White v. Hispanic cognitive gap across 39 studies with 5,696,529 sample size

From Philip L. Roth et al's 2001 article Ethnic Group Differences in Cognitive Ability in Employment and Educational Settings: A Meta-Analysis (PDF), where d stands for a standard deviation (such as roughly 15 points on an IQ test or 100 points on an SAT subtest scored 200 to 800):

In other words, this meta-analysis considered overall 39 studies with a combined sample size (all ethnicities) of 5,696,519 (only age 14 and above). The overall white-Hispanic gap was 0.72 standard deviations or (expressed on an IQ scale) 10.8 IQ points. This is a larger gap than Linda Gottfredson's 1988 estimate of a 7.5 IQ point gap. 

The 95% confidence interval (i.e., excluding the single lowest and highest findings out of 38) range from gaps of 0.60 to 0.88, so there isn't a lot of disparity. (The one outlying 0.40 finding was for occupants of particular jobs, and thus had severe restriction of range issues.)

Language familiarity is of course an issue. For example, East Asians tend to score much better on American tests in math subtests than in verbal subtests. While there is some evidence for this pattern among Hispanics in America, it is not at all as clear as among Asians.

Here some interesting subtests included in the overall meta-analysis:

Obviously, restriction of range is a problem with many of these. For example, the Graduate Record Exam test is taken by people who either have graduated from college or about to graduate from college and are considering an advanced degree. But the overall gap on the GRE is still 0.72 standard deviations.

If you are interested in the meta-methodology:

Literature Review: Data Sources  

Articles on Black-White and Hispanic-White differences on tests of cognitive abilities were gathered from several sources including Psych-Lit of the American Psychological Association, Educational Resources Information Center (known as ERIC), Dissertation Abstracts Interna-tional, and Abstracted Business Information (known as ABI Inform). Reference lists and studies used by several narrative literature reviews and meta-analyses of related concepts were also examined (Dreger & Miller, 1968; Herrnstein & Murray, 1994; Jensen, 1980; Osborne & McGurk, 1982; Schmitt et al., 1996; Toquam, Corpe, & Dunnette, 1989). Attempts were made to overcome the “file drawer problem” by contact- ing test publishers and researchers active in the field. Letters were written to 6 major publishers of cognitive ability tests and 16 prominent researchers working in the area. 

To be sure ...

As such, it does not suggest that there are not high scoring individuals in both groups.