The No.1 Website for Pro Audio
At last: new Mac Pro!
Old 30th December 2019 | Show parent
  #5221
Lives for gear
 

Quote:
Originally Posted by khollister View Post
I suspect we all need to take a breath before throwing the 2019 MP under the bus. Setting aside the Diva behavior (which I'm increasingly certain is some programming anomaly with respect to the Xeon's and MacOS), it seems all we are proving is that for similar numbers of cores (and reasonably equivalent architectures), all-core turbo clock speed wins. It is also pretty clear that for most practical music making purposes, more cores certainly can trump a slight clock disadvantage. After all, I can't imagine a musical scenario where I'm using 48+ voices of Diva (or Alchemy) anyway. Something is seriously amiss when 8 instances of a plugin on the playback buffer uses WAY more CPU than 1 instance on the live buffer x 8 in Logic.

Furthermore, I think it is a very real possibility that the low clock speeds noted on the new MP's may be due to increased use of AVX512 instructions which throttle back the turbo speeds, not a bug in Apple's SpeedStep/C-state or VRM implementation. Maybe a SW/FW fix will be forthcoming, but I doubt it. It would be a pretty big thing to miss in testing. Aside from cost, this is the biggest thing that would keep me from swapping my iMP for a MP (going from 3.8-4.0 GHz per core to 3.2-3.4). Of course going from 10 to 16 cores would probably more than offset that in real-world use.

IMHO, saying a MBP or even i9 iMac "outperforms" an iMP or 2019 MP is missing not only the availability of a lot more cores, but peripheral bandwidth (2 TB3 busses instead of 1 plus PCIe3 slots on the MP) and cooling (ability to run high duty cycles at near full capacity without throttling or noise).
Agree wholeheartedly with your post, except the very last part.

We’re saying the MBP or iMac i9 outperforms the new Mac Pro or iMac Pro in single-core performance ONLY. Which they do. The i9 is Intel’s best single-core performer. The Xenon is server grade, which means single core performance is not its intended goal. The Xenon is about multi-core performance, which it excels at. Your points about the other design choices in the Mac Pro are absolutely right, and there’s no debating that. Thermals play a big role too. That’s why having an i9 in the new Mac Pro would be nuts.

In real time applications, like playing VIs live with low buffer sizes to record, the i9 does have an advantage over the Xenon. This is a fact, and neither Intel nor Apple would fight you, because their intentions were different for the Xenon. Apple is targeting intense multi-core needs like graphics rendering and video. For real time performance, the i9 is the winner. A Mac Pro with a high core count i9 option, with a high base clock like the 9900KS, WOULD be the ultimate music production rig.

You clearly have a lot more knowledge than me about certain specifics on how the Xenons work vs. the i9, which I appreciate.
Old 30th December 2019 | Show parent
  #5222
TNM
Lives for gear
Quote:
Originally Posted by MattyJoe View Post
This is a fair point, and one thing to realize is that every single CPU offered by Intel today fluctuates constantly. They all have turbo boost. This solution is Intel’s response to needing more processing power while keeping thermals and power consumption down. They decided to innovate more in this area rather than continuing to just push clock speeds higher and higher the way we were 5-7 years ago. This may end up being the wrong move. We’ll see.

Read about the 9900KS if you haven’t already. It’s basically an i9 9900K that comes factory overclocked at 4GHz, with the highest turbo boost speed possible:

https://www.pcmag.com/review/371606/...core-i9-9900ks

Intel did this for gamers, who are simply demanding higher clock speeds, and overclocking their CPUs manually because they’re fed up with Intel. The fact that they released a factory-overclocked i9 is an interesting development, and might indicate they’re questioning their current base clock/turbo boost approach, especially since AMD is wrecking them.

The gaming community will dictate processing needs more than any other market, because it’s just a huge market. Apple is considering a GAMING Mac now for crying out loud! Us audio guys will benefit from this large market demanding better options from manufacturers, because on our own we’re just not big enough to make a difference.



This is another good point to remember. Optimization for CPUs is a very important part of how all software functions.

What’s happening here is people are saying “THIS power hungry synth brings the Xenon to its knees, so Intel needs to step up their game.”

But really, it should be the other way around. Diva is so poorly optimized that it’s overloading your cores easily, with the same functionality that other very sophisticated synths do WAY more efficiently. Honestly I think there are many that offer far MORE functionality than Diva, and don’t wrecklessly overload your CPU. In 2012, Diva almost had no equal, sonically. But in 2019? I’m not so sure. This is a bad look for U-He, not Intel or Apple.

Someone was saying it looks to them like Diva hasn’t been updated much since 2012 or something. That gives you your answer right there, it does for me anyway.

But I’m also someone who isn’t always looking for the most sophisticated analog-emulating synth out there. In your opinion, does Diva still offer something that other options in 2019 can’t match?
Don't even think about telling U-HE his synth is poorly optimised.. We had an argument about his reason plugins being poorly optimised, from actually being friends we more or less became foes, and the next thing I know I am permanently banned from KVR from a forum topic that was arguing with him. Of course, zero profanity or personal attacks were given, so I maintain to this day he told the mod to ban me or he'd pull his KVR support. I have zero actual proof though.

I understand everything you are saying but it doesn't change the fact that windows systems can be set to turbo disabled and ONE constant clock speed.. even with turbo disabled, if EIST is enabled, the cpu will clock low when not being stressed, for efficiency/lower power use. Of course I know all this but I do appreciate you giving a detailed reply and taking the time.
My point remains, that even without a bios, in bootcamp I was able to set it so my iMac had zero fluctuation, and that caused a double performance increase in a certain situation, using the very same project file in the same OS respective version of the same DAW.
There are tools like throttle stop, intel's own XTU, and the bios as well as motherboard manufacturer specific apps that can do all sorts of things we can never do in OSX.
Of course, with a TOSH, we can disable all speed stepping in BIOS and OSX can't do a thing about it but that's a complicated route to go.. Most of us just want a turnkey computer out the box.

Honestly? If apple build the gaming mac properly it will likely be the best performing for us SLUTZ as well.

Anyway... I digress.. yes, all intel CPU's fluctuate, but in Windows they can be made not to. That's the TLDR; version of it in a nutshell.

I have no clue if Diva is poorly optimised.. but even just loading it and playing a few notes on the default patch, doing nothing else, you can hear it literally sounds like a real analog synth.. The characteristics and "goose pimple on arm" is all there. So no matter what, it sounds fab. I am too scared to buy it in case my order gets rejected LOL (seriously though), but I will never deny Urs Heckmann of his ability to make a wonderful sounding synth.. I hate Zebra, always did, yet 99% of people love it, so I am in the minority there.. I do like Hive as an EDM synth for sure..But for me, the second he released Diva, I was convinced immediately by the sound.
Do you have any links that show it hasn't been optimised at all since 2012? Not doubting YOU or that you saw it, I just find it hard to believe it's the same basic code with OS compatibility updates once a year..I think he would want to optimise it as best as possible. Sure, his reason plugins got 5x the instances in a VST or AU daw on the same computer that they did in Reason, but at least 3/4 of that problem was a major efficiency problem with the RE format itself, as almost all RE devs suffered from it.
I NEVER meant that as an attack on Urs personally, it just so happened because those were the ones that had the biggest performance gap between the RE and their VST brethren, that they (the Uhbiks) were the suite of effects I decided to use to showcase all my benchmarking in the "RE is unbelievably inefficient" tests, and I think he took it really personally unfortunately. It was aimed more at propellerhead than anyone. Even when they gave VST support to Reason, it was the most inefficient DAW on the planet still, but apparently they fixed that now. But that's another, WAY off topic, long story.

PS I know all about the KS.. 5GHZ factory all core turbo, so basically with a good PSU/motherboard/cooling combo, can be set to just run at 5ghz permanently, turbo disabled, one stable clock speed.. has the highest single core performance of any current CPU I believe.
Old 30th December 2019 | Show parent
  #5223
TNM
Lives for gear
Quote:
Originally Posted by MattyJoe View Post
Agree wholeheartedly with your post, except the very last part.

We’re saying the MBP or iMac i9 outperforms the new Mac Pro or iMac Pro in single-core performance ONLY. Which they do. The i9 is Intel’s best single-core performer. The Xenon is server grade, which means single core performance is not its intended goal. The Xenon is about multi-core performance, which it excels at. Your points about the other design choices in the Mac Pro are absolutely right, and there’s no debating that. Thermals play a big role too. That’s why having an i9 in the new Mac Pro would be nuts.

In real time applications, like playing VIs live with low buffer sizes to record, the i9 does have an advantage over the Xenon. This is a fact, and neither Intel nor Apple would fight you, because their intentions were different for the Xenon. Apple is targeting intense multi-core needs like graphics rendering and video. For real time performance, the i9 is the winner. A Mac Pro with a high core count i9 option, with a high base clock like the 9900KS, WOULD be the ultimate music production rig.

You clearly have a lot more knowledge than me about certain specifics on how the Xenons work vs. the i9, which I appreciate.
Hang on.. then why don't Apple release something for us that is great for ALL ROUND use in a DAW? They are pimping the mac pro as a DAW beast, not just a video editor's dream, as they clearly use Logic as a selling point for it in the mac pro's webpage.

Next, it's not just at low latency the single core performance is making a difference.. we can see with the tests that at 1024 playback buffer, the Xeon still falls over..

I want to use another perspective.. if Diva really wasn't updated since 2012, that means it was made in 2012 with 2012 speed CPUs.. shouldn't a modern cpu then eat it for breakfast?

Is there anyone here who would say no to a desktop mac, just called, "the new mac", no pro designator required, with a 9900KS inside WITHOUT power limit throttling? (as discussed, different from thermal throttling and seemingly an OSX staple right now). And for those wanting a sort of best of all worlds scenario, why not a nice 10940X, *allowed* to run at it's intel rated all core turbo? I bet you most of us would choose that in a heartbeat.. It would pass the Diva test, as WELL as having a good amount of cores, and excell in all scenarios.

The fella on the previous pages (forgot his username, so sorry) may have been on to something with the idea of a multi mac setup for those who need more cores but also speed..The 9900K gets replaced very soon with a new 10 core comet lake flagship consumer part.. If apple were to release a "non pro", non "screen required even if you don't want it" desktop with that, allowed to run at full specification, two of those chained, albeit only if someone were willing to use a VE Pro slave, would kill the 28 core mac pro in VI count. I am 100% certain in this case that 20 cores would beat 28 and I'll even bet it would be better as a single 10 core mac than the 16 core mac pro, "overall" for DAW use.. However, if they release a 10 core 5GHZ all core turbo version, we will probably get it in the 27" iMac throttling down to 3.5 under load
Old 30th December 2019
  #5224
TNM
Lives for gear
https://www.guru3d.com/news-story/co...ke-s-leak.html

Hmmmmm, to say the least..

That six core part with a base of 4GHZ will go into the lower end iMac, and kill the iMac pro for DAW use.

2020 is going to be very interesting indeed.

Everyone and his dog using a decent cooling setup will have all 10 cores at 5.2ghz on the 10900K.

I really think, and I can only speak personally, that if I sold the iMac pro and used the 16" MacBook for now (i'd just spend a couple hundred bucks on a 1440P screen in the meantime), and exercise some patience, that with any luck, apple will release a refreshed iMac line to accomodate these, and maybe with better thermals. If so, this could be the one to watch out for. Need 2 screens? Buy 2 and chain them. I just wish there was a way to slave the current iMac gen as a screen like you could with the thunderbolt/thunderbolt 2 models.. Then you could slave VE Pro, use that when required, then press command F2 to put it in target display mode as a second screen for your DAW, with VEP running in the background just doing playback. Need to edit a synth again? command F2. That would be brill. I'd happily buy two 10 core iMacs and work this way (if they were quiet).

I think the best way to use VE Pro is to compose all the stuff on your DAW (so you are not subject to VEP latency), then simply change the output of the track to the VE Pro server and let the playback grunt be offloaded there..
Which brings me to another idea..

Ok.. bear with me.. I am not trying to be stupid here..

VE PRo is platfrom indépendant... if you use plugins that are both OSX and Windows it has total inter compatibility with projects..

Why not the 16 core NMP, and put together a dirt cheap (in comparison) 10900K pc box just as a slave to handle all the Diva style VI's, rather than spend thousands extra on the 28 core? Most of your work would be done in your DAW of choice in OSX still.. I mean it's an idea to get around the core speed problem, not ideal, but an idea.
Old 30th December 2019 | Show parent
  #5225
Lives for gear
 

Quote:
Originally Posted by TNM View Post
[url]
Why not the 16 core NMP, and put together a dirt cheap (in comparison) 10900K pc box just as a slave to handle all the Diva style VI's, rather than spend thousands extra on the 28 core? Most of your work would be done in your DAW of choice in OSX still.. I mean it's an idea to get around the core speed problem, not ideal, but an idea.
That sounds like a workflow killer. I wouldn't personally entertain this unless I was composing for film/tv.
Old 30th December 2019 | Show parent
  #5226
Gear Maniac
Quote:
Originally Posted by mat1 View Post
That sounds like a workflow killer. I wouldn't personally entertain this unless I was composing for film/tv.
Been there, done that. And most of the film/TV/game guys over on VI-Control are trying to get back onto a single box and not have to use VEP. It works incredibly well, but I hated the workflow. The reason I dropped the coin on the iMP was to get rid of the VEP slave machine.

My previous DAW was a 5.1 6-core MP.
Old 30th December 2019
  #5227
Lives for gear
I see things have not much changed since the Mac Pro 2013 version. Several plug in instruments are not set up for multi core use so single core performance is important.

This was why I went with the 6 Core Mac Pro 2013 back in 2014.

Is Kontakt the only sample player to use multi core processing?

It seems just a matter of these Plug In creators to write software to actually use the multi cores. WHY don't they fix this? Otherwise, we just have a bottlenck on speed use.
Old 30th December 2019 | Show parent
  #5228
Gear Addict
 
Greg Heimbecker's Avatar
Quote:
Originally Posted by MichaelDroste View Post
Here it is
That seems like a hell of a lot of 2buss processing! I'm not surprised it's balking.
Old 31st December 2019 | Show parent
  #5229
Tui
Gear Guru
 
Tui's Avatar
Happy New Year, guys.

Let's make it a good one.

Old 31st December 2019 | Show parent
  #5230
TNM
Lives for gear
Quote:
Originally Posted by Tui View Post
Happy New Year, guys.

Let's make it a good one.

11 hrs 23 to go here..

2020 is a major one IMO, and for me personally.. Lots of self betterment and musical goals I have..

Absolutely.. let's make it an AWESOME start to the decade!
Old 31st December 2019 | Show parent
  #5231
Gear Maniac
 
derdiedasscheusa's Avatar
 

What I would want is a „dynamic cpu“ which can switch between high core counts and lower speed, to lower core count and higher speed, depending on the task or user setting. It needs to have the ability of deactivating a couple of cores and instead bump up the frequency or vice versa.
Old 31st December 2019 | Show parent
  #5232
Lives for gear
 
juiseman's Avatar
Quote:
Originally Posted by TNM View Post
yes, you must have missed it in all posts here.. (huge topic!)

my very own iMac pro in pro tools gets 12 in the diva test on windows vs 6 on OSX.
Windows has been set to high performance mode, which forces the minimum processor speed to actually be it's maximum speed and doesn't fluctuate. So it was at 3.93ghz even with no DAW open and browsing the web LOL. Waste of power but the best situation for a DAW.

So even on the iMac pro, with the newer OSes, apple is limiting the power draw under sustained load..
The guy that tested the new mac pro literally shows in his video that windows is giving 20 whole more watts to the CPU vs OSX. It's really ridiculous when you think about it.. which is why the CB score was basically the best that cpu could get. 4000 is spot on for that 8 core running full bore without any throttling.

I really don't know why Apple do this, the only thing I can come up with is to save on energy bills but still..why not give the user the option? They used to do that in the power pc G5 by the way.. in system preferences/power settings, you could disable frequency hopping!
Sorry, I did miss that. That is crazy; I would think it would be 10 in Mac Os vs 12 in Windows...not double...

Also; side note, I thought Core Audio has extra safety buffers or something? if so this is more significant.
Old 31st December 2019 | Show parent
  #5233
Lives for gear
 
juiseman's Avatar
Quote:
Originally Posted by TNM View Post
10 cores at a constant 5.2 won't just nip the heels of the NMP 12 core, it will destroy it for DAW purposes.. Between those two choices, if apple offered it, i'd not even have to thunk about it.. I'd take the less cores at full speed.
I actually agree with that. I think it would destroy the Xeon in all areas. The only advantage the Xeon would have is 2 more cores (4 more threads) and 6 Memory
channels plus the extra PCI-E lanes.... those 10 faster core's clocked over 1GHZ faster will score higher than the 12 of the slower clocked Xeon.

The only thing left to would be the 1090k is far beyond its power efficiency curve;
but we don't care about here; not for our uses....
Old 31st December 2019 | Show parent
  #5234
Lives for gear
 
juiseman's Avatar
Off topic; but has anyone tried this Intel XTU utility on a mac booted into windows?
https://downloadcenter.intel.com/dow...lity-Intel-XTU

As long as the mac was using an unlocked CPU; it could work.

I tried it on LGA2011 based server chipsets with locked Xeon E5-2600 series CPU's;
you cant change the CPU multiplier; but you can set the power limit at max.

I'm wondering if it would work with a MP 6,1 (trashcan) because they use unlocked Xeon E5-1600 series CPU's (minus the 12 core, that's a E5-2697v2)
I only mention this for testing purposes;
or maybe for a few guys running cross platform DAWs & regularly dual boot.who may want to squeeze some life into an older Mac Pro

Ok, sorry. But that idea just popped into my head....

Also; if the made a utility like this that would work for Mac Os; that would solve
the thermal limiting issues or at least get the most performance out of these CPU's....
Old 31st December 2019 | Show parent
  #5235
Lives for gear
 

Quote:
Originally Posted by TNM View Post
Lots of self betterment and musical goals I have..
Those two things run contrary to each other, don't they? Not possible to have both!

:-)
Old 31st December 2019
  #5236
Gear Maniac
I've been doing some more experimenting with my 10 core iMP in light of the Diva discussions. What I have found is a few synths exhibit very non-linear behavior/scaling on the iMP Xeon, and I'm guessing they would behave more normally on recent i7/i9 CPU's based on the Diva testing. By non-linear I mean that the percentage of CPU used (as per Logic's CPU meter) increases dramatically more for each instance of the VI added than the CPU load for a single instance would indicate. E.g., 4 instances of a VI might consume 75% of 5 threads where a single instance (record armed even) consumes only 25% of a single thread.

So far, Diva, Repro-5 and Massive X exhibit this behavior.

Omnisphere, Zebra, Alchemy, Hexeract, Absynth & TAL-UNO (just the ones I own and have tried so far) behave "normally" - the total amount of CPU increases by approximately the number of instances compared to a single instance. The threading models vary - some use multiple threads when not record armed in Logic and some don't. But on average, the number of threads X indicated percentage = instance count X single instance CPU.

My Kontakt sample libraries behave as expected as well, with the added characteristic that the indicated CPU due to scripting often decreases dramatically when played on the playback buffer (1024) instead of the record armed (normally 128 in my case). Most of the synths do not demonstrate nearly as large a change.

It would be really interesting for folks that have recent 8 core i9 Mac's to check for this as well. I'm wondering if we are seeing
a) improper compiler options being used that don't scale to Xeons
b) compiler bugs
c) use of programming techniques that don't scale well on Xeons

While I understand TNM's argument that Diva only gets 6 instances because the CPU is overloaded and can't spin off another thread for instance 7, that really sidesteps the root cause of why 6 instances max out 7 threads (700% total CPU thread use) when 1 instance only uses 25-30% of a single thread. It is this non-linearity that is the root cause. The decrease in the clock speed of the Xeons just means this bites you in the ass even sooner.

The question I have is whether this also exists on an i9 and the increase in instances is just due to a bit more overhead so additional threads can be spun off, or is the scaling fundamentally different? I'm guessing the later since an extra 10-15% of clock speed should only get you another 1 or 2 instances - not 10.

And BTW, I see this on Cubase as well as Logic on Catalina, so I refuse to believe it is related to the DAW.
Old 31st December 2019 | Show parent
  #5237
TNM
Lives for gear
Quote:
Originally Posted by Resonater View Post
Those two things run contrary to each other, don't they? Not possible to have both!

:-)
Not even remotely.. I don't see it!

The reason they directly relate TO one another for me, is that I didn't create enough music in 2019.. So bettering myself actually means making more music and generally trying to get along better with others.. they are my two main goals.. but, I won't go further OT. Sorry about that!
Old 31st December 2019
  #5238
TNM
Lives for gear
@ khollister , very interesting post.. I never looked at it that way. I was always under the assumption that because of the way virtual cores work, each new track that allocates itself to a new core, will add a bit of overhead to others, which is why Diva eventually tips over with so many unused cores.. But I guess there's a lot more to it!
Old 31st December 2019 | Show parent
  #5239
TNM
Lives for gear
Quote:
Originally Posted by juiseman View Post
Off topic; but has anyone tried this Intel XTU utility on a mac booted into windows?
https://downloadcenter.intel.com/dow...lity-Intel-XTU

As long as the mac was using an unlocked CPU; it could work.

I tried it on LGA2011 based server chipsets with locked Xeon E5-2600 series CPU's;
you cant change the CPU multiplier; but you can set the power limit at max.

I'm wondering if it would work with a MP 6,1 (trashcan) because they use unlocked Xeon E5-1600 series CPU's (minus the 12 core, that's a E5-2697v2)
I only mention this for testing purposes;
or maybe for a few guys running cross platform DAWs & regularly dual boot.who may want to squeeze some life into an older Mac Pro

Ok, sorry. But that idea just popped into my head....

Also; if the made a utility like this that would work for Mac Os; that would solve
the thermal limiting issues or at least get the most performance out of these CPU's....
Yeah I used it in the 8 core 15" i returned.. i couldn't undervolt as it was the locked cpu, the 9880H, but when i get around to installing bootcamp on my 16" I think I will be able to as it's the 9980HK. However, i was able to limit the total watts which gave me much more sustained performance and a pretty stable clock speed for gaming, rather than trying to go to 4.8ghz then overheat and drop to 2.8 and stuff like it did in mac OS. I was basically seeing a consistent 3.5hghz.. I am hoping I can undervolt the 16" and tune it to see 3.8ghz across all cores in bootcamp at all times.. It will be a perfect balance if I can pull that off for gaming. Because of the PT scaling issues in Windows 10 on high dpi screens, I actually won't be using it for DAW work, unless I eventually get an external monitor for it without any scaling. PT and Reaper were both horrible to use on the iMac pro visually (and yes I tried every scaling override setting and there was always a compromise). The only way for it to "just work" was to set the 27" 5K screen to it's actual scaled display of 2560x1440 but it was super blurry. UAD doesn't have high DPI support on windows, neither did PT.. So even with the performance bump on windows, once again it shows how much easier and just "plug and play" OSX is. I never had such a scaling issue with one single app in the entire history of using retina macs since 2014.. even if they were not a high res app, they just worked at the correct size. In reaper i would get plugins that were the size of a matchbox if i stayed at 5K in windows LOL! (just an example).
Gaming is find cause i can just choose my resolution per game.
Old 31st December 2019 | Show parent
  #5240
TNM
Lives for gear
Quote:
Originally Posted by juiseman View Post
Sorry, I did miss that. That is crazy; I would think it would be 10 in Mac Os vs 12 in Windows...not double...

Also; side note, I thought Core Audio has extra safety buffers or something? if so this is more significant.
well technically it does but the UA driver has similar real latency in both OS. Ultimately it doesn't really matter if core audio has extra buffers, it only matters what the real latency of the interface is and if windows is lower.. with Motu interfaces for example, the latency is higher in windows.
Old 31st December 2019
  #5241
Gear Guru
 
henryrobinett's Avatar
Does anyone know whether I can use the firewire Satellite UAD box with a firewire PCIe card with the new Mac Pro? I'm thinking that I have three UAD accelerators: Thunderbolt satellite, PCIe satellite and a firewire box. I was assuming I'd have to do away with one and I've never run all three together.
Old 31st December 2019 | Show parent
  #5242
Lives for gear
 
basehead617's Avatar
In my experience, all of my audio apps performed worse on my MBP under windows in Boot Camp than natively in MacOS.
Old 31st December 2019 | Show parent
  #5243
Gear Maniac
Quote:
Originally Posted by henryrobinett View Post
Does anyone know whether I can use the firewire Satellite UAD box with a firewire PCIe card with the new Mac Pro? I'm thinking that I have three UAD accelerators: Thunderbolt satellite, PCIe satellite and a firewire box. I was assuming I'd have to do away with one and I've never run all three together.
https://www.apple.com/shop/product/M...rewire-adapter

Apple FW->TB adapter. If your TB Satellite is the previous TB2 version, I suspect you could daisy chain this on the back of it since the adapter is TB2.
Old 31st December 2019 | Show parent
  #5244
TNM
Lives for gear
Quote:
Originally Posted by basehead617 View Post
In my experience, all of my audio apps performed worse on my MBP under windows in Boot Camp than natively in MacOS.
The iMac pro had exceptionally low DPC latency.. usually bootcamp has terrible DPC latency which makes for bad DAW performance.. It's just one of those freak things with the iMac pro.. The MacBook will be a completely different story, same with mac mini I believe.. The new mac pro should be good like the iMac pro.

Anyway, my iMac pro had lower dpc latency than some of the best PC Motherboards, which means it was entirely suited for low latency DAW work in windows..

32 buffer in PT playing 16 notes of Diva is unheard of on mac, PT on mac goes nuts at 32 buffer.. but of course it also depends on the DAW you use. I was playing away at 32 buffer in PT without a single dropout in bootcamp.

S1 definitely performs WAY better on windows, but other DAWs are all open for discussion.. What I can guarantee is that PT and S1 will perform better on windows on the same spec hardware, as long as your windows build has low DPC latency.
Old 31st December 2019 | Show parent
  #5245
TNM
Lives for gear
Quote:
Originally Posted by henryrobinett View Post
Does anyone know whether I can use the firewire Satellite UAD box with a firewire PCIe card with the new Mac Pro? I'm thinking that I have three UAD accelerators: Thunderbolt satellite, PCIe satellite and a firewire box. I was assuming I'd have to do away with one and I've never run all three together.
Personally I would get a thunderbolt hub with a firewire port which will just work.. and everyone should have one good thunderbolt hub on the mac pro considering it's dearth of standard USB ports. A PCIE card should work though, I don't see why not. Note, if you have a firewire UAD in your system, even though you CAN mix and match, all your UAD plugins default to the slowest possible latency, i.e even the plugins hosted on your thunderbolt and PCIe devices will have the additional latency of the firewire satellites as if they were being hosted on that. If you happen to use Logic, this makes it even worse for the latency compensation automation timing issue when using UAD plugins.. I'd try fob it off and replace it with another thunderbolt.. They have halved in price these days, but I understand that may be impractical for you. That said, the thunderbolt hub will be a good purchase regardless.. on my iMac pro, I use two hubs.. the native thunderbolt 3 hub from cal digit, and a thunderbolt 2 OWC hub using the apple TB3 to TB2 adapter, and the firewire port on that works perfectly with the motu ultralite at 32/96.

Cheers!
Old 31st December 2019 | Show parent
  #5246
Lives for gear
 

Quote:
Originally Posted by khollister View Post
I've been doing some more experimenting with my 10 core iMP in light of the Diva discussions. What I have found is a few synths exhibit very non-linear behavior/scaling on the iMP Xeon, and I'm guessing they would behave more normally on recent i7/i9 CPU's based on the Diva testing. By non-linear I mean that the percentage of CPU used (as per Logic's CPU meter) increases dramatically more for each instance of the VI added than the CPU load for a single instance would indicate. E.g., 4 instances of a VI might consume 75% of 5 threads where a single instance (record armed even) consumes only 25% of a single thread.

So far, Diva, Repro-5 and Massive X exhibit this behavior.

Omnisphere, Zebra, Alchemy, Hexeract, Absynth & TAL-UNO (just the ones I own and have tried so far) behave "normally" - the total amount of CPU increases by approximately the number of instances compared to a single instance. The threading models vary - some use multiple threads when not record armed in Logic and some don't. But on average, the number of threads X indicated percentage = instance count X single instance CPU.

My Kontakt sample libraries behave as expected as well, with the added characteristic that the indicated CPU due to scripting often decreases dramatically when played on the playback buffer (1024) instead of the record armed (normally 128 in my case). Most of the synths do not demonstrate nearly as large a change.

It would be really interesting for folks that have recent 8 core i9 Mac's to check for this as well. I'm wondering if we are seeing
a) improper compiler options being used that don't scale to Xeons
b) compiler bugs
c) use of programming techniques that don't scale well on Xeons

While I understand TNM's argument that Diva only gets 6 instances because the CPU is overloaded and can't spin off another thread for instance 7, that really sidesteps the root cause of why 6 instances max out 7 threads (700% total CPU thread use) when 1 instance only uses 25-30% of a single thread. It is this non-linearity that is the root cause. The decrease in the clock speed of the Xeons just means this bites you in the ass even sooner.

The question I have is whether this also exists on an i9 and the increase in instances is just due to a bit more overhead so additional threads can be spun off, or is the scaling fundamentally different? I'm guessing the later since an extra 10-15% of clock speed should only get you another 1 or 2 instances - not 10.

And BTW, I see this on Cubase as well as Logic on Catalina, so I refuse to believe it is related to the DAW.
I have the 16" MBP with the i9 9980HK, and I tested in Logic with Diva, Massive X, Omnisphere, and Absynth to see if I get similar results to you. Buffer size of 1024, and I A/B'd between arming an instrument track and selecting an audio track as I went to see the differences. For the results I'm about to discuss, keep in mind that in every case, Logic always left one thread completely open. Also, I was using a MIDI chord progression that had 16 voices the whole way through.

Omnisphere did the most logical CPU management as I added instances. It very systematically loaded up one thread at a time proportionally, showing roughly the same % on each one down the line until it used them all, then started utilizing more headroom on each thread down the line until I started maxing it out around 28 instances. I also specifically chose a patch that used a lot of headroom on a thread to start with (which was hard to do, most are very efficient).

Diva exhibited the most erratic behavior, and it even seemed patch dependent. Some patches utilized each thread somewhat logically. Some wouldn't even load up more than 7 threads!! It would just hog more headroom on those 7 until they started overloading with just 6 instances. Other patches allowed me to have more instances, because they used more threads before overloading individual ones.

This is the best evidence I've seen yet that Diva is just poorly optimized. I had 8 more threads open and that particular patch wouldn't use them. Other patches exhibited similar behavior, but seemingly without any rhyme or reason. Some would load up all the available threads, some would only do a handful. And yes, it seemed to manage the CPU much more poorly as I added instances, very disproportionate.

Massive X and Absynth behaved mostly the same. They're a little less proportionate than Omnisphere, but did pretty well. I didn't see the more dramatic results you're reporting with Massive X, so it's possible there are some differences with how it's handling my i9 vs. your Xenon, as you suspect.

But with the lower single core performance in the Xenon, a synth like Diva will overload much more quickly than on the i9, since it seems unable to utilize more threads with certain patches. The patch that performed the worst for me maxed out at 6 instances, using only 7 threads. Which means on your new Mac Pro, you'd probably max out at just 4 or 5 instances!! Even with all those extra cores.

Sorry, TNM, but Diva is just poorly optimized . I would even go so far as to say it's a matter of laziness on the part of U-He; not accounting for higher core count CPUs as time has gone on. Spectrasonics does a way better job. The good news is your CPUs are far less sh*tty than it appears.

EDIT: TNM, your points about overclocking an i9 and maintaining a constant base clock in a machine that would allow you to are valid. It would get you far more instances of Diva before overloading. But again, that's a backwards solution. You're altering your CPU out of deference to Diva, instead of Diva being programmed to take better advantage of the threads you already have!

Last edited by MattyJoe; 31st December 2019 at 08:39 PM..
Old 31st December 2019
  #5247
TNM
Lives for gear
I hear you Matty but I still believe that because virtual cores do not provide the same power as real cores, and DAW's do their best to basically divide all the threads into an "even management" (plus the fact that I almost got one on every thread in bootcamp) actually gives a bit of weight to my thinking.. In OSX the fluctuating clock speed doesn't help, 1, but 2, it's not just a case that if one core can take an instance, then all cores can, because when any plugin I have tested that hammers a core, the previous core gets loaded more when you add the next instance on a new core...and I believe this is due to hyper threading. I am going to disable hyper threading with CPU setter and see if it proves my theory.. If suddenly I can get 8 on the iMac pro instead of 6, then, well it does. Diva is simply right at the cusp of overloading the xeon core from the very first instance and as you add more and more instances, the tinsy tiny bit of headroom left over on each previous core gets eaten away, therefore overload.

On the flip side to u-he optimising it more, why can't we say to apple, "stop this ridiculous speed fluctuation when cores ARE under load"? I have literally seen on both my main macs that the speed will drop 400 to 500 mhz on a whim, when thermals are in check and there is enough cpu loading that they should *absolutely* be at their maximum turbo.. Sample based synths are just not as hungry as component modelling like this.. I will admit, I was reading a topic where people were asking for AVX optimisation and u-he said there was no benefit for Diva.. I am not a programmer.. you said his coding is poorly optimised (and truth be told, his two flagship synths are two of the hungriest plugins in existence), he says it's not. Since I am not a programmer, how do I know what's correct? How do I know if it is possible for him to achieve such a level of quality more efficiently?

Today I was playing around with a very highly rated analog sample library.. and it just sounded so "lifeless" compared to something "organic" and "alive" like Diva. Sure, it used 1/20th the CPU, but samples simply can not properly capture the feel of analog. I have quite a few real analog synths and Diva sounds authentic.. Not many synths do. OK, the legend does, and it basically uses almost no cpu, it's ridiculous (like 12x the diva instances on the same computer), but it's limited to 8 voices.. Maybe U-HE could do what Synapse did with AVX and do an 8 voice high performance mode? Again, I don't know. Anyway, I am not going to buy it nor use it for anything other than DSP testing, as it's just too heavy for a real world application for me, and I have my analog hardware and the Legend plugin which I adore.. I have options...

I will say, there are many plugins out there that say they are VA's and they sound digital.. Like that ZONE synth everyone is raving about.. I didn't hear one sound that I couldn't do with plugins ten years ago as far as quality is concerned, and it sounded entirely digital (not that that's bad, but it claimed to be an analog style sound), yet it uses tons more cpu than those older plugins do.. So yeah, to me, that's poorly optimised for the sound you get vs the cpu required..But Diva's quality surely has to come at at least some sort of cpu cost... could it be better? Again, i'm no programmer. But i DO know it sounds like the real thing with the same kind of warm (ugh that word) gut punch. If you know something about programming, why not alert him to this topic and ask him why it can not be optimised to use less cpu? As I have said, I am banned from KVR for years now, so someone else would have to do it. I could create a second account but I don't do that stuff.. when I am not wanted somewhere, i make sure not to grovel, as that's even more pathetic than being banned. I don't like U-HE as a person at *all*, I think he's unbelievably vain and can not handle criticism of any kind, basically, like a child.. and here I am defending his product, so surely that says how sincere I am about how good it sounds.

PS, I just bought softube modular on the holiday sale, I have wanted it for some time, and I will not be using that either.. I will not use something that hammers 50+ percent of one core for a monophonic bass (for example)..Could that be better optimised too?? Don't know! It's an absolute cpu pig. Lesson learned, heavily demo from now on and don't buckle under the pressure of a sale timeframe.
Old 31st December 2019 | Show parent
  #5248
Lives for gear
 

Quote:
Originally Posted by MattyJoe View Post
.
This is the best evidence I've seen yet that Diva is just poorly optimized. I had 8 more threads open and that particular patch wouldn't use them.
Isn't this just cpu cost of circuit modelling? I'm guessing the cpu behavior will be based on which modules are present in the patch
Old 31st December 2019 | Show parent
  #5249
Lives for gear
 

Quote:
Originally Posted by TNM View Post
... it's not just a case that if one core can take an instance, then all cores can, because when any plugin I have tested that hammers a core, the previous core gets loaded more when you add the next instance on a new core...and I believe this is due to hyper threading.
This is not the case with every plugin or VI. Omnisphere added no appreciable additional load on its existing threads until all 15 were used. I was pretty impressed at how logical it behaved. The other threads only started going up once all available threads were used.

Quote:
Originally Posted by TNM View Post
Diva is simply right at the cusp of overloading the xeon core from the very first instance and as you add more and more instances, the tinsy tiny bit of headroom left over on each previous core gets eaten away, therefore overload.
This depends on the patch. Some start at around 50% utilization on one thread, some start at 75%+. Take a look at Logic's CPU thread monitor, and you'll see what I mean. Patches vary wildly. The Xenon would also fare a lot better than you're making it seem. Some patches will only get you 4 to 5 instances, this is true, but many will get you more.

But again, this is Diva's fault, and I'm standing by this. Many other synths do a much better job

If you had the 28 core Mac Pro, it looks like you should be getting 80+ instances of Omnisphere before you get overloads, based on how it was utilizing my i9. (90+ if you're at 1024 buffer).

Totally understand your points about the sound of Diva. Since it's trying to deliver faithful analog-style synthesis, it doesn't hurt to think of it like an analog synth, where you might get 4-10 instances max, and then start printing them down. It still works, you just can't load up as many of them simultaneously. If I was using Diva, that's how I'd think of it, instead of stressing over not having the ability to run 20+ instances.

Quote:
Originally Posted by mat1 View Post
Isn't this just cpu cost of circuit modelling? I'm guessing the cpu behavior will be based on which modules are present in the patch
This is probably true. I'll confess I don't know a whole lot about the specifics of what Diva is doing vs. other synths that try to deliver analog-emulation. Happy to learn more from people here.

But even on that one patch that was using a max of 7 threads, since I had another 8 open, why couldn't it utilize another 7 at least to get me a few more instances? It simply would not use more than 7, and then overloaded those threads at 6 instances. Every other synth fills up all the threads with healthy load.
Old 31st December 2019 | Show parent
  #5250
Lives for gear
 

Quote:
Originally Posted by MattyJoe View Post

This is probably true. I'll confess I don't know a whole lot about the specifics of what Diva is doing vs. other synths that try to deliver analog-emulation. Happy to learn more from people here.

But even on that one patch that was using a max of 7 threads, since I had another 8 open, why couldn't it utilize another 7 at least to get me a few more instances? It simply would not use more than 7, and then overloaded those threads at 6 instances. Every other synth fills up all the threads with healthy load.
Were the other instances the same patch? Maybe change the preset.

I think these kind of heavy modeled plugs are the future so it's something we're going to have to keep any eye on.
📝 Reply

Similar Threads

Thread / Thread Starter Replies / Views Last Post
replies: 4594 views: 587162
Avatar for Arseny
Arseny 3 weeks ago
replies: 97 views: 27989
Avatar for mowmow
mowmow 15th September 2010
replies: 835 views: 170859
Avatar for jlgrimes11
jlgrimes11 30th August 2020
replies: 56 views: 8824
Avatar for Aiyn Zahav
Aiyn Zahav 10th August 2012
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
🖨️ Show Printable Version
✉️ Email this Page
🔍 Search thread
🎙️ View mentioned gear
Forum Jump
Forum Jump