With current Canon cameras (up to the 7D at least), reports put the fastest aperture that increases light to the sensor at f/2. If you shoot at f/1.4, the ISO is "invisibly" bumped (it still reports as say ISO 100 but the files are noisier than they should be).
I'm sorry but this is simply not true in any way shape or form.
Just so nobody is harmed by this breathtakingly assured response:
I must affirm that what I write is true. I can mention some useful resources for understanding related phenomena: Expose-to-the-right
and the non-linear response curves of digital camera sensors
; the most relevant information for persuading you is in this thread
and especially here
. In short, you, sir, are completely wrong.
If you would like to test it out, it is simple to see it in action:
Find a recent Canon camera, set ISO manually, and switch to A/v mode. Slap a lens faster than f/2.8 on it (I thought it was f/2; it might actually be wider, I'll have to look at this again). Shoot at f/2.8 and then shoot at the widest setting. You will notice that the image is noisier (and, depending on the lens used, there may be vignetting as well, but this is not the important effect to notice). By the theory, image brightness and noise should remain constant, since we are holding ISO constant, and exposure is simply light intensity * duration. But quite obviously that is not what happens; the image becomes noisier.
In truth, the manufacturer's declared ISO settings are not scientific measurements - the scientific measurements given by DxOMark show that for every camera (regardless of manufacturer) the ISO sensitivities are not linear or even very predictable compared to what they "should" be.
I do not consider mentioning this issue to be a burden to other users, since it's fairly easy to comprehend (I gather that it is due to the narrow angle of current photosites restricting the angles at which light is effectively gathered, but in any case the effect is clear).
Digital SLRs are not film cameras, and the final image brightness and grain is affected by much more than just the user's selected settings.
To the gentle users of the forum in general, I will leave with this final thought - DSLR "best practice" can be confusing to master. The main thing to consider, if you want to have clean files, is to get as much information into the highlight section of your RAW by keeping the image bright without burning out highlights. The second thing, of course, is to use low ISOs when possible. Finally, there is a limit to how high you should crank ISOs in most Canon DSLRs; it is not necessary to get the image looking bright enough on the camera's preview screen, because that is not the final version of the image. Instead, you may brighten the image afterward to retain your settings and a reasonable ISO.
How do you find the point at which brightening in post is better than pumping the ISO (obviously, if you just shoot JPEGs or can't be bothered, don't worry, but your image quality won't be as good as possible)? Look at the sensorgen charts from the DxOMark data, specifically the point at which the "read noise" curve goes flat, no improvement (or even shoots upward). On the 7D, that point is ISO 800 or 1600. On the T1i, it was ISO 800.
How do you brighten the image? Not with DPP. Daniel Browning recommended RawTherapee last year; there is also The GIMP, Lightroom, or Darktable (a Lightroom replacement for Linux boxen).
It is worth mentioning that there are other losses at High ISO: Dynamic range and saturation capacity (I believe saturation capacity is essentially a measure of how even the image is, since it measures the number of pixels that reliably reflect a gray target, instead of appearing in the final image to be some other shade).
The culprit, as Daniel Browning mentions, and also Gregg Siam, is that the photosites of Canon DSLRs don't respond equally well to light coming in from unusual angles. In fact, slight purple and green fringing in some lenses has been blamed on this, as well.