Lower buffer settings mean lower latency (for real-time throughput), and faster updates for parameters As a result, a higher demand on the CPU is incurred since it must respond faster and more often. With RTAS plug-ins running on disk tracks, the audio is "pre-buffered" and processed ahead of time, so the throughput latency doesn't really matter. With the HTDM and DirectConnect plug-ins case, they aren't processing audio "ahead of time" (similar to regular TDM plug-ins), so the StreamManager buffer size does affect how much the audio gets delayed when running thru them. In this case, I prefer using the 128 StreamManager for running just 1 or 2 HTDM or DirectConnect tracks, but using the "normal" 512 buffer if you are using more and placing more demand on the CPU. If you have read Abase ID# 25553 you will remember the distinction between RTAS plug-ins and StreamManager (HTDM/DirectConnect) being 2 separate "engines." As a result, they can conflict with each other if both are trying to use maximum CPU resources, and especially if they are using different buffer settings. For example, examine the case of the RTAS HW buffer setting is set for 1024, but you are using the 128 StreamManager. Every 1024 samples, the RTAS engine must process its chain of plug-ins, and if the CPU usage limit is set for 75%, that means it can get up to 1024*0.75 = 768 samples worth of time. But, the StreamManager needs to get some CPU time every 128 samples! Since there is not an easy way on Mac OS 9 for us to "pre-empt" the RTAS engine in the middle of its job, that means the StreamManager can't keep up, and you have dropouts/errors with your HTDM/DirectConnect plug-ins. The reverse situation could also happen of course. This is why we recommend not mixing RTAS plug-ins with HTDM/DirectConnect.I know, this is not really a good design (this is an example of the downside of nicely object-oriented "modular" design where components don't know much about each other. At the time we built the StreamManager/DirectConnect, we didn't have RTAS plug-ins in a TDM system, so there wasn't an issue. We're looking at ways to make this "nicer" in future releases.Of course, mileage varies with each system (and varies widely depending on the algorithms being used). With the 128 Stream Manager, be on the lookout for possible dropouts/errors when using too many HTDM/DirectConnect clients.