Intel Computex 2024 Keynote: The Best that was Yet to come is Here

[btarunr]

We are live from the Intel Computex 2024 Keynote address in Taipei. CEO Pat Gelsinger, along with the heads of the company's various business groups will take stage to unveil the latest in PC processors, server processors, generative AI accelerators, datacenter infrastructure gear, and lots more! In particular, we're interested in their Core Ultra "Lunar Lake" and "Arrow Lake" processors, and next-generation Xeon 6 chips. Can Gaudi 3 break into the duopoly between NVIDIA and AMD? Join us to find out in this live-blog!



02:55 UTC: Something really cool from the show: The camera is pointing at Intel's industry partners in the audience, and a Gaudi 3 is turning them into superhero images!


03:02 UTC: Pat Gelsinger takes center stage.


03:02 UTC: "Whatever has been done, can be outdone" -- Gordon Moore


03:04 UTC: 8286 from 1981. Moore's Law is alive and well "unlike what Jensen would have you believe"


03:05 UTC: "The AI era is like the Internet era 25 years ago."

03:07 UTC: Intel Foundry is the world's first systems foundry for the AI era.


03:08 UTC: AI is everywhere, and will be as pervasive as the Internet


03:09 UTC: There are 130 million Xeons running data-centers around the world right now--Gelsinger.


03:09 UTC: Xeon 6 with pure E-cores, based on Skymont cores. Intel 3 foundry node, 144 cores.


03:10 UTC: Significant rack-space savings.


03:12 UTC: A rack full of 28-core Xeon Scalable Gen 2 compares to just 2/5th of a rack with Xeon 6 E-core, also a 4.6x performance upgrade.


03:13 UTC: The rack and power savings are incredible thanks to E-cores.


03:14 UTC: Xeon 6 E-core with 288 cores! Coming later this year.


03:15 UTC: Xeon 6 has a sizable ecosystem.


03:16 UTC: Xeon 6 "Granite Rapids" pure P-cores, Intel 3 node.


03:17 UTC: Gaudi 2 is cost-effective and open.


03:18 UTC: Xeon 6 + Gaudi 2 running a medial LLM.


03:20 UTC: Pricing of Gaudi 2 and Gaudi 3


03:21 UTC: Gaudi 3 pricing is compelling.


03:22 UTC: More details on Gaudi 3. It's competitive with NVIDIA H100 at 2x the perf/$


03:23 UTC: Gaudi 3 pre-built servers showcased.


03:24 UTC: Inventec provides end-to-end baremetal servers powered by Xeon 6 and Gaudi 3.


03:25 UTC: Ultra Accelerator Link (UAL) and Ultra Ethernet Consortium. Intel is a founding member (AMD is, too).


03:28 UTC: 200 million CPUs and 90,000 edge deployments.


03:29 UTC: AI edge example: Samsung Medicine.


03:31 UTC: AI PC is the Centrino-like moment (productivity anywhere)


03:33 UTC: 8 million Core Ultra PCs shipped since December.


03:33 UTC: Over 500 PC design wins.


03:33 UTC: 1/3 of the above are Lunar Lake.


03:34 UTC: Satya Nadella looks exactly the same as he did from the AMD keynote.


03:34 UTC: Behold, the Intel Lunar Lake.


03:35 UTC: TSMC makes both chiplets on Lunar Lake.


03:36 UTC: Lunar Lake posts significant performance and efficiency gains.


03:37 UTC: 67 TOPS from the iGPU


03:38 UTC: Skymont is the star of the show as you'll soon see. Are you reading our Lunar Lake architecture article?

03:39 UTC: Jonney Shih of ASUS. It's never a Computex keynote without him!


03:41 UTC: Lunar Lake is a revolution—Shih of ASUS


03:44 UTC: Arrow Lake, desktop coming in Q4 (won't come out before October)!


03:45 UTC: Gelsinger holds up an Intel 18A wafer with Panther Lake dies. Coming to life in 2025.


03:46 UTC: Acer CEO Jason Chen says, people love the NPU that Meteor Lake offers.


03:48 UTC: Acer sees a trend change between Search to Ask.

03:48 UTC: Lunar Lake-powered Acer notebook.


03:50 UTC: Intel is powering AI across the industry.


03:51 UTC: The Intel Advantage—scale and trust.


03:54 UTC: And that's a wrap. Good show.

View at TechPowerUp Main Site

 

[Chaitanya]

More Ai nonsense(from everyone this year).

 

[Sabotaged_Enigma]

AI isn't something good. I just watched the start of the show, and there's a scene Moore talked which is generated by AI and with permission of his family as Pat says...
Dead people can talk now, huh? Real trouble is, if AI is in the wrong hand, fake edivence like that, say "withOUT permission", would cause real chaos.

"Whatever has been done, can be outdone" -- Gordon Moore

this sentence

 

[Noyand]

Lion cove is slower than Redwood cove in some cases ?

 

[Sabotaged_Enigma]

Huh, looks like a sitcom, well-designed conversation

 

[R0H1T]

Lion cove is slower than Redwood cove in some cases ?
View attachment 349998

It an happen in edge cases, like with lack of HT, what's more concerning for Intel is that at higher power it's losing efficiency! Upto(?) 10% faster near the end of the curve while IPC is 14% on avg I assume, so poor returns with that.

 

[Sabotaged_Enigma]

More Ai nonsense(from everyone this year).

At least they're showing AI helping medical which is good, by contrast Nvidia came up with some "AI-help-you-play-games" crap. I mean what kind of people are lazy enough to even find a helper to play their games for them...

 

[natr0n]

Artificial Intensifies

So no new consumer gpu to compete?

 

[Chaitanya]

At least they're showing AI helping medical which is good, by contrast Nvidia came up with some "AI-help-you-play-games" crap. I mean what kind of people are lazy enough to even find a helper to play their games for them...

There are companies out there doing drug discovery using Machine learning tools, and from early results it seems like quite a few pharma companies have started layoffs in drug discovery departments putting researchers out of jobs.

 

[usiname]

Summary of the new core:
Meteor Lake 10% less IPC compared to Raptor Lake
Lunar Lake 14% IPC over Meteor Lake => ~4% IPC over Raptor Lake
~10% lower clocks compared to Raptor Lake
Lack of HT

So the new core is slower than Raptor Lake in ST and MT?

 

[persondb]

Summary of the new core:
Meteor Lake 10% less IPC compared to Raptor Lake
Lunar Lake 14% IPC over Meteor Lake => ~4% IPC over Raptor Lake
~10% lower clocks compared to Raptor Lake
Lack of HT

So the new core is slower than Raptor Lake in ST and MT?

Redwood cove/Meteor Lake actually had greater IPC over Raptor Cove. It wasn't big, I think was 4% or so(I think?), and it was more than negated due to clock speeds.
So it's possible that it's more like 18% vs Raptor Cove.

 

[R0H1T]

Raptor cove? Don't remember about IPC numbers but since MTL never launched on desktop it wasn't discussed as much. And yes OCing your chips beyond reasonable levels resulting in massive(?) unstability overtime was a bad move!

 

[usiname]

Redwood cove/Meteor Lake actually had greater IPC over Raptor Cove. It wasn't big, I think was 4% or so(I think?), and it was more than negated due to clock speeds.
So it's possible that it's more like 18% vs Raptor Cove.

No, the IPC is lower

Intel Meteor Lake P-cores Show IPC Regression Over Raptor Lake?

Intel Core Ultra "Meteor Lake" mobile processor may be the the company's most efficient, but isn't a generation ahead of the 13th Gen Core "Raptor Lake" mobile processors in terms of performance. This isn't just because it has an overall lower CPU core count in its H-segment of SKUs, but also...

www.techpowerup.com

 

[john_]

This is probably the first time in the last few(many?) years that I feel that Intel has something substantial in it's hands and not just a bunch of marketing slides.

 

[ncrs]

Phoronix has benchmarked two of the new 144-core Xeons with E-cores.
Significantly improved power efficiency compared to previous Xeons, but only in tests not using AVX-512 or AMX.
I'd call it a solution for lighter server workloads.
Unfortunately this is a staggered launch:

Today it's the Xeon 6700E series launching for up to 144 E cores. Next quarter will be the Intel Xeon 6900P launch for the top-end Granite Rapids processors. Meanwhile it won't be until Q1'2025 of seeing the rest of the Xeon 6 family -- including the Xeon 6900E series where there can be up to 288 E cores per socket. That Q1 target is also when seeing the lower-tier Xeon 6300P / 6500P / 6700P parts.

 

[theouto]

03:51 UTC: The Intel Advantake

Small detail, but I spotted a typo

 

[Daven]

No mention of Arrow Lake?

 

[phanbuey]

It an happen in edge cases, like with lack of HT, what's more concerning for Intel is that at higher power it's losing efficiency! Upto(?) 10% faster near the end of the curve while IPC is 14% on avg I assume, so poor returns with that.

Supposedly they’re setting up for rentable units - which will be faster than ht but not ready yet.

 

[mouacyk]

cringatars

 

[InVasMani]

Supposedly they’re setting up for rentable units - which will be faster than ht but not ready yet.

It's a piano roll music sequencer approach to threading. The diagram even looks like a music sequencer Synthesia or rotate that and you've got FL Studio. Remains to be see how it compares to the old approach, but in terms of programmable flexibility of different event type threads it's a very proven methodology that works great at a high intricacy level. Seems like they reshuffled the deck a bit on HT approach, but it should be for the better and more in line with like G-Sync/Freesync/Reflex technologies for example and of course MIDI setting the gold standard in event timing technology.