Search results for 'General' - Page: 7
| RadioNZ - 29 Oct (RadioNZ)SIS Director-General Andrew Hampton said `a small number of foreign states` were actively trying to steal emerging technologies. Read...Newslink ©2024 to RadioNZ | |
| | | RadioNZ - 29 Oct (RadioNZ)Applying CHOGM outcomes on climate change, oceans, blackbirding offers new secretary-general shot at relevance, writes Tess Newton Cain. Read...Newslink ©2024 to RadioNZ | |
| | | sharechat.co.nz - 25 Oct (sharechat.co.nz)General Capital Limited (NZX: GEN) today announced the resignation of Paul Zingel as a non-executive independent director of the General Capital Board effective 31 October 2024 Read...Newslink ©2024 to sharechat.co.nz | |
| | | sharechat.co.nz - 25 Oct (sharechat.co.nz)General Capital Limited advises that its subsidiary General Finance Limited, a licensed Non-bank Deposit Taker, has uploaded its quarterly unaudited report for the quarter ended 30 September 2024 to the Disclose Register Read...Newslink ©2024 to sharechat.co.nz | |
| | | sharechat.co.nz - 25 Oct (sharechat.co.nz)Starting from 24 October 2024 to 26 November 2024 (inclusive), take advantage of our exclusive “General Capital presents Long Lunch with Boris Johnson” Promotion and automatically “Be In The Draw to Win” tickets to “one of th Read...Newslink ©2024 to sharechat.co.nz | |
| | | PC World - 25 Oct (PC World)Intel’s latest Arrow Lake processor for desktop PCs fuses its “Meteor Lake” and “Lunar Lake” architectures together, carrying over Meteor Lake’s NPU and Lunar Lake’s abandonment of hyperthreading. Yes, hyperthreading has been banned from Intel’s desktop chips, based on a similar rationale for excluding the feature from Lunar Lake.
Arrow Lake, also known as its Core Ultra 200S processor lineup, is Intel’s first “disaggregated” desktop processor, built on tiles, meaning each part of the chip is individually fabricated on a different process. In a twist, Intel unveiled a deep dive into the architecture of Lunar Lake and the models, prices, and performance of the Core Ultra 200S processor. A key omission? Hyperthreading, which also was not part of Intel’s Lunar Lake mobile processor.
The story of Arrow Lake is a simple one: More performance and yet substantially less power than the 14th-generation Core chips. And to get there, Intel executives said they applied the same thinking to both Lunar Lake and its next-gen desktop counterpart: Make its cores as efficient as possible, both for power and for space.
Update: Intel’s Arrow Lake processors have now launched. How did the efficiency and performance claims stack up? Find out in Core Ultra 9 285K tested: 10 must-know facts about Intel’s radically new CPUs. Gordon Mah Ung also dived deeper into the 285K’s performance in productivity workloads in the deep-dive video below:
What is hyperthreading?
Hyperthreading (also known as simultaneous multi-threading) is a fairly simple concept: While each processor core is designed to execute one instruction thread, hyperthreading creates a second “virtual” processor inside the single processor core. With hyperthreading, the idea is that the individual processor core is always executing instructions on at least one of the two cores, keeping it in operation the whole time. The last thing enthusiasts want is a CPU core sitting idle when it could be performing useful work.
Intel
The problem is that the second core is a virtual core, and not a “true” second physical core. That can lead to some contention of resources and additional overhead, enough that the question of whether to leave hyperthreading on or off while gaming has been a source of debate for years.
Intel, meanwhile, has gone back and forth on the feature: Some of Intel’s 9th-, 10th-, and 11th-gen Core processors have excluded hyperthreading, such as the Core i7-9700K, and Intel’s Atom chips never used it. Most of Intel’s Core chips do, however. AMD has pretty consistently used hyperthreading, however, and still does. The question has always been: Does hyperthreading deliver a performance increase that surpasses the toll it takes in terms of system latency, the controller die cost, and the power hyperthreading consumes?
In Lunar Lake, the answer was “no,” and that has carried over to Intel’s latest desktop chips, too. In part, that’s because Arrow Lake cribs heavily from Lunar Lake, with the same Lion Cove performance cores and the same Skymont efficiency cores that appear in Lunar Lake.
Robert Hallock, a vice president and general manager of client AI and technical marketing for Intel, said that Intel basically comes out ahead in terms of power and performance by not using hyperthreading. Arrow Lake includes both desktop and mobile processors, and Hallock was being asked about the desktop implementation of hyperthreading. But it sounds like Hallock’s response applies to both desktop and mobile chips.
“It’s a combination of a couple things, actually,” Hallock told reporters. “First, we knew that we can actually save the wattage for hyperthreading by not including it on the product, and you see that we’re still coming out net ahead by roughly 15, 20 percent in [multicore performance] without it. So we’re able to bump up efficiency and still hit our goals in overall compute performance.
“The other thing that I would say is, you know, these are the same designs as leveraged from Lunar Lake,” Hallock added. “We took those cores, those designs, and were able to immediately integrate them because of [Intel’s] Foveros [technology]. So that’s the kind of one-two punch that influenced our decision: speed to market and maximizing performance per watt.”
Will hyperthreading ever return? It’s possible it could. But it would have to justify itself in terms of performance, power, and die space, and it appears right now that it isn’t making the cut.
Editor’s note: This article originally published on October 10, but was updated to include links to Core Ultra 200S review materials. Read...Newslink ©2024 to PC World | |
| | | PC World - 25 Oct (PC World)Intel’s new Arrow Lake architecture, aka the Core Ultra 200S series, brings AI capabilities onto Intel desktops. But the chip doesn’t use the Copilot+ capabilities of Intel’s mobile Lunar Lake chip — its designers used the older NPU found on Meteor Lake instead.
For now, this means that if you buy an Arrow Lake chip, you won’t be able to use it with some of the new AI enhancements found within Windows 11’s 2024 Update, like generative AI and the controversial Recall function. Arrow Lake’s optimized NPU 3 only provides 13 TOPS, while Microsoft set 40 TOPS as the bar for Copilot+ status. Intel still plans to ship more than 40 million AI PCs in 2024, using the vaguer “AI PC” definition that accompanied the launch of Meteor Lake.
Arrow Lake is the not the first desktop PC architecture to include an NPU for AI. Last January, AMD announced the Ryzen 8000 series of desktop processors with an NPU capable of 39 TOPS. However, the blink-or-you’ll-miss-it Ryzen 8000 was quickly replaced by the Ryzen 9000 processor six months later in June — without an NPU, but ushering in Zen 5 with a potent 16 percent performance improvement over the prior generation. Neither the Ryzen 8000 nor Arrow Lake meets the threshold to be called a Copilot+ PC.
Arrow Lake, like Meteor Lake, is a disaggregated architecture — a fancy name for a modular design. Theoretically, couldn’t Intel have just made more of the 45 TOPS NPU 4 found within Lunar Lake, and added it to the Arrow Lake package? Yes…and no, Intel executives said.
Update: Intel’s Arrow Lake processors have now launched. How did the efficiency and performance claims stack up? Find out in Core Ultra 9 285K tested: 10 must-know facts about Intel’s radically new CPUs. Gordon Mah Ung also dived deeper into the 285K’s performance in productivity workloads in the deep-dive video below:
Keeping it simple
First off, Arrow Lake’s NPU is the same NPU 3 as Meteor Lake, Robert Hallock, an Intel vice president and general manager of client AI and technical marketing at Intel, told reporters.
“So we’ve had a lot of time to learn it and optimize it, and it made sense that we would fit this in,” he said. (Meteor Lake’s NPU generated 11.5 TOPS, while Arrow Lake’s optimized version produces 13 TOPS.)
However, Intel’s customer base of enthusiasts also indicated that they didn’t want to give up certain features, like a potent GPU, to check the box of AI. And Intel, which faced pressure to deliver Arrow Lake on time, found it easier to just use an established design.
“We actually had a long, a really long chat internally about how to allocate the transistor budget on this part,” Hallock continued. “To be clear, yes, it was fully possible to put a 50 TOPS, 40 TOPS NPU on this product, but to do so would also require shrinking the core count, changing the GPU core count. You start making sacrifices in sort of fundamental performance dimensions that enthusiasts really care about — that didn’t feel like the right mix. And we also talked at length about sort of the enthusiast market’s disposition on AI as a whole. And I think it’s fair to say it’s somewhat reluctant.”
Arrow Lake contains rudimentary AI, for a reason.Intel
Instead, Intel believes that software developers aren’t always using AI hardware effectively, that a combination of components (CPU, GPU, and NPU) is better than an NPU alone, and various AI models are being effectively compressed small enough that they don’t need a massive NPU. All told, Arrow Lake-S (the Core Ultra 200S series) has a total of 36 platform TOPS. Although Arrow Lake has an integrated GPU, many customers will pair the chip with a discrete GPU, which provides far more AI horsepower than just the NPU alone.
“We’ve proven that Meteor Lake’s 13-TOPS NPU is more than enough,” Hallock said. “We’ve got enthusiast users who are somewhat skeptical, and so we sized an NPU that kind of fits all of those constraints while preserving and protecting the CPU performance that people care about most and still gives enough AI to embrace workloads that are coming down the pipe.”
Interestingly, Intel will debut two mobile chips in the Arrow Lake family during the first quarter of 2024. One, code-named Arrow Lake-HX, will essentially be a mobile copycat of the Core Ultra 200S chips that Intel will begin shipping in a few weeks. A second family of Arrow Lake-H chips for notebooks provides many more TOPS overall than Arrow Lake-S or -HX: 99 platform TOPS in all. However, those will come from a beefed-up GPU that uses XMX extensions and more Xe cores. The NPU will still provide 13 TOPS.
Roger Chandler, vice president and general manager for enthusiast PC and workstation product marketing at Intel, reiterated what Hallock said — that software developers aren’t using the full capabilities of the existing NPUs effectively. Intel’s goal, he said, was to offer a “balanced platform.”
“When I look at AI right now, we’re like, 10 seconds into a 20-hour movie,” Chandler said.
Editor’s note: This article originally published on October 10, but was updated to include links to Core Ultra 200S review materials. Read...Newslink ©2024 to PC World | |
| | | PC World - 25 Oct (PC World)Intel’s new desktop CPUs are radically different than the Intel CPUs you’re used to.
Today marks the launch of Arrow Lake, the company’s latest architecture for desktop processors, formally sold as the Core Ultra 200S Series. According to Intel’s claims, this batch of Core chips is faster and far more power efficient than last-gen Raptor Lake CPUs, while still beating out the competition.
However, Arrow Lake isn’t notable just for what it does, but also for what it is. As the first set of desktop x86 processors shipped but not fabricated by Intel, TSMC’s involvement in their production is but one new standout aspect. Intel has reworked its chip design, too—and the overhaul comes with some quirks, as we discovered during our testing.
For a deeper dive into productivity benchmarks, check out Gordon’s video review of the Core Ultra 9 285K below.
Here, let’s dig into the top ten things you should know about Intel’s Core Ultra 200S CPUs, after extensive testing of the flagship Core Ultra 9 285K.
New names for a new era
Alex Estevez / Foundry
Last year, Intel retooled its naming system for Core processors, bidding farewell to the “i” designator before 3, 5, 7, and 9, and dropping the reference to the generation. It also squeezed in the word “Ultra” for CPUs containing Arc graphics and a Neural Processing Unit (NPU) for AI-based tasks, as well as reset the numbering system to start in the hundreds.
This change first rolled out to laptops with Meteor Lake CPUs in late 2023. Arrow Lake’s release is the first time we’re seeing the new names on desktop—a big shift after a decade-plus of the old style of names. Intel also opted to begin Arrow Lake’s numbering in the 200 range, despite being the first on desktop—likely so that laptop and desktop architectures from the same year match. (Lunar Lake, which just began shipping in laptops in September 2024, labels its processors with numbers in the 200 range as well.)
Intel
These are the five CPUs in the new Core Ultra 200S series, which includes two variants without integrated graphics, as denoted by the “F” designator (that convention is still the same as before):
Intel Core Ultra 9 285K: 24 cores (8 P-cores, 16 E-cores, 5.7GHz max); 4 GPU cores, 13 TOPS NPU, $589
Intel Core Ultra 9 265K: 20 cores (8 P-cores, 12 E-cores, 5.5GHz max); 4 GPU cores, 13 TOPS NPU, $394
Intel Core Ultra 9 265KF: 20 cores (8 P-cores, 12 E-cores, 5.5GHz max); 0 GPU cores, 13 TOPS NPU, $379
Intel Core Ultra 9 245K: 14 cores (6 P-cores, 8 E-cores, 5.2GHz max); 4 GPU cores, 13 TOPS NPU, $309
Intel Core Ultra 9 245KF: 14 cores (6 P-cores, 8 E-cores, 5.2GHz max); 0 GPU cores, 13 TOPS NPU, $294
Hyperthreading is dead (again)
Intel introduced the masses to hyperthreading, its technology that allows a CPU core to run two threads (that is, set of instructions) simultaneously, rather than the default of just one per physical core. But in more recent years, the company has flirted with killing it off. For Arrow Lake, the feature’s gone overboard once again.
As Intel explained recently, it dropped hyperthreading as part of efforts to improve power and space efficiency in Arrow Lake—that is, how much power a chip uses, and what can be fit into each processor’s package.
So the core counts you see for Core Ultra 200S processors are what you get—no additional threads. That said, the number of Lion Cove performance cores (aka P-cores), and Skymont efficiency cores (aka E-cores) packed into Arrow Lake aren’t stingy. The top Core Ultra 9 285K chip still boasts 24 cores overall, with 8 P-cores and 16 E-cores.
Performance gains for content creators
Arrow Lake’s performance will be a touchy subject among chip enthusiasts, who’ve become accustomed to huge gains over the last couple of years. But that was possible due to high power ceilings—recent flagship processors essentially came overclocked as the default. In contrast, a key attribute of the Core Ultra 200S lineup is power efficiency, which Intel emphasizes as a selling point.
With this change, Arrow Lake provides creatives improved performance, but the gains aren’t spectacular enough to compel most Core i9-14900K owners to upgrade just yet.
Across our rendering and encoding benchmarks, the Core Ultra 9 285K consistently pulled ahead of last generation’s 14900K flagship, with improvements ranging between 2 to 21 percent depending on the task. It also mostly edged ahead of the Ryzen 9 9950X. The one exception: It trailed slightly behind both older CPUs in DaVinci Resolve.
But gaming is less of a win
Gamers love to see new CPUs crush the existing competition—and for those in that crowd, the Core Ultra 9 285K may seem disappointing. It’s a chip that trails behind the 14900K and 9950X in some games, holds about equal with them in others, and edges them out in yet others.
In our small selection of gaming benchmarks, the 285K generally held even with the 14900K, but often trailed by a double-digit margin behind AMD’s Ryzen 9 9950X. In Cyberpunk 2077, the 9950X kept a cushy 18 percent lead over the 285K, with similar performance in F1 2023 and Rainbow Six Siege. The 285K only held its own in 3DMark’s Steel Nomad, a synthetic gaming benchmark that simulates a more system-intensive game at 4K resolution.
Even when paired with faster, more performant CU-DIMM memory (instead of DDR5), gaming performance doesn’t dramatically improve—in F1 2023, we saw a rise to 410 frames per second, or about a 7 percent increase. That narrows the gap between the 285K and 9950X, but the latter still comes out on top.
By no means is the 285K a bad chip for gaming—but it won’t blow anyone’s socks off. So while Intel has been open about Arrow Lake’s gaming performance, the internet pitchforks will likely be out in force.
Power efficiency is improved, but…
As for the reason that Arrow Lake’s smaller performance gains—Intel claims its new chips match 14th-gen Raptor Lake-R’s performance at half the power, but you may not see quite such a dramatic improvement.
Compared to the 14900K, our tests showed a 17 percent decrease in power consumption during our Handbrake AV1 benchmark (a difference of 65 watts). Similarly, during Cinebench 2024’s single-core benchmark, energy use dropped by about 16 percent (a difference of 22 watts). At idle, the 285K’s power draw actually increased slightly, nudging up about 3 percent (about 2.5 watts). These three benchmarks cover the general spectrum of usual PC behavior.
The 285K fares better against AMD’s best, with a 4 percent decrease in power consumption in Handbrake, an almost 25 percent decrease in Cinebench 2024, and a roughly 28 percent decrease at idle compared to the 9950X.
Overall, Intel’s improvements give it an edge if you’re worried about heat or your power bill. But the story doesn’t end there. Turns out, when you dig into power draw, these new chips reveal an unexpected peculiarity.
…Windows optimizations may be a work in progress
Desktop users don’t often pay much mind to Windows power settings—they’re less relevant when you’re plugged into a wall all the time. At least, that’s the case normally.
However, being a curious bunch, we checked the 285K’s power consumption on different Windows 11 power plans. And oddly, the 285K’s performance dropped on the Balanced and Power Saver plans.
In our Cinebench 2024 single-core test, the 285K slowed down notably while completing its task, stretching out to 30 minutes on the Balanced plan. In contrast, the 14900K and 9950X finished with more usual times—meaning we could start those chips’ benchmarks later than the 285K (15 minutes into the 285K’s run) and still see them complete the task faster.
The resulting score showed a 55 percent decrease in performance compared to the 14900K and 9950X on the same Balanced power plan. The difference was even higher on the Power Saver plan, with a 67 percent drop.
Meanwhile, on High Performance settings, the 285K actually outperformed its predecessor by about 5 percent, and the 9950X by about 2 percent. This outcome is all the more puzzling, given that our power draw measurements showed little energy savings across the various power plans.
Does this mean Intel’s new chip could be nerfed by your power settings? For the moment, likely yes, especially since the Balanced power plan is the Windows default. Time will tell if this behavior can and will be fixed via better Windows optimization.
Intel revamped its processor design
Intel
Growing pains with a new chip design aren’t too surprising—Arrow Lake is considerably different compared to 14th-gen Raptor Lake-R.
Until now, Intel released monolithic processors for desktop; everything related to the CPU existed on a single die. But starting with Arrow Lake, Intel has switched to a chiplet design, a move that rival AMD made years ago.
Intel calls Arrow Lake a “disaggregated” processor, in which separate chiplets (“tiles”) for different functions are linked together as a single package. If you pull off the lid of an Arrow Lake processor, you’ll find compute, GPU, SOC, and I/O tiles, along with a “filler” and “base” tile for stability. Intel’s Foveros technology connects them all together. Each of these chiplets are fabricated (produced) on different processes—you can read more details about them in our initial overview of Arrow Lake.
For chip nerds, this change in approach represents a huge shift for Intel, which had previously championed a single die as better performing. For most home users, however, what matters more is how Intel had to juggle the layout of the package.
You’ll need a new motherboard
Intel
Arrow Lake requires a new socket—LGA 1851—and accordingly, that means it won’t be compatible with existing LGA 1700 boards.
The first chipset to be announced is the Z890, which supports up to 192GB of DDR5-6400 memory (up to 48GB per DIMM). You can use more common SO-DIMMs, but the newer CU-DIMM memory modules work as well. Intel says that if you enable XMP for faster RAM speeds, DDR5-8000 will be the ideal.
For connectivity, Intel’s 800 series chipsets offer up to 24 PCI-e 4.0 lanes, as well as 10 USB 3.2, 14 USB 2.0, and 8 SATA 3.0 connections. Combined with an Arrow Lake CPU, you’ll get up to 48 PCIe lanes, with 20 of them PCIe 5.0.
Thunderbolt 4 and Thunderbolt 5 integration are also features, as are technologies like Thunderbolt Share (which lets you share screens, peripherals, storage, and files with another PC), Intel Killer Wi-Fi 7, Bluetooth 5.4, and 2.5Gbps ethernet (depending on configuration).
DDR5 RAM & existing coolers still work
IDG
While you can pair Arrow Lake with CU-DIMM memory, it doesn’t mean you must. DDR5 RAM will still work with a Core Ultra 200S processor, and as we’ve seen above, you’re not losing that much of a performance boost.
In fact, with general availability for CU-DIMM memory not slated to heat up until early next year, if you buy into Arrow Lake now, your option really will be only DDR5 memory.
As for coolers—if you already have a cooler compatible with 13th or 14th generation processors, they should still work with Arrow Lake. However, Intel says you’ll likely new mounting hardware to achieve the correct pressure for optimal cooling.
No upgrades to AI capabilities on desktop
Intel
If you were hoping that the launch of Arrow Lake meant desktop access to generative AI and the divisive Recall feature in Windows 11, prepare yourself for disappointment. These new processors incorporate an older NPU into the package, which only provides 13 TOPS of computational performance, well below the minimum of 40 TOPS Microsoft requires to unlock Windows 11’s Copilot+ AI features.
Why? Decisions were made to keep gamers happy—as they’re one of the primary groups of buyers for desktop chips. You can read more about Intel’s approach in our rundown of Arrow Lake’s AI capabilities, but know that systems with these chips will still likely be marketed as AI PCs. Read...Newslink ©2024 to PC World | |
| | | PC World - 25 Oct (PC World)If you’re buying a new GPU (graphics processing unit), you should definitely have an understanding of how it all works. Although the terms GPU and graphics card are often used interchangeably, technically speaking, they’re not the same thing. We’ll go over all the technical information and break it down in an easily digestible way for you. Read on to learn more.
Further reading: 4 things to consider before buying a GPU
A typical graphics card contains a GPU die (chip) soldered onto the circuit board and surrounded by memory modules. This is built into a heat sink, cover, and fan, creating the actual graphics card.
Companies like AMD and Nvidia use multiple GPU dies in each generation to create different models and customize the specifications as needed. The GPU die designation often follows a system where a lower number on the spec sheet indicates a larger and more powerful die. For example, Nvidia’s AD102 GPU utilizes the flagship RTX 4090 graphics card while the RTX 4080 and RTX 4080 Super use the AD103 GPU. Newer GPUs utilize smaller manufacturing processes to offer more performance in the same physical space.
Strictly speaking, GPU is only the name for the graphics processor (here from AMD). Only with other components does it become a graphics card.
IDG
The computing units on a GPU are often referred to as cores. However, this is misleading, as the cores of a GPU do not correspond to the physical cores of a CPU. GPUs have thousands of small cores or processing elements, which are grouped into clusters and then into compute units.
These compute units are labelled differently by each GPU manufacturer–compute units (AMD), Xe cores (Intel), and stream multiprocessors (Nvidia). Faster GPUs have more Cuda cores (Nvidia) or stream processors (AMD) than slower models. However, these specifications are only comparable within the same generation and the same manufacturer.
VRAM (video RAM) is a specification that almost every PC user is familiar with. VRAM is the memory on a graphics card that is available to hold the so-called frame buffer, texture information, and other graphical data.
If your game settings and video resolution utilize the entire VRAM of your GPU, this can lead to graphics errors and a drop in performance. In addition to the size of the VRAM, other specifications are also decisive such as the memory type (e.g. GDDR6 or GDDR6X), the memory bus width (384, 256 or 128 bits), or the memory bandwidth (in GB/s).
Graphics cards are complex units with many technical specifications. To be able to compare them, it helps to familiarize yourself with GPU terminology.IDG
Just like a CPU, a graphics processor also has a clock rate both for the GPU core and the memory. The core clock rate indicates how quickly the GPU can process instructions. The memory clock rate, on the other hand, determines how fast the memory (VRAM) can send data to or receive data from the GPU. In general, the following applies:
The higher the core clock rate and the memory clock rate of a graphics card, the higher the performance. You can also change the core clock rate and the memory clock rate–this is known as overclocking.
Just like a CPU, a GPU has a maximum power consumption, which is referred to as TGP (Total Graphics Power). The TGP value indicates the maximum power that a graphics card may consume during operation and is measured in watts. TGP is an indicator of how power-hungry your graphics card is and also provides information about the expected temperature of the GPU during gaming or other GPU-intensive tasks.
FP32 or Floating-Point Single Precision is a mathematical method for measuring the theoretical performance of a GPU. It indicates how many floating-point operations a GPU can perform in one second and is measured in TFLOPS.
FP32 is useful for comparing different GPUs, even across generations, although the method does not cover all aspects of performance. Other specifications such as the supported PCIe generation and compatibility with graphical APIs such as DirectX, Vulkan, and Open GL can also be important, especially for designers or developers.
Software features such as Nvidia’s DLSS, AMD’s FSR, and Intel’s XeSS differ in performance and gaming support. You should therefore check which upscaling technology and version your GPU supports.
The graphics memory (shown here in green) of a graphics card is located in the immediate vicinity of the die to avoid delays in communication.
IDG
When buying a new GPU, you should always consult real-world tests and benchmarks from reputable sources to make an informed decision. Check the performance figures in the games and applications you intend to use. You should consider checking out Reddit posts, YouTube benchmarks, and reviews from trustworthy publications. Read...Newslink ©2024 to PC World | |
| | | sharechat.co.nz - 24 Oct (sharechat.co.nz)Pursuant to NZX Listing Rule 3.20.1(a), SkyCity Entertainment Group Limited (SkyCity) advises that Jo Wong has resigned and will step down from the role of General Counsel and Company Secretary on 22 January 2025 Read...Newslink ©2024 to sharechat.co.nz | |
| | |
|
|
| Top Stories |
RUGBY
The All Blacks have produced an error-ridden 29-11 victory over Italy in Turin to finish their season with ten wins and four losses More...
|
BUSINESS
Warnings of missed opportunities ahead of next year's Companies Act reforms More...
|
|
| Today's News |
| News Search |
|
|