Ericsson First to Market with Processors Made on "Intel 4" Technology

[Fouquin]

Ericsson, a massive player in networking and cellular technology, has announced their new lineup of RAN Compute (Radio Access Network) processors as well as new high bandwidth routers built around in-house silicon designs fabbed on Intel 4, beating Intel's own Meteor Lake processors to market. These new processors are not something the average person is going to be using in their PCs or home routers, but they are pivotal in the reliability and speed of current and future generation 5G networking as the devices they power handle the ever increasing traffic and bandwidth demands of modern networks. In its press release Ericsson notes that the new RAN Processor 6672 and Radio Processor 6372 offer four times more capacity at twice the efficiency compared to their previous generation. They claim that the power draw of their new processors on Intel 4 is between 30% and 60% lower than the industry benchmarks.

Ericsson signed on as one of Intel's largest customers when Intel announced their "Intel Foundry Services" initiative under the IDM 2.0 strategy to offer chip designers the ability to fab their processors at Intel fabs. The strategy has - at least outwardly - appeared to be a boon to Intel as they've signed on large partnerships ranging from big budget defense contractors to datacenter clientele and even ARM. The new RAN Compute systems from Ericsson packed full of technology built on "Intel 4" even ahead of Intel's own designs exemplifies that Intel is at the very least committed to the strategy, and Ericsson has already announced plans for even more chips on Intel's "18A" process slated for 2025.



View at TechPowerUp Main Site | Source

 

[mahirzukic2]

built around in-house silicon designs fabbed on Intel 4, beating Intel's own Meteor Lake processors to market

Love that they are beating intel to the market on their own (Intel's) FAB. Intel snap out of it already.

 

[dgianstefani]

Love that they are beating intel to the market on their own (Intel's) FAB. Intel snap out of it already.

It's an Intel custom chip, so Intel is "beating" Intel, in a major loss for Intel and a win for Intel.

Intel 4 isn't offered through IFS.

 

[pressing on]

Intel 4 isn't offered through IFS.

As Ericsson signed up as an IFS customer and the chips are made on that process, that is clearly not the case. Intel seem to be ready to make any of their process nodes available to IFS customers, and if IFS is going to go anywhere that is what they need to do.

 

[mahirzukic2]

It's an Intel custom chip, so Intel is "beating" Intel, in a major loss for Intel and a win for Intel.

Intel 4 isn't offered through IFS.

Oh I thought it was developed by Ericsson and built by Intel. My bad.

 

[dgianstefani]

As Ericsson signed up as an IFS customer and the chips are made on that process, that is clearly not the case. Intel seem to be ready to make any of their process nodes available to IFS customers, and if IFS is going to go anywhere that is what they need to do.

Except IFS is for companies to make their own IP on. This is just Intel making a custom chip for a partner.

 

[Six_Times]

Shared progress in technology. Good to see it.

 

[BArms]

Networking huh?

We had cheap fanless 1GbE Intel NICs 20 years ago.

Maybe we'll get cheap and fanless 10GbE NICs and Switches, finally?

 

[Squared]

Did they really beat Meteor Lake to market? I don't see any information about availability and a quick search didn't reveal a way to buy this.

I had wondered if IFS could even produce enough chips to bring Meteor Lake to market by itself. That an IFS customer and Intel appear poised to launch Intel 4 products almost simultaneously does seem promising for Intel 4.

 

[lemonadesoda]

Networking huh?

We had cheap fanless 1GbE Intel NICs 20 years ago.

Maybe we'll get cheap and fanless 10GbE NICs and Switches, finally?

^^ This. What is it that a mobile phone can manage 5G or wifi6 but a network card / switches cant cope without large power draw, heat, fans.

 

[dgianstefani]

^^ This. What is it that a mobile phone can manage 5G or wifi6 but a network card / switches cant cope without large power draw, heat, fans.

It's a difference in bandwidth by a factor of ten.

 

[Minus Infinity]

future generations of 5G, how nice for some. I still have no access to first gen 5G and even my 4G sucks ass.

 

[R-T-B]

Maybe we'll get cheap and fanless 10GbE NICs and Switches, finally?

If the laws of physics give way a bit, or we wait for 10 more nodes, then maybe.

Progress speed, she ain't what she once was.

 

[BArms]

If the laws of physics give way a bit, or we wait for 10 more nodes, then maybe.

Progress speed, she ain't what she once was.

I suspect it's because ML/AI and the Automotive industry's ravenous hunger for cutting edge chips are probably the reason that foundries are making too much money off those industries to allow the humble NIC and Switch to get the later nodes. Idk though.

It's funny because In 2003 the largest SD cards were 512MB or at most 1GB for like $500. Twenty years later you can buy a 1TB Micro SD card for around $100.

Somehow consumer NAND grew 1000x but Networking speeds are affordably 2.5x faster than 2003. Also I guess most consumers don't need more than 1Gbit anyway so there's that angle to consider but still I wish 10GbE gear was ubiquitous at this point.

 

[R-T-B]

Somehow consumer NAND grew 1000x

This is no secret. NAND went vertical. Layers. Too hot to work elsewhere honestly. Though DRAM might benefit someday.

 

[lemonadesoda]

Not withstanding the comments above, from 2003 to 2022 we managed to go from PCIe x1 0.250 GB/s to PCIe x5 3.938 GB/s without orders of magnitude additional power or cooling requirements. It would have been nice to see a little more progress on NIC on speeds and prices.

 

[dgianstefani]

Not withstanding the comments above, from 2003 to 2022 we managed to go from PCIe x1 0.250 GB/s to PCIe x5 3.938 GB/s without orders of magnitude additional power or cooling requirements. It would have been nice to see a little more progress on NIC on speeds and prices.

That's on very short distances and with very wide/many traces/wires.

 

[TumbleGeorge]

If the laws of physics give way a bit, or we wait for 10 more nodes, then maybe.

Progress speed, she ain't what she once was.

What is problem with laws of physics attached to only 10Gbps?

 

[R-T-B]

What is problem with laws of physics attached to only 10Gbps?

It's attached not only to 10gpbs, it's attached to moving 10x the info through the same space without significant advances in process technology as were present in the past. The answer? Wait, or stick a fan on it.

That's on very short distances and with very wide/many traces/wires.

Also if you look at pcie 5.0 ssds cooling isn't exactly looking great there either.

 

[lemonadesoda]

Can I make my car go faster by sticking a fan on it?! /s

 

[80-watt Hamster]

Can I make my car go faster by sticking a fan on it?! /s

Yes.

 

[Squared]

I suspect it's because ML/AI and the Automotive industry's ravenous hunger for cutting edge chips are probably the reason that foundries are making too much money off those industries to allow the humble NIC and Switch to get the later nodes. Idk though.

I would imagine that's actually backwards. Moore's Law stayed alive even as foundry costs grew exponentially because the computer market grew exponentially, providing enough revenue to pay the ever-increasing costs. Now that the computer world is growing slowly, the ML/AI world is helping to drive foundry development, and every time a new node becomes available, all the most expensive chips are made with it and in the wake of all this is left cheap nodes which used to be the best.

Somehow consumer NAND grew 1000x but Networking speeds are affordably 2.5x faster than 2003. Also I guess most consumers don't need more than 1Gbit anyway so there's that angle to consider but still I wish 10GbE gear was ubiquitous at this point.

This is frustrating. It might have to do with scale; like new foundry nodes, networking chips are expensive to develop, but since 2003 wired networking has been sidelined in favor of wireless, so consumer hardware development has focused on that. It's finally gotten fast enough that the ethernet connection between the modem and router is becoming the weakest link, so maybe there's reason to invest in consumer ethernet development again, at least a little.

 

[Minus Infinity]

It's finally gotten fast enough that the ethernet connection between the modem and router is becoming the weakest link, so maybe there's reason to invest in consumer ethernet development again, at least a little.

Isn't that why 2.5GB-E is now fairly common at least on MB's? Most consumers don't need 10GB-E, but 2.5GB-E is a worthwhile upgrade (bad Intel implementations aside). Alas Synology won't get on board and still push out NAS with garbage 1GB-E ports and a lot of routers only have one 2.5GB-E port at best.

 

[Squared]

Isn't that why 2.5GB-E is now fairly common at least on MB's? Most consumers don't need 10GB-E, but 2.5GB-E is a worthwhile upgrade (bad Intel implementations aside). Alas Synology won't get on board and still push out NAS with garbage 1GB-E ports and a lot of routers only have one 2.5GB-E port at best.

Yeah between that and 2.5GB-E was created just a few years ago as a response to the never-diminishing cost of 10GB-E. I bought what I think was the first 5-port unmanaged 2.5GB-E switch and checking prices today they're only like 10% cheaper, although I guess after inflation that's like 30% cheaper. But I am getting the impression that it's slowly replacing gigabit ethernet in the motherboard market.

 

[pressing on]

This technology seems to be already working for Ericsson. The Financial Times UK reported on December 5 that "...Ericsson boosted by $14bn ‘open RAN’ deal with AT&T. Swedish group triumphs over rival Nokia by being selected to help US giant roll out new kind of 5G network...".