Intel strikes back: Pat Gelsinger outlines Intel's future at Computex 2024

Bob O'Donnell

Posts: 88   +1
Staff member
In context: In an era of impressive innovation and intense competition in the semiconductor market, it's fascinating to observe how the major players have refined their strategies and messaging at this year's Computex show. Both in relation to major trends like generative AI and against their competition, the big tech companies have been working to position themselves in unique and differentiated ways.

For Intel, whose CEO Pat Gelsinger delivered the fifth major CEO keynote here, part of the task was to reassure the industry that the long-time leader was back in full force.

In his typical energetic fashion, Gelsinger laid out a wide-ranging vision for the company, offering deep dives on new server architectures (Xeon 6), AI Accelerators (Gaudi 3), and new Copilot+ AI PC architectures (next-gen Core Ultra, codenamed "Lunar Lake").

Along the way, he also highlighted the company's efforts to transform its foundry chip manufacturing business, both for Intel's own products and for other potential partners – many of whom also gave keynotes here at Computex.

The overriding theme linking these points was speed of execution. Gelsinger provided several examples of how the company had accelerated timetables on various products – particularly its Lunar Lake PC architecture – to offer proof that Intel had moved beyond its recent challenges and into a new era of tech leadership and innovation.

In truth, we're still years away from Intel regaining the kind of undisputed leadership position that the company once had – and which Gelsinger has been driving towards since returning to take over as CEO in 2021. However, this year's Computex keynote did feel like a turning point, with enough impressive products and technologies to make people start to believe the Intel of old is ready to strike back.

On the server and data center front, the company unveiled their Xeon 6 architectures, with the first of these coming out just six months after the fifth-generation Xeon. In line with other major computing architectures, Intel's Xeon now offers both efficiency and performance cores, each designed for different types of workloads.

The new Xeon 6 E-core product, codenamed "Sierra Forest," was officially launched here in Taiwan, focusing on the increasing data center challenge of power efficiency. Intel illustrated its improvements by showing how a few servers with the new chips could match the performance of a full rack of older Xeon-based servers. Intel also noted that while power efficiency is the key focus of Xeon 6 E-core, an upcoming version with 288 cores can tackle massive workloads as well.

As the name suggests, Xeon 6 P-core based CPUs, codenamed "Granite Rapids," are optimized for higher performance (and higher power) for the most demanding workloads. While initially expected to debut in 2025, Gelsinger commented that they're now going to be available by the end of this year.

Intel's Gaudi 3 – the company's GPU-based AI accelerator chip – has been announced and its performance capabilities discussed at previous events, but during the Computex keynote the company unveiled pricing compared to competitive Nvidia cards. Gaudi 2 is only one-third the price, and Gaudi 3 is two-thirds the price, emphasizing the point that they believe it to be a better value for organizations looking for alternatives.

The biggest news from Gelsinger's Intel keynote was the unveiling of Lunar Lake. The first Intel chip compatible with Microsoft's Copilot+ AI PC requirements, Lunar Lake features a new CPU architecture, a greatly enhanced Xe2 GPU, and an NPU with 48 TOPs. Equally, if not more important, Intel emphasized the power efficiency of the overall SoC.

As the company explained, the power efficiency of a chip isn't necessarily determined by its instruction set. In fact, Intel believes the x86-based Lunar Lake part will be able to compete on a power level with Qualcomm's Arm-based Snapdragon X products. Unfortunately, there are still no benchmarks available to prove this either way since neither part is officially shipping, but that's an area that will certainly get a great deal of attention once they both are.

Part of the reason Intel feels confident about the technology is the unique design of Lunar Lake. As Apple has done with their M Series processors for the Mac, Intel's Lunar Lake includes memory directly on chip. This offers the advantage of lower power consumption and faster access to memory, apparently playing a big factor in the chip's overall performance and efficiency.

In addition to product details, another big surprise regarding Lunar Lake is its availability, which is now expected this fall. As AMD did with their new Ryzen AI 300 chips, Intel dramatically moved up their release date to compete with Qualcomm's offerings.

The net result is we will have an amazing array of different Copilot+ AI PC systems available in the fall and holiday buying season. In fact, several PC makers will end up selling systems with all three different chips. How they will intelligently segment those different offerings – and not completely confuse potential buyers – remains to be seen, but it's going to be an exciting second half of the year for those in the PC market.

One challenge that Intel may face is the availability of the x86 and Intel NPU-specific versions of the Copilot+ software. Microsoft still needs to complete that work, and there is no firm date yet as to when that will be available. Interestingly, because Intel will be later to market than AMD, they may be able to avoid having Copilot+ "Ready" systems that need a software update before they have the full AI functionality, but we won't know for sure until a few months from now.

Another key factor to remember in all these developments is that despite some of Intel's recent challenges, they are still a behemoth in the computing (and particularly PC) industry with incredibly wide and deep roots across the entire supplier ecosystem. It's a point they've been emphasizing with all their partners in that ecosystem here in Taiwan, and it will end up influencing things like the number of systems available, total systems sold, and much more.

Intel also has a long history of laying out long-term roadmaps for their critical products, and during his keynote, Gelsinger did that for future PC architecture. Next year's Panther Lake chip, in particular, looks to be interesting as it builds on the improvements from Lunar Lake but will also be the first part manufactured in Intel's 18A process technology. This will mark what Intel believes will be a return to semiconductor manufacturing process and performance leadership.

While the final story on that remains to be told, the good news coming out of the Intel keynote is that competitive confidence is clearly back, inspiring the entire computing industry to raise their game.

Bob O'Donnell is the president and chief analyst of TECHnalysis Research, LLC, a market research firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow Bob on Twitter

Permalink to story:

 
I hope they wont release 500W processor for PC gamers . lol
With numbers like that, I guess they wont have any other choices...

Intel_Tech%20Tour%20TW_Next%20Gen%20E-core%20The%20Skymont%20Architecture-17.png
 
With numbers like that, I guess they wont have any other choices...

Intel_Tech%20Tour%20TW_Next%20Gen%20E-core%20The%20Skymont%20Architecture-17.png
@trparky @Burty117
I think that's for the E cores. One of their slides said 14% IPC improvement for the Lunar Lake P core vs Meteor lake. But if I'm reading the slide right, Intel might have clock speed regression for the high end as the new core scales better at lower TDP.

But final performance numbers are just speculative right now.
 
@trparky @Burty117
I think that's for the E cores. One of their slides said 14% IPC improvement for the Lunar Lake P core vs Meteor lake. But if I'm reading the slide right, Intel might have clock speed regression for the high end as the new core scales better at lower TDP.

But final performance numbers are just speculative right now.
The other thing to consider is the lost of Hyperthreading. To be honest, I feel I prefer Intel to start focusing on chip power efficiency because it has gone way out of control with Raptor Lake refresh being a prime example. You can surely win over hardcore PC enthusiasts, but those insane power requirement is a no go for 99.8% of the population.
 
The other thing to consider is the lost of Hyperthreading.
The way they explain things, it seems it won't be a loss at all. I'm ok with it. Frequently turn HT off anyway.
To be honest, I feel I prefer Intel to start focusing on chip power efficiency because it has gone way out of control with Raptor Lake refresh being a prime example. You can surely win over hardcore PC enthusiasts, but those insane power requirement is a no go for 99.8% of the population.
Agree with you on this point. The wattage/performance ratio for modern Intel CPU's is crazy.
 
The other thing to consider is the lost of Hyperthreading. To be honest, I feel I prefer Intel to start focusing on chip power efficiency because it has gone way out of control with Raptor Lake refresh being a prime example. You can surely win over hardcore PC enthusiasts, but those insane power requirement is a no go for 99.8% of the population.
The way they explain things, it seems it won't be a loss at all. I'm ok with it. Frequently turn HT off anyway.

Agree with you on this point. The wattage/performance ratio for modern Intel CPU's is crazy.
We don't know if the desktop version will disable HT or not, just that the mobile Lunar Lake will do it.

Even if the E cores are better now, losing 15% MT performance from the P cores is nothing to scoff at. (15% according to intel)
 
We don't know if the desktop version will disable HT or not, just that the mobile Lunar Lake will do it.

Even if the E cores are better now, losing 15% MT performance from the P cores is nothing to scoff at. (15% according to intel)
They explained elsewhere that the restructuring of the inner workings of the new core negate the effectiveness if HT and the improvements made up for it. So if that is correct, HT isn't effective like it once was and thus not needed. There should be no reason for these changes not to be shared with other CPU lines.

Who knows yet if these changes will be a good thing or a flash in the pan, but they are an interesting development. Benchmarks done objectively will tell the performance story.
 
Intel made the 2500K and then went to sleep.

They really, really thought they would be able to milk the market through 2500K derivatives ad infinitum and thought they would sell 2 Core CPUs mainstream and 4 Core CPUs to enthusiasts for the next 50 years.

Such socking lack of vision. Intel is made up of clowns. Bring back the 2010's Intel people.
 
Intel made the 2500K and then went to sleep.

They really, really thought they would be able to milk the market through 2500K derivatives ad infinitum and thought they would sell 2 Core CPUs mainstream and 4 Core CPUs to enthusiasts for the next 50 years.

Such socking lack of vision. Intel is made up of clowns. Bring back the 2010's Intel people.
Don't forget Core2Duo / Core2Quad, They were beasts back in their day and beat AMD Phenom.

CPU Release Date
Q6600 2007 (Q1)
2600K 2011 (Q1)
7700K 2017 (Q1)
8700K 2017 (Q4)

It literally took them over 10 years to move past 4 cores. All it did was give AMD time to build Ryzen and now, AMD actually beat them on performance and power effciency.

I've recently been spec'ing up servers for clients, I've started recommending AMD variants as the price difference is close to nothing, whilst getting a better performing CPU that eats less power.

The only thing that does frustrate me is the lack of Business Laptops and Desktops with AMD chips in, they do exist but not enough out there in my opinion. They have the better product, Ryzen has been proving itself since 2017 and has been the performance and effciency leader for the last few generations.

I find it strange that HPE produce Servers with the latest AMD CPU's but not their laptops or desktops. For example, HPE ProLiant DL385 Gen11 has the latest AMD 4th gen Epyc 9XX2 CPU's yet, checking the ProBook's and EliteBook laptops, None of them are using AMD's latest 8040 series CPU's whilst Intel's latest and "greatest" are always there.
 
Efficiency hasn't really been a goal at Intel recently. Their goal is to be performance king. Performance at all cost and power-boost the crap out of your PC. Run as hot as you can to squeeze every ounce of performance. It would take Qualcomm stealing some market share before Intel starts catering to customers that want better battery life. Intel can do it. It just hasn't been a priority for them.
 
Don't forget Core2Duo / Core2Quad, They were beasts back in their day and beat AMD Phenom.
Only if using software that was small enough to fit on L2 cache.

Practically Phenom offered much smoother experience, because it had IMC. Again I don't care what benchmarks said, I care about real world situations. At least one site noticed same: https://www.anandtech.com/show/2715/4

After playing through the several levels on each platform, we thought the Phenom II 940 offered a better overall gaming experience in this title than the Intel Q9550 based on smoother game play. It is difficult to quantify without a video capture, but player movement and weapon control just seemed to be more precise.
 
Only if using software that was small enough to fit on L2 cache.

Practically Phenom offered much smoother experience, because it had IMC. Again I don't care what benchmarks said, I care about real world situations. At least one site noticed same: https://www.anandtech.com/show/2715/4
You’re comparing a Phenom II? That came out years after the Q6600, the first gen Phenoms, which came out much closer to the Q6600, were worse, in everything, Phenom II was barely any better, years later.
 
I think Intel is making a MAJOR miscalculation building RAM onto the thing and not allowing expandable memory. At least the amounts are OK (16GB and 32GB), but, like, with most ARM systems having soldered memory, having expandable RAM is a major advantage of running an AMD or Intel system over ARM. Well, it was until Intel threw this advantage away. Will I buy a system without expandability? Yeah if it's VERY inexpensive, like one of those $100-$150 Chromebooks. As an Ubuntu user, I've run Ubuntu on ARM Chromebook before and it was a pleasant experience; and that was several years ago when the x86/x86-64 emulation existed but was nowhere near as mature as now, now it's even seamless if you want to run (x86/x86-64) Windows stuff under wine and steam. So my next notebook will likely be an ARM (after someone gets sick of their Windows Qualcomm system and wants to sell it cheap), but Lunar Lake not having expandable RAM makes it 100% chance it won't be Lunar Lake. My last notebook cost $300 and I found one with M.2 slot, space for a SATA drive, and a laptop DIMM slot. They shipped it with like a 256GB SSD and 4GB RAM, it now has a 1TB SSD, 1TB HDD, and 20GB RAM.

LOL my Dad was using a Dell with a Q6600 until... like 6 months ago? Ubuntu 22.04 on there. That poor machine -- he was loading, printing and editing multi 100 page documents full of charts and graphs; heavy amounts of web surfing, watching baseball among other videos on it, heavy scanning and weekly Zoom conferences. The power use was brutal, it'd peg out 2 out of 4 cores watching a 1080 video or doing Zoom, 55W usage (it's a 110W CPU). Firefox and Chrome are both multithreaded enough (and it's per-thread performance low enough) that it'd just peg out all 4 cores for a few moments loading up any significant web page. Now has a Coffee Lake, needless to say he's enjoyed having ~8x the RAM and ~10-12x the CPU performance.

The power use -- I have an 11th Gen Core i3-1115G4. And had a Ryzen 3450U immediately before that. The Ryzen was a 4C/8T and this is a 2C/4T. The per-core speed is impressive, overall the i3 gets about 2/3rds the performance (both CPU and GPU) of the Ryzen. But power use? I wondered why the fan was ramping up on this system, find out it's got like a 35W TDP. The Ryzen was 15! 2/3rds the performance using just over double the power is not particularly good. (Both these chips came out within a month of each other.)

 
Really makes AMD look like they're a bunch of wizards over there with their 16% increase in IPC.

Note AMD is doing that their gpus with skelleton staff, their new epic am5 chips & more while only a fraction of the size of inhell or invidia AMD is kicking *** ;)
 
The other thing to consider is the lost of Hyperthreading. To be honest, I feel I prefer Intel to start focusing on chip power efficiency because it has gone way out of control with Raptor Lake refresh being a prime example. You can surely win over hardcore PC enthusiasts, but those insane power requirement is a no go for 99.8% of the population.
Not to mention honest benchmarks now just show how pathetic intelligence home heaters lol
 
You’re comparing a Phenom II? That came out years after the Q6600, the first gen Phenoms, which came out much closer to the Q6600, were worse, in everything, Phenom II was barely any better, years later.
Phenom too. When Core 2 run out of L2 cache, then it have very slow memory access. Remember Celeron without L2 cache? It ran out of L2 cache, well, always, and every memory intensive task was painful to use.

No matter what benchmarks said, AMD CPUs with IMC offered much smoother experience because Core 2 L2 cache was too small to offset lack of IMC.
 
Phenom too. When Core 2 run out of L2 cache, then it have very slow memory access. Remember Celeron without L2 cache? It ran out of L2 cache, well, always, and every memory intensive task was painful to use.

No matter what benchmarks said, AMD CPUs with IMC offered much smoother experience because Core 2 L2 cache was too small to offset lack of IMC.
Look, I ain't going to argue with you, It was a long time ago, I don't remember it as clearly as you do, however, I do remember Crysis running better on an Intel chip vs an AMD chip. Could I be wrong? Probably, but the internet does seem to be backing me up here. AMD didn't lose market share so drastically for no reason, it was worse.

Besides, that wasn't the point of my post, my original post was to poke fun at Intel for basically doing nothing for 10 years and then is surprised by how well AMD not just caught up, but has overtaken them, not in sales, but does produce the better product.
 
Look, I ain't going to argue with you, It was a long time ago, I don't remember it as clearly as you do, however, I do remember Crysis running better on an Intel chip vs an AMD chip. Could I be wrong? Probably, but the internet does seem to be backing me up here. AMD didn't lose market share so drastically for no reason, it was worse.

Besides, that wasn't the point of my post, my original post was to poke fun at Intel for basically doing nothing for 10 years and then is surprised by how well AMD not just caught up, but has overtaken them, not in sales, but does produce the better product.

As you can see behind link I posted, https://www.anandtech.com/show/2715/4

Crysis Warhead, that uses same engine as Crysis do, runs smoother on AMD vs Core 2 Quad although benchmarks do not agree. Nehalem runs even better because it also has IMC like AMD does.

It's somewhat funny how Intel lost everything they were leading force just because, despite warnings, they thought 10nm process would work. First went process leadership and because future architectures were made for process that didn't work, als architectural leadership was gone.
 
Back