Launch event for Core Ultra and 5th Gen Xeon CPUs. 'AI is everywhere' says Intel

[filpo]

Summary

Intel announced their launch event 'AI is everywhere' for the release of the Meteor Lake CPUs (only for laptops) and the 5th Gen Xeon CPUs for servers and databases

 

Quotes

Quote

On December 14th, Intel is set to unveil two product series, as announced by the company in a recent media alert. Intel has officially confirmed an upcoming livestream scheduled for this day, during which they will introduce the Intel Core Ultra CPUs, codenamed Meteor Lake, and the 5th Gen Xeon Scalable processors, known as Emerald Rapids

 

Intel’s pre-launch presentation and teasers focused on AI-driven experiences in today’s operating systems and applications that can harness machine learning computing. Such AI enhancements include improved video conferencing, improved realism in in-game 3D animations, real-time transcription and translation for gaming, and AI assistance, all contributing to enhanced multi-app productivity and audio-visual capabilities

 

Quote

With a clear focus on AI, this launch aims to showcase how Intel’s new CPUs can enhance AI workloads. The utilization of Intel’s 4 process technology in Meteor Lake, coupled with a disaggregated CPU, GPU, and AI-dedicated VPU design, underscores their commitment to AI support

 

Quote

Join Intel Chief Executive Officer Pat Gelsinger and other Intel leaders for Intel’s “AI Everywhere” event. Marking a momentous year for the company, Intel will accelerate its execution engine and power AI workloads across the data center, the cloud and the edge with the launch of 5th Gen Intel® Xeon® processors and Intel® Core™ Ultra processors.

When: Dec. 14 at 10 a.m. EST (7 a.m. PST)

Where: Live from Nasdaq in New York

 

My thoughts

I reckon the further introduction into AI for CPUs is a good thing but that we probably shouldn't rely on it too much. For example, in gaming we feel we NEED to use upscaling for certain games such as Remnant II and Avatar: Frontiers of Pandora. I hope that just an AI core or thread bump for the consumer chips isn't everything we get on both the meteor lake and new Xeon cpus since that'd be pretty lukewarm and a quite typical way to end the year, considering all the game launches we had this year (looking at you Cities: Skylines II). Alas all we can do is hope now

 

Sources

Intel announces December 14 "AI Everywhere" launch event for Core Ultra "Meteor Lake" and 5th Gen Xeon CPUs - VideoCardz.com

Intel Will Unveil Meteor Lake, Emerald Rapids CPUs On December 14 | Tom's Hardware (tomshardware.com)

[jaslion]

34 minutes ago, filpo said:

I hope that just an AI core or thread bump for the consumer chips isn't everything we get on both the meteor lake and new Xeon cpus since that'd be pretty lukewarm and a quite typical way to end the year, considering all the game launches we had this year (looking at you Cities: Skylines II). Alas all we can do is hope now

Don't worry ai stuff is gonna stay in the big boi pants market a bit longer Thats where the money is at after all (and potentially where they will maybe find some use for it because as of now a pretty low end gpu cooks it with the only advantage being that these can use a lotta ram).

 

If it does find footing well. Amd get your shit togheter, get the ai cores on your gpu to do something and well make some apu abomination of a threadripper to go ham

 

Probably just another core bump as for now with some frequency. Not much else to do really since we are only nearing mid lifecycle of this console generation and really seeing that we do not need much more cores than an 8c/16t cpu except for some exceptions like every single Paradox game (no shade at all here).

 

 

[GOTSpectrum]

Honestly, for the Xeon stuff I'm really interested to see how they stack up to AMD's EPYC offerings... I have a feeling they will be lacking in cores against AMD's offering but we will see 

[jaslion]

9 minutes ago, GOTSpectrum said:

Honestly, for the Xeon stuff I'm really interested to see how they stack up to AMD's EPYC offerings... I have a feeling they will be lacking in cores against AMD's offering but we will see 

I do think amds gonna have them beat in terms of single core and multi core for the price once more if the trend stays true. Amd is still more energy efficient in the IPC/watt category in the consumer space especially when slowed down so unless intel really got their act together I feel like this will be another buy intel because of long trackrecord of it work good purchase mainly whilst amd and different arm cpu makers keep eating into their marketshare

[porina]

1 hour ago, filpo said:

I hope that just an AI core or thread bump for the consumer chips isn't everything we get on both the meteor lake and new Xeon cpus since that'd be pretty lukewarm and a quite typical way to end the year, considering all the game launches we had this year (looking at you Cities: Skylines II). Alas all we can do is hope now

A lot of official info has already been released about Meteor Lake short of actual product configurations. It will have Intel's newest architecture, main cores will be on Intel 4 process. It gets Arc GPU on N5 which should get it closer to AMD's best APUs. You're not getting threads though, as leaks/rumours suggest it only goes up to 6 P-cores which may be in part why there will not be desktop offerings. The NPU follows AMD's release of one in specific CPUs. It's going to be a more power optimised unit than using a dGPU which could still be used for more heavy lifting. But you're not getting dGPUs in thin laptops which is where it is more targeted. Hardware needs to get out there to enable software support. One part of the AMD demo previously was about removing the background from the video on a zoom call, and how efficient it was compare to throwing a dGPU on it, especially if you're running on battery.

 

Main Xeon side gets a newer core (but not as new as Meteor Lake), more cache and faster memory controller. Server stuff historically lags consumer by about a generation so Sapphire to Emerald is like Alder to Raptor. Hopefully it gets the updated 7 process too as Sapphire was on the older one at launch. Wonder if it is also time for Sierra Forrest, which is the E-core only model going for core count. Supposed to ship 1H 2024 so announcing December could fit with that in best case.

 

Given the December date, the launch is likely to be Intel saying they shipped on time this year with the real volume to follow later on 2024.

[leadeater]

36 minutes ago, jaslion said:

I do think amds gonna have them beat in terms of single core and multi core for the price once more if the trend stays true. Amd is still more energy efficient in the IPC/watt category in the consumer space especially when slowed down so unless intel really got their act together I feel like this will be another buy intel because of long trackrecord of it work good purchase mainly whilst amd and different arm cpu makers keep eating into their marketshare

The only thing saving Intel right now in the server market are two things, lack of supply of EPYCs and hardware acceleration paths on some Intel options people do actually use and want.

 

https://www.tomshardware.com/news/amd-will-hold-20-of-server-cpu-market-in-2023-analysts-say

https://wccftech.com/amd-grabs-over-30-percent-market-share-with-epyc-cpus-intel-continues-to-decline/

[starsmine]

December is such a weird launch point for mobile only chips. 

To late for Q1 purchases because OEMs need to roll out models, to late for Back to school purchases. 

Just seems like a missed launch window. But its still going to be interesting. See if it still has wind in its sales for back to school 2024 when new zen 5 AMD laptops will (hopefully) be out it has to compete with

[Just A Name]

6 hours ago, starsmine said:

to late for Back to school purchases

Also, it could be too late for Christmas if OEMs don't get their hands early to build machines.

10 hours ago, filpo said:

I reckon the further introduction into AI for CPUs is a good thing but that we probably shouldn't rely on it too much

 

9 hours ago, jaslion said:

Don't worry ai stuff is gonna stay in the big boi pants market a bit longer

Judging what Microsoft is doing AI-wise on Windows 11, I have a bad feeling that the new Core series will have AI cores built-in

[porina]

5 minutes ago, Just A Name said:

Judging what Microsoft is doing AI-wise on Windows 11, I have a bad feeling that the new Core series will have AI cores built-in

Meteor Lake has an NPU, which appears as a separate device in Task Manager, like the GPU is separate from CPU. AMD got there first though with a release in some CPUs.

[jaslion]

9 hours ago, Just A Name said:

Also, it could be too late for Christmas if OEMs don't get their hands early to build machines.

 

Judging what Microsoft is doing AI-wise on Windows 11, I have a bad feeling that the new Core series will have AI cores built-in

I mean as in the actual learning and stuff. Local processing is also just noy happening as its all cloud based since as of right now you need WAY too much processing power to even have a basic ok experience with a language modrl

[RejZoR]

Ai this and Ai that and computer controlled enemies are still acting like drunk baboons in almost all games. Like, you'd think this would be the first thing to apply "Ai" to, to an actual "Ai". Yet everyone is raging over Ai generated porn or something. Priorities!

[Kisai]

On 11/1/2023 at 12:58 AM, porina said:

Meteor Lake has an NPU, which appears as a separate device in Task Manager, like the GPU is separate from CPU. AMD got there first though with a release in some CPUs.

That's the first NPU part Intel put in their CPU's. I don't think anything uses it. Even trying to find the driver for it, required getting it from a different intel product driver package.

https://docs.openvino.ai/2023.1/openvino_docs_OV_UG_supported_plugins_GNA.html

Quote

GNA Device

 

The Intel® Gaussian & Neural Accelerator (GNA) is a low-power neural coprocessor for continuous inference at the edge.

Intel® GNA is not intended to replace typical inference devices such as the CPU and GPU. It is designed for offloading continuous inference workloads including but not limited to noise reduction or speech recognition to save power and free CPU resources. It lets you run inference on Intel® GNA, as well as the CPU, in the software execution mode. 

 

It's been available since the Intel 10th gen. I'm not aware of any software that uses it that someone might actually use, and the implication is that Windows speech recognition might actually use it (in a Siri/Alexa manner.) It's a 400Mhz processor in the Intel CPU. So it could run while the OS has spun down the P cores or something on the 12/13/14th gen.

 

[porina]

13 minutes ago, Kisai said:

That's the first NPU part Intel put in their CPU's. I don't think anything uses it.

Curious. I looked on my 11700k system and I don't have GNA present. I'll check bios later in case it needs to be explicitly enabled. I don't have any items showing with missing drivers in Device Manager. I was wondering if perhaps it was mobile side only, which wouldn't be the case if 11700k includes it.

 

13 minutes ago, Kisai said:

I'm not aware of any software that uses it that someone might actually use, and the implication is that Windows speech recognition might actually use it (in a Siri/Alexa manner.)

From memory the AMD AI demo used it during a zoom video call to do background blurring.

[igormp]

On 11/1/2023 at 2:06 PM, jaslion said:

I mean as in the actual learning and stuff. Local processing is also just noy happening as its all cloud based since as of right now you need WAY too much processing power to even have a basic ok experience with a language modrl

Consumer devices are not meant to do the "learning" part. Training is done of big boy places as you mentioned, but the product of those training rounds are what are expected to be run on consumer devices (after optimization, ofc).

Those small neural units are really helpful for local inferences, best example is the gallery stuff in both android and ios that can detect specific people, objects and whatnot.

With that said, language models are not going to be run locally in such devices the near future indeed, but AI is not limited to that kind of application.

50 minutes ago, RejZoR said:

Ai this and Ai that and computer controlled enemies are still acting like drunk baboons in almost all games. Like, you'd think this would be the first thing to apply "Ai" to, to an actual "Ai". Yet everyone is raging over Ai generated porn or something. Priorities!

The amount of people that use their computers for excel or social media browsing far outweighs the amount of people that use it for games, so yeah, priorities.

12 minutes ago, Kisai said:

That's the first NPU part Intel put in their CPU's. I don't think anything uses it. Even trying to find the driver for it, required getting it from a different intel product driver package.

https://docs.openvino.ai/2023.1/openvino_docs_OV_UG_supported_plugins_GNA.html

 

It's been available since the Intel 10th gen. I'm not aware of any software that uses it that someone might actually use, and the implication is that Windows speech recognition might actually use it (in a Siri/Alexa manner.) It's a 400Mhz processor in the Intel CPU. So it could run while the OS has spun down the P cores or something on the 12/13/14th gen.

 

I wouldn't call the GNA a NPU since it's not general purpose, it can only be used for audio stuff and has an awful SDK with barely no docs. IIRC, MS made use of it on skype or teams, can't remember which, but it has a really limited adoption (again, not like any dev can easily make use of it due to the lack of a proper API).

[porina]

4 minutes ago, igormp said:

I wouldn't call the GNA a NPU since it's not general purpose

Looking at the openvino link earlier, it does list GNA as a separate category from NPU, so that would agree.

[porina]

16 minutes ago, porina said:

Curious. I looked on my 11700k system and I don't have GNA present. I'll check bios later in case it needs to be explicitly enabled.

Found it, default disabled in bios. Turned it on and there it appeared in Device Manager. I don't get NPU tab in Task Manager. So it doesn't seem to fill that exact role even if there is overlap. 

[jaslion]

22 minutes ago, igormp said:

best example is the gallery stuff in both android and ios that can detect specific people, objects and whatnot.

Thats still cloud based for now. Some basic functions are local but proper analysis is cloud based.

 

Really does depend where the industry goes. If full cloud ever becomea a thing or if local with cloud acceleration becomes the main thing (or well stays as its quite there already)

 

[igormp]

2 hours ago, jaslion said:

Thats still cloud based for now. Some basic functions are local but proper analysis is cloud based.

Most of that is actually done locally on both android and ios:

https://machinelearning.apple.com/research/recognizing-people-photos

https://research.samsung.com/blog/Advancing-Privacy-Preserving-Techniques-for-Machine-Learning

 

Image classification is pretty much a solved problem and we managed to make it really optimized, even a simple mcu like an esp32 can run some simpler models.

[leadeater]

7 hours ago, igormp said:

Image classification is pretty much a solved problem

Could websites please stop asking which images have palm trees then please lol

[starsmine]

31 minutes ago, leadeater said:

Could websites please stop asking which images have palm trees then please lol

My VPN is on by default. So EVERY  time I google. I get asked about which squares contain a bike, and I, as a human, do not know if they want me to include the bit of tire that is 10 pixels into another square. 

so every single time. I do like 4 in a row, and it goes, NOPE not proven human enough.

At this point I think the captchas are harder for me then a computer. 
What counts as part of stairs? the cement under the steps? 
Does the person obstructing the seat of a motorcycle count as a motorcycle?
Does the exaust pipe in the bottom right square count as part of the bus when none of the body is in the square?

Ill split tunnel apps that dont like VPNs and I dont care. But default on for everthing else is how I want it to be so I dont to have my ISP getting letters because I turned on my torrent client for 10 seconds once a week and forgot the VPN. 

[igormp]

2 hours ago, leadeater said:

Could websites please stop asking which images have palm trees then please lol

At this point I guess they evaluate how bad humans are and see your margin of error, because machines are pretty much able to solve those without much issues lol

[williamcll]

AI anywhere either implies either a NPU designed for network use or low power for edge users (uncommon for Intel since this is mostly dominated by ARM-based smart devices).

 

They must keep the power draw low if they want any competition.