[Network] Theory-crafting: CPU ≈ access point performance

One of the fact I’ve learned recently during WiFi 6 access point comparison testing is that not all access point are created equal even if with the same PHY. So how do we tell which AP is worth buying when looking for the best performance? Perhaps, CPU specification maybe the answer.

Background

Manufactures list the maximum theoretical throughput on their access point specifications. While these numbers are useful to estimate the minimum performance, they do not offer, which AP can perform better than the other beyond the minimum.

If we compare Ubiquiti UniFi’s U6 Pro and U7 Pro AP’s 5GHz radio specifications, U6 Pro lists significantly higher number than U7 Pro on the throughput rate. However, for majority of users, U7 Pro will give better throughput on 5 GHz than U6 Pro even on the same WiFi 6 client device. Why?

Client bottleneck

The key principle of networking is “bottleneck determines the throughput”. Since most WiFi clients are 2×2 MIMO, 4×4 would not give practical throughput advantage in most situations*. Additionally, 160 MHz channel width in 5 GHz radio is often impractical or unusable for many**.

Therefore, the actual 5 GHz WiFi link (PHY) for WiFi 6 client and U6 Pro is 1200 Mbps (1/4 of advertised).

The bottleneck principle also applies to WiFi generation as well. So if we have WiFi 6 client connected to WiFi 7 capable AP, U7 Pro, the PHY is at the rate of WiFi 6. So with WiFi 6 client, U7 Pro’s PHY is also 1200 Mbps.

*There are situations where 4×4 can be useful such as in mesh setup, those who owns 3×3 or 4×4 capable client devices, or have ability to utilized MU-MIMO etc.
**Apple has not supported 160 MHz channel width until the most recent iPhone 15 series. 160 MHz in 5 GHz radio requires to use DFS channels i.e. DFS strikes make the channel unstable.

WiFi Efficiency

The graph above is from my recent U7 Pro review article. The comparison graph shows the average throughput of U7 Pro is significantly higher than that of U6 Pro. Neither is anywhere close to 1200 Mbps. This is related to what I call “WiFi efficiency”. Typically, the maximum Wi-Fi efficiency ranges between 50-70% of PHY. This 20% range is what we are seeing the difference between U6 Pro and other APs on the graph.

As a network hobbyist who like to keep trying new access points, and looking for the best I can afford, I like to have a rough idea of Wi-Fi efficiency of the new AP before purchasing one. After staring at the specifications for a while, I came to a brilliant (not really) hypothesis. The relative WiFi efficiency can be predicted by the chipset used by each access point. With each generation, chipset/platform category changing, I need a step further. So within the chipset, I hypothesize more specifically the “CPU” specification as the best predictor.

Let’s me show how this hypothesis holds true for various testing I have done so far.

Although each venders had their proprietary, perhaps marketing, technology, I hypothesize the real difference in maximum throughput performance comes from the chipset used by each AP.

This should be a fairly easy guess if we think about other electronics e.g. computer, smart phone, and tablet. However, just looking at the chipset did not quite explain a difference between certain AP performance comparison. It also won’t help when new chipset is released as I have no reference to compare from previous generations. So I looked one step further than the chipset.

So my current hypothesis is “the best (specification parameter) predictor for wireless access point performance (max throughput) is CPU”.

The best test to prove this point would be to have two matched setup that is identical for everything, but CPU and measure the max throughput as output. However, this is not practical because there are many other variable such as vender specific software and other hardware difference.

So rather than scientifically proving my hypothesis, I will show my hypothesis holds true in multiple real world tests of my own.

Before going into the actual test result comparison, we need to know the basic fact that not all chipsets are made equal. There are several companies that make access point WiFi chipsets. Two big names are Broadcomm and Qualcomm. Qualcomm chipset was used by all APs that I have tested recently. So this is the focus of current article.

In the last 3 WiFi generations (WiFi 6, 6E and 7), there have been two platforms in Qualcomm WiFi Chipset: immersive home and networking pro series. The latter is top of the line platform while the former is more budget friendly platform.

Although two platforms differ in the features, the specifics change between generations. So let’s just focus on the CPU. We should see some difference amongst CPUs; otherwise, my hypothesis immediately fails.

I’ve listed below Networking Pro platform’s the highest and the lowest end models, and Immersive Home platform’s chipsets for each generations.

16206203210326
WiFi Version7777
PlatformNetworking ProNetworking ProImmersive HomeImmersive Home
CPUQuad Cortex-A73Quad Cortex-A73Quad Cortex-A53Quad Cortex-A53
Clock2.2 GHz2.2 GHz1.5 GHz1.5 GHz
WiFi 7
1610610318316
WiFi Version6E6E6E6E
PlatformNetworking ProNetworking ProImmersive HomeImmersive Home
CPUQuad Cortex-A53Quad Cortex-A53Dual Cortex-A53Dual Cortex-A53
Clock2.2 GHz1.8 GHz1.0 GHz1.0 GHz
WiFi 6E
1200400216214
WiFi Version6666
PlatformNetworking ProNetworking ProImmersive HomeImmersive Home
CPUQuad Cortex-A53Quad Cortex-A53Dual Cortex-A53Dual Cortex-A53
Clock2.2 GHz1.0 GHz1.0 GHz1.0 GHz
WiFi 6

In WiFi 6 and 6E generation, immersive home platform used dual core CPU at 1.0 GHz. In contrast, networking pro series used quad core CPU with variable clock speed depending on the specific chipset model.

In the latest, WiFi 7 generation, Immersive Home now uses Quad core CPU, but Networking Pro has moved to from A53 to A73. If we assume manufacture’s claim for CPU performance improvement on each Cortex ARM generation, A73 should be more than twice that of A53.

So in every generation, we can expected twice the CPU performance difference between networking pro and immersive home on the paper. Conceptually, this is like comparing Intel’s Celeron/Pentium chips against Core series. This drastic difference in CPU specification looks to be a good target to use for performance prediction.

Now let’s look at several different scenarios to see if CPU difference indeed can be an explanation for throughput test result observed.

Throughput Test Setup

  • Client device: iPhone 14 Pro Max (WiFi 6, 2×2, 80 MHz)
  • AP to client position: 5 ft with full line of sight
  • Test Methods: iPerf3 to local NAS server with 10 Gbps connection. The tests were run at least twice on each, and if there is significant discrepancy, I had ran more to ensure stability is achieved. Took the best.
  • [math]WiFi Efficiency = \frac{Practical Throughput}{PHY}[/math] is calculated for better of the two (download vs. upload). PHY in all test cases are 1200 Mbps as that is the max limit of WiFi 6 for given client specification.

Immersive Home vs. Networking Pro Test

Purpose:

To prove Networking Pro (CPU) offers significantly better throughput than Immersive Home (CPU).

Method:

Ubiquiti UniFi U6 Pro vs. Ruckus R350.

Ubiquiti UniFiRuckus
U6 ProR350
Hardware
ChipsetQ Home 216Q Pro 400
CPUDual A53 1.0 GHzQuad A53 1.0 GHz
5 GHz
WiFi Version/MIMO/Max Channelization (MHz)6/4×4/1606/2×2/80
Max PHY rate (Mbps)48001200
Practical Throughput
Download (Mbps)659874
Upload (Mbps)622799
WiFi Efficiency55%73%

Expectation:

U6 Pro uses Immersive Home chipset while R350 uses Networking Pro. While both AP’s CPU have 1.0GHz clock speed, R350 equipped with Networking Pro 400 has quad core whereas U6 Pro with Immersive Home 216 is only a dual core. With such significant CPU specification difference, I expect significant throughput gain of R350 over the U6 Pro.

Result:

R350 showed nearly 20% improvement in WiFi Efficiency over the U6 Pro. This performance gain is equivalent to estimated throughput gain from WiFi 6 to WiFi 7 by new 4K-QAM technology i.e. it’s quite a significant gain.

High vs. low-end Test

Purpose:

To compare different models of the same class chipset (CPU).

Methods

Ruckus R750 vs. Ruckus R350.

RuckusRuckus
R750R350
Hardware
ChipsetQ Pro 1200Q Pro 400
CPUQuad A53 2.2 GHzQuad A53 1.0 GHz
5 GHz
WiFi Version/MIMO/Max Channelization (MHz)6/4×4/1606/2×2/80
Max PHY rate (Mbps)24001200
Practical Throughput
Download (Mbps)861874
Upload (Mbps)760799
WiFi Efficiency72%73%

Both uses Qualcomm Networking Pro series from WiFi 6 generation, but R750 uses Networking Pro’s highest end chipset 1200. On the contrary, R350 uses the lowest end model 400.

Expectation

While both have quad core CPU, R750’s CPU has twice that of R350 in clock speed. However, 1200 supports up to 12 streams while 400 supports only 4 streams. So my expectation is the primary purpose of the extra CPU’s clock difference is to support more streams rather than individual throughput performance difference. Thus, for a single 2×2 client maximum throughput comparison test, I do not expect significant difference.

Result

As expected, the throughput results are essentially identical.

This fact indicates high quality hardware design of Qualcomm and Ruckus. When a hardware is well design from ground up, there should be no real bottleneck. If there is a significant bottleneck, it means remaining hardware powers are wasted.

Beyond CPU Test

Purpose:

To prove, impact of non-chipset (CPU) factor on performance is relatively small.

Methods:

Ruckus R750 vs. HPE Aruba AP-535

RuckusHPE Aruba
R750535
Hardware
ChipsetQ Pro 1200Q Pro 1200
CPUQuad A53 2.2 GHzQuad A53 2.2 GHz
5 GHz
WiFi Version/MIMO/Max Channelization (MHz)6/4×4/1606/4×4/160
Max PHY rate (Mbps)24002400
Practical Throughput
Download (Mbps)861813
Upload (Mbps)760791
WiFi Efficiency72%68%

Two APs use the same chipset; hence, the same CPU.

Expectation

Given the same CPU, I expect similar throughput performance between two APs, but it does not have to be identical. In fact, I expect other factors can influence throughput, but the impact should be relatively small when compared to CPU.

Result

The practical throughput was indeed very close. By WiFi efficiency point, it’s merely 4% difference. After multiple additional tests on different days, all Ruckus WiFi 6 APs were able to achieve over 850 Mbps throughput; however, HPE 535 has not been ale to do it even a once. Thus, I believe, the difference is not by a chance, but there is a real max throughput superiority of Ruckus over AP 535 despite using the same chipset/CPU. This slight advantage of Ruckus AP maybe coming from Ruckus’s proprietary technology such as Beam Flex+.

Purpose:

To prove, WiFi generation difference have minimal influence when compared to CPU.

Method:

U7 Pro vs. HPE Aruba 535 vs. Ruckus R350

Ubiquiti UniFIHPE ArubaRuckus
U7 Pro535R350
Hardware
ChipsetQ Home 3210Q Pro 1200Q Pro 400
CPUQuad A53 1.5GHzQuad A53 2.2 GHzQuad A53 1.0 GHz
5 GHz
WiFi Version/MIMO/Max Channelization (MHz)7/2×2/1606/4×4/1606/2×2/80
Max PHY rate (Mbps)288224001200
Practical Throughput
Download (Mbps)871813874
Upload (Mbps)764791799
WiFi Efficiency73%68%73%

Ubiquiti UniFi has recently released their first WiFi 7 access point U7 Pro. It uses Qualcomm immersive home chipset, but interestingly this is now Quad core A53. This makes U7 Pro’s CPU specification comparable to WiFi 6 Networking Pro chipsets.

Expectation

Since this particular test uses WiFi 6 client, there is no WiFi 7 technology advantage of U7 Pro against WiFi 6 APs. Since three APs have similar CPU specifications, I expect U7 Pro’s throughput to reach close to HPE Aruba and Ruckus WiFi 6 APs.

Result

U7 Pro indeed achieved, WiFi efficiency compatible to Ruckus APs. This is the number, I could never achieve with HPE Aruba.

*The reason I did not put UniFi 6E access point here is because it’s unclear what exact CPU they use. Community members on the Ubiquiti official forum suggest they may be using same as U6 Pro with some little add-on. In which case, I do not expect any gain over U6 Pro when using WiFi 6 client. Since I do not own any U6 Enterprise (WiFi 6E) AP, I can’t confirm the actual numbers, but I have read some users on reddit confirming this.

Summary

Ubiquiti UniFIUbiquiti UniFiRuckusHPE ArubaRuckus
U7 ProU6 ProR750535R350
Hardware
ChipsetQ Home 3210Q Home 216Q Pro 1200Q Pro 1200Q Pro 400
CPUQuad A53 1.5GHzDual A53 1.0 GHzQuad A53 2.2 GHzQuad A53 2.2 GHzQuad A53 1.0 GHz
5 GHz
WiFi Version/MIMO/Max Channelization (MHz)7/2×2/1606/4×4/1606/4×4/1606/4×4/1606/2×2/80
Max PHY rate (Mbps)28824800240024001200
Practical Throughput
Download (Mbps)871659861813874
Upload (Mbps)764622760791799
WiFi Efficiency73%55%72%68%73%

Although throughput of wireless access points depends on many factors, CPU specification can be useful predictable factor.

Reference