This study estimates 985,000 deaths.
The report I referred to was authored by the IAEA, the WHO, several UN agencies, and several other international organizations. It estimated the number of deaths due to an increased rate of cancer from exposure to radiation from the Chernobyl accident. The total number of deaths that have already or will occur and be caused by Chernobyl, according to that report, is ~9,000.
The vast majority (all but 50 or so) of those 9,000 deaths would come among the (comparatively large) population that received a low dose of radiation. The report arrives at 9,000 deaths by making an assumption about how increased cancer risk depends on exposure to radiation at very low doses. The assumption is that the dependence is linear: if exposure to 100 units of radiation increases cancer risk by 1%, then - under this linear assumption - 1 unit of radiation exposure increases cancer risk by 0.01%.
That assumption was made because it is regarded as very conservative - meaning that it is believed to
overestimate the risk. Studies of various populations exposed to radiation have not confirmed it. On the contrary, they have shown that below a certain threshold, there appears to be no risk (in fact a few studies have shown that below a certain threshold, exposure to radiation actually decreases risk of cancer, i.e. it's good for you). The reason it's hard to be sure if this is that the change in cancer rates at these low doses (whether positive, negative, or zero) is very small compared to the background (cancer rates in the normal population).
If you assume a relationship of cancer rates to exposure that includes that threshold - so you use a relationship based on data, rather than the conservative and simplistic relation assumed in the IAEA et al report - you find that there are and will be something on the order of 100 deaths as a result of exposure to radiation from Chernobyl, in total (rather than 9,000). If nothing else, that tells you that the total deaths predicted by such studies are almost entirely determined by what they assume about increased cancer rates at very low doses of radiation.
Biologically a threshold is intriguing, but not entirely surprising - the relationship between cancer and exposure to low levels of environmental radiation turns out to be complex and very interesting. Just as a quick example, most organisms get cancer at about the same rates humans do - if you divide by their typical lifetime. In other words a 2-year-old rat is likely to have cancer, just like a 70 year-old human - but a 2-year-old human baby is extremely unlikely to have it. So cumulative with time (linear) exposure of mammal cells to environmental/cosmic ray radiation cannot account for cancer rates (in fact it's completely wrong), even though such an assumption might seem reasonable a priori.
Coming back to your link - 1,000,000 deaths? I don't think so. Based on what I know, that's completely out of the question. It must be based on some wild extrapolation of cancer rates at extremely low levels of radiation. It exceed a conservative estimate by a factor of 100, and a more reasonable estimate by a factor of 10,000.
