# EPON vs GPON: Compare PON Standards and Choose an ISP Upgrade Path

**Title tag:** EPON vs GPON 2026 Comparison | Draftech  
**Meta description:** EPON vs GPON: compare IEEE and ITU-T line rates, wavelengths, XG-PON, XGS-PON and 10G-EPON, with optical budgets and migration decisions for ISP networks.  
**Author:** Omar Molina  
**Published:** October 17, 2026  
**Last updated:** October 17, 2026  
**Category:** FTTH Design  
**URL:** https://draftech.com/blog/epon-vs-gpon  
**Primary keyword:** epon vs gpon  
**Word count:** 2936  
**Read time:** 12 minutes

![A worker in a safety vest kneels at an open green sidewalk cabinet, holding a handheld meter to its yellow cables.](../../blog/img_epon_vs_gpon.webp)

---

A faster PON port can leave an ISP with the same service problem. The shared upstream may still fill during busy hours, or the installed optical path may lack the margin needed for the replacement equipment.

We start with the installed network. An operator with a working EPON estate faces a different decision from an operator buying equipment for an empty central office.

This comparison separates the standards' physical specifications from the operating choices that follow. The decision turns on which services the ISP needs to deliver and what its existing plant can support.

## EPON vs GPON: what the standards change

EPON vs GPON compares Ethernet-based passive optical access under IEEE 802.3ah with the ITU-T G.984 family. EPON uses a 1.25 Gbit/s optical signaling rate in each direction; the GPON profile compared here uses 2.48832 Gbit/s downstream and 1.24416 Gbit/s upstream. Both share port capacity among connected subscribers.

Those figures describe the optical interface. They do not describe a guaranteed download or upload speed at a customer's router. We separate line coding from Ethernet framing overhead, then account for the upstream scheduling behavior that determines when an ONU can transmit. A rate printed beside a port cannot answer those questions alone.

### Ethernet framing and GPON encapsulation

EPON carries Ethernet frames and uses multipoint control to coordinate access to the shared upstream. Its 8B/10B encoding turns the Ethernet-side gigabit stream into the higher optical signaling rate. GPON uses the GPON encapsulation method, or GEM, to transport service traffic within its transmission convergence structure. Ethernet services run over either family.

We would reject a selection memo that calls GPON an ATM-only system or describes EPON as a dedicated Ethernet circuit. Those shortcuts hide the relevant behavior. An ISP must establish how the chosen equipment maps subscriber traffic into queues and how the scheduler serves those queues when several customers upload at once.

The same distinction applies to management. A standards label identifies a protocol family, while an operational service also depends on the supported software and provisioning profile. We want the replacement terminal to recover its configuration after a restart and report a useful alarm when the optical signal disappears.

### Compare usable capacity before service tiers

A subscriber plan can be symmetric even when the PON has unequal aggregate upstream and downstream rates. Whether that plan remains defensible depends on the loading of the shared port. We evaluate the busy upstream independently, because an upload-heavy business connection can create a different constraint from a residential download peak.

We also separate activated customers from addresses passed. An empty splitter output contributes optical loss through the splitter's design, but it contributes no traffic. Forecasting both populations as though they were identical can lead an ISP to buy capacity too early or overlook the port that fills first.

Our comparison baseline is a documented optical distribution network, or ODN. For the relationship between a splitter tree and its serving equipment, the [GPON network design guide](/blog/gpon-network-design-guide) provides the design context. The standards table below addresses the optical interface; the route design must still prove each branch.

Also identify the intended customer handoff. A fast optical port feeding a slower premises Ethernet interface will encounter a limit that a splitter redesign cannot remove. We test through the intended handoff and record the frame sizes used, so application throughput and optical line rate remain separate measurements.

The port's uplink is another boundary. We follow traffic from the PON into the aggregation network before assigning a performance problem to the access standard. If the aggregation path is already congested, increasing the access line rate moves more traffic toward the same queue. The upgrade plan needs to identify which interface becomes the next limiting point.

Latency also needs a loaded test. A quiet port can make different scheduling policies appear equivalent, while simultaneous uploads expose the way each service receives transmission opportunities. We compare the required service behavior under contention and record the configuration that produced it. That gives the operator an acceptance target it can repeat after a software change.

## PON line rates and wavelength ranges

The table uses downstream first, from the OLT toward the subscriber. Upstream runs back toward the OLT. Wavelengths are operating bands rather than rounded center labels. We retain each standard's rate notation and identify the selected GPON profile so the table does not imply that G.984 defines only one rate combination.

| PON and standard | Downstream / upstream line rate | Downstream / upstream wavelengths | Selection consequence |
| --- | --- | --- | --- |
| EPON, IEEE 802.3ah; 1000BASE-PX | 1.25 / 1.25 Gbit/s signaling | 1480 to 1500 / 1260 to 1360 nm | Gigabit Ethernet rate before PON overhead |
| GPON, ITU-T G.984.2; selected asymmetric profile | 2.48832 / 1.24416 Gbit/s | 1480 to 1500 / 1260 to 1360 nm, basic single-fiber bands | Specify the installed upstream band for coexistence |
| XG-PON, ITU-T G.987.2 | 9.95328 / 2.48832 Gbit/s | 1575 to 1580 / 1260 to 1280 nm | Higher downstream with asymmetric upstream |
| XGS-PON, ITU-T G.9807.1; basic wavelength set | 9.95328 / 9.95328 Gbit/s | 1575 to 1580 / 1260 to 1280 nm | Symmetric optical line rates |
| 10G-EPON, IEEE 802.3av; 10GBASE-PR | 10.3125 / 10.3125 Gbit/s signaling | 1575 to 1580 / 1260 to 1280 nm | Symmetric 10G Ethernet family |
| 10G-EPON, IEEE 802.3av; 10/1GBASE-PRX | 10.3125 / 1.25 Gbit/s signaling | 1575 to 1580 / 1260 to 1360 nm | Asymmetric Ethernet family |

### GPON: name the profile being purchased

ITU-T G.984.2, Gigabit-capable passive optical networks (GPON): Physical media dependent (PMD) layer specification, August 2019 edition, approved 8/29/2019, defines the GPON optical parameters used here. G.984.2 includes other rate combinations; the selected asymmetric profile must therefore appear explicitly in the equipment schedule.

Because the basic upstream band overlaps the band used by newer PON generations, coexistence cannot be inferred from the word GPON on an asset list. We require the actual terminal wavelength specification and the filtering arrangement before accepting shared-ODN operation. An incomplete terminal inventory leaves the migration design unresolved.

### EPON: keep Ethernet rate and signaling rate separate

IEEE 802.3ah-2004, Ethernet in the First Mile, approved in June 2004, introduced the EPON specifications now maintained within IEEE 802.3. The table's optical signaling rate includes its line coding. It would be misleading to treat the extra encoded bits as additional subscriber payload.

EPON remains a practical retention choice where its installed equipment currently meets the service requirement, so we would assess remaining software support and replacement-terminal availability before proposing a family change. A working provisioning system has operational value, even when another standard looks faster in a headline comparison.

### XG-PON: examine the upstream separately

ITU-T G.987.2, 10-Gigabit-capable passive optical networks (XG-PON): Physical media dependent (PMD) layer specification, February 2023 edition, approved 2/22/2023, defines the asymmetric rates shown. Its in-force components include Amendment 1, approved 6/13/2023, and Erratum 1, approved 9/28/2026. The erratum corrects the expansion of OLT, without changing these rates.

We treat XG-PON as an asymmetric option even though its downstream belongs to the ten-gigabit class. A downstream-led upgrade can fit that structure. An ISP planning substantial symmetric demand should compare the upstream requirement against XGS-PON before committing to another asymmetric generation.

### XGS-PON: identify the wavelength set

ITU-T G.9807.1, 10-Gigabit-capable symmetric passive optical network (XGS-PON), February 2023 edition, approved 2/22/2023, with Amendment 1 from May 2025, defines equal nominal line rates. The basic wavelength set appears in the table. Its optional set instead uses 1480 to 1500 nm downstream and 1300 to 1320 nm upstream.

That optional set overlaps GPON bands. We would therefore specify the basic set when evaluating a GPON coexistence design, then verify compatibility across the actual ODN. Our preference for symmetric capacity has a limitation: extra upstream headroom does not repair a marginal optical path or make an unsupported terminal interoperable.

### 10G-EPON: distinguish PR from PRX

IEEE 802.3av-2009, Physical Layer Specifications and Management Parameters for 10 Gb/s Passive Optical Networks, approved in September 2009, introduced symmetric 10GBASE-PR and asymmetric 10/1GBASE-PRX. The ten-gigabit directions use 64B/66B coding. These historical amendment identifiers identify the technology's origin; procurement should cite the applicable consolidated IEEE 802.3 clauses.

PRX retains a gigabit upstream Ethernet rate, so an order that says only 10G-EPON leaves a consequential choice open. We specify the PMD type before comparing offers. For an existing EPON operator, the upgrade question is whether the selected OLT supports the intended mixed population and its scheduling requirements.

## What matters to an ISP beyond the PON label

First, establish the ISP's planned service mix. We ask which traffic needs a committed allocation and which traffic may share the remaining capacity. Subscriber counts alone cannot settle that question, particularly where a port serves business connections whose upload behavior differs from the rest of the tree.

### Traffic loading and splitter placement

A splitter choice affects both the optical path and the number of customers sharing a port. We model those constraints separately. Reducing the split can improve the loss calculation, but it can also change the serving-port count and the cabinet patching arrangement. That change belongs in the design before equipment quantities are fixed. The [HLD-to-LLD design handoff](/blog/difference-between-hld-and-lld-fiber-design) should carry that revised serving arrangement into the construction details.

We use the proposed take rate to examine which distribution area fills first because, if that area contains the upload-heavy customers, a network-wide average will hide the likely bottleneck. Capacity planning should preserve the association between the service forecast and its particular splitter branch instead of spreading demand evenly across the entire footprint.

That is where Draftech's [FTTH design engineering](/services/ftth-design) connects the PON choice to the route. Our engineering is 100% in-house. The equipment schedule and branch assignments need to agree before a capacity model can become a buildable design.

### Optical margin at the correct wavelengths

Optical loss must be evaluated for each operating direction and the selected interface class, so we include the loss of any coexistence element in the path instead of treating it as an accessory outside the calculation. A design can pass before that element is added and fail after the installation changes.

For the ITU-T options, the [PON Power Budget Calculator](/tools/pon-power-budget-calculator) is a free check of whether a GPON or XGS-PON link closes against the ITU-T class budget, covering total loss from OLT to ONT, margin and maximum reach. Use the actual branch inputs when checking a proposed equipment change.

A wavelength-compatible splitter still needs a verified loss specification. We would not approve reuse from its housing label alone if the component record is missing. The same caution applies to unexplained splice loss: changing the PON family does not establish why the installed path differs from its design calculation.

> **Check the short branch too.** The longest path tests receiver sensitivity, while a low-loss path can challenge the receiver's overload limit. Both ends of the supported operating window belong in acceptance.

Where route records and field conditions disagree, reconcile them before deciding the electronics are unsuitable. The [as-built records needed for GPON operations](/blog/as-built-fiber-for-gpon-network-operations) connect terminal assignments to the physical fiber path. Those records let an operator investigate a weak branch without disturbing a healthy neighboring service.

## Prove migration and daily operation before rollout

A GPON terminal cannot register as an EPON terminal simply because its connector fits. The protocols differ. Even within a family, we require a tested pairing of OLT software and terminal firmware before treating an equipment list as an approved service configuration.

### Coexistence needs an explicit operating design

GPON and basic-set XGS-PON can share suitable passive plant through a designed coexistence arrangement. That statement does not mean every installed GPON terminal qualifies. We check the upstream emission band and receiver filtering against the proposed arrangement, then calculate the insertion loss added by the shared optical equipment.

EPON migration has its own constraints. Older and faster upstream transmissions occupy overlapping wavelength ranges, so wavelength separation alone cannot coordinate them. The serving equipment must support the relevant burst reception and time allocation. We would require the proposed mixed terminal population in the acceptance setup, rather than test each generation in isolation.

The rollback path should remain usable until the replacement service passes, which means preserving the previous configuration and knowing which patch connection still restores the original service. A successful laboratory migration does not establish that a field technician can identify the correct branch during the allowed maintenance window.

### Acceptance must reach the subscriber service

We test registration before throughput. A terminal that appears online can still carry the wrong service mapping, and an apparently successful download can conceal an upstream allocation problem under contention. The acceptance record should show behavior when competing traffic is present, with the promised service profile applied at the customer interface.

Because operations staff need to distinguish loss of optical signal from loss of customer power, we include alarm behavior in the service review and confirm that it reaches the monitoring system in a useful form. Otherwise an equipment change can increase capacity while making the fault desk slower at identifying the affected branch.

Firmware maintenance belongs in the purchase decision. We ask how the operator validates a replacement image before deploying it to subscribers and how it recovers a terminal that does not return after the change. Spare hardware needs the same accepted configuration as the devices it will replace.

The commercial comparison should cover the complete supported service. We compare the installed equipment requirement against the operating cost of maintaining it, including the effect of a required customer visit. Without current offers and an agreed scope, declaring one family universally cheaper would conceal assumptions that belong in the procurement record.

Record the upgrade in the network model as it happens. A terminal swap that changes the serving port must also change the associated inventory entry. We favor a migration procedure that closes that record during acceptance, because a delayed update can send the next fault investigation to the former port.

## EPON vs GPON: choose by the installed network

**Existing EPON operators:** retain EPON where it meets the service requirement and remains supported. Evaluate symmetric 10G-EPON when the upgrade needs more upstream capacity. Make the mixed-population test a purchase condition, and specify PR or PRX explicitly rather than accepting a generic ten-gigabit description.

**Existing GPON operators:** keep healthy GPON services operating while qualifying an XGS-PON overlay for customers whose demand justifies it. Use XG-PON only where its asymmetric upstream fits a defined requirement. We would not add that intermediate step merely because the downstream rate matches the larger service ambition.

**Greenfield ISPs:** make symmetric capacity the baseline comparison between XGS-PON and symmetric 10G-EPON. Select the family that passes the service and operational acceptance plan with a supportable terminal estate. We would choose legacy gigabit access only when the narrower service case and whole-life cost clearly justify it.

For a route whose splitter assignments or optical margin still prevent that decision, Draftech's [free route engineering offer](/free-design) covers the first 20,000 linear feet of a qualifying route at no cost, from feasibility and field survey through permit approval. The owner reviews each request before Draftech commits the package.

If the capacity forecast points to an upgrade but the branch records cannot prove reuse, send the route scope to [info@draftech.com](mailto:info@draftech.com). We can address the design mismatch between the proposed service and its optical path before that uncertainty becomes a field cutover problem.

## Frequently Asked Questions

### Is GPON faster than EPON?

For the asymmetric GPON profile compared here, ITU-T G.984.2 specifies 2.48832 Gbit/s downstream and 1.24416 Gbit/s upstream. IEEE 802.3ah EPON signals at 1.25 Gbit/s in each direction, including line coding. GPON therefore has a higher downstream optical rate in this comparison, but subscriber throughput still depends on framing overhead and the traffic sharing the serving port.

### What is the difference between XG-PON and XGS-PON?

ITU-T G.987.2 XG-PON uses 9.95328 Gbit/s downstream and 2.48832 Gbit/s upstream, while G.9807.1 XGS-PON uses 9.95328 Gbit/s in both directions. That upstream difference matters when the service forecast includes substantial uploads. An ISP should also confirm the selected wavelength set and optical class, because equal downstream rates do not establish that the proposed equipment can reuse the installed path.

### Can an ISP replace a GPON terminal with an EPON terminal?

A shared connector does not make those terminals interchangeable. IEEE 802.3ah EPON and ITU-T G.984 GPON use different access protocols, so the serving OLT must support the terminal's actual mode. Even a terminal advertised for multiple modes needs an accepted software pairing and service profile. The ISP should verify registration and service delivery before scheduling customer replacements.

### Does 10G-EPON always have a ten-gigabit upstream?

IEEE 802.3av defines both symmetric 10GBASE-PR and asymmetric 10/1GBASE-PRX. The symmetric option signals at 10.3125 Gbit/s in both directions; PRX retains 1.25 Gbit/s upstream signaling for its gigabit Ethernet direction. An ISP comparing service capacity should therefore request the exact PMD type and supported operating mode rather than rely on a general 10G-EPON label in an equipment offer.

### Can XGS-PON reuse a GPON splitter network?

Reuse requires an optical design for the installed components and terminal population. G.9807.1's basic XGS-PON wavelength set uses 1575 to 1580 nm downstream and 1260 to 1280 nm upstream, enabling suitable coexistence arrangements. The ISP must verify legacy terminal compatibility and added insertion loss, then establish that the resulting paths remain within the selected interfaces' operating limits.

## Related Resources

- [How to Size Fiber Distribution Hubs for FTTH PON Networks](/blog/fiber-distribution-hub-sizing-ftth-pon) - FTTH Design
- [GPON Network Design: Architecture, Split Ratios, and OSP Planning That Holds Up in the Field](/blog/gpon-network-design-guide) - FTTH DESIGN
- [What Is FTTH Design](/blog/what-is-ftth-design) - FTTH DESIGN
- [FTTH Low-Level Design](/blog/ftth-lld-splice-point-placement-guide) - FTTH Design
- [5 FTTH High-Level Design Mistakes That Cost Fiber Providers Millions](/blog/ftth-hld-design-mistakes) - FTTH Design
- [LLD Quality Control Checklist](/blog/lld-quality-control-checklist-fiber-design-errors) - LLD

---

**About Omar Molina:** Leads FTTH HLD design, PON architecture, construction coordination, and fiber engineering engagements. [info@draftech.com](mailto:info@draftech.com)