The memory bandwidth calculator uses the memory interface width (bus), memory clock speed and the memory type in order to calculate the memory bandwidth.
The equation is as follows: Memory Bandwidth = number of times the memory type can send data per clock cycle x memory interface width (in bits) x memory clock (in MHz). This will give you the memory bandwidth in mbps. To view the number of times each memory type can send data per clock cycle see the table below. To get this in GB/s which is how the memory bandwidth is usually displayed as, you need to divide by 8 to get the answer in MB/s and then divide by 1000 to get the answer in GB/s. Remember that there are 8 bits in 1 byte.
Using the effective memory clock
If you use the effective memory clock in the calculation, then you don’t need to multiply by the number of times data can be sent per clock cycle. This is because the effective memory clock already includes this number within itself. The effective memory clock is simply the normal memory clock multiplied by the number of times data can be sent per clock cycle.
Example Calculation - Memory bandwidth Calculator
The Nvidia GeForce RTX 4090 has a memory clock speed of 1313 MHz or 21 Gbps effective. It also has GDDR6x memory and a memory interface width of 384-bit. GDDR6X memory can send 16 times per clock cycle (see table below). So the Memory Bandwidth = 16 x 1313 MHz x 384 = 8067072 mbps. Then divide by 8 gives us 1008384 MB/s. Then divide by 1000 to get 1008.34 GB/s.
How many times data can be sent per clock cycle table
| Memory Type | How many times data can be sent per clock cycle |
|---|---|
| GDDR6X | 16 |
| GDDR6/GDDR5x | 8 |
| GDDR5/LPDDR5 | 4 |
| SDR (or other single data rate memory types) | 1 |
| Everything Else including DDR,DDR2,DDR3,DDR4,HBM,HBM2,HBM3 etc | 2 |