Difference between revisions of "ER-301/CPU Load"

From O|D Wiki
Jump to: navigation, search
(Created page with "= CPU Load= You can query the CPU load at any time from the admin menu. File:admin-maintenance.pngFile:cpu-load.png CPU load will **usually** be the sum of the fixe...")
 
(CPU Load)
Line 12: Line 12:
 
* All random-access sample playing units and delay units with a speed control: L1/L2 cache misses caused by non-sequential access to sample memory will result in increased CPU load (stalling). Layman translation: Randomly jumping around in the buffer at audio rates, or playback at really high speeds (>10x). This forces the ER-301 to repeatedly query the RAM (slow) for necessary data because it is not in the L1/L2 memory cache (fastest).
 
* All random-access sample playing units and delay units with a speed control: L1/L2 cache misses caused by non-sequential access to sample memory will result in increased CPU load (stalling). Layman translation: Randomly jumping around in the buffer at audio rates, or playback at really high speeds (>10x). This forces the ER-301 to repeatedly query the RAM (slow) for necessary data because it is not in the L1/L2 memory cache (fastest).
  
The number of (local) connections does not matter either.
+
The number of (local and global) connections does not matter.

Revision as of 05:07, 30 April 2019

CPU Load

You can query the CPU load at any time from the admin menu.

Admin-maintenance.pngCpu-load.png

CPU load will **usually** be the sum of the fixed costs of all units in use (plus ~3% of overhead). A unit’s cost does not depend on where you insert it except to the extent that mono vs stereo might affect it. In general expect, a stereo unit's cost to be from 50% to 100% more than its mono version.

In general, units are designed to have a fixed CPU cost that does not depend on their inputs. The cases where this is violated (in order of severity) are:

  • Manual Grains: CPU load grows as more grains are overlapping (i.e. polyphony). This happens when you trigger grain production faster and/or increase grain duration.
  • All random-access sample playing units and delay units with a speed control: L1/L2 cache misses caused by non-sequential access to sample memory will result in increased CPU load (stalling). Layman translation: Randomly jumping around in the buffer at audio rates, or playback at really high speeds (>10x). This forces the ER-301 to repeatedly query the RAM (slow) for necessary data because it is not in the L1/L2 memory cache (fastest).

The number of (local and global) connections does not matter.