Wednesday, August 14, 2024

Intel -- Lunar Lake -- Core Ultra 200V

Locator: 48417INTEL.

Original Post
August 14, 2024

September 3, 2024: Lunar Lake now shipping. Re-named: Core Ultra 200V.

Intel's Lunar Lake, wiki, August 14, 2024.

Intel's Lunar Lake changes everything about X86. Forbes, July 3, 2024.

Intel sells stake in chip designer Arm Holdings, Yahoo!Finance, August 14, 2024.

Could Intel be booted from the Dow? Asking for a friend, August 14, 2024.

From Forbes, wow, this sounds like Apple's M4 -- SoC --

This year’s Computex in Taiwan, arguably the largest and most influential PC trade show in the world, saw Intel unveil more details about its Lunar Lake processor during CEO Pat Gelsinger’s keynote. I was personally surprised that Intel didn’t lead with Lunar Lake and PCs during Gelsinger’s presentation, considering that Computex is predominantly a PC-oriented show. Instead, it led with data center and server AI products, which I believe was a mistake because most people’s focus at the show is on PCs. But it seems clear there was an executive decision on Intel’s part to talk up its competition with NVIDIA.

Just before Computex, I had the opportunity to attend Intel’s Tech Tour in Taiwan, the company’s third such architectural briefing series around the world. The purpose of these tours is to showcase Intel’s latest innovations and shine a light on the local Intel teams that make its global success possible. Last year’s Meteor Lake processor briefing was held in Penang, Malaysia, where the company does a lot of its packaging and testing. Taiwan is strategically important for Intel because it is the heart of the PC ecosystem, and all PC vendors have a significant presence there.

Lunar Lake is Intel’s best bet to compete with the rising threat of Arm-based architectures, especially those coming from Apple and Qualcomm, which deliver both high performance and efficiency for laptops. While AMD also offers some competition, its AI processors have been positioned more for high performance than low power consumption. Lunar Lake’s specs and design also align with Microsoft’s vision of what makes a Copilot+ PC.

To accomplish all this, Lunar Lake is not just another incremental improvement over Intel’s previous designs; it is a radical departure for the traditional x86 paradigm that has dominated the PC industry for decades. At a platform level, Lunar Lake is an evolution from last generation’s Meteor Lake with further improvements to performance and efficiency and a continuation of its advanced packaging design. This includes the introduction of on-package memory.

One of Intel’s most important ingredients in accomplishing this task was radically changing how it implemented its different types of CPU cores. Lunar Lake is a significant departure because it focuses on power efficiency and does so by prioritizing a cluster of four Skymont efficiency cores (sometimes called E-cores) over Lion Cove performance cores (P-cores) for most tasks. By isolating the efficiency cores into a four-core cluster, Intel is also able to keep as many workloads as possible on the smaller, more efficient cores before lighting up the bigger, more power-hungry performance cores.

One of Intel’s engineering achievements for Lunar Lake is that in certain workloads Skymont’s single-threaded performance is anywhere from 38% to 68% faster than the last generation. Not only that, but the new Skymont cores are much more dynamic, delivering the same performance at one-third the power, 70% greater performance at the same power or even 2x the performance at marginally higher power. The four-core cluster can deliver up to 2.9x the performance of two of Meteor Lake’s E-cores at the same power.

These new E-cores are so performant that they are effectively on par with earlier Raptor Cove performance cores found in Intel’s 13th and 14th Gen desktop CPUs. The Lion Cove P-cores also deliver a 14% uplift over their Redwood Cove counterparts in Meteor Lake, making them the fastest cores in the CPU. Intel has architected these cores’ power and performance curves to overlap perfectly, maximizing the performance and power efficiency of the SoC. The next generation of Thread Director is designed to make sure that workloads are appropriately sent to the right cores, optimizing for efficiency and prioritizing the E-cores for the first time ever.

Lunar Lake is Intel’s first implementation of its second-generation GPU architecture, dubbed Xe2. This is the same GPU architecture planned for the company’s much-anticipated discrete Battlemage GPUs that are expected to ship soon. Intel says that the new Xe2 architecture is 50% faster than its predecessor inside of Meteor Lake and features new XMX engines, which increase AI performance to 67 TOPS. Intel has also spent the last year working on improving graphics drivers for its Xe family of GPUs, which means that the Xe2 architecture will come out of the gate much more optimized for most games, including older titles.

This, paired with the expected 48 TOPS from the NPU on the same SoC, delivers well over 100 TOPS of platform AI performance, which makes for a nice statistic but will almost never be possible to utilize all at once. The important part is that the NPU and GPU are each quite performant and can deliver high-end AI experiences that developers want their users to have. What’s more important is how efficiently the GPU and NPU deliver this level of performance—and how that preserves battery life. Intel says that its fourth-generation NPU is twice as performant at the same power as the last generation and capable of four times as much peak performance.

Much, much more at the link. 

By the way, articles like this in Forbes were the reason I finally subscribed to Forbes

List of Intel CPU microarchitectures.

I have no idea why wiki does not list Intel Lunar Lake at this page.

Meteor Lake.

Comment: personal investing --

  • I have no positions in Intel --
  • I'm overweight in tech, right now --
  • I have positions in seven incredibly good tech companies and don't see the need for any more --
  • I would not replace any of my tech positions with Intel --
  • but, having said all that, I'm going to be watching Intel like a hawk for the next twelve months and if there is any hint that Intel starts to move like I think it could -- I'll be the first to start a new position --
  • if I rode Intel down to $20 with the recent sell-off, I might:
    • hold, hoping for better times; or,
    • sell it all, take the tax loss harvest, and then buy back in 30 days
***********************************
Disclaimer Briefly 
Reminder
  • I am inappropriately exuberant about the US economy and the US market, 
  • I am also inappropriately exuberant about all things Apple. 
  • See disclaimer. This is not an investment site. 
  • Disclaimer: this is not an investment site. Do not make any investment, financial, job, career, travel, or relationship decisions based on what you read here or think you may have read here. All my posts are done quickly: there will be content and typographical errors. If anything on any of my posts is important to you, go to the source. If/when I find typographical / content errors, I will correct them.
  • Reminder: I am inappropriately exuberant about the US economy and the US market, 
  • I am also inappropriately exuberant about all things Apple. 

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.