here: https://www.storagereview.com/review/samsung-980-pro-pcie-4-0-nvme-ssd-review

 

100 microsecond latency while handling 100k reads per second sounds pretty good to me.

 

Granted, if NOR flash is closer to RAM then it has read latency more like tens of nanoseconds. Also, that particular drive may be out of budget, and I'm pretty sure that latency number is an average, when what we care about is the worst case. And you probably need to wait on more than one read to handle a key-on event, depending on how the on-disk data structures look. And you need some buffering and time for effects processing and D/A conversion and other stuff, I'm sure.

 

Still, I don't know, in theory it seems doable to meet a deadline of a few milliseconds without needing to preload a lot of data?

[AnotherScott]

Related, NAND/SSD isn't actually capable of high speed truly random access

 

See, this is the part I'm not convinced of.

Again, I'd refer you to the link I mentioned: https://www.embedded.com/flash-101-nand-flash-vs-nor-flash/ - though as techy as it is, it also appears inconsistent, saying that NAND doesn't do random access, then later saying it does, albeit at a rate that is 250 times slower than NOR. But I think there's also a distinction to be made between random access of entire blocks (which NAND can do), and random access of individual bytes (which I believe only NOR can do). But I'm at the edge of my understanding here. (Either way, the point remains that NOR is seen as RAM, and NAND data must be copied into RAM, with whatever additional hardware and software infrastructure that requires.)

 

Still, I don't know, in theory it seems doable to meet a deadline of a few milliseconds

Not necessarily related to what I was talking about above but just in general, I think even a few milliseconds could be a lot when you're talking about high polyphony and a serial protocol. Then it's not necessarily a single instance of a few milliseconds, but could be numerous cumulative instances of a few milliseconds. And unlike MIDI, where (typically) no note-specific data is being transmitted between a given Note On and a Note Off, sample data must continuously be read for each note for the duration it is held. So for example, with the sustain pedal down, you can quickly be in the position of needing to be transferring lots of different data streams simultaneously.

[bfields]

Not necessarily related to what I was talking about above but just in general, I think even a few milliseconds could be a lot when you're talking about high polyphony and a serial protocol. Then it's not necessarily a single instance of a few milliseconds, but could be numerous cumulative instances of a few milliseconds. And unlike MIDI, where (typically) no note-specific data is being transmitted between a given Note On and a Note Off, sample data must continuously be read for each note for the duration it is held. So for example, with the sustain pedal down, you can quickly be in the position of needing to be transferring lots of different data streams simultaneously.

 

Definitely true. But one thing SSDs seem to be very good at is high parallelism. Like I said, that SSD benchmark is showing the server responding in less than a ms even while serving 100,000 reads per second. A few hundred notes of polyphony shouldn't be a big deal.

 

So the bigger problem is whenever you have to wait for one read before doing the next one--e.g. if you're storing your sample data in a regular filesystem at /samples/grandpiano/ppp_A0toA1 or whatever, you might have to do multiple reads to traverse to the right directory, then once you get to the file read some metadata to find out which blocks on the disk actually hold that files data, etc. So you'd need avoid that.

[AnotherScott]

Definitely true. But one thing SSDs seem to be very good at is high parallelism. Like I said, that SSD benchmark is showing the server responding in less than a ms even while serving 100,000 reads per second. A few hundred notes of polyphony shouldn't be a big deal.

Oh yes, SSDs definitely work, as proven in countless laptops and Korg Kronos. But they are never played from directly (as you would with NOR flash), you still must copy the data into RAM, with, as I said, the attendent hardware and software infrastructure. (Again, take a look at http://sandsoftwaresound.net/new-workstation-namm-2016/ to see what Yamaha came up with, and in particular, how that contrasted with some of what would have been needed to support an SSD.) So you can't just take out 2 GB of NOR flash, drop in a 60 GB SSD and call it a day. And then to get back to the query that started this, neither can you drop in a dollar's worth of cheap NAND flash.

[Fleer]

A/Scott, you"re on a roll.

[nadroj]

Always up for more 76/73 weighted options, though it does seem that the incremental changes Nord bring to each generation are getting smaller and smaller every time.

[Nathanael_I]

The thing is you don't need to copy ALL the data into RAM. All the orchestral libraries in Kontakt use a pre-load buffer of only 60kb by default. This gives enough that everything else CAN be streamed from normal SSD. Nord will do whatever they think is best, of course. Their instruments tend to be narrowly defined. They are NOT going after the Kronos/Kurzweil space....

[miden]

Nord's are no better than the rest. They trade on a name, much like Pro Tools in the DAW area....and folks keep lining up to buy them. Well, as long as gullible people (who keep maintaining Nord are dope) are doing that Nord will not change the strategy, same as Apple.

 

Personally my reaction to Nords, and I've played a few, is "meh", take it or leave it.

[AnotherScott]

The thing is you don't need to copy ALL the data into RAM. All the orchestral libraries in Kontakt use a pre-load buffer of only 60kb by default.

All data does get copied into RAM, at some point. The pre-load buffer is there to give it the time it needs to load the subsequent pieces, but everything you hear must be read (streamed) off disk into RAM before you can hear it. So it all ends up copied into RAM, it just doesn't all end up in RAM at the same time. But again, the RAM and other hardware infrastructure and operating system support to do this does not exist in the Nord or most other keyboards. I think Kronos is still the only board that has the ability to stream large sound sets off SSD, and again, they do it by building the keyboard around a full PC motherboard running Linux, because that's probably the most cost-effective way to get the hardware and OS support you need in order to accomplish this.

[bfields]

All data does get copied into RAM, at some point. The pre-load buffer is there to give it the time it needs to load the subsequent pieces, but everything you hear must be read (streamed) off disk into RAM before you can hear it. So it all ends up copied into RAM, it just doesn't all end up in RAM at the same time. But again, the RAM and other hardware infrastructure and operating system support to do this does not exist in the Nord or most other keyboards. I think Kronos is still the only board that has the ability to stream large sound sets off SSD, and again, they do it by building the keyboard around a full PC motherboard running Linux, because that's probably the most cost-effective way to get the hardware and OS support you need in order to accomplish this.

 

I'm a little surprised no one else is does it yet. The ability to use commodity parts and software seems like a big advantage. All sorts of things are suddenly easier once you've got a full operating system. Boot time and other issues seem totally solvable....