It is currently Fri Mar 29, 2024 10:33 am


Impact of PCIe4 SSDs vs PCIe3 on HW6 load/cache build times

Buying or building computers for Hauptwerk, recommendations, troubleshooting computer hardware issues.
  • Author
  • Message
Offline

bobhehmann

Member

  • Posts: 74
  • Joined: Sun Dec 01, 2019 6:08 pm
  • Location: Moorpark, California

Re: Impact of PCIe4 SSDs vs PCIe3 on HW6 load/cache build ti

PostWed Aug 18, 2021 7:05 pm

Hadn't heard the Linus rant re AVX-512, but general press on it was that it was far from ready for prime time, though seemingly a good goal. Some speculation that that was the reason AMD chose to not implement it for the moment, as there is another related extended instruction set that AMD says has most of the bang for a large set of problems, without the heat penalty. Better use for the heat/power/silicon budgets. AMD may have finally felt pressured to add AVX-512 to the Zen-4 architecture, to keep "brand-marketing" parity with Intel. I certainly would recommend AVX2 capability for HW use, but I think that almost all present-day CPUs mid-range and up from either main manufacturer have it. I think AMD adopted it with the 1st-gen Ryzen.

Re: IR - I'm interpreting your comments to be that you are looking for sample-set producers to explicitly take greater advantage of HW6 Convolution Reverb. Interesting thought. Perhaps specific IR samples taken from the various microphone locations, and a matching set of IR-optimized samples? I know it has been popular of late for the producers to implement 3 or more stereo perspectives (6 or more channels per rank) with a HW mixer interface, and building-acoustics baked in to assumptions about the user-adjustable mix - pick your reverb by adjusting the perspective mix. That's what led to Alessandria being a 106GB load full out - 6 channels with extensive duration / native reverb / et al in the samples.

Would be interesting to see a "best implementation" of both approaches for perhaps two or three mid-size instruments of different character, side-by-side (each instrument presented in a best-of-breed IR based sample, and an otherwise identical best-of-breed native acoustics sample, both with perspective mixing.) Alas, this probably falls in the same category as Martin's response to a distributed HW: time/money may not be in the cards. I adore Gernot's Anloo take, and play it with his recommended IR config - and it is wonderful. But I would never be able to compare that against Augustine's Tihany or Piotr's Alessandria, both following a 6-chanel / 3 perspective / natural reverb approach. The underlying instruments are too different, and I'm guessing so are the acoustics - too many non-linear variables interacting for me to tease it apart!

Regardless, unless there is a hardware break-through for consumers, I'd hate to see otherwise desirable future sample-sets exceed the capabilities of a 128GB PC, or go beyond the capabilities of the readily available but not exotic high-end consumer CPUs of the day (for AMD, 5900/5950 class, or their equivalent successors, or Intel class equivalents). High-end Workstations and Server class CPUs are just too far off the chart for what it still essentially a consumer-tool. Just my opinion...

Thought provoking! Thanks kindly.
Cheers, Bob
Offline

bobhehmann

Member

  • Posts: 74
  • Joined: Sun Dec 01, 2019 6:08 pm
  • Location: Moorpark, California

Re: Impact of PCIe4 SSDs vs PCIe3 on HW6 load/cache build ti

PostWed Aug 18, 2021 7:38 pm

The fast-boot topic caught my attention. Unless I'm away long-term, I generally just sleep my PC, for example, overnight. But, out of habit I suppose, I generally close down anything important, such as an IDE, or any editor (Word, Video...), and HW. And I power off most external devices, such as speakers/amps, external D/A, monitors, keyboards... So if I want to play with HW, I have to start it up and load an instrument first time each day, or perhaps more frequently.

Inspired by you guys, I tried an experiment - leaving HW up, loaded, ready to play while I sleep the machine. I'm careful to not shut down any peripheral until after the PC is snoring, so as not to trigger a precipitous loss of device from underneath running HW. Do not want to annoy the Giant at bedtime...

And it almost works! I power everything external back on first, then wake the PC. From first triggering mouse-jiggle to playing on a keyboard, about 6 seconds. One rub is with my external D/A, a Focusrite Clarett - while it engages just fine with the OS upon waking, HW may or may not produce sound on it, at first. Visually, everything looks fine, no errors, HW believes it is generating audio - but no sound. If I tweak the Window's general volume control slider (the little task-bar speaker icon), Window's sounds its feedback ping, and then HW can be heard. About 50/50 if I need to kick it to allow HW to sound on wake-up. I'm guessing its a race condition between HW and the Focusrite driver, which does synch-up with the Clarett when the PC wakes up - perhaps if HW starts communicating with the driver before the driver is synched, something gets lost. And external display panels are not refreshed. Wait, got an idea, sleeping... Testing... Aha, doing a HW Engine:Restart Audio/Midi clears it, and everything refreshes fine! So far, no MIDI-via-USB devices have gotten confused as to their identity.

OK, let's call it 10 seconds, as I'll always do the HW Restart Audio/Midi just to re-synch the LCD Panels even if the sound was OK.

Thanks for the inspiration guys!
Cheers, Bob
Offline
User avatar

mdyde

Moderator

  • Posts: 15446
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Impact of PCIe4 SSDs vs PCIe3 on HW6 load/cache build ti

PostThu Aug 19, 2021 6:33 am

vpo-organist wrote:For Hauptwerk, the next stage will be an extensive extension to use Convolution Reverb (I own HW VI).
There I would like to ask Martin Dyde to push this topic. Especially Gernot Wurst from Prospectum is an excellent contact person for this, because he creates samplesets based on IR and has professional knowledge about it. At the moment it is probably not possible to deliver an IR configuration, is that right?
The use of IR will solve some problems for the user: Much less memory consumption, much shorter loading times, much less polyphony requirements, etc.

We will also see with the upcoming Ansbach/Wiegleb sample set how fantastic a sample set with Convolution Reverb will sound.

The better and faster Hauptwerk will support IR, the better sound experiences we may enjoy.


Thanks, vpo-organist.

There have been lots of lengthy discussions on this forum over the years of the advantages/disadvantages of wet samples vs. dry samples + convolution reverb (including per-pipe convolution reverb, with a separate impulse response per pipe). I know that you're strongly in favour of the latter approach.

Briefly:

vpo-organist wrote:At the moment it is probably not possible to deliver an IR configuration, is that right?


Hauptwerk v5/v6's convolution reverb engine is designed to be extremely high-performance and to take full advantage of lots of CPU cores specifically for handling large numbers of simultaneous convolution reverb instances, for per-pipe convolution, etc..

You can indeed configure 'virtual acoustics'' in v5/v6, with ranks/groups routed amongst up to 1024 separate reverb instances (each with separate reverbs) for (almost) per-pipe convolution. Here's the excerpt from the Hauptwerk user guide about that:

Multi-channel virtual acoustics

If you use just a single stereo audio output from Hauptwerk (such as for headphones), then you can still obtain most of the benefits ‘virtually’ that multi-channel audio would bring (in terms of perceived spatial distribution of pipes/ranks around an acoustic space, and increased perceived clarity as a result of it) by configuring Hauptwerk exactly as you would for conventional multi-channel audio output (as described in the previous sub-sections), but not actually selecting any audio device channels for the primary mixer buses, and instead selecting different impulse response reverbs for each of them, preferably with those reverbs having been created for different virtual sound source positions within a single (real or modeled) acoustic space. You will then be positioning the primary mixer buses (and thus pipes routed to them)
accordingly within the virtual acoustic space. In that case, you would just select device channels for master mix bus 1 (which is fed a stereo mix from all primary buses by default).


For an example showing how to route ranks to groups of mixer buses (to which you could apply reverbs individually) please see the 'Example 6: multi-channel audio with 24 (or more) speakers configured as stereo pairs in 3 groups of 4 pairs each, with different organ divisions routed to different groups ...' section in the user guide (starting on page 216 in the current v6.0.2 version). In that example, for setting up a 'virtual acoustic' you would simply select different IRs for each of the primary mixer buses, instead of selecting device channels for them.

At the moment (v6) there are only 8 mixer presets, and you would need to configure any such virtual acoustic manually (assigning reverbs to mixer buses, putting them into groups, and assigning ranks to the appropriate groups), but as I mentioned in reply to one of your previous posts in another thread:

viewtopic.php?f=1&t=19921&p=149532#p149517

mdyde wrote:Yes -- the intention always was/is for the limit to be much higher than 8, so that you would effectively use one mixer preset per organ if you wish. There wasn't time within Hauptwerk v5's development to include the ability to filter the screens by mixer preset, so the limit was set to 8 as a temporary measure, in order to keep the screens reasonably usable in the meantime.

However, as high priority enhancements we do intend to increase the number of mixer presets, and also to allow filtering the screens by mixer preset, and also to allow mixer presets to be copied/imported/exported, so that mixer presets could conveniently be used on a per-organ basis (and suitable configurations provided by sample set makers or other users to import) for those people wishing to use per-organ convolutions schemes.


Getting back to the subject of PC hardware, since convolution reverb is particularly important to you (and especially if you want to use larger numbers of separate reverb instances, e.g. for virtual acoustics), if I was in your position I would probably opt for an Intel CPU that supported AVX-512, since it might potentially benefit the performance of Hauptwerk's convolver. (All CPUs that support AVX-512 also support AVX2.)

I hope that helps.
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

vpo-organist

Member

  • Posts: 306
  • Joined: Wed Apr 29, 2020 6:49 am

Re: Impact of PCIe4 SSDs vs PCIe3 on HW6 load/cache build ti

PostThu Aug 19, 2021 11:44 am

Hi Martin,

I think times are changing right now and old discussions about releases/IR are no longer relevant.
The reason is that e.g. 8 channel sample sets are released, which overtax most PC's.
So the topic should be refreshed with the new experiences.
After all, it should be a worthwhile goal to enable as many customers as possible to have a great sound experience with their not quite up-to-date PC's. And this is made possible by IR in its simplest form.

I read the section in the HW manual, but didn't really understand it.
I've also seen the tutorials from users, but it's still complicated to understand everything.
Maybe it's a little bit because of the English that I have a hard time with it.

But I now understand that by virtual acoustics is meant audio routing with multiple IR's.
That would be nice if I could set up an IR for each division for the Caen-Dry. But I have no idea how to do that.

AVX-512 was reset after your problems after all. Now you're saying an AVX-512 CPU might be worthwhile.
Why? If only the SIMD extensions of AVX-2 are supported, then AVX-512 is useless because the larger vectors are not used.

Is there a web page of the FFT library purchased? Will this vendor support the AVX-512 instruction set?

With AMD Zen 4 AVX-512 should be supported. But that will probably take at least another year until corresponding hardware is on the market.

DDR5 will then also be a positive aspect. Faster is always good.

Best regards
Offline
User avatar

mdyde

Moderator

  • Posts: 15446
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Impact of PCIe4 SSDs vs PCIe3 on HW6 load/cache build ti

PostThu Aug 19, 2021 12:23 pm

Hello vpo-organist,

vpo-organist wrote:But I now understand that by virtual acoustics is meant audio routing with multiple IR's.
That would be nice if I could set up an IR for each division for the Caen-Dry. But I have no idea how to do that.


Yes -- that's absolutely possible. The section I mentioned in the user guide covers (by way of example) how to put primary mixer buses into groups, and to route ranks to those groups, and you can assign IRs to mixer buses. That's how you apply different IRs to different ranks (or even to different pipes within ranks).

If you have difficulty understanding how to use audio routing from the user guide, perhaps these topics may be of help:

viewtopic.php?f=17&t=17916
viewtopic.php?f=17&t=19055
viewtopic.php?f=17&t=19438

If it still isn't clear, you could contact Francois at support [at] hauptwerk.com and tell him what you want to set up and he could potentially help you to configure it.

vpo-organist wrote:AVX-512 was reset after your problems after all. Now you're saying an AVX-512 CPU might be worthwhile.
Why? If only the SIMD extensions of AVX-2 are supported, then AVX-512 is useless because the larger vectors are not used.


This is what I said when you first asked about AVX-512:

mdyde wrote:AVX-512 might possibly benefit convolution reverb performance (which relies on a third-part library for the DFFT function), but Hauptwerk (v6.0.2) won't currently benefit from it above AVX2 in other respects. (AVX2 is definitely worthwhile.)


I.e. the convolver may indeed benefit from AVX-512 (currently, and/or in future versions). I don't have specific benchmarks to offer on the performance gain (if any) that AVX-512 will give with it, but in your case I would recommend opting for an Intel CPU with AVX-512, since it may benefit you if you want to use convolution reverb extensively.
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

vpo-organist

Member

  • Posts: 306
  • Joined: Wed Apr 29, 2020 6:49 am

Re: Impact of PCIe4 SSDs vs PCIe3 on HW6 load/cache build ti

PostThu Aug 19, 2021 2:12 pm

Thanks, Martin.
Previous

Return to Computer hardware / specs

Who is online

Users browsing this forum: No registered users and 1 guest