World famous organ of Martinikerk Groningen will come

Possibly not the point, but it's FUN!!!
I bought the Zwolle, Caen and Groningen responses today to play with the semi-dry.
The Zwolle responses seem to work best.
The Caen Jeklin's are also good (All this evaluation is on the laptop and headphones (712's, gigaport HD+), until I have the time and money to switch all the 1T drives to 3T ones. 2 down, 3 to go).
Whenever I've time in the new year and get the main HW machine all transferred, it should be fun to have the Zwolle or Caen responses with the cardio to front, omni to side and rear to rear speakers.
Now if only there were Rotterdam responses........ (especially for the party horns!!)
(Although I might try to emulate Zwolle with an extremely narrow source in Impulse creator).

Jiri,

I have been eagerly awaiting this release so that I can finally see what the price and hardware requirements are. As I expected, neither are trivial.

With HW5 and the Martinikerk organ now available, this seems perhaps a good time to upgrade my computer system to something a bit more powerful than the Dell Precision M6700 laptop (4-core i7 and 32Gb) that went into service at short notice when my MacPro died (and working much better than I ever expected). I have been considering a Dell Precision T3431 with an 8-core i9 processor, buying the RAM and SSD elsewhere to keep the costs down. However, the maximum RAM usable with this system is 64Gb, and I understand from another user in another thread that the maximum polyphony is about 6,000. I see from your Web site that with the full surround version in 24-bit the requirement is 85Gb and you recommend a polyphony of 8,000. You very helpfully explain the many ways of reducing the RAM requirement, but inevitably as more organs are sampled in this way in the future we are going to require more powerful computers to use them.

Of course with unlimited funds it is possible to buy a workstation with an 18-core Xeon and 256Gb RAM but that is a bit beyond what I can afford, and probably most users. Therefore what computer, or at least what specification of computer, do you recommend as a reasonable compromise for this and similarly demanding organs?

Best Wishes,
Julian

Just in case you got the 6000 polyphony number from me, that was with my old i7 4 core T3420, not my new i9 8 core T3431. I haven't moved to that yet, but I'm sure it will higher. I'm too busy playing with HW 5.

The T3431 i9 should handle Groningen 4 channels easily, 6 channels if you're careful to choose ranks that fit in 64GB. Actually loading all 8 channels and all tremmed ranks won't fit in memory.

Therefore what computer, or at least what specification of computer, do you recommend

Where to start? Indeed, looking at the feature tab of the St. Martini sample set. The listed RAM figures are the decisive factor.
For a two channel variant (distant, semidry) a 32GB computer is good, for the diffuse variant (4 channels), 64 GB computer is good, for the 8 channel, 96GB computer would be a good choice.
But I am myself running 8 channels on 64 GB computer at 16 bits and it is OK for the purpose. You can always tweak the rank loading so that the sample set fits in your RAM.

For polyphony, i.e. the CPU, it is not so dramatic. Especially in HW5. I do not know if I am correct, but it seems to me that you can achieve higher polyphony on the same hardware compared to HW4. So, do not through your older computers away. (Has anyone tried to run HW5 on an XP computer, or on a very old Mac OS? MDA does not actively support it, obviously, but still, is there anything preventing that?)

When playing "normal" music, you hardly need 3000 pipes of polyphony for the 8 chanels sample set anyway. Recording the Bach BWV 532, I experimented and set the polyphony low, only about 3200 pipes. And Hauptwerk did not even hit the last green bar on the polyphony meter! It is only some tutti and heavy chords when faster repeated which eat the polyphony.

But the problem is how to optimize the computer (if on Windows) for audio. There were some threads about it on this forum. One of my testers has a machine, 96GB RAM, Intel Core i7 7800 X CPU @ 3.50GHz, and he is getting a lot of cracks and pops when running St. Martini. I have considerably older computer and St. Martini works smooth. The polyphony is set to about 5400 pipes and it is enough for a real-world music. So, it is not only about the computer specifications, but more about tweaking the computer for real-time audio.

Unfortunately, the experience tells me that not every computer can be tweaked well. I remember having a nice Dell laptop in past years, where it was impossible to make it work reliably with real-time audio. The glitches were never gone, no matter what I tried. So, I cannot make any recommendation.

OK, I could not help myself and I did an experiment. The same machine, but two different Hauptwerks. Hauptwerk 4 performing against Hauptwerk 5. The screenshots explain it all.

This picture shows the maximum polyphony achieved on my machine when Hauptwerk 4 i used.
You can count the pressed keys and multiply by 500.
It is about 8.500 simultaneous pipes.

This pictures shows the maximum polyphony achieved on the same machine when Hauptwerk 5 is used.
It is about 11.000 simultaneous pipes. And I could still add some little more if I had more hands to use.

This is a dramatic improvement. Is there still anybody here who wants to tell us, that there is no reason for the upgrade? It can basically make a new machine from your old one.

By the way, the testing machine was a compute about 5 years old now, running now rather old i7-5820K processor.

For this, and future organs: will the corresponding IR be included with the semi-dry set?

No. That is not the point. Semi-dry+IR do not provide a "cheap" substitute for the surround sample set.

Maybe not the point. But couldn't it be?

I do not own it, but I hear very good results about Altiverb.
So if the IR's are of very high quality, theoretically one could have a high quality organ set up. And even save some money on the PC itself because we wouldn't need 64 or 128G anymore..?

But basically the same effect is achieved with any suitable IR. You do not need an IR of the particular church to have "very good results". You can have any IR to have very good results. We recorded about 50 churches and their IRs. You can find almost any desirable reverb time already in our collection. Select the IR with similar reverb time to your desired church and go.

Looks fantastic Jiri! A couple of questions if I may...

1. Will there be an upgrade available later, from wet to surround?

2. For those of us who will have to load some stops in 16bit and some in 20bit do you have any recommendations as to which stops benefit most from loading in higher quality?

In the demo, with multiple output channels in a bus group and using the static "everything" cycled default bus allocation algorithm, I get all ranks of the demo organ playing through a single output on the same note. I can see this on the RME mixer display.

This doesn't happen with other SP organs I use. Ranks cycled doesn't seem to work with the Groningen demo.

That is, with one slider, direct or diffuse or distant or rear, at full and the others silent, all ranks choose the same pair of speakers for the same note. With more sliders raised, more speakers sound, but simply drawing more stops doesn't use more speakers for the same note as usual.

The same routing and algorithm on Caen sends different ranks to different speakers on the same note.

Is this an artifact of the rank ids in the demo, or is it the behavior of the full sampleset?

Thanks.

1. Will there be an upgrade available later, from wet to surround?

Yes, the eshop should do this automatically. When you have, for example, the vol. 1 on your account, it should display to you an upgrade product for volumes 2+3.

2. For those of us who will have to load some stops in 16bit and some in 20bit do you have any recommendations as to which stops benefit most from loading in higher quality?

You should try to load the most often used ranks in 20-bits. Less important stops in 16-bit.

Is this an artifact of the rank ids in the demo, or is it the behavior of the full sampleset?

I am sorry, but the sample set creator does not have any tools how to influence the mixer behavior. It is completely up to Hauptwerk, if I understand Hauptwerk correctly. There is no settings in the ODF which would tell, how the Hauptwerk cycling algorithm should work. That is exclusively the job of your mixer. No, there is no "artifact" in the sample set.

"Artifact" may have been an unclear word choice when "effect" or "outcome" was what I meant.

The rank ids assigned by the sampleset producer affects the rank cycling algorithms, controlling the starting channel within the audio group for a specific note. Ranks with consecutive rank ids use distinct speakers until the group runs out of outputs, then start over.

I would see this lack of rank cycling behavior if the demo's rank ids were separated by multiples of 4, for example. That's what I was trying to ask. I don't know how to view rank ids in HW.

I could add different note offsets to ranks to fix this, but doing that level of fiddling shouldn't be necessary.

Thank you.

Aha, if so, then, you may have matched the common denominator of the rank ids and your cycling buses. This would be perfectly understandable, because of the 8 channels, while Caen or other sample sets have only 4 or 6 channels. If that is the cause, then, you should probably go and add some not offsets to your mixer. I do not know why you would call it "fiddling", it would appear to me a standard procedure.

I have uploaded a view of the ids of ranks directly from the database, just a specimen of some first ranks. Perhaps it may be useful to you to see how the rank ids are created. With the tremmed ranks (and double tremmed ranks for RW stops) and 8 channels, everything becomes really complex.
https://ibb.co/mXZ9prC

Thank you for the upload. It makes it clear that the problem isn't as simple as the differences between rank ids within one microphone perspective being zero modulo my audio group size.

But you're right, using note offsets with the cycling algorithm fixes it.

Indeed. Just imagine, that the virtual ranks for the St. Martini full surround, are alltogether 726! Multiply by about 50 and you get the sum of virtual pipes, multiply the result by about 4 and you get the number of samples.