Part of: OCUDU on COSMOS · 4G / 5G & O-RAN tutorials
Measured performance of every OCUDU build on COSMOS, per CPU platform and radio, with one reference modem and with 10–60 emulated UEs. The OCUDU overview keeps only the best single-UE result; this page has every test. The current and previous build page says which image to load.
lteue UE stands in. Multi-UE: Amarisoft lteue emulating 10, 20, 30, 40, 50 and 60 UEs on one USRP N310, each UE in its own network namespace. The current UE host resets under traffic above 10 UEs (a hardware investigation is open), so points above 10 are sparse until that is fixed.iperf3, 10 s, all UEs at once; the figure is the sum of the receiver rates.iperf3 at a rising ladder of offered loads (starting at 0.8× the TCP figure, +30 % per step, stopping at 5 % loss or when the link saturates); the figure is the highest received rate with less than 5 % loss, the definition the capacity sweep uses.docs/results/perf-trend/) are folded in with tools/ran-bench/backfill_from_ledger.py.| CPU platform | Cores / threads |
|---|---|
| A | 2× Xeon Gold 6226, 24C/24T |
| B | 2× Xeon Gold 6126, 24C/48T |
| C | 1× AMD EPYC 9355P, 32C/32T |
| D | NVIDIA GH200: Grace, 72× Arm Neoverse V2, 72C/72T, + Hopper GPU running the NVIDIA Aerial layer 1 |
Latest build ocudu-20261009 · image ocudu-20261009 · 2026-10-09. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | 1/1 | 25.9 ▲10% | 195 ▲5% | 75 ≈0% | 564 ▲3% | 83 ≈0% |
| lteue, 10 UEs | 10/10 | 25.9 ≈0% | 249 ▲32% | 32 ▼2% | 276 ≈2% | 33 ≈1% |
| lteue, 20 UEs | not measured | — | — | — | — | — |
| lteue, 30 UEs | not measured | — | — | — | — | — |
| lteue, 40 UEs | not measured | — | — | — | — | — |
| lteue, 50 UEs | not measured | — | — | — | — | — |
| lteue, 60 UEs | not measured | — | — | — | — | — |
Figure — Trend by build. One line per UE configuration (red = the reference modem). A gap means the point was not measured on that build.
Latest build 26.10 · image ocudu-20261020 · 2026-10-23. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 33.0 ▼7% | 110 ▲101% | 59 ≈1% | — | — |
Figure — Trend by build. One line per UE configuration (red = the reference modem). A gap means the point was not measured on that build.
Latest build ocudu-20260916 · image ocudu-20260916 · 2026-09-17. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 36.5 ▼4% | 113 ▲3% | 79 ≈1% | — | — |
Figure — Trend by build. One line per UE configuration (red = the reference modem). A gap means the point was not measured on that build.
Latest build 26.10 · image ocudu-20261020 · 2026-10-23. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 24.3 | 206 ≈2% | 75 ≈1% | 591 ▲88% | 83 ▲7% |
Figure — Trend by build. One line per UE configuration (red = the reference modem). A gap means the point was not measured on that build.
Latest build ocudu-20260909b · image ocudu-20260909b · 2026-09-14. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 30.6 ▼2% | 26 ≈1% | 4.9 ▼9% | — | — |
Figure — Trend by build. One line per UE configuration (red = the reference modem). A gap means the point was not measured on that build.
Latest build 26.10 · image ocudu-20261020 · 2026-10-26. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 24.5 | 179 ▲620% | 140 ▲775% | 316 | 168 |
Figure — Trend by build. One line per UE configuration (red = the reference modem). A gap means the point was not measured on that build.
Latest build ocudu-20261005b · image ocudu-20261005b · 2026-10-05. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 28.0 | 208 ▲128% | 61 ▲64% | — | — |
Figure — Trend by build. One line per UE configuration (red = the reference modem). A gap means the point was not measured on that build.
Latest build 26.10 · image ocudu-20261020 · 2026-10-23. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 39.7 | 143 ▼3% | 144 ▲4% | 711 | 166 |
Figure — Trend by build. One line per UE configuration (red = the reference modem). A gap means the point was not measured on that build.
Latest build aerial-26-2 · image aerial-26-2-gh200 · 2026-11-09. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | — | 474 | 144 | 590 | — |
Latest build aerial-26-2 · image aerial-26-2-gh200 · 2026-11-09. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | — | 464 | 147 | 600 | — |
Latest build ocudu-20260909b · image ocudu-20260909b · 2026-09-14. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | — | 182 | 138 | — | — |
Latest build aerial-26-2 · image aerial-26-2-gh200 · 2026-11-09. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | — | 190 | 26 | — | — |
Latest build ocudu-20260916 · image not recorded · 2026-09-18. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 25.0 | 24 | 11 | — | — |
Not measured with the current benchmark yet.
Latest build ocudu-20260915 · image ocudu-20260915 · 2026-09-15. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | — | 96 | 38 | — | — |
Latest build ocudu-20260909b · image ocudu-20260909b · 2026-09-14. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 30.8 | 26 | 4.7 | — | — |
Latest build ocudu-20260909b · image ocudu-20260909b · 2026-09-14. Each cell is the median of the build's rounds; the arrow compares with the previous build measured at the same point (⚠️ = more than 10 % worse).
| UE | Attached | Ping p50, ms | TCP DL, Mbit/s | TCP UL, Mbit/s | UDP DL, Mbit/s | UDP UL, Mbit/s |
|---|---|---|---|---|---|---|
| Single reference UE | — | 29.6 | 27 | 5.0 | — | — |
Author(s): COSMOS team · Generated: 2026-11-09 by tools/ran-bench/render.py from results/ran-bench/ · Tags: ocudu, performance, benchmark