It would be nice to get NAEP type data from every ethnic group because of the large representative samples of test prepared people who take these tests. It would be especially nice to get such data from post-apartheid South Africa and the statistics below come close.

According to a report

The outcomes of black learners are poorer compared to their coloured, Indian, and white counterparts. For example, in 2016 the proportion of black learners who passed maths with 60% or more was 9%, By contrast, the proportion of whites passing with more than 60% was 52% (the
proportions for coloured and Indian pupils were 20% and 40% respectively). In maths literacy, the figures are even starker – only 8% of black learners passed with a mark of above 60%, compared 20% of
coloured learners, 44% of Indians, and 73% of white learners. Figures for other subjects show similar trend

If we assume maths literacy is a proxy IQ test and force the distribution of all groups onton normal curves with an SD of 15, we get:

South African white IQ = x

South African Indian IQ = x – 11

South African Coloured IQ = x – 21

South African Black IQ = x – 30

But what is x?

We know that South Africa is about:

And we know that their average score on international math tests is about 73. Thus solving for x:

0.807(x – 30) + 0.088(x – 21) + 0.079x + 0.026(x – 11) = 73

0.807x – 24.21 + 0.088x – 1.848 + 0.079x + 0.026x – 0.286 = 73

x = 73 + 24.21 + 1.848 + 0.286

x = 99

So if the white Math IQ is 99, then the IQs of the other groups can be deduced.

 Ethnicity IQ (based on the percentage of each population scoring 60%+ on maths literacy relative to the white rate) whites 99 Indians 88 Coloureds 78 Blacks 69

Of course given the legacy of apartheid and the huge differences in living standards and education quality between these groups, it would be very stupid to take scores on a math test as reflective of actual intelligence, let alone genetic ability. As of 2015 it looks like the percentage of Blacks, Coloureds, Indians and Whites living in poverty is about 40%, 24%, 2% and 0.5% respectively (crude guesses based on eyeballing the below graph):