AI: Nvidia, AMD, Intel Similar

Databricks tests have similar results

Databricks, a large service bureau, has the scale, expertise, and incentive to do things carefully. There are small differences that would change with a different set of tests. These numbers are strongly suggestive of very rough parity between the chips on the market in December 2023. Intel & AMD are not far behind Nvidia,

Continue reading

AT&T’s Scary 300% Hurdle Model

AT&T is cutting capex, usually a mistake in telecom. Chief Data Officer Andy Markus told WSJ “AT&T aims for at least a ‘3x return on any AI investment,’ ideally within the first year of deployment. …“Every dollar you invest with us to create an AI solution … should return three or more back.”

A while back a friend at AT&T complained that he was required to get his money back in a year. This is much worse. Even spread over 2 years, this is a higher hurdle rate than I recall seeing.

If Markus didn’t misspeak, AT&T is screwing shareholders who don’t sell very quickly.

Continue reading

From 2017: Maintaining the Internet connection with autonomous networks

As a propaganda campaign about Internet splintering grows, I thought to reprint this

The question: how to arrange “robust interconnection” of the Internet without being dependent on ICANN. Vladimir Putin is creating an alternate root, with support from India and China. Columbia Professor Eli Noam convinced me a “network of networks” was possible and could be a good thing in theory. I doubt ICANN will actually shut out the Russians, but it’s reasonable for Russia to protect itself.* My question here is technical. Thousands of people, especially at the IETF, have worked to build the Internet we have today. The principles are simple; the implementation is demanding. So I’m asking engineers, “What technical systems must be built to ensure robust interconnection, assuming everyone wants to work in good faith?”

     ICANN CEO Fadi Chehadi confirmed to me there was no technical reason the Chinese, Hebrews, Verizon or any other competent party couldn’t set up independently. Vladimir Putin intends to test that, creating a new root that is not controlled by a California organization. The primary benefits of “The Internet” could be maintained so long as there was “robust interconnection.” Fadi added the rub was how to ensure that robust interconnection. 

    I think Fadi was worried about censorship, a real issue.

Continue reading

Prof Eli Noam’s 1, 2, Many Internets

“Would having multiple Internet styles be a good thing?” Columbia Professor Eli Noam asked a while back. The Internet, after all, is just a “network of networks.” Noam wrote Interconnecting the Network of Networks back in 2001, and has long been the leading public intellectual in communications. (I don’t have notes on how he Pleaase consider all of the following as by me, strongly influenced by Eli.)

Since then, network science has evolved the concept of “multi-layer” network concepts. Each network is autonomous, determining its own structure. What is crucial is that they are linked.

Continue reading

Massive MIMO explained in depth

Eom Marzetta invented Massive MIMO in 2010 and the first production units shipped in 2017 in Japan & China. As the build of 5G mid-band picked up in 2019 and 2020, M MIMO became the standard choice. As I write in late 2022, China has deployed over two million cells.

MIMO uses multiple antennas to send several signals in the same spectrum. With 4 antennas, typical speeds double. Results vary based mostly on the terrain.

M-MIMO was the next step. The typical M-MIMO uses 32-64 relatively small antennas. Lower frequencies (below ~2300 MHz) require larger antennas so it is generally impractical to have more that 4 or 8 antennas so rarely implement M-MIMO.

Continue reading

5G Small Cell $3,000? Yes

“I can even bring it lower than that,’ said <one who knows> but who couldn’t go on the record by company policy. At the Brooklyn 6G conference, everyone was excited by the performance of “systems on a chip.”

You can’t buy one at Amazon today. If you order less than 100,000, the price will be higher.

Continue reading

Element: Customers love fiber

Customers around the world are still choosing fiber despite cable upgrades. Improved cable is essentially equal to fiber for 99% of practical uses. All of Comcast will get cable uploads of 100 or 1,000 megabits. (DOCSIS 3.1, not 4.0)

Most other cablecos have similar plans to quickly upgrade. It’s cheap: less than $200/home. Comcast doesn’t even have to increase capex.

Continue reading

India, France, US Partial Winners in US DOD Open Ran Tests

Lack of complete interoperability is the biggest obstacle to Open RAN adaptation by the major carriers. Even enthusiastic analyst Stefan Pongratz expects O-RAN to need 3 or 4 more years to get to 5% market share. The major carriers are going very slowly.

The US Department of Defense has a 5G Challenge for ” a fully integrated multivendor end-to-end 5G network.” Radisys (owned by India’s Jio), Cap Gemini (France), and Mavenir (US, India*), each won a share of the $3 million in prizes. Signal System Management (from the US defense world) and Fujitsu (Japan) also got modest grants.

Continue reading

Bringing Evidence to US Spectrum Policy

Verizon capacity margin

Almost everyone in DC has missed the dramatic change in spectrum needs. With the current spectrum, capacity is going up at perhaps 40%/year, per Verizon and AT&T. Demand is going up 15-25%year. (43 OECD countries 15% in 2021.)

Ergo – Carriers don’t need spectrum for years. More spectrum will l do almost nothing for US broadband deployment anytime soon. T-Mobile is already going to 99% 5G, 90% 100 Mbps. Verizon and AT&T have to match. How much more could they build?

Continue reading

Can America add 10M lines of fiber in 2023? Moffett says no.

Top analyst Craig Moffett adds up announcements for US fiber homes passed and finds plans for 10 million in 2023. AT&T alone plans 4 million.

He finds that unrealistic due to the well-publicized staffing and supply problems. But soon after, I discovered that the UK passed 4 million homes the last year. Britain has 68 million people, the US, 330 million. If the US built at the UK level, we could do 17 million per year.

Continue reading

Network Slicing: So far, few will buy (Q)

Network slicing “still remains many years away in most markets. A very complicated undertaking” @DHSchoolar @disruptivedean points out ” I haven’t met a Slice salesperson yet, or a Slice-procurement team.” It will be difficult to monetise.

Telcos tell me they haven’t seen any demand for network slicing except the security people. Regular “best efforts” networks s do a good job for almost all.
Customers so far don’t want to pay for “slices” or QoS.

Three years ago, Henning Schulzrinne warned of disappointment with slicing. “In 3G & 4G, we were told QoS would sell well. It may very well go the same way in 5G.”

Telcos outside China aren’t investing because so far there’s no market.

Don’t Believe The Hype

  • Nothing is a bigger scam than 5G. Latency is little improved from 4G. 5G: It still does nothing important. The 5G Emperor has no clothes.
  • Open RAN will probably be a good thing, but sales will be modest for years. Still many challenges. Nokia & Ericsson blocking essential interop testing.
  • Telco Edge Networks are minuscule outside Asia.
  • 5G will not change life in India. The pundits and politicians have been scammed.
Continue reading

Hans Vestberg: Liar of the month for false latency claims


Verizon’s “low latency” 5G Edge network is specified at 20-35 ms. Its 4G network averages ~32 ms. That’s an insignificant difference that makes almost no difference to any application. It is not “low latency” nor does it improve the user experience.

For example, I and many others get dizzy in Virtual Reality unless the latency is close to 10 ms. 20-35 ms doesn’t cut it.

Vestberg claimed in 2022 “You cannot do Metaverse without the network that Verizon is building today with low latency.” 20-35 ms doesn’t improve the Metaverse.

Continue reading

World’s Best 5G Network Starts Building in New York

Gigabit 5G is millimeter wave but the reach is closer to 100 meters than 1000 meters. To get the gigabit speeds reliably, you need a heckuva lot of cells, perhaps every block. Paul Baran showed the way years ago with Richocet but he was ahead of his time.

LinkNYC has just installed its first 32 foot kiosk in the Bronx, designed for Wi-Fi and 5G from up to 4 carriers. True 5G gigabit will usually reach 100-250 meters.

Continue reading

Maybe Believe: Ones to watch

  • RISC-V chip designs are accelerating. The processor is open source and royalty-free.
  • Michael Waring, possibly the most effective marketer in telecom, now CEO of Calix. The stock has tripled since Michael pivoted the company to systems that help its customers sell.
  • Accelercomm Professor Rob Maunder improves wireless performance through better coding software
  • Tarana Wireless Andrea Goldsmith tells me wireless has plenty of room for improvement. Tarana claims to have done it, with $100 million in sales to prove it.
  • Positron Access works well. It’s a very fast and cheap way to use existing wires to connect fiber to the basement or wireless to the rooftop
Continue reading

We’re Already Watching All The TV We Want. Traffic growth was clobbered

People mostly are not watching more or higher-bandwidth video.

Video is about 70% of traffic,

Ergo: Unless something crazy happens, traffic will only grow moderately.

I’ve been reporting falling traffic growth rates for years but the 2021 OECD 15% growth still startled me. The likely explanation: people have run out of time to watch more video. In fact, hours watching has actually gone down.

Continue reading

China reaching 2,000,000 5G Cells

China had 1,968,000 5G base stations at the end of July and should reach 2,000,000 in August. I believe that is more than twice the total in the rest of the world. Users are at 475,000,000, more than the entire rest of the world.

Huawei is pulling ahead in technology, deploying a distributed Massive MIMO system. DM MIMO, also called cell-free, is the next big technological advance. It uses multiple MIMO radios all working together to optimize performance. DM MIMO does more for network builders than anything likely from “6G” in a decade.

Continue reading

5G Phones Have Taken Over

5G phones are now so cheap that all but the poor are buying them. They still do nothing important but for $0 (Apple) to ~$40 more (Most others, falling) most people don’t want to buy a phone soon to be obsolete.

80-90% of the phones sold in China & Korea are 5G. >70% in the US. Similar figures for Japan, Australia, Canada, and Western Europe, although I don’t have data.

Continue reading

Soon, Almost Every Mobile Phone Will “Have” to Pay Huawei Royalties

China’s largest tech giant almost doubled its R&D budget over the past half-decade to $22.1 billion in 2021 — more than any company in the world outside America. Bloomberg

2021 R& D (billions) from Bloomberg

Huawei is the driving force behind Sisvel’s Wi-Fi 6 patent pool and will collect a significant % of the 50-60 cents Sisvel is demanding from every phone, Wi-Fi router, etc. That could easily amount to over $100 mi;llion per year. It will be hard to circumvent Sisvel, which also includes Phillips and Mediatek.

Continue reading