Pulsar Performance Overview

System Performance Notes
top | system | audio | midi | dsp | synth

The Pulsar environment eats system memory like M&M's. Sure, maybe it does all of its processing on those SHARCs, but the Pulsar environment (all the user interface stuff, and system drivers), requires a TON of memory. On a Pentium II 400, with 64 MB memory, the system often slows to a crawl as Windows 95 swaps like mad. Frank Hund, President of Creamware explains that the ultra-flexible modular object-oriented nature of the Pulsar/Scope environment just simply requires more memory than simpler less flexible architectures.

With only 64 MB, simple tasks such as saving a project file or trying to exit Pulsar, can take minutes to accomplish due to Win95 swapping. The good news is that upgrading with more memory improves the situation dramatically. Also, with the version 1.1 Pulsar software release, Creamware has significantly reduced the overall amount of memory required by the software.

For details on individual module memory usage, as well as a comparison of memory usage in version 1.0 vs. 1.1, see Martin Saleteg's memory usage chart.

Audio Latency Notes
top | system | audio | midi | dsp | synth

I performed a series of tests to verify Creamware's "zero-latency" claim. These tests were confined to the non-ASIO world - ASIO is a completely different mess (an infrastructure for passing audio data from harware to music applications and back), and an area in which Creamware makes no such "zero-latency" claims. My tests simply determined how fast Pulsar could take in audio, possibly do something to it, and spit it back out again, in various different I/O situations.

I've explained my testing and results in detail on a separate page, but to summarize the results briefly, I did indeed find that Pulsar's I/O latency was very very small. In situations involving analog conversions, I found the latency to be in the range of 20-80 samples, which at 48k is 0.4 - 1.7 ms. In an all digital bounce, the latency was 9 samples, which at 48k is 0.2ms or 2 microseconds. Even in a multi-channel mix, with effects, EQ, etc, I found the latency to be a mere 13 samples (excluding analog conversion), which is truly remarkable, and exceeds the standards set by expensive digital mixing consoles like the Yamaha 02R.

Please see the Audio Latency page for further details.

MIDI Latency Notes
top | system | audio | midi | dsp | synth

Many people have asked about Pulsar's MIDI responsiveness. So I rigged up some tests to determine precise MIDI latencies for the Pulsar synth modules. I focused on the synths, as it is unclear to me how to test Pulsar's responsiveness to MIDI parameter control or automation.

The results are detailed on a separate page. I found the Pulsar synths' MIDI latencies to be fantastic (typically average less than 2ms with standard deviation of significantly less than 1ms)! If you are running Pulsar software versions pre 1.1, you will definitely have some MIDI latency problems (high latencies and jitter in the Sample Player, and the dreaded voice stealing bug) - so definitely install the latest version of the Pulsar software!

The Pulsar synths' MIDI latencies were in most cases significantly better than my Alesis HR-16 and Waldorf Microwave XT.

Please see the MIDI Latency page for further details.

DSP Performance Notes
top | system | audio | midi | dsp | synth

Regardless of its voracious appetite for system memory, the Pulsar's DSP power really packs a punch, without depending much on the system CPU. Creamware claims that the 4 SHARC DSPs on the Pulsar are equivalent to 2½ Pentium II 300's.

If you're hoping to run several synths, as well as mix hard disk audio with effects, etc, etc, you'll need to become adept at juggling your DSP load, reducing synth polyphony where you can, adding delays to cover up for lost polyphony, etc. It's therefore important to understand how much DSP each Pulsar module requires. With this in mind, I put together the DSP usage charts:
synth | mixer and I/O | effects | modular

I've observed the Pulsar's DSP performance using the DSP Load window, comparing before-and-after DSP usage percentages when loading in modules, connecting wires, etc. Here's what you see in the DSP Load window:

The Sync values are the percentage of each DSP used.

I've noticed that the DSP load statistics are never perfectly consistent - this is probably an artifact of sharing DSP processes over multiple DSPs with some amount of overhead for multi-processing. So please keep in mind that the numbers in the charts are not intended to be exact.

In general, when I speak of DSP percentage, I'm refering to the percentage of a single DSP, not a percentage of the whole system (unless your system only has one DSP!) In other words, if a BlueSynth voice takes 43% of one DSP, and you have four DSP's, then given that 10% per DSP is lost in initial DSP overhead, you have a maximum of 360/43 voices = 8 voices. (360 is the 4 DSPs at 90% each). It appears there is a little bit of additional overhead per voice (or per module), so in fact, in the BlueSynth case, I was only able to reach 7 voice polyphony on 4 DSPs, with nothing else running but the mixer.

As you add more DSPs (adding a second Pulsar gives 4 more DSPs, while a Scope adds 15), your total DSP power obviously increases.

Synth Performance Notes
top | system | audio | midi | dsp | synth

Creamware is careful to point out that their synths are optimized for awesome sound, not for awesome polyphony. This will be obvious as you scan the synth chart.

Some synths require more DSP for their first voice, compared to subsequent increases in polyphony. I have noted these on the synth chart.

The amount of DSP used by a synth generally depends on the preset - especially with FM One, and obviously with the Modular. With the miniscope, and BlueSynth, the variances are trivial (+/- 5%), but with the FM One, the difference between the init preset and the Technoid preset was over 50% of a DSP!

It appears that the maximum polyphony of any synth module (if you have enough DSP to reach the max) is 16 voices. Polyphony is selected on the preset panel for the synth. In the table, I noted the maximum I was able to reach with a single 4 DSP Pulsar card in a simple project containing only the Big Mixer and the synth. Keep in mind that when a synth is maxed out on polyphony, the Pulsar won't have much DSP left over for other things!

It is apparent with the version 1.1 Pulsar software release that Creamware is getting better at optimizing the syths. The new U Know 007 synth sounds better than the previous synths, is more responsive to realtime control, and requires less DSP (therefore giving you higher max polyphony!)