thinQmaster Virtual testing partner Adaptive practice exams that find your weak spots and drill them. Every answer explained. From $69
Enroll

thinQmaster

Find your exam

Five questions free, before you decide.

Step 1 · choose a vendor

In one of our camps? Your practice exams come with the tuition — they are on your study page already.

AI & Machine Learning · Intermediate

FP16

16-bit half-precision floating point, halving memory versus FP32 with a narrower numeric range.

Definition

FP16 accelerates Tensor Core math and reduces memory, but its limited range can cause underflow or overflow, so loss scaling is used in training. It is common for inference.

Exam tip

FP16 needs loss scaling during training because of its narrow range; BF16 usually does not.

Related terms