Pre-configured PC Computers now available

[mdyde]

Hello Tom,

The technical publication from Hauptwerk says that '2400 pipes of polyphony will play the biggest organs with the longest release times.'

That was true of the sample sets available at the time that document was written (early 2006), but some of the larger (and/or surround-sound) sample sets released more recently do really need quite a lot more than that, and need a minimum of 4 (or even 8+, in the case of the Caen surround sample set) CPU cores for their tuttis to be fully usable.

(Of course, those are requirements of the particular sample sets, rather than Hauptwerk per-se.)

Hence we should probably get that document updated - sorry for any confusion there.

I also think that HW 3.23 has the benefit of some optimization coding that makes the most of any processor above the minimum.

Yes - since v2.00, Hauptwerk has been highly optimised for multiple CPUs and/or multi-core CPUs, and can take full advantage of them in terms of performance.

I just noticed that the HW 3.23 manual has dropped the discussion of polyphony from the manual when talking about processors perhaps the increased speed from 1 GHz to 2.5 GHz has been benficial.

Can you clarify which section of the user guide? I've just done a quick comparison of the current version of the user guide with the original v2.00 version, and I can't find any relevant differences / specific polyphony recommendations.

Since v2.00 we're stated that Hauptwerk will work fine with a single 1 GHz processor with small sample sets and/or realism features disabled (for those people that want to use it on older existing PCs), but we've never actually *recomemnded* a 1 GHz processor.

The 'technical data' PDF has the more thorough dicussion of polyphony and CPUs (although it's perhaps in need of some updating in some areas, as above):

http://www.hauptwerk.com/clientuploads/documentation/PDF/TechnicalData.pdf

I too personally think that it would make sense for you to offer 4-core and/or 8-core i7/Nehalem systems with 8 or 16 GB for customers planning to use the Hauptwerk Advanced Edition, since those are the specs that most Hauptwerk customers seem to be buying at the moment, in order to be able to take advantage of the recent large sample sets, as well as for some degree of future-proofing for forthcoming sample sets.

For the Basic Edition a Core 2 Duo system with 4 GB is probably more than ample, but it's probably a good idea to make it clear which of your models would be suitable for which Hauptwerk editions and broad classifications of sample sets (e.g. clarify that a dual-core 4 GB system is suitable for the Basic/Free editions with medium-sized sample sets).

Perhaps you could target your OT-1 for the Basic Edition (Core 2 Duo, 8 GB), your OT-2 as a mid-range offering for the Advanced Edition (4-core Intel i7, 8 GB), and the OT-4 for the largest sample sets for the Advanced Edition (two 4-core Nehalem CPUs, 16 GB)?

As Peter mentioned, touch-screens are extremely popular with Hauptwerk users (since they provide a very easy, convenient and cost-effective way to control it when using several different sample sets, even for people with just a single MIDI keyboard and no understanding of MIDI), so considering 17/19" touch-screens as your default monitor option could make sense?

Tom,
I think Martin's reply is for you.
Best

Peter

[mdyde]

Hello Peter,

Tom,
I think Martin's reply is for you.

Sorry Peter! That was a typo. I've corrected my post.

Martin,

Thank you for the update on the polyphony issue. I lost my Hauptwerk folders in a move so I cannot find them to quote page numbers but I remember a listing of CPU models and the approximate polyphony they would produce. Now I cannot find it.

I just do not have an 8 core 16GB machine available to me yet. I have found some 4 core machines that have 6 GB and 8 GB of memory but they are 4 slot motherboards so I must wait for the 4GB modules to come down a bit before the 16GB machines are available or wait for the i7s.

I am offering a limited number of Quad Core Q8200 to replace the Dual Core offerings but they are truly limited availability. The price is the same as the dual core.

I am also quite willing to leave the last 10-15% performance market to those who have the smarts to assemble their own super system for Caen and the larger sets. If we have three categories of Good, Better, Best, I am content to operate in the Good and Better for now and leave the Best to others. To me a machine that performs adequately on over 85% of the sets, is reliable, easy to use, has a good in home warranty program and costs less than an Ahlborn module is fine.

I think it an excellent idea to tell the prospective customers what will play in a dual core 4GB/8GB machine and what will not. But unfortunately there is no indicator on the sample set except the amount of memory that is required to load it into the computer. How do I know what to say to the customer about a particular sample set and computer combination? If you could help me here I would appreciate it. Perhaps it would be easier to know which ones won't play in a quad core 8 GB machine. How about a table of available sample sets and the memory and polyphony requied for each? What are we up to now about 100 sample sets and variations?

Thanks again for the helpful advice.

Tom

Peter, Matin,

Touch screen monitors seem to be hanging out in the $500-$600 price range while the overlays are only about $200.
The single touch screen overlay lcd is the OT-3. I will remove the non touch monitor from the OT-2 and reduce the price.

Tom

Hello Tom

Re Touch screens here's one $285

http://cgi.ebay.com/ebaymotors/S02-17-I ... ioQ5fVideo

Here's Another at $350

http://cgi.ebay.com/NEW-POS-TOUCHSCREEN ... 286.c0.m14

And One more at $360

http://cgi.ebay.com/NEC-MultiSync-LCD17 ... 286.c0.m14

You can even have a wide 19 inch touch screen for $360

http://cgi.ebay.com/NEC-ASLCD194WXM-19- ... 286.c0.m14

So plenty of choice well below $500-600

Best

Peter

[mdyde]

Hello Tom,

I think it an excellent idea to tell the prospective customers what will play in a dual core 4GB/8GB machine and what will not. But unfortunately there is no indicator on the sample set except the amount of memory that is required to load it into the computer. How do I know what to say to the customer about a particular sample set and computer combination? If you could help me here I would appreciate it. Perhaps it would be easier to know which ones won't play in a quad core 8 GB machine. How about a table of available sample sets and the memory and polyphony requied for each? What are we up to now about 100 sample sets and variations?

We currently leave that level of detailed specifications up to the sample set producers, since it would be quite a bit of additional work for us to produce and maintain that for every third-party sample set (although I agree it would certainly be handy to have it all in one place and one standard format).

In our sample sets database if you click on a sample set and then click on its name in the 'details' page, the link will take you to the sample set producer's own sample set info page, which should have detailed specs and hardware recommendations.

As a very, very rough rule of thumb, you could perhaps get an idea of the static polyphony a computer ideally needs to be able to handle in order to play the tutti for a given classical (non-unified) sample set with a formula something along the following lines:

---------------------

Total number of pipes in sample set

x average reverb time in seconds (since pipes continue to consume voices of polyphony until their reverb tails have completely finished).

x 0.1 since the tutti won't use all ranks and ten fingers and one foot won't play all pipes in a rank.

x 5 since when playing the tutti fast it's unlikely you'd often play more than about 5 chords per second.

x 1.4 if the sample set's samples are stereo, rather than mono, to allow for some or all ranks being loaded in stereo. Most modern sample sets are stereo and Hauptwerk's polyphony is measured for uncompressed, unenclosed, mono samples. Stereo samples have higher CPU overheads than mono so Hauptwerk gives extra polyphony weighting to stereo pipes when calculating its dynamic polyphony load to compensate. (Loading sample sets with all ranks uncompressed in mono, and with enclosure filters disabled globally will give the highest polyphony, but possibly at the expense of realism.)

x2 if the sample set is surround-sound. Surround sample sets use two voices of polyphony per pipe.

x 1.4 since the polyphony limit setting typically needs to be set at about 70 percent of the measured static polyphony the computer can handle, to allow it to absorb fluctuations in CPU load for transitions between attack/release samples etc.

x 1.3 since the polyphony limit mechanism starts to fade out the least prominent pipes when about 75 percent of the user-specified polyphony limit is reached, so as to ensure the CPUs never get fully loaded, thus guarding against audio glitches, whilst also trying to give priority to newly-played pipes so that the audible effects of polyphony management are as unobtrusive as possible.

x 1.15 if memory compression is enabled for all ranks. Memory compression adds some CPU load, so Hauptwerk gives extra polyphony weighting to compressed pipes when calculating its dynamic polyphony load to compensate.

x 1.25 (unless the swell box filters have been disabled globally) to allow for some ranks being enclosed in swell boxes. Swell box filters require additional CPU overheads, so again Hauptwerk gives extra polyphony weighting to enclosed pipes when calculating its dynamic polyphony load to compensate.

---------------------

So for a classical (non-unified), stereo, non-surround, compressed sample set with all realism features enabled that gives:

Computer's ideal minimunm measured static polyphony =very, very approx= (total number of pipes in sample set) x (sample set's reverb time) x 1.83

Arbitrarily, the Sonus Paradisi Forcalquier wet sample set (happens to be to hand at the moment) has about 1750 pipes and an average reverb time (release sample length) of about 4.8 seconds, so to allow plenty of headroom for playing its tutti fast without release samples ever fading out you might ideally want a computer capable of handling a measured static polyphony (measured with the polyphony testing organ, which is mono, uncompressed, unenclosed pipes) of something like 1750 x 4.8 x 1.83 = 15,372.

In practice, extremely wet sample sets like that could almost certainly get away with quite a lot less because Hauptwerk's polyhony management system could be relied upon to fade out the least prominent releases safely without the effects being noticeable. E.g. just multiplying the number of pipes by the reverb time gives 1750 x 4.8 = 8,400.

So for that sample set let's say you would ideally want a computer capable of a measured static polyphony very roughly in the range of 8,000 - 15,000 (mono, unenclosed, uncompressed) pipes.

If you don't use extreme registrations very fast, then it could be correspondingly less. E.g. 4,000 would probably be fine for normal playing.

In summary, polyphony requirements for a sample set depend a lot on:

- The sample set (reverb time, mono/stereo/surround).

- How you load it (mono/stereo, compressed/uncompressed, enclosure filters enabled/disabled).

- How you play it (very large registrations played fast need very high polyphonies).

- How tolerant you are to the least-prominent release samples being faded out occasionally in times of high polyphony load.

Peter,

I see that most of your 'good buys' are off eBay which means a long way from here. I prefer to deal with local suppliers with whom I already have a good working relationship and know if something goes bad I can get help immediately. Perhaps I can figureout how to configure for a touchscreen I do not have in my possesion and take the touch screen ovelays out of the pricing for that person.

For now, we will stay with the overlays and work with those who want to provide their own touchscreen monitor.

Tom

Martin,


Thanks for the explanation on what kind of polyphony I need for a specific sample set.
The document for the link you sent is copyright 2001-2009 and still says 2500 polyphony is good for big organs but makes no mention of monster needsw like 17,000! Wow that's a lot of polyphony.

Let us reverse the view and determine how much polyphony a certain computer configuration can produce.

Can you come up with some multipliers for that approach or do I need to hold down 132 keys at 125 pipes per key and see what happens?

Tom

[mdyde]

Hello Tom,

The document for the link you sent is copyright 2001-2009 and still says 2500 polyphony is good for big organs but makes no mention of monster needsw like 17,000! Wow that's a lot of polyphony.

The document was transferred to the new website early this year, but that section hasn't been updated for a while, so yes - sorry - that particular statement probably isn't very relevant given some of the larger instruments released fairly recently.

Let us reverse the view and determine how much polyphony a certain computer configuration can produce.

Can you come up with some multipliers for that approach or do I need to hold down 132 keys at 125 pipes per key and see what happens?

Since there are so many potential factors that affect polyphony (CPU type/revision, CPU clock speed, L2 cache, memory speed, memory controllers, motherboard, audio interface, quality of audio interface's drivers, operating system, firewire port performance, to name a few), the only way to get a very accurate benchmark with any given combination of PC parts is to try it and measure it using the polyphony testing organ (although that currently only has about 8,000 pipes).

My 2006 Mac Pro (two dual-core Xeon 2.66 GHz 5160 CPUs) currently manages a static polyphony with PCIe interfaces and a 64-bit operating system in the region of 7,000 pipes. Very generally speaking, most of the Intel quad-core desktop CPUs with good motherboards and good audio interfaces typically seem to manage polyphonies in the region of 5,000 to 7,000.

Brett has a 2008 (I think) 8-core Mac Pro (two quad-core 54xx-series Xeons) which I believe can play all 8,000-ish pipes on the polyphony testing organ whilst showing fairly low CPU activity, so would probably manage something of the order of 12,000 - 17,000.

From my recollection the better-performing Core 2 Duo models typically managed polyphonies in the region of 2,500 - 4,000.

The bottom of page 7 in that PDF document gave some exact measurements from our computers at the v3.00 was released:

http://www.hauptwerk.com/clientuploads/documentation/PDF/TechnicalData.pdf

This page also has a few user-submitted PC hardware benchmarks:

http://www.inspiredacoustics.com/wiki/index.php/Hardware_configuration

I didn't read that far. That is exactly what I remembered.

I think I will peruse the sets and findout which ones the computers can handle.

I have looked at about 6-7 of the smaller ones and they don't seem to say how many pipes and how much delay but perhaps that will change as I look into it more. It appears that if you can find pipes x reverb.time and multiply by 1.83 you get a good approximation. That is helpful.

Thanks
Tom

[mdyde]

Hello Tom,

I have looked at about 6-7 of the smaller ones and they don't seem to say how many pipes and how much delay but perhaps that will change as I look into it more.

Some sample set producers' websites give the reverb times in their detailed sample set info sections, but not all. Numbers of pipes probably needs to be estimated based on numbers of manual/pedal ranks and compasses in most cases.

Perhaps in the longer term we could consider trying to get that information from all sample set producers, and coming up with some sort of agreed standard reference 'ideal polyphony' benchmark formulae (something along the lines of my very rough attempt above) so that we could list the resulting benchmark values alongside the existing reference memory requirements for each sample set.

It would definitely be useful info to have - the main issue is just the work/logistics involved in coordinating and maintaining the information for so many sample sets from so many sample set producers.

There's probably no easy way you could currently get and maintain detailed lists of specific sample sets suitable for given CPU specs yourself, so maybe the most useful thing to do would be to state something along the following lines:

- High-spec Core 2 Duo-based systems should be sufficient for small and medium dry or moderately-wet classical sample sets (given sufficient memory).

- High-spec Intel quad core-based (e.g. higher clock-rate i7s, with plenty of L2 cache) systems should be sufficient for all current sample sets (given sufficient memory), except for some recent extremely-wet, extremely-large or surround-sound sets.

(Theatre organ sample sets are usually significantly more demanding in terms of CPU resources than their classical counterparts for a given number of pipes/ranks, due to unification and the fact that extensive tremulants and additional real-time per-pipe filters typically add quite a bit to the CPU overheads. It seems the existing largest TO sample sets (MDA 3-31 and KeyMedia 3/35 really need the highest-spec current quad-core CPU models for best results.)

(Theatre organ sample sets are usually significantly more demanding in terms of CPU resources than their classical counterparts for a given number of pipes/ranks, due to unification and the fact that extensive tremulants and additional real-time per-pipe filters typically add quite a bit to the CPU overheads. It seems the existing largest TO sample sets (MDA 3-31 and KeyMedia 3/35 really need the highest-spec current quad-core CPU models for best results.)

The Paramount Theatre Organs currently under development employ sampled tremulants and therefore there is no additional processor overhead for ranks under trem than there is for their non-tremmed counterparts. Trem is engaged by substituting one set of samples for the other.

Potentially more memory can be needed because of the duplication of most samples; however, I anticipate that the largest model planned (Paramount 440) will fit into 8 GB. The 310 can run in 1.5 GB and the 320 requires 2.0 GB for stereo (with lossless compression engaged).

Unification issues are the same as for other VTPOs. A coupler can double the number of notes used. On the other hand, an 8' Flute keyed simultaneously at middle C in all divisions will only use but one note (or two for stereo) due to unification.

Joe

Martin,

The idea of a 'P-factor' occurred to me too. I have the spread sheet started but in looking at the sites usually the best I could do was look at the specifications of the computer required in relationship to what I have available.

The good news is; I am up to number 68 in the number of sample sets that WILL fit on the base model OT-1 with 6GB. Some of those 68 sample sets are 24 bit compressed loads and I need to calculate the 'P-factor' to see if they truly fit. I believe the DualCore 2.8GHz, 3 MB cache, probably has a 'P-factor' of around 3000 and the Quad Core Q8200 2.33 GHz, 4MB cache perhaps 4500, In the 8 or so high end sample sets (those that require >8GB), none of them will run loaded as 24 bit 48 kHz uncompressed surround samples. So those go in the 'if you want one of these sets this computer is not for you' category but in the group of 60 that are listed as 'loads in 4GB', it appears that a Dual Core 2.8 Ghz, 3MB cache, and 8GB of system ram will be sufficient for all. In the group that requires 8GB of system ram to load, it is clear that some will run if loaded compressed and dry but it is uncertain if they will exceed the polyphony capability under some conditions.

If you have any notes on the number of pipes and reverb delays on any of the available sets that you could share, I would be glad to put them in the spread sheet and send it to you. I think we probably should work from the top down in this case and get the bigger ones first. I also will volunteer to keep it up to date if you could provide me with the pipes and reverb for new ones as they come out.

Thanks for your advice in this area.

Tom

[B. Milan]

If you have any notes on the number of pipes and reverb delays on any of the available sets that you could share, I would be glad to put them in the spread sheet and send it to you. I think we probably should work from the top down in this case and get the bigger ones first. I also will volunteer to keep it up to date if you could provide me with the pipes and reverb for new ones as they come out.

Hello Tom,

We wouldn't have the time for that. You should be able to just visit each vendors website for the particular sample set at hand and be able to judge the figures based on the information provided on those websites. We do not have an internal list for each sample sets specs, so we would need to review the websites just the same as you. If you want to take it upon yourself to undertake that project please feel free, but unfortunately we just don't have the time to spend on that for you.