u-he

VitaminD · Post by **VitaminD** » Fri Jan 20, 2012 2:48 am

There has been some discussion on wondering what performance one can expect out of DIVA with different computer processors.

I (Not officially sanctioned by U-he) propose using a REAPER playback file (since anyone can download and trial REAPER) that I have created: as_Diva_CPU_Test_1.RPP Google Docs file

There are 3 patches individually played back, each more intensive than the next.

I wish there was a way to automate this to export a text file with results.. but I'm not quite that smart. So I load the Performance Meter in REAPER and right click on it to enable RT (realtime) CPU status. Then I eyeball it for the highest % per piece played. Yes, I realize this isn't exactly scientific. But at least it is a rough estimate no?

Please let me know if you have any trouble with it. I saved it in REAPER 32-bit v4.13 trial client.

Example System:

AMD Athlon64 x2 5600+ (2.8 GHZ stock), Win7 64bit, REAPER 32bit client 4.13, DIVA Rev 517

RT CPU Percentage:
1: 59.8
2: 134.6 (scratchy stutter)
3: 182.6 (slower scratchy stutter)

Intel i5 2500 (3.3 GHZ stock), Win7 64bit, REAPER 32bit client 4.13, DIVA Rev 517

RT CPU Percentage:
1: 40.3
2: 61.3
3: 74.9

ObsoleteAcc99 · Post by **ObsoleteAcc99** » Fri Jan 20, 2012 7:51 am

[DELETED]

VitaminD · Post by **VitaminD** » Fri Jan 20, 2012 2:18 pm

ttoz wrote:i really want a standardized test we can use our own host for Vit.

How do you propose this happens?

We need the least amount of effort by the tester, because the tester is capable of flaws (human). I really wish I could have automated this one yet even more!

I guess I could export the MIDI file, and we could just agree to a few sounds in a certain quality mode. Then that is putting the burden on each person to ensure they are selecting the correct patches with the correct quality mode, that nothing else in the song is playing back or active at the same time, etc.

Which is why I chose a REAPER file... since it's available to both Windows and Mac users, and is free to trial.. so no one is left out.

Plus each host may handle DIVA a bit differently. Which would make this a test of each host. I, and I thought the others??, wanted a test of DIVA on each CPU. So I think one host used by all would make for a more accurate benchmark.

If all goes well

I'll be benching an i5 later tonight or tomorrow in the same manner.

ObsoleteAcc99 · Post by **ObsoleteAcc99** » Fri Jan 20, 2012 3:46 pm

[DELETED]

xh3rv · Post by **xh3rv** » Fri Jan 20, 2012 4:29 pm

I would guess the hosts shouldn't be making such a difference, not much overhead in terms of resources relative to Diva. (I'm trying to think if I've ever heard of a properly configured host failing to run a single not-sampling VA well).

If there were any issues with just using the INIT-ALPHA as in the previous benchmark thread, it would be around the fact that Diva's filters are sort of multi-pass with a variable number of passes. The number of passes I *think* correlates somewhat to characteristics of the input. I've experienced and seen people mention some quirky deviations (diva-ations?) between patches, and Urs mentioned white noise, feedback, high resonance all affecting the number of passes needed to achieve the desired accuracy. So the numbers may not reflect a fixed number of voices to count on (I noticed some variance just by shifting a few octaves w/ INIT-ALPHA).

Still the relative performance of CPUs should be decent, I know a couple figures I wondered about but most seemed reasonable & matched other metrics of similar hardware.

ObsoleteAcc99 · Post by **ObsoleteAcc99** » Fri Jan 20, 2012 4:55 pm

[DELETED]

xh3rv · Post by **xh3rv** » Fri Jan 20, 2012 6:18 pm

ttoz wrote:The host makes a difference in this case unfortunately. I have confirmed i can play in realtime certain divine or great patches in logic which i can NOT in studio one.

You don't notice these differences as much with single instances of other instruments, but diva has changed the game.

You used to notice it collectively, "overall", which host would handle which project with multiple instruments better.

But with Diva, it could in many cases have the cpu usage of many other instruments combined. Make sense?

Not really

I just don't think the variables line up that way.

H( g ) Host global overhead
H( n ) Host native plugs
H( p ) Host overhead per plug
V( p ) VST overhead per plug

I'd be very interested if H( p ), the resource costs of hosting a plug, really changed a whole heck of a lot between hosts, at least when RAM isn't overly scarce. Likewise for V( p ). But I have to imagine people would sniff that out pretty quickly and a host with poor performance on these metrics would catch a lot of flack for it.

Plugs across many cores might be a little different (I think hosts have to manage that to some extent?), but since Diva isn't multi-core ATM, it's really neither here nor there for testing Diva.

H( g ) and H( n ) certainly can change, and can even change on host configuration - e.g. buffer size, sample rate, etc.

Since the performance relies on all these factors, I would tend to believe similarly configured hosts may have sort of a different offset related to H( g ) and H( n ), which could cause some difference in voice count.

This is testable and I know I tested Live, GarageBand, and Reaper with Diva in Divine and found no real differences; IIRC Reaper could handle one less voice on 'Divine' but I was nearly maxed out (like, 98%) on the CPU meters with those voices in Live and GB, so a small bump in the overall overhead of Reaper could've easily put that last voice just barely out of reach. This doesn't really mean to me that a DAW has a problem, just a little more overhead used.

That's my story and I'm sticking to it ... unless I'm totally wrong

VitaminD · Post by **VitaminD** » Sat Jan 21, 2012 1:44 am

Updated OP with i5 results. Welcome anyone else who wants to try it.

Kriminal · Post by **Kriminal** » Sat Jan 21, 2012 9:43 am

VitaminD wrote:There has been some discussion on wondering what performance one can expect out of DIVA with different computer processors.

im not trying to put a downer on your test, but i have a question or two

why are you doing this test? ppl can just try the demo of Diva on their pc/mac and try in in their owns host(s) they use, and therefore see what the cpu usage is

having evryone DL reaper and try it just to put results up here seems a bit pointless, especially if ppl dont use reaper

its fairly obvious that Diva can be a cpu hog, and therefore fairly obvious you need fast computer to use it

a list of test results isnt going to change that

VitaminD · Post by **VitaminD** » Sat Jan 21, 2012 3:30 pm

Kriminal wrote:
VitaminD wrote:There has been some discussion on wondering what performance one can expect out of DIVA with different computer processors.

im not trying to put a downer on your test, but i have a question or two

why are you doing this test? ppl can just try the demo of Diva on their pc/mac and try in in their owns host(s) they use, and therefore see what the cpu usage is

having evryone DL reaper and try it just to put results up here seems a bit pointless, especially if ppl dont use reaper

its fairly obvious that Diva can be a cpu hog, and therefore fairly obvious you need fast computer to use it

a list of test results isnt going to change that

You are trying to be a downer, but in a nice way. But that is ok.

To answer your question -- For those considering upgrading computers, of course. They obviously can't test something they don't have (new computer).

There was a big discussion on this awhile ago where people wanted to see how DIVA performed on various systems. Everyone talked, no one acted, so I acted. Reaper, again, seems the good generic that everyone can use. I don't use it either but I'm not afraid of downloading and trying it. Neither should you.. maybe you will enjoy it.

Actually, you've just given me an idea to make this work even better!

djsubject · Post by **djsubject** » Sat Jan 21, 2012 6:35 pm

VitaminD wrote:
ttoz wrote:i really want a standardized test we can use our own host for Vit.
How do you propose this happens?

midi files & diva presets?

will test your... Ummm.. test project later & post results!

Win7 x64 Reaper x64 Diva x64 Intel Q9550 4gig Ram

Subz

whet · Post by **whet** » Sun Jan 22, 2012 3:14 am

Mhh reaper doesnt like my E-mu card that much (never noticed, until diva), gives me alot of crackles on the last patch and some spikes on the second, my cpu usage stays under 50% for each core though, and stays most of the time at idle clockspeed. With flstudio i can play the same patches in devine mode with more voices. Iam on an overclocked 2500k.
So ya it realy depends on the system config what works and what not.

Edit: could be the bit bridge in reaper since i ve reaper x86 installed and it doesnt find the x86 .dll of diva somehow and loads the x64.

But not realy up to inverstigate that further tbh. I posted my "INIT ALPHA result" in the other thread.

VitaminD · Post by **VitaminD** » Sun Jan 22, 2012 3:28 am

djsubject wrote:
VitaminD wrote:
ttoz wrote:i really want a standardized test we can use our own host for Vit.
How do you propose this happens?
midi files & diva presets?

hehe.. yes.. essentially that is what a saved project file is.. to remove many of the abilities for the end user to fubar the process.

For instance, who is to say Tester B is ensuring all other sound sources are not playing? The project file accounts for this

What is stopping them from playing short stabs and then claiming they can play 19 instances?

The project file accounts for this

What is to say they even use the patches required? The project file accounts for this

But in all actuality Krim's comments made me realize the test needs to test the most demanding possible sound (just one patch, all options enabled, full cutoff etc) and then add very long notes in succession until the cpu maxes. That would make for a more telling study. I still and will continue to argue that the test needs to occur in a single host and in a project file to minimize the fubar effect of human hands.

I didn't feel like doing anything today though so I didn't get on it. Especially after the wall of static here.

whet · Post by **whet** » Sun Jan 22, 2012 3:52 am

I just realized the last "crackpipe" patch is mono, wondered why i can stack 32 voices with devine mode in 2 instances in flstudio.

Tbh i think its impossible to make a standardized benchmark for Diva. There are to much variables like driver or latency beheviors, this all interacts with the actual cpu usage in the host.

Mhh, in some cases here it crackles hard and the cpu is still at idle speed, just clocks to full speed in between. Not sure if its divas or the hosts fault or something else like the asio driver.

Kriminal · Post by **Kriminal** » Sun Jan 22, 2012 9:15 am

VitaminD wrote: To answer your question -- For those considering upgrading computers, of course. They obviously can't test something they don't have (new computer).

the answer is simple then really, buy the fastest computer you can afford, isnt?

or, put this test on the reaper forum, then post the results here? you'll prob get more takers.

u-he

KVR Audio

DIVA CPU Usage benchmark