### HOW TO THINK OF COMPUTERS Imagine you (the computer) are presenting information to an audience (the User) using a whiteboard (the Display). You have access to a library of books (Storage Device, such as an SSD or the Cloud) that contain all the information you need. However, you haven’t looked at these books before, so you need to first retrieve them and make sense of their contents. Your brain (the CPU) is responsible for understanding the information in the books. But instead of simply reading the book, you extract the necessary sections (processing the data) and prepare it to be displayed on the whiteboard. Since your brain has only a limited amount of immediate processing space (CPU cache), you use a notepad (RAM) to jot down key information temporarily. This notepad allows you to remember what you’ve just processed long enough to write it on the whiteboard. However, the notepad can only hold so much at a time, so after you've written down one section, you quickly clear space on the notepad to prepare for the next book. You repeat this process for hours with dozens of books, efficiently retrieving, processing, and displaying each piece of information for your audience. As soon as you’re done with one piece, you erase it from the notepad (clear RAM) and move on to the next, ensuring that your brain can keep up with the demands of processing and displaying information continually. The motherboard is the desk or workspace where everything happens. Just as a desk provides the surface and structure for your notepad, brain, and books to function together, the motherboard connects and coordinates all the components in a computer The operating system can be thought of as the rulebook or guide that tells you know exactly what to do and how to do it. A set of instructions and protocols that guide how you retrieve, process, and display the information, it makes sure everything runs smoothly. As you focus on the text (CPU processing), your talented artist assistant (GPU) is responsible for any complex drawings or graphics you want to display on the whiteboard. The assistant can handle many, but less complicated tasks at the same time, so you don't slow down. The assistant also has a super-genius baby brother (NPU); while the skilled artist with a strong body (hardware), the sibling shows remarkable talent in areas like pattern recognition and fast learning. Though not yet as physically powerful, the baby geniuses unique abilities (AI and machine learning) make it an emerging talent worth keeping an eye on. Or perhaps the artist will grow jealous and in a tragic twist devour the little boy to absorb all his powers. You receive your many books from different shelves and locations in your library. Some shelves (SSD) are closer and easier to reach, so you can retrieve those books faster, while others (HDD or cloud) might be in the back of the library or in a remote location, making them slower to access. You also have sticky notes (cache) you pin to the side of the whiteboard for the most important information that you need to access repeatedly or quickly. Of course, you need electricity to keep the lights on and power your work. But because your room (the computer) uses a lot of energy, you have your own generator (the power supply unit) that carefully draws electricity and distributes it efficiently to all the devices (components) you need to keep everything running smoothly without overloading the system. The audience (the User) interacts with you (the computer) using various tools like pens or pointers (input devices such as a keyboard, mouse, etc.) to give you instructions or provide information. They use these tools to point out which book they want you to retrieve from the library (storage), or to specify what they want you to write on the whiteboard (display). They work you really hard, so you need a fan to keep you from slowing down or worse (cooling system). Your Artist (GPU) also has their own fan. METAPHOR OVER, BORING SPECIFICS NEXT REMEBER, DON'T HUMANISE COMPUTERS. In computing, every operation can be viewed as a series of calculations that result in data manipulation. Every task a computer performs, whether it's executing a program, processing input, or managing data, involves calculations. These calculations are carried out by the CPU, GPU, or other processing units. The results of these calculations are data points. Data points can represent anything from numerical values and text to images and complex datasets. Storage provides the necessary infrastructure for data points to exist and be managed, just as the material world provides the conditions required for human life. When data is stored on a medium such as an SSD, HDD, or cloud storage, it remains in a static state. This means that the data is preserved as-is until it is accessed or modified. The storage medium holds the data without actively processing it. In the case of non-volatile storage like SSDs and HDDs, the data remains intact even when the power is off. This contrasts with volatile memory, such as RAM, where data is lost when the power is turned off. Data points are only processed or modified when the computer retrieves them from storage and performs calculations. Until this retrieval happens, the data remains in its stored form, unchanged. CPUs, NPUs, and GPUs are fundamentally all processors that use cores to perform calculations to manipulate and arrive at data points Why use 3 different calculators? As laid out in the metaphor, they have different specialisations. CPUs are used to process instructions, such as running applications or performing calculations. Each CPU consists of multiple cores, with different models offering either 1 or 2 threads per core (a feature known as hyperthreading). Most CPUs today have between 4 and 16 cores, but high-end models like the AMD Threadripper can have up to 96 cores and 192 threads. The clock speed, usually between 3 and 4.5 GHz, indicates how many cycles (or calculations) each core can perform per second. Each thread is dedicated to executing a single instruction or task, meaning that more threads can increase a CPU's ability to handle multiple tasks simultaneously. Therefore, the potential speed of a CPU is often estimated by multiplying the number of cores by the clock speed (in GHz), reflecting its ability to perform billions of cycles per second. This makes CPUs particularly efficient for processing single, complicated tasks, such as CPU-based rendering or video processing. Supercomputers also prioritize CPU performance because they are designed to solve complex problems that benefit from high processing power across multiple cores. Potential Speed ≈ Number of Cores × Clock Speed (in GHz) GPUs (Graphics Processing Units) are specialized processors designed to handle large volumes of data and perform many calculations simultaneously. Unlike CPUs, which focus on sequential processing with fewer, more powerful cores, GPUs have thousands of smaller cores that can work in parallel. While a GPU typically has a lower clock speed than a CPU (often about half), its architecture is optimized for parallel processing. GPUs also come with their own dedicated RAM, known as VRAM, which ranges from 6 to 24 GB for gaming GPUs and can exceed 90 GB for high-end data center GPUs. This dedicated memory is crucial because, despite the lower clock speed and smaller memory per core, the vast number of cores allows GPUs to execute thousands of trillions of calculations per second. The potential speed of a GPU is often measured in TFLOPS (teraflops), representing trillion floating-point operations per second, highlighting its ability to perform extensive parallel computations. This makes GPUs ideal for tasks such as rendering graphics in gaming, processing large datasets in data centers, and training AI models, where handling many operations simultaneously is crucial. NPUs (Neural Processing Units) are designed to handle parallel tasks efficiently, even with a relatively small number of cores. Their architecture varies widely, with core counts ranging from a few to thousands, depending on the specific design goals. NPUs often use SIMD (Single Instruction, Multiple Data) architectures, allowing a single instruction to be applied across multiple data points simultaneously, which is ideal for processing large datasets with repetitive mathematical operations. NPUs manage workloads intelligently, distributing tasks across cores and dynamically adjusting to optimize efficiency. They share the system’s dynamic RAM with the CPU and GPU but are currently limited by memory bandwidth, which makes them less effective compared to GPUs for tasks like AI training, where processing as many tasks as possible is more critical than efficiency. Memory is essential for completing calculations in your system, providing the space needed to store and manage data during processing. While NPUs (Neural Processing Units) are designed to perform basic tasks with lower power consumption compared to GPUs, they are not typically suited for running large local language models (LLMs) or performing highly intensive AI tasks. LLMs, which are advanced AI models trained on vast amounts of text data, are currently the most effective and optimized approach for complex AI applications. Some advanced GPUs incorporate NPU-like components to offload specific tasks such as background blurring or application-specific filtering. NPUs also help extend laptop battery life by handling GPU tasks with lower power consumption. Future advancements might see NPUs becoming more integral or GPUs adopting NPU technologies, potentially leading to significant improvements in AI processing and energy efficiency. Technically any type of storage can be used for a computer to calculate. Back to the metaphor; think of your computer’s storage options like different types of workspaces. An SSD is like a large desk where you can keep a lot of files, but it fills up quickly if you try to use it for everything. You could designate some of that desk space as a “quick-access” zone (VRAM), but it only helps a little with speed. Cloud storage is like using a distant warehouse for storing large amounts of data— it's great for long-term storage and big projects but not ideal for tasks needing instant access. RAM, on the other hand, is like a super-efficient, quickly wiped tablet right in front of you. It wipes away unnecessary items to make room for the next task, ensuring fast, continuous access to the data you need. However, this tablet needs constant power to stay functional and is far smaller than the other options. An operating system manages hardware and software resources on a computer, providing a user interface and enabling applications to run. It handles tasks such as Resource Management, allocating CPU time, memory, and storage to applications. File Management to organize and control access to files and directories. Process Management to manage running applications and multitasking. Device Management to control hardware peripherals (input devices). AN OS requires Instruction Set Architecture (ISA), the set of instructions a CPU can execute. It defines the CPU's capabilities and how software communicates with hardware. An ISA includes instructions for operations like arithmetic, data movement, and control flow, as well as how to encode and decode these instructions. Examples include x86 and ARM. ARM (Advanced RISC Machine): ARM architecture is used in many mobile devices and embedded systems. ARM processors are designed for efficiency, with a focus on low power consumption and lower heat generation. They use a Reduced Instruction Set Computing (RISC) approach, which means they execute a smaller set of instructions more efficiently. x64 (64-bit x86): x64 refers to the 64-bit extension of the x86 architecture, used in most modern PCs and servers. It supports a larger addressable memory space and is designed for higher performance. x64 processors use Complex Instruction Set Computing (CISC), which includes a broader set of instructions to handle more complex operations. SO, FINALLY TO THE SURFACE LAPTOP 7 PROCESSORS All your PUs are integrated into the CPU CPU Qualcomm Snapdragon X Elite X1E-80-100 (12 cores) CPU performance about average for price class. Note that stress tests reduced clock speed significantly. ARM chips do use less power whilst maintaining good performance, however this newest generation of chips has underwhelmed in terms of the promised performance. Windows on ARM (WoA) might be an issue. Overall, this second generation of Snapdragons is a very solid, if slightly underwhelming and definitely not headline breaking attempt. iGPU Qualcomm SD X Adreno X1-85 (3.8 TFLOPS) Stable, more than sufficient for playing videos. Very few Games run as there are very few native ARM games and no driver support. Stable performance. Solid for AI tasks. Note that running intensive AI tasks won't utilise the NPUs power efficiency and impact battery life. Due to the iGPU not being powerful enough, any video editing at high resolutions really struggles. However, this doesn't affect video playback at all. It is somewhat strange that a selling point is the admittedly amazing screen to edit videos on. QUALCOMM gave a lot of hype to davinci (visual effect studio) but users are reporting frustrating slowness when working with high resolutions. Obviously to the relevant points: Fine for playing videos at 4k, not to relevant to office. iNPU Qualcomm® Hexagon™ (45 TOPS) Qualcomm's release of their NPU's 45 TOPS gives a glimpse of its speed but lacks detailed specs, similar to evaluating a car based on its top speed alone. While 45 TOPS is fast for an NPU, with AMD and INTEL looking at exciting launches current models are expected to be replaced by more advanced versions soon. NPUs are beneficial for low-power, less demanding tasks, but for high-performance AI applications, the CPU, GPU, and dedicated VRAM are crucial. On this laptop all components share the same RAM, so the absence of dedicated VRAM means the iGPU could dominate resources, potentially making the NPU less relevant for intensive tasks. The real selling point of the NPU is its power efficiency stacked on ARM, rather than its future-proofing or performance gains. It does improve battery life by offloading tasks from the iGPU, but it's still early for evaluating its long-term impact. If you're spending on a NPU, see it as half of your money being an investment into allowing companies to further research, not the current hardware you're getting. OS Windows on ARM (WoA) In everyday situations, the Surface Laptop is very responsive. Nevertheless, it’s important to note that ARM versions aren’t available for every application or driver. Compared to Windows on x86, you are limited. Despite QUALCOMM building Microsoft processors since 2018, there is still no clear plan for ARM from Microsoft. Microsoft is not tied to ARM, as performance still matters most to consumers and desktops simply perform better on x86. Some disappointing hardware launches and missing driver support might make developers weary to port/develop software for ARM. They will always develop for x86. However WoA exclusivity with QUALCOMM ends this year and AMD and INTEL are looking to dip their toes. This might encourage devs to invest into porting apps to ARM. Apple, who ported their OS to ARM, have shown it can be done successfully. They however strictly control their entire App eco system and don't need x86 as they don't make power-focused computers. Note that many old input devices (printers, scanners etc) lack driver support on ARM and not every app is ported over. STORAGE Samsung PM9B1 version m2 ssd 416gb storage left after installation. Write speed @3500. Fine for everyday use. Stable, nothing special. Also swappable. 16GB LPDDR5x RAM Fast and current-generation, but 16GB is relatively low for modern laptops. The RAM is also not expandable. This amount of RAM can become a bottleneck, especially since all processing units (CPU, GPU, and NPU) share the same RAM without dedicated VRAM. If your laptop slows down, it’s likely due to RAM becoming a bottleneck, leading to a backlog of calculations and reduced performance across all tasks. Get 32gb. HOUSING Fan and Case Stays deactivated for long periods of time and even when under a little load it stays very quiet. Case average temperature in an idle and low demand state is 24 degrees (average in class), when under high stress it heats up quickly up to 44.6 on bottom (slightly higher than average) and 47.1 on the upper side (highest in class, 15% over average). Paired with metal case is uncomfortable and can lead to typing issues. Another confusing choice as simply running the fan faster would prevent this. A potential BIOS update to the fan could increase the speed slightly, making it slightly louder however. Also, you will probably never stress the system to that level (as long as you get 32gb RAM). The Case is top quality. Absolutely a selling point. Touchscreen 15” PixelSense Flow™ Display A 3:2, IPS touchscreen with a resolution of 2,304 x 1,536 pixels. Sufficiently sharp and the high 120Hz refresh rate allows for a smooth image and video viewing experience. Despite not being 4k the resolution is far superior to FUll HD (1920x1080p), but doesn't quite measure up to OLED panels seen in other devices, such as the current surface pro tablet. Both the brightness and the color temperature can be adjusted to the ambient light via Sensor. The colour depth and high brightness lend themselves to image and video editing. The touchscreen is very reflective and can cause issues outside. Overall another very good component that doesn't quite finish top of its class. INPUT The Surface Laptop 7’s keyboard provides a comfortable typing experience. Haptic touchpad doesn't have any moving parts, so the click is created via the vibration motors. This allows the "click" to be quiet but still gives feedback to the user. The touchscreen works flawlessly. Speakers Do their job fine, comparable to other models in class. BATTERY 66wh, unknown PSU Kind of hard to find solid data on the 15' version of this laptop. The 13.8' had a weak PSU, this isn't an issue with the 15' model. 66wh battery is average sized. The battery life is impressive. Around 15-17 hours browsing internet on 25% brightness. It does drop off compared to competitors under high stress. The charging is fast at just over 2 Hours. Full brightness does have an impact on the performance when compared to other laptops, but never goes bellow the 10th percentile. Really impressive. ALL IN ALL Laptops, maybe hardware in general seem to be in a weird place right now. For some years, software was being outpaced by hardware. Then in the last 2 years INTEL have stumbled, NVIDIA have had underwhelming launches and broken promises, whilst AMD strengthens their position in the market with consistent performance at friendlier pricing. Now the dawn of AI, rising energy prices and never ending consumer demands for faster, stronger machines is seeing the tech companies pushing for new innovations and facing increasing competition across the board. So we have the surface laptop, Jack of most trades, master of One. The battery is Outstanding. So is the build quality. The screen is very good. Its sleek, quiet and comfortable to use. Its just not outstanding at doing anything when actually turned on. Sure the CPU is solid, the iGPU does its job, the laptop can be bought with 32 or 64gb and the SSD is swappable, but the NPU is still gimmicky, selling a 3-in1 chip with 16gb total TAM as future proof is astounding and for the price its not going to blow your socks off in the performance department, but by the standards we are used to now what would. Then again that doesn't mean Microsoft should market their laptop with a focus on AI and performance, though you could blame QUALCOMM for their overhyped and now underwhelming CPU and WoA eperience. Again here its promising but still a couple generations from competing with Intel and AMD. The main selling points should be the insane battery life, the quality design and stable performance using comfortable inputs including the great screen. No bloatware and energy efficient ARM Windows. Be mindful of ARM limitations, poor performance under Stress, current NPU models being more marketing bark then bight, no dedicated GPU leaving you dependant on the inbuilt RAM and potential struggle to perform with complex AI. #PC #Computer #GPU #NPU #CPU #RAM #AMD #INTEL #QUALCOMM #COMPUTING