Intel Plans To IPO Programmable Chip Unit Within Three Years

An anonymous reader quotes a report from CNBC: Intel said it will treat its programmable chip unit as as a standalone business, with an aim to spin it out through an IPO in the next two to three years. Intel's Programmable Solutions Group will have its own balance sheet as it heads toward independence. The company will continue to support the business and retain a majority stake, and could also seek private investment. Sandra Rivera, who leads Intel's broader Data Center and AI group, will become PSG CEO. Intel will manufacture the group's chips.

The move also highlights the strong demand in the semiconductor industry for field programmable gate arrays, or FPGAs. [...] FPGAs are simpler than the powerful processors at the heart of servers and PCs but are often more flexible, respond faster and can be more power-efficient. They're "programmed" after they're shipped for specific uses in data centers, telecommunications, video encoding, aviation and other industries. FPGAs can also be used to run some artificial intelligence algorithms.

Intel's FPGAs are sold under the Agilex brand. Intel doesn't break out PSG sales yet, but said in July that the unit had three record quarters in a row, offsetting a slump in server chip sales. PSG has been part of Intel's Data Center and AI group, which generated $4 billion in sales in the second quarter.

Re:

[Comboman]

Intel is too big to fail. The US will go to war with China before letting domestic chip makers go under.

Re:

[boulat]

Your logic is both lacking and laughable.

TSMC is in Taiwan and is not domestic. Also fun fact: its not Intel.

There is also AMD in case you are still confused.

Re:Funny stuff

[ThurstonMoore]

AMD doesn't make chips they design them. Intel actually makes physical chips.

Re:

[Comboman]

I was not talking about TSMC or Taiwan. I was talking about the US subsidizing it's domestic chip makers and at the same time using sanctions and other tactics to prevent China from building advanced 7mm and 5mm chip fabs (which is a source of increasing tension between the two nations). It is you who are confused.

Re:

[boulat]

Ohh.. I'm sorry I thought we were in 2023 where we have 7 nanometer chips, but you are still in the mm land.

FYI China already has 8nm chip via Rockchip RK3588, and 5-7nm in development.

Its not rocket surgery, they will get there either by stealing or reverse-engineering.

Fool is the one that goes to war over chips that barely give you 20% improvement over existing (and much cheaper) technology.

Re:

[Comboman]

Never said I was fool enough to risk war over it; I said the US was fool enough to risk war over it. They've gone to war over much less.

Re:

[gweihir]

While they are clearly dying, that process may take a few decades.

Re:

[TomGreenhaw]

LOL, I'll take everyone's bets. Intel's upcoming Meteor Lake generation will leapfrog everybody in the server business. Processors with 256 cores and 64GB of high speed memory on one device will be a game changer in the datacenter.

Intel's vast foundation of inventions backed up by patents alone secure their long term future. They are one of the few companies that design and fabricates devices.

They have made great strides in process engineering. The number of nanometers in a process is like the old schoo

What about Altera?

[dskoll]

Doesn't Intel own Altera? Why does it need two FPGA businesses? Or is Agilex just a rebranding of Altera?

Re:

[ArchieBunker]

The suits have determined this is the best way to extract the most amount of profit.

intel is run by fucking idiots

[Anonymous Coward]

Intel buys Altera. AMD goes, oh shit, we better buy Xilinx. Xilinx integration goes great and AMD now has EPYC CPUs with FPGAs. Intel apparently had its tiny, also-ran division which also makes FPGAs swallow up the Altera acquisition and botch it completely. Intel gives up on FPGAs and gets ready to spin the whole kit and kaboodle off for less than it paid. And of course, to reiterate what I said, AMD now has an even more compelling product.

You see this shit happen at other companies like MSFT, where the ex

Re:

[jacks smirking reven]

Agilex was Intel's house brand for FPGAs so when they bought Altera the Altera name was dropped and they were folded into Agilex.

Re:

[ShanghaiBill]

Yes, Intel acquired Altera in 2015.

AMD responded by acquiring Xilinx.

Intel has now realized the acquisition was a mistake and is spinning it out again.

Back in 2015, many were assuming that integrating FPGA fabric into CPUs would be the Next Big Thing. That assumption was incorrect. There is no mass market for CPUs with integrated programmable logic.

The new assumption is that the mass market wants tensor processors, which is why Nvidia's market cap is seven times Intel's.

Re:

[larryjoe]

Yes, Intel acquired Altera in 2015.

AMD responded by acquiring Xilinx.

Intel has now realized the acquisition was a mistake and is spinning it out again.

Back in 2015, many were assuming that integrating FPGA fabric into CPUs would be the Next Big Thing. That assumption was incorrect. There is no mass market for CPUs with integrated programmable logic.

The new assumption is that the mass market wants tensor processors, which is why Nvidia's market cap is seven times Intel's.

The new assumption is that the mass market wants tensor processors, which is why Nvidia's market cap is seven times Intel's.

No need for an assumption about the current mass market for AI processors, as Nvidia's annualized data center revenue is already $40 billion. The big question is whether this number (for Nvidia and the entire market) is sustainable, and if so, for how long. The answer to this question will largely depend on the marketization of killer apps based on AI.

We got to our current situation because the market continued AI research ten years ago and developed many AI improvements to existing apps. These improveme

short answer: it's not sustainable

[Anonymous Coward]

nVidia is already finding that demand is less than supply for its H100 chips. Like with anything else (e.g. the 4090 for gaming consumers) there is a certain amount of demand for chips that will pay almost any price for the best but that's only a small part of the market.

The other issue with generative AI is that it is very hard to maintain a market position because there's no secret sauce in there. Any other company can get similar results by investing in training and then the price can be undercut. All th

Re:

[mcgett]

Intel didn't have an internal develop FPGA (well not since the very early 1990's). Altera was folding into Intel as Intel Programmable Solutions Group (PSG) and Agilex is the next generation FPGA post Stratix 10.

Re:

[DrMrLordX]

They bought Altera, presumably to integrate FPGAs into their Xeon products. Unless they sign a cross-licensing deal with their independent unit, that dream will be dead. This new unit will essentially be what remains of Altera.

Re:What about Altera?

[flashpoint31415]

According to this article , the Agilex line is the new generation after the Stratix, which was legacy Altera. I would guess that, since the 2015 acquisition of Altera, Intel has poured cash into integrating or updating FPGAs to a smaller feature size. This plan to spin off what appears to be the Altera assets (no mention of the Stratix, Arria, or Cyclone families) suggests to me that Intel 1) couldn't figure out how to integrate FPGAs into their processors; 2) couldn't figure out how to get the FPGAs to play nice with UEFI; or 3) didn't see enough profits in keeping them in the fold.

Here's hoping the Agilex, Stratix, Arria, and Cyclone families stay together... a family that stays together, programs together.

Re:

[flashpoint31415]

need the link - https://www.anandtech.com/show... [anandtech.com]

Re:

[mcgett]

Intel doesn't have two FPGA businesses. Intel bought Altera in December 2015 and that became Intel Programmable Solutions Group (PSG) and Intel PSG is being spun out as a separate company. Agilex isn't rebrand of Altera, it is the next generation FPGA family name for post Stratix 10.

Re:Marriage of the best of two worlds.

[dskoll]

A CPU with embedded FGA fabric is cool. They already exist; you can get high-end FPGAs from Xilinx and Intel with embedded ARM cores. I don't know about FPGA fabric on a really high-performance CPU, though... I doubt that currently exists.

I don't think adding FPGA fabric to a CPU core would mean it's the "last CPU they'd ever design" because FPGAs run at much lower clock speeds than typical CPUs. Also, reconfiguring an FPGA takes quite a bit of time, on the order of hundreds of milliseconds to load a new bitstream, so it's not something you could reconfigure each time you switch to a new process, for example. We'd still need new high-end CPU designs.

Re:

[Guspaz]

Considering how comically overinflated Intel's FPGA pricing is compared to new competitors like Efinix (who will sell to smell and medium sized customers at like 10% of the price of equivalent Intel FPGAs), I shudder to think how much an Intel CPU with a big enough FPGA to be useful would cost.

Intel's FPGA prices aren't terrible if you're a huge company buying in massive quantities, but if you're only trying to buy a few thousand FPGAs, they're going to charge you multiple times if not orders of magnitude m