GS-2 delay

Are you talking about AAS' Strum GS-2? Or what? Clueless here, and not even in Seattle.

If you bypass all effects (Press "FX" button in the MIX module), do you still get the latency?  If doing this makes the latency go away, then one of your effects is introducing latency (maybe a linear phase EQ, for example).

Could try the PDC button  near the FX button in the Mix module http://www.cakewalk.com/Documentation?product=SONAR%20X3&language=3&help=Recording.23.html
 

If the FX button makes no difference the PDC button won't either, they are both related to plug-in induced latency, Running with ASIO buffers set to 1024 will create considerable delay when trying to monitor a softsynth while recording. If you can run ASIO at 64 buffers that is your solution. You can probably get away with 128 buffers if 64 is too low for the DAW to run reliably.

On my system, I can record at 128 without trouble, so haven't felt the need to drop to 64. But when mixing, I raise it to 1024 to reduce CPU load and save it for the plugins.

Excepting the creative stuff that has to happen first, the typical process of converting those sound and lyric ideas into a complete song is: record, mix, master. My reference to mixing above was to the second phase of the process. There's a ton of great info on the web on this subject, but I would highly recommend two books by Mike Senior. They're not cheap, but they are definitely worth the money.
 
Recording Secrets for the Small Studio
https://www.amazon.com/Recording-Secrets-Small-Studio-Senior/dp/0415716705/
 
Mixing Secrets for the Small Studio
https://www.amazon.com/Mixing-Secrets-Small-Studio-Senior/dp/0240815807/
 
Long story short, though...
 
Recording is capturing your musical performances on "tape" (digital today, but I sometimes miss my old JVC 4-track reel-to-reel).
 
Mixing is combining each of the individual performances (drums, bass, guitar, keys, synths, vocals, etc) into a cohesive whole. That's when you use most of your heavy CPU plugins (Linear phase EQs and compressors being hogs, for example; OK, they're typically used in the mastering phase, but I'm just making a point).
 
So, when tracking (recording), where you want "instant" feedback of your playing, you set the buffer small (128 or 64). This gives you low latency, at the expense of tying up the CPU processing the small buffer.
 
When mixing, where you don't want glitching in the audio you're listening to, you set the buffer larger (2048 or 1024). This gives the CPU a LOT more time to spend on processing plugin code instead of interrupts.