Some say the D3 is 2 stops better,
[a href=\"index.php?act=findpost&pid=167684\"][{POST_SNAPBACK}][/a]
Those are probably the guys who shoot in jpeg mode, Bernard . I think the clue to the D3's low noise performance at high ISO is found in the first field report of that camera on LL by James Russell who was very impressed with the D3 but very oddly used it in only in jpeg mode.
I took my 5D to the Canon Service Centre in Bangkok for repair a couple of weeks ago. I collected it today. It so happens that the main Nikon agent in Bangkok is just a few minutes walk from the Canon Service Centre.
Apparently the D3 is so popular they didn't have one in stock, except one for demo purposes. I told them that was fine. I just wanted to compare high ISO performance with my 5D.
I arrived a bit late in the afternoon, about an hour before closing. The place was crowded and I didn't feel like setting up everything with tripod and being completely meticulous, so I just took a few dozen hand-held shots with each camera of the same dark corner in the showroom using the same focal length of 50mm but at a variety of f stops and shutter speeds ranging from f5.6 at 1/8th to f11 at 1/100th.
I autobracketed each 5D shot +/- 1/3rd stop and each D3 shot +/- 1/3rd & 2/3rds of a stop. The D3 can take a series of 5 autobracketed shots. The idea was to make sure I'd have a number of equal exposures from each camera, at ISO 3200 and above, that I could match and compare, which I'm in the process of doing.
The results are pretty much as I expected so far. With RAW images converted with the latest ACR, I see only a marginal improvement in noise in the D3 images, from ISO 3200 to 32,000. Passing both images through Noise Ninja at the same default settings narrows the gap further to pixel-peeping proportions.
Here's an example of the 5D at ISO 3200 (underexposed) compared with the D3 at ISO 12,800, both shots at 25th sec and f11. I show the ACR window for each shot then 100% crops that have been filtered using Noise Ninja.
[attachment=4709:attachment] [attachment=4710:attachment] [attachment=4711:attachment]
Here's another pair of 100% crops with a more agressive noise reduction applied equally to both images. I think the gap has now narrowed to virtual invisibility. I might buy that Noise Ninja program .
[attachment=4712:attachment]
Here's the highest ISO I could match from the shots I took. F11, 1/50th, IS or VR used with both lenses. The 5D shot is 3 stops underexposed at ISO 3200 (really ISO 4000). That works out at ISO 32,000. The D3 was set at ISO 25,600 with a -0.67 EV adjustment in exposure bracketing. Both sensors have received the same amount of light at false ISO settings.
After passing both image through Noise Ninja with the same settings, this is the result.
[attachment=4713:attachment]
Unfortunately, I botched the shots slightly. I was too close with the 5D shot and have had to enlarge the D3 image on screen by 108% instead of 100% to get the image the same size. So, if you think both images look about equal, you can claim the D3 has the edge because it was taken from a very slightly greater distance.