latency calculator
A professional, two-in-one latency tool: compute DRAM true latency (ns) from CAS & data rate, and estimate network one-way/RTT latency by combining propagation, serialization, and processing delays. Built for engineers, performance analysts, overclockers, SREs, and curious learners.
Latency calculator modes
True CAS latency
—
tCK: — ns
Estimated random-row access
—
CL + tRCD + tRP (if provided)
Data Source & Methodology
- JEDEC DDR Standards (e.g., JESD79 series for DDRx timing definitions & tCK). “All calculations strictly follow the formulas and data provided by this source.”
- Optical Fiber Physics: group velocity \( v = c/n \) with typical \( n \approx 1.468 \Rightarrow v \approx 2.04\times 10^8 \,\mathrm{m/s} \).
- Latency Engineering: ITU-T practice for delay budgeting (e.g., conversational guidelines in G.114) and standard serialization formula.
The Formula Explained
DRAM true latency (nanoseconds):
\( t_{\mathrm{CL,ns}} = \dfrac{2000 \cdot \mathrm{CL}}{\mathrm{DataRate\ (MT/s)}} \)
Clock period: \( t_\mathrm{CK} = \dfrac{2000}{\mathrm{DataRate\ (MT/s)}} \) ns
Optional random-row access (cycles): \( \mathrm{CL} + t_\mathrm{RCD} + t_\mathrm{RP} \Rightarrow t_{\mathrm{ns}} = (\mathrm{cycles}) \cdot t_\mathrm{CK} \)
Network propagation delay (one-way):
\( t_\mathrm{prop} = \dfrac{d}{v} = \dfrac{d}{\alpha c} \), with \( d \) meters, \( \alpha \in \{0.68,0.77,0.99\} \)
Serialization: \( t_\mathrm{ser} = \dfrac{\mathrm{packet\ bits}}{\mathrm{bandwidth\ bps}} \)
Processing: \( t_\mathrm{proc} = \mathrm{hops} \times t_{\mathrm{per\ hop}} \)
Total (one-way): \( t_\mathrm{oneway} = t_\mathrm{prop} + t_\mathrm{ser} + t_\mathrm{proc} \)
RTT (approx.): \( t_\mathrm{RTT} \approx 2 \times t_\mathrm{oneway} \)
Glossary of Variables
- Data rate (MT/s): DDR transfer rate in mega-transfers per second.
- CL: CAS latency cycles.
- tCK: DRAM base clock period, in ns.
- Distance (km): One-way physical or conduit distance.
- Medium: Fiber, copper, or air/free-space (affects signal speed).
- Bandwidth (Mbps): Link throughput, used for serialization delay.
- Packet size (bytes): Payload size used for serialization.
- Processing per hop (µs): Device forwarding delay excluding queueing.
- Hops: Number of intermediate devices, one-way.
How It Works: A Step-by-Step Example
DRAM
Given DDR4-3200 (DataRate = 3200 MT/s) and CL = 16:
\( t_{\mathrm{CL,ns}} = \dfrac{2000 \cdot 16}{3200} = 10 \,\mathrm{ns} \).
With \( t_\mathrm{RCD}=18 \) and \( t_\mathrm{RP}=18 \), random-row = \( (16+18+18)\cdot t_\mathrm{CK} \) with \( t_\mathrm{CK}=\dfrac{2000}{3200}=0.625 \) ns → \( 52 \cdot 0.625 = 32.5 \) ns.
Network
Distance = 1000 km over fiber (\( \alpha=0.68 \)), 1 Gbps link, 1500-byte packet, 6 hops @ 5 µs each:
\( t_{\mathrm{prop}} \approx \dfrac{1{,}000{,}000\,\mathrm{m}}{0.68c} \approx 4.9 \,\mathrm{ms}\),
\( t_{\mathrm{ser}}=\dfrac{1500\times8}{10^9} \approx 12 \,\mathrm{\mu s} \),
\( t_{\mathrm{proc}}=6\times5\,\mathrm{\mu s}=30\,\mathrm{\mu s} \).
One-way ≈ 4.942 ms; RTT ≈ 9.884 ms.
Frequently Asked Questions
Is CL the only timing that matters?
No. For random-row access, tRCD and tRP also contribute. Compare total ns, not cycles alone.
Why doesn’t the network result include queueing?
Queueing is workload-dependent and stochastic. This tool provides a deterministic baseline.
Does air always beat fiber?
Air has higher propagation speed, but fiber enables straighter long-haul routes and far higher bandwidth.
How do jumbo frames affect latency?
Bigger packets increase serialization delay; impact is notable at lower link rates (e.g., 100 Mbps).
Can I use this to estimate gaming ping?
You can estimate the physical lower bound. Actual ping includes queueing, OS/network stack, and server processing.
Tool developed by Ugo Candido. Content verified by CalcDomain Editorial Board.
Last accuracy review: