My current guitar signal path generates a low-level noise signal of between -80 and -90 dbFS, where 0 dbFS is the non-clipping limit of the DAW main output bus.  This is with the guitar volume turned all the way down.

I'm going to investigate the source of this noise in greater detail to find out where it's coming from with a view to eradicating/improving it, there are one or two likely candidates.

How would you expect a guitar amp/cab sim to react to this level of signal?  Granted there are a great many factors which this depends on, but assume that the sim is running a high gain amp with all controls set to 50% ('default', '12 o'clock').

The particular sim in question is reacting to this level of noise by producing a noise output of -36 dbFS.  This is a gain increase of approx. 50db and seems disproportionately large compared with other amp/cab sims that I use.

The sim is perfectly quiet with a 0 level input signal, but I'm not sure how to judge how it is reacting to very low level noise.  The reaction seems excessive to me, but I don't have sufficient knowledge to know whether this is perfectly reasonable, or a symptom of poorly implemented software.  What do you think?
 
Also, would you say my signal path is excessively noisy and needs to be improved?
 
TIA :-)

A cab sim is often essentially the same as EQ.
 
A high gain amp severely distorts the signal by adding huge amounts of gain and clipping the peaks.

True enough, but how should it react to very, very small values/numbers?  I'm thinking along the lines of an anti-denormalisation problem perhaps(?)
 
I did some more experimentation last night.  My usual input signal path for guitar is as follows:
 
passive humbuckers->good q guitar lead->PSA-1->balanced out->mixer in->alt out (balanced)->soundcard
 
The quiescent noise level averages out to about -89.5dbFS, about 38% of the signal is at [-inf] and the amp sim reacts to this by producing noise at a level of about -36dbFS.  i.e. the noise floor is being boosted by a good 50db.
 
If I bypass the mixer like this:
 
passive humbuckers->good q guitar lead->PSA-1->balanced out->soundcard
 
...then the quiescent noise level averages out to about -90.3dbFS and about 56% of the signal is at [-inf].  The amp sim reacts to this by producing noise at a level of about -42dbFS.  A boost of about 48db.
 
Note: averages taken are of the non-[-inf] samples
 
The reaction of this particular amp sim is much noisier compared to others that I use.  I just don't what's 'normal'.

That does seem on the excessive side, but several things can contribute esp. with high gain scenarios.

Does the amp sim have an input noise gate? If set to -60 dB or so is it effective at all?

mettelus
That does seem on the excessive side, but several things can contribute esp. with high gain scenarios.

Does the amp sim have an input noise gate? If set to -60 dB or so is it effective at all?

It doesn't, but I have tried inserting a gate plugin just before it.  However, when I start playing and the gate opens the noise floor seems much more noticeable than other plugins e.g. Guitar Rig 5 and TH2.  It's not that big a deal for me as I actually bought the plugin for its cabinet emulations (which is what it's best known for).  The current release is the first one to also incorporate an amp sim.
 
Another thing is that there's a 10db signal level drop between completely bypassing the plugin, and not bypassing it but disabling all of its sections.

Being in its infancy may be the bigger issue unfortunately. Changes like you are seeing are a bit to obvious to ignore (and the lack of a noise gate seems to be an oversight).

Did you happen to try unplugging the guitar input and see what it was doing? If truly giving it nothing, is it giving you something back?

If the input signal is truly nothing, then the output is nothing (i.e. it's not generating noise out of thin air).
 
I'm visiting a friend who knows a lot about such things this weekend (and who also still has my Eventide H9!), so hopefully I'll find out more then :-)

ston
True enough, but how should it react to very, very small values/numbers?  I'm thinking along the lines of an anti-denormalisation problem perhaps(?)
 

Very very small values is a relative term.
 
Here's what I'd try:
 
Feed it a low level pure sine wave with no background noise and see what you get.