New, Updated SonarTest

This one is more applicable to Sonar 3.x. It uses the track EQ and full Lexicon Reverb, and so requires Producer Edition to run. In an effort to isolate the CPU, front side bus, and memory, the file consists of purely input-monitored tracks with various plugins/softsynths. The layout represents one possible mixing/tracking scenario.

Here it is:

Sonar3Test

The file contains:

19 audio tracks
3 DXi audio tracks
3 MIDI tracks (routed to the DXis)
5 Busses (3 setup as auxes)

19 Track EQs (4 bands enabled for each)
6 Sonitus Gates
6 Sonitus Compressors
1 Sonitus Delay
1 Sonitus Multiband
1 Sonitus Reverb
2 Lexicon Pantheon Reverbs

2 Dreamstation instances (each setup for 16-note polyphony)
1 Cyclone instance

Testing is simple...

Open Sonar... Set default sample rate to 44.1KHz (important to do this). Set latency to, say, 46.4ms (For consistency's sake, try to use ASIO drivers if you have them available... otherwise, WDM is okay)... and then open the test file. Make sure the audio engine is active, and note the CPU meter reading. Do the same at 23.2ms, 11.6ms, 5.8ms, 2.9ms, and 1.5ms. Keep in mind that on many systems, the audio engine may refuse to engage at the lower latencies. Make a note of that as well.

If you have a Dual-CPU system... or one with Hyperthreading... or both... post results with Sonar's Multiprocessor Engine both on and off.

Along with your results, post detailed system specs, and I'll try to put together some detailed charts mapping out the performance of various CPUs at various latencies.

Thanks ahead for all of your help with this!



Now... this is a nice bonus to having a web-based forum. I can update this top post at will in order to keep updated results readily available to all. Here's what we have so far (this will be updated as new data comes in):

CPU Chart
CPUs are listed in order of low-latency performance.
Blue Lines represent Intel systems.
Green Lines represent older-generation single-CPU AMD systems.
Red/Pink Lines represent new-generation single-CPU AMD systems.
Yellow Lines represent multi-CPU systems.
Dotted lines represent multi-CPU systems with Sonar's MP capabilities disabled.



-Scott

SONY VAIO PCG-NVR23 laptop, Athlon 1.6 GHz, 512 RAM, 80 GB HD, Win XP.

2 buffers in playback queue

50 ms. 55-56%
40 ms. 57-59%
30 ms. 57-61%
20 ms. 57-63%
10 ms. 65-78%

I also get about -34db of noise on the main outs, mostly on the right channel.

It doesn't seem to make any difference, but is it deliberate that
the Master bus is muted? Anyway...

HARDWARE:

AMD Opteron 146 (2.0GHz)
ASUS SK8N nForce 3 mobo
2x512MB PC3200 Corsair registered ECC DDR SDRAM @400MHz
Matrox P750 AGP graphics card
RME Hammerfall DSP 9632 audio card
Zalman HSF and PSU
(Disk, CDROM, NIC not used in benchmark)

SOFTWARE:

Windows 2000 SP4
Sonar 3.1
RME ASIO drivers v2.62

RESULTS:

1.5ms ....... 64-72%
2.9ms ....... 44-47%
5.8ms ....... 36% +/- 1%
11.6ms ..... 31-32%
23.2ms ..... 28%
46.5ms ..... 26%
92.9ms ..... 25%
185.8ms ... 25%

I'm not an overclocker, and the Opteron isn't set up to play
games with the clock multipliers, but just for grins, I overclocked
the whole CPU by 5%, for a 2.1GHz processor clock and 210MHz
(x2) FSB. What's interesting is that there was actually *less*
variation in the CPU display at short latencies than at the standard
operating frequency. In some cases, the speedup came out to more
than 5%, presumably because the measurement isn't really exact.
The results *seem* to show that the Opteron at 2.1GHz is very
slightly faster than the Athlon FX 51 at 2.2GHz at low latencies.
That could be due to measurement error, or it could be a real
phenomenon of speeding up the FSB as well as the CPU.

1.5ms ....... 61-63%
2.9ms ....... 43-44%
5.8ms ....... 34%
11.6ms ..... 30%
23.2ms ..... 27%
46.5ms ..... 25%
92.9ms ..... 24%
185.8ms ... 24%

Hi Scott

Intel 2.4gig O/C to 3.0gig @ 250fsb
Asus 865PE Chipset DDR
**NO** H/T
Memory @ 1:1 @ 3/4/4/8
2 x 512 DDR500
WinXP SP1
Sonar 3.1 **NO** MP

Asio 46ms %25 <-> %26
Asio 23ms %27 <-> %28
Asio 11.6ms %32 <-> %33
Asio 5.8ms %40 <-> %42
Asio 2.9ms %58 <-> %60
Asio 1.5ms no cpu reading

Thx,
timboe
PS: My H/T and MP is turned off as no matter what I try, as Kontakt is not thread-safe, it is simply too erractic to run it in multiple instances - with MP and H/T off, all seems very happy.

Ron Kuper mentioned previously how the meter reading would vary betwen ASIO and WDM, although the actual CPU usage is the same. So you'll need to compensate for one or the other before graphing.

Also the choice of audio card/driver makes a noticable difference. If you get enough data perhaps you could remove this variable (i.e. only include M-Audio Deltas, Echo Mias, or RME Hammerfalls, or...).

Good points. This is why I'm going to try to use ASIO results for the most part. They may also be more consistent than WDM overall... since the buffer size is set at the driver. With WDM... a latency of 11.6ms might mean the driver has either a 64, 128, 256, or 512 sample buffer. ASIO keeps things a bit more honest.

-S

I just did a 5% overclocking of my Opteron (brute force,
FSB speedup) and edited my earlier Opteron system results
posting to include the overclocked numbers. The system seems
happy and is running just as cool, but I'm gonna back it
down to spec, 'cos I'm that kind of guy...

For those of you with MP systems, do you mind comparing results with and without MP enabled in SONAR? Thanks.

EDIT: Ok I got it. ;) Nevermind.

I would like to try the test. Is the file down? Can't seem to download it.

cAPS

My desktop machine
WDM mode only available

50ms 37%
40ms 39%
30ms 40%
20ms 44%
10ms 50->55%

This is
2.4ghz P4
Soundmax Integrated Audio
Intel GEBV2 (MOBO)

Andy C

Scott,
Here's my system (as best as I can describe it):
Dell Dimension 8300
Pentium 4 HT 2.6ghz
512mgb DDR RAM
2 Delta 66's (WDM Mode)
Midiman 4X4 usb midi interface
2 Western Digital HD's at 7200rpm programs-120gig, wavedata-80gig
Windows XP Pro SP1
Sonar 3.1

Here are my results:
44.6 = 31%
23.2 = 33%
11.6 = 38%
5.8 = 49%
4.4 = 58%
2.9 = 74%
1.5 = chokes

Thanks!

Intel P4 3.0 HT (3.0GHz)
ASUS P4C800 Deluxe nForce3 Motherboard
1GB PC3200 CAS2.0 Registered DDR (2x512MB)
Maxtor 160GB 7200rpm SATA drive
Sony DW-U14A DVDR
Antex Truepower 380W PS
UAD-1 (2)
WindowsXP Professional SP1
RME Hammerfall Dsp Digiface

At first I was surprised... I wondered how on Earth could I be beating the 64 bit Athlons? I then realized that my default sampling rate is 88.2k:

46.4ms........25%
23.2ms........26-27%
11.6ms........28-29%
5.8ms..........31-33%
2.9ms..........40-43%
1.5ms..........56-63%

So at 44.1 I should be more in line with your graph:

46.4ms........26-27%
23.2ms........28-29%
11.6ms........31-33%
5.8ms..........41-42%
2.9ms..........57-62%
1.5ms..........CPU WARNING ;)

So.. I would suggest editing the instructions to include the test sample rate at 44.1 (or multiple rates I guess).

cAPS

Sometimes ya gotta laugh!

It would be better to exclude OC-ed CPUs or busses, not every stock CPU can be overclocked to the same amount.

Much more importantly, can somebody post with a really slow PC, my DAW is currently sticking out a mile as the slowest one in existence.

My Dual Proc DAW

SMP

46.4 msec- 24%
23.2 msec- 26%
11.6 msec- 33%
5.8 Msec- 47%
2.9 msec- 71%
1.5 Msec- Would not Engage

Single Proc mode

46.4 msec- 37%
23.2 msec- 42%
11.6 msec- 50%
5.8 Msec- 67%
2.9 msec- Would not Engage
1.5 Msec- Would not Engage


This is a dual AMD 2600+
Tyan Tiger 2466 MOBO
Delta 44 Soundacard, ASIO

The results *seem* to show that the Opteron at 2.1GHz is very
slightly faster than the Athlon FX 51 at 2.2GHz at low latencies.

I'm pretty sure this is because you are using lower-latency memory than I am. That Corsair RAM has much more agressive timings than my Kingston memory. Otherwise... an AthlonFX51 -is- an Opteron 148.

-S