Theoretical memory bandwidth from speed and channel count — the number that explains single vs dual channel.
Updated 11 Nov 2026 · 6 min read
Quick answer
Bandwidth (GB/s) = transfer rate (MT/s) × 8 bytes × number of channels. A DDR5-6000 dual-channel setup therefore peaks at 96 GB/s; the same kit on one stick is 48 GB/s.
RAM Bandwidth Calculator
Theoretical memory bandwidth from speed and channel count — the number that explains single vs dual channel.
Contents
Where the formula comes from
Every memory channel is 64 bits (8 bytes) wide and completes one transfer per MT/s step. Multiply rate × width × channels and you have the ceiling every real benchmark chases (expect 85–92% of it in practice). This is the maths behind single vs dual channel.
The formula has three parts, and each one is a physical fact about the memory bus. The transfer rate (MT/s) is how many million transfers the bus completes per second. The channel width is 64 bits — 8 bytes — for every standard desktop and laptop memory channel, DDR3 through DDR5 alike. And the channel count is how many of those 64-bit channels the memory controller can use at once: one for a single stick, two for a matched pair in dual channel. Multiply them together and divide by 1,000 to get gigabytes per second:
Bandwidth (GB/s) = MT/s × 8 bytes × channels ÷ 1000
How to use the calculator
Enter two numbers: the transfer rate in MT/s and the number of channels. The transfer rate is the figure printed on the box — 3200, 4800, 6000 and so on — labelled ‘MHz’ by most manufacturers. The channel count is 1 if you have a single stick, 2 if you have a matched pair running in dual channel, and 4 only on workstation platforms such as Threadripper or Xeon W.
Enter the speed the memory actually runs at, not the headline speed. A DDR5-6000 kit with XMP disabled runs at JEDEC 4800 MT/s, and its real bandwidth is 76.8 GB/s in dual channel, not 96 GB/s. Check CPU-Z’s Memory tab or Task Manager → Performance → Memory for the actual speed before calculating. Laptops almost always run at JEDEC speed, since most have no XMP option at all.
Worked examples
Example 1 — DDR5-6000, dual channel. 6000 × 16 ÷ 1000 = 96 GB/s. This is the figure quoted in the quick answer, and the current mainstream DDR5 sweet spot.
Example 2 — the same kit, single channel. 6000 × 8 ÷ 1000 = 48 GB/s. One stick instead of two halves the bandwidth. This is why a single 16GB stick can feel slower than two 8GB sticks of the same kit, and why integrated graphics — which share system memory — suffer most in single channel.
Example 3 — DDR4-3200 dual vs DDR5-4800 dual. DDR4: 3200 × 16 ÷ 1000 = 51.2 GB/s. DDR5: 4800 × 16 ÷ 1000 = 76.8 GB/s. The DDR5 setup moves about 50% more data per second, which is the real generational gain — not latency, which is roughly the same for both.
| Kit | Single-channel | Dual-channel |
|---|---|---|
| DDR4-2666 | 21.3 GB/s | 42.7 GB/s |
| DDR4-3200 | 25.6 GB/s | 51.2 GB/s |
| DDR4-3600 | 28.8 GB/s | 57.6 GB/s |
| DDR5-4800 | 38.4 GB/s | 76.8 GB/s |
| DDR5-5600 | 44.8 GB/s | 89.6 GB/s |
| DDR5-6000 | 48.0 GB/s | 96.0 GB/s |
| DDR5-6400 | 51.2 GB/s | 102.4 GB/s |
What the result means in practice
Bandwidth matters most where large amounts of data move in order: integrated graphics, video editing, 3D rendering, scientific computing and file compression. Integrated GPUs are the clearest case — they have no dedicated video memory, so every texture and frame passes through system RAM, and doubling bandwidth can raise frame rates by 20–40% in some games. That is why the single most effective upgrade for a laptop with integrated graphics is often a second, matching stick rather than a faster one.
Common mistakes
- Counting four sticks as quad channel. Consumer platforms have two channels even with four sticks fitted. Quad channel exists only on workstation platforms (Threadripper, Xeon W).
- Using the headline speed instead of the real speed. A kit running at JEDEC 4800 because XMP is off delivers 76.8 GB/s dual, not 96 GB/s. Check the actual speed first.
- Expecting the full figure in real use. 85–92% of peak in a synthetic stream test is a healthy system. If a benchmark shows far less, something is wrong — usually single channel, a mismatched pair, or the memory running at JEDEC speed.
- Chasing bandwidth when latency is the problem. Random-access workloads such as gaming at high frame rates care more about latency. Use the speed calculator to see both numbers together before spending.
Frequently asked questions
Why do reviews show less than my calculated number?
Real workloads can’t sustain the theoretical peak; timings, refresh and access patterns take their cut. 85–92% of peak in a synthetic stream test is a healthy system.
Does quad channel exist for desktops?
Only on workstation platforms (Threadripper, Xeon W). Consumer dual channel is 2 channels even with 4 sticks fitted.
Will faster RAM speed up my whole computer?
Only if memory bandwidth is the bottleneck. For most desktop tasks it is not — the CPU, storage or graphics card limits speed first. Integrated graphics and data-heavy work are the exceptions where bandwidth gains are real.
My two sticks are different sizes. Am I still in dual channel?
Partly. In flex mode, the matched portion runs in dual channel and the rest in single channel — for example, 8GB + 16GB gives 16GB dual and 8GB single. It is much better than single channel alone.
Does DDR5 have wider channels than DDR4?
Not per channel — both are 64 bits wide. DDR5 splits each stick into two independent 32-bit sub-channels, which improves efficiency, but the total width and the bandwidth formula are the same.