Skip to content
Omnikalc

DDR memory bandwidth calculator

Enter the data rate in MT/s, the bus width, the number of channels, the CAS latency and the efficiency to get the peak and sustained bandwidth, the module rating and the true latency in nanoseconds.

Updated on
Your inputs
MT/s

Value out of range

Value out of range

%

Value out of range

cycles

Value out of range

GB

Value out of range

Results

Peak bandwidth (all channels)

51.20 GB/s

Peak bandwidth per channel
25.60 GB/s
Sustained bandwidth
40.96 GB/s
Module rating (as named PC-xxxx)
25,600 MB/s
True CAS latency
10.00 ns
Time to read the data
0.391 s

The DDR memory bandwidth calculator gives the peak and sustained bandwidth of a memory system, the module rating and the true latency in nanoseconds, from the data rate, the bus width, the channels, the CAS latency and the efficiency.

How the calculation works

  • Enter the data rate in MT/s and pick the bus width.
  • Enter the number of channels, the efficiency and the CAS latency.
  • Enter the amount of data to read.
  • The calculator gives the peak bandwidth, the bandwidth per channel, the sustained bandwidth, the module rating, the true latency and the read time.

Example

DDR4-3200 transfers 3,200 million times a second on a 64-bit bus, which is 3,200 × 8 = 25.60 GB/s per channel, and 25,600 MB/s as named in PC4-25600. With 2 channels the peak is 51.20 GB/s, and 40.96 GB/s sustained at 80 % efficiency. At CL16, the true latency is 16 × 2000 / 3200 = 10.00 ns. Reading 16 GB takes 0.391 s.

Why it matters

The calculator helps computer builders, gamers, engineers and students who compare memory kits or estimate whether a program is limited by memory bandwidth.

Good to know

Bandwidth matters most for integrated graphics, scientific computing and the inference of large language models, where the weights must be read from memory at each token: the maximum token rate is roughly the sustained bandwidth divided by the model size. Latency matters more for programs that chase pointers or access memory irregularly. Dual-channel setups double the bandwidth of single-channel, so two modules are better than one of twice the size. The speeds above the processor’s rated limit may need manual tuning. DDR5 splits each module into two 32-bit subchannels, but the module still has 64 data bits, so the formula is unchanged. Graphics memory uses much wider buses, such as 256 bits or more.

The average memory access time calculator weights hits and misses, the roofline model calculator relates bandwidth to compute limits, the LLM memory calculator sizes models in memory and the frames per second calculator converts frame rates to frame times.

Frequently asked questions

How is the peak bandwidth computed?

DDR memory transfers data twice per clock cycle, which gives the data rate in megatransfers per second. Each transfer moves as many bits as the bus is wide, 64 for a standard DIMM. So the bandwidth per channel is MT/s × width / 8, which gives 25.6 GB/s for DDR4-3200, and channels add up.

What is the true latency?

The CAS latency counts clock cycles, and the clock runs at half the data rate. The real delay is CL × 2000 / MT/s nanoseconds: 10 ns for DDR4-3200 CL16 and 8.3 ns for DDR5-6000 CL30. Faster memory has a larger CL number but often a similar delay in nanoseconds.

Why is the sustained figure lower?

The peak assumes data moves on every cycle. Real access patterns pay for opening rows, refresh cycles, switching between reads and writes and bus turnaround, so 70 to 90 % of the peak is typical in a streaming test, and random access gets much less.

Was this calculator helpful?

Thanks for your feedback!

Results are estimates provided for information purposes. They do not replace professional advice.