AMD Zen CPUs "IPC" uplifts (2017 - 2024)
https://www.extremetech.com/computing/c ... b-globallyChina-based Semiconductor Manufacturing International Corporation (SMIC) has become the world's third-largest contract chip manufacturer, according to an industry research firm. SMIC's ascension has seen it join US-based GlobalFoundries in the competition for market share in the collection of firms that make chips for "fabless" companies such as AMD, Nvidia, Apple, and others.
The reordering of the global rankings of contract foundries such as SMIC comes from Counterpoint Research, which just released its industry report for the first quarter of 2024. In the most recent quarter, SMIC's global market share rose one percentage point from the previous quarter to 6%, allowing it to join GlobalFoundries and topple Taiwan's UMC, according to CNBC. Although SMIC moved up on this list during the most recent quarter, it is still far behind industry heavyweights TSMC and Samsung Foundry.
Foundry market share
https://www.extremetech.com/computing/t ... intel-cpusIntel has already revealed its next-gen mobile architecture known as Lunar Lake, and in a first for Intel, its compute tile is made by TSMC on its 3nm process. Intel will also be tapping its biggest rival for its upcoming Arrow Lake desktop processors as well, and now a report from Taiwan states TSMC has begun production on the 3nm chips that will be used for both of Intel's upcoming platforms.
TSMC has been cranking out 3nm chips for a little over a year now, but its initial run was reserved for Apple's A17 Pro chip for the iPhone as well as its M3 and M4 processors. Now TSMC is turning its attention to Intel, which will be using its 3nm process for both Lunar Lake and Arrow Lake. As Techpowerup notes, although we know Lunar Lake's compute tile is made on TSMC's 3nm process, it's unclear at this time how Intel will be using these chips for Arrow Lake. It is a huge deal for Intel to turn to a competing foundry to make its CPUs, so it's unclear if Lunar Lake is just a one-off deal. Plus, Arrow Lake has been touted as the first Intel CPUs to use its cutting-edge Intel 20A process.
https://www.extremetech.com/gaming/inte ... pus-report
After many delays, Intel launched its first-generation Arc Alchemist GPUs in late 2022, and they were still somewhat behind the 8-ball even at launch. Intel launched its cards using the mature TSMC 6nm node, whereas rivals AMD and Nvidia were already one node ahead, using TSMC 5nm for Ada Lovelace and RDNA 3 GPUs. Intel appears to have learned a lesson for its follow-up architecture codenamed Battlemage. It will use TSMC 4nm, which is roughly the same node as Nvidia's next-gen graphics and likely the same as AMD's upcoming RDNA 4.
Word of AMD's plans comes from a paywalled article on DigiTimes and flagged by TechPowerUp. The article notes Intel has selected TSMC's 4nm node for Battlemage, which should provide a significant uplift from its previous generation GPUs while putting it in the ballpark of its biggest rivals. Intel will also move from deep ultraviolet lithography (DUV) on the 6nm node to extreme ultraviolet (EUV) at 4nm as part of the node upgrade. The move to 4nm could allow Intel to practically double the number of Xe cores in Battlemage due to the inherent advantage of the node switch. Intel is also expected to offer sizable gains in ray tracing and upscaling with Battlemage.
https://www.extremetech.com/computing/i ... th-13-tops
As you may have heard, the era of the "AI PC" is upon us (ugh). The first wave of these machines has already arrived in the form of Copilot+ PCs using Qualcomm CPUs, and now this phenomenon might also bleed into desktops. According to a new leak, Intel's upcoming Arrow Lake desktop CPUs will offer a neural processing unit (NPU) to improve its AI performance, which is notable for two reasons: First, Intel has never offered an NPU on a desktop chip before, and second, AMD has already announced four Zen 5 Ryzen 9000 CPUs, which lack NPUs.
https://www.extremetech.com/gaming/amd- ... technologyAMD has announced that the second generation of its frame-generation technology is now available via a new preview version of its Adrenalin software. The company says the latest version contains many improvements and is a "significant upgrade" over the first version. The biggest changes include more fine-grain controls for those who want to tweak their settings, support for more gaming APIs, and options for folks with older hardware. AMD is releasing it to gamers today in technical preview form, so proceed with caution.
The company calls this technology AMD Fluid Motion Frames 2, so when abbreviated it's now AFMF2—a bit clunky, but we'll allow it. A lot of changes are in version two, with AMD adding AI to the mix, support for more APIs, more controls for those who prefer to tweak their settings, and better overall performance, according to AMD. When the new version is enabled for a game, the optimal settings will be automatically applied, but so-called advanced users can now fine-tune some settings. This is not how the previous version worked, where it was like Nvidia's frame generation in that it was a binary switch; either it was on or off.
AMD STRIX POINT
https://www.extremetech.com/computing/i ... -on-sept-3Intel unveiled its next-gen Lunar Lake mobile architecture at Computex in May, but it didn't get into specifics about pricing and specs. It was also unclear exactly when it would launch other than "later this year." Intel has now announced that everything will be revealed on Sept. 3, as the company will formally launch the entire range of Lunar Lake CPUs at the IFA 2024 conference in Berlin, Germany.
Intel said company executives will be on hand to unveil the next generation of what it now calls Core Ultra mobile processors. Since these are laptop CPUs, the presentation will include the Lunar Lake CPU models Intel is launching and an array of Lunar Lake laptops. These notebooks will theoretically be the first x86 "AI PCs" because the existing models certified by Microsoft as Copilot+ are all powered by Qualcomm's Arm-based Snapdragon X CPUs. AMD's Zen 5-based Ryzen 9 AI mobile CPUs are also awaiting certification and will join Intel in the arena of x86-based AI PCs shortly.
brief Intel Granite Rapids server CPUs table
https://www.extremetech.com/computing/s ... 65mm-thickAI has begun encroaching on every tech world nook and cranny. A lot of the time, it's used as a buzzword without any identifiable benefit—like having AI in your thermal paste, for example. However, Samsung has unveiled ludicrously thin LPDDR5X memory that could provide a tangible benefit to all phones, not just those running some kind of AI application.
Samsung stated it had shrunk its 12nm LPDDR5X chips in both 12GB and 16GB capacities by 9%, so they're just 0.65mm thick. The advantages of this development are easy to understand: Thinner memory in a mobile device allows for more room inside a tightly packed phone, which could allow for more airflow inside the device.
Samsung LPDDR5X
Ryzen 9600X compared to multiple other CPUs
Ryzen 9700X performance in AI
Zen4 Threadripper PRO vs Xeon-w9 for science and engineering
https://www.extremetech.com/computing/f ... regressionIntel is preparing to launch its next-generation Arrow Lake desktop platform later this year, so the trickle of early benchmark results has begun. Given the company's current problems with Raptor Lake, hopes are high it can put all that nonsense behind it with Arrow Lake. It's on a new node with a new socket and an all-new, tile-based design. However, the first benchmark for Arrow Lake has been revealed, and it's close enough in performance to Raptor Lake to raise questions about Intel's upcoming platform.
High Numerical Aperture (NA) lithography machines just became available earlier this year. Intel was the first global foundry to purchase one in an attempt to one-up its rivals in the race to ever-shrinking nodes. Now, Samsung foundry has reportedly jumped on board the High-NA train; a new report states the company is joining Intel in the High-NA arms race. With Samsung now onboard, TSMC is the only major foundry still holding off on adopting the new technology.
Word of Samsung's plans comes from a site called Seoul Economic Daily, which includes a photo of Samsung executives alongside a High-NA lithography machine at ASML's headquarters in the Netherlands. It was previously reported that Intel had purchased all of ASML's High-NA machines for 2024, but perhaps that is no longer the case. The Korean reports say ASML made eight machines, and Intel purchased seven, leaving one for Samsung to put in its cart.
https://www.extremetech.com/computing/i ... for-ai-pcsToday, Intel is officially launching its Lunar Lake architecture at the Innovation For All (IFA) conference in Berlin. These CPUs announce the company's arrival to the AI PC market thanks to the CPUs' embedded NPU. They also mark a radical shift in design philosophy, with many notable firsts for the company. They're the first Intel CPUs with a compute tile made on a TSMC node, the first to include embedded memory like Apple's M-series chips, the first with its all-new P-core and E-core design, and the first CPU to feature its second-generation Battlemage GPU architecture. In other words, it's practically an entirely new CPU from the ground up.
