Heavy and Light: The Hardware Philosophy Behind My 2025 Setup
power and restraint

power and restraint
Within nova, as you’ve come to expect, we pursue balance, knowledge and alignment. Choosing muscle at the desk and silence in the bag isn’t just smart – it also keeps costs under control. Reviewers lean hard on benchmarks, nudging buyers toward overspending or chasing FOMO. My take is different: a quiet 12700K tower with an RTX 5080 (priced the same as a 4080 in my market), paired with a base MacBook Air M4 (16/256) – light, instant. One handles 3D, exports, and games; the other covers writing, code, and daily media edits with absurd battery life. Together they act as one system: AtlasOS + portable apps keep Windows lean; macOS stays clean and predictable. Nothing is overbuilt for show. This is the why, the parts, and the workflow.
See my setup guide for a lightweight, debloated, and functional Windows 11 build, here🔗. 
Also, check my nuclear-wiped macOS Tahoe installation and debloat guide (without disabling SIP) guide here🔗.
1.1) Longevity & support
Apple Silicon ages well, and I wanted the longest runway. I won’t consider Macbook Pros, as they are too expensive. The M4 Air finally ships with 16 GB unified memory as standard, which matters for newer macOS visual effects and GPU-accelerated bits on MacOS 26 Tahoe (which remain even if you “Reduce Transparency” and “Disable Motion”. I wanted to have an Apple Macbook for a while now, and I’ve studied market trends and peer-to-peer marketplaces for more than a year.
Photo: The M2-M4 Air design. 3D render by Rever_Art🔗
When the M4 Air launched on 12 March 2025, I couldn’t believe that Apple actually made the 16GB RAM variant standard (maybe a push for Apple Intelligence – a feature I never used, keeping it disabled). “Good thing I was this patient“, I thought, “Let’s see if I can find a good deal.“
Around June of 2025, browsing through a local peer-to-peer marketplace, I experienced a phenomenon that usually happens in the EU. Let me explain it plainly. I had three “saved” Airs, with the following list prices:
| Spec / Feature | M2 Air | M3 Air | M4 Air | 
|---|---|---|---|
| Battery | 52.6 Wh | 52.6 Wh | 53.8 Wh | 
| Memory | 8 GB unified | 8 GB unified | 16 GB unified | 
| Memory Bandwidth | 100 GB/s | 100 GB/s | 120 GB/s | 
| SSD | 256 GB (single-NAND bottleneck) | 256 GB (single-NAND bottleneck) | 256 GB (dual-NAND, ~2× faster) | 
| Wi-Fi | Wi-Fi 6 | Wi-Fi 6E | Wi-Fi 6E | 
| Camera | 1080p FaceTime HD | 1080p FaceTime HD | 1080p + Center Stage & Desk View | 
| Media Engine | H.264 / HEVC | + AV1 decode | + AV1 decode | 
| CPU / GPU | 8c CPU / 8c GPU | 8c CPU / 8c GPU | 10c CPU / 8c GPU | 
| Price | 690.00 EUR | 885.00 EUR | 900.00 EUR | 
June 2025 peer-to-peer Marketplace SH Prices (very good or excellent condition)
1.2) The EU Price Paradox
All 3 deliver great performance, for what I planned to use them for. They also look the same. The only dealbreaker here is software support. When you line them up like that, the “choice” looks obvious: spend +200 EUR above the M2, respectively +15 EUR more than the M3 (which is the price of a meal around these parts) and get a brand-new M4 with double the memory, faster storage, more years of updates, and even better battery life. But the real question is: why does the market even allow this kind of overlap?
So you get paradoxical listings: an “older but bigger SSD” M2 asking almost the same as a brand-new M4 base. On paper, the M4 wins, but the market logic runs on perception, not specs. That’s why patience + knowledge pay: once you know the bottlenecks (8 GB RAM, single NAND SSD), the choice narrows itself.
| Device | Typical New / Local Price | Overlap / Close Range To | 
|---|---|---|
| M2 Air (8/256 or 8/512) | ~ €1,000 – 1,100 list new | Overlaps with low-end new M4s in EU | 
| M3 Air (8/256) | Hard to find new. SH ~ €850 | Blurs with M4 pricing in used market | 
| M4 Air (16/256) | ~ €1,199 EU list new | Anchors the high end of used M2/M3 prices | 
October 2025 List Prices
The 256 GB SSD is no longer the boat anchor it was on M2. Real-world tests on the M4 256 GB show ~1.9 GB/s write / 2.9 GB/s read – roughly ~2× the base M2 256 GB – so the “single-NAND bottleneck” problem is effectively gone at this tier.
Scrap the benchmarks – the differences are there but they are small. I suggest only thinking about Software Support, especially when OCLP will most likely never work with Apple Silicon – so, when the brick becomes Obsolete, you either stay with it like that till browsers don’t work, or switch to Asahi Linux🔗.
1.3) Battery that Feels
Review cycles consistently report all-day (15-16 h light use) with improved efficiency over prior gens. I can confirm, the runtime is really good. At home, I run AlDente, capped to 60% to preserve the pack. I disabled Apple Intelligence and keep it minimal, with my recommended Mac debloat settings and essential apps here🔗.
2.1) Performance per watt
M4 stays silent and cool while pushing daily writing/media/dev/entertainment (yes, you even forget sometimes that you’re on battery). 
It can even game up to a point: I managed to reach a playable-ish 40fps on Cyberpunk 2077 with mixed-low setting at 900p (MetalFX used). I ignored the “For This Mac” setting, because it is too optimistic for the M4’s 8-core GPU. Meanwhile, World of Warcraft consistently maintains a smooth 60 fps (or more on external display).
Yes – it’ll throttle under load. Use a cooling pad, give it breathing room, and expect fps drops until it cools. But even with that, the 8-core GPU inside is likely the usual ceiling, not the temps. Battery drains faster under gaming stress, of course – but it still outlasts nearly any x86 Ultrabook under normal use (I won’t talk about current ARM Windows laptops, they’re very bad for games). The catalog of supported games is smaller, but what is supported runs remarkably well on the M4.
Anyway, if I record any of my own in-game videos, I’ll post them here in the future. For now though, I’m sticking to playing Balatro in my garden – you actually feel free for a moment. By the way, check out my in-depth Balatro review here🔗.
Video: CP2077 on M4 Air – close to what I’m actually getting on mine.
2.2) Ports & Constraints
This Macbook Air M4 features two Thunderbolt 4 ports, MagSafe charging port, and a 3.5mm headphone/mic combo jack. Two external displays now work properly with lid open, which removes a long-standing friction point for the Air. You can connect two external displays with up to 6K resolution at 60Hz or 4K at 144Hz, or one high-resolution display up to 8K. To achieve this, we use Thunderbolt/USB-C hubs or adapters, as the MacBook Air lacks built-in HDMI ports. It sucks really, but here we are – you can’t have it all.
I didn’t really need two and a half displays on my desk – I’ll just turn on my PC if I have that kind of needs, so I bought a cheaper USB-C Hub: the Anker 364 10-in-1. It’s everything I need: 2 x HDMI 4K60, 1 x Ethernet 1Gbps, 2 x USB-C, 1 x SD Card Reader, 3 x USB-A.
While it markets that it supports two HDMI-out, on Mac, it doesn’t support Multi-Stream Transport, so it will only mirror the two external displays. Reading online, you need two proper USB-C to HDMI / DP, or a more expensive dock / hub.
Again, for an on-the-go laptop, I really didn’t care too much.
2.3) Weight and Form Factor
I like editing stuff, writing while I’m on a trip (and I’m not the type that endlessly swipes his smartphone). No really, how much time can you spend at a beach, on a peak, or visiting an old city center until it’s dark and everything closes? Getting wasted in another city/country bar at night is not really my thing. So here we are, another important point that makes this Macbook Air win.
At 2.7lbs / 1.24kg and 11.97 x 8.46 x 0.44 inches / 30.41 x 21.5 x 1.13cm, with a 13.6″ screen and all the advantages stated above, I can safely conclude that the Macbook Air M4 is the best ultraportable for 2025. Just pack your stuff, and most likely there will be a spot left for the laptop.
For the paranoia, I grabbed a clear plastic case, a screen protector, silicone keyboard protector (care to get the correct thickness so the caps don’t print on the screen – not permanent, can be wiped but it’s a hassle. I got a good one here🔗).
In conclusion, the M4 Air feels like a return to the old Apple – crafting great products at thoughtful prices, designed for everyone. It helps me save power (high energy cost in the EU if you don’t have home-renewable), it’s a fast and light a machine that I can carry everywhere.
An ultrabook that I hope will stay with me for many years to come. I just wish they had swappable SSD or RAM, but if this is the compromise, I’m willing to take it.
1.1) Economics over ego
The EU price paradox strikes again.
Locally, the RTX 5080 was only ~€+100 over a 4080S at the time I bought it. 
Paying nearly the same for newer silicon (Blackwell), DLSS 4 + 4 x Multi-Frame Generation, GDDR7 and a bit more CUDA cores is the rational move, not a flex.
Recent EU price trims🔗 (~10%) on the 50-series can strengthen the value case for you.
Coming from an RTX 3070 OC (8 GB GDDR6, ~448 GB/s, 265 W TDP), choosing a Palit Gamerock RTX 5080 wasn’t a flex – it was more of a reliever and time saver. It gave me double the VRAM 8 – 16GB, double the bandwidth, double the CUDA cores, at around 360W TDP. When I saw double, double, double, I waited a couple of months for these price anomalies to appear and I jumped in.
1.2) The silicon that matters
Before diving into workloads, I like to line specs side by side. It doesn’t tell the whole story (architecture, driver maturity, and game optimization all matter), but it helps frame why I moved from the RTX 3070 -> 5080 instead of stopping at the 4080 SUPER.
| Spec | Gigabyte GeForce RTX 3070 Gaming OC | GeForce RTX 4080 SUPER | Palit Gamerock GeForce RTX 5080 | 
|---|---|---|---|
| Chip name, nm node | GA104 – Samsung 8 nm (8N) | AD103 – TSMC 4N (5-nm class) | GB203 – TSMC 4N (5-nm class) | 
| CUDA cores | 5,888 | 10,240 | 10,752 | 
| Mem bandwidth + GDDR | ~448 GB/s – GDDR6 @14 Gbps (256-bit) | ~736 GB/s – GDDR6X @23 Gbps (256-bit) | ~960 GB/s – GDDR7 @30 Gbps (256-bit) | 
| VRAM | 8 GB | 16 GB | 16 GB | 
| Boost clock max (reference) | ~1.73 GHz | ~2.55 GHz | ~2.91 GHz | 
| TDP / Max board power | 220 W | 320 W | 360 W | 
| Other tech AI / APIs | DLSS SR (no FG) Vulkan 1.4, OpenGL 4.6, DX12U | DLSS 3 FG (DLSS 4 features via app) Vulkan 1.4, OpenGL 4.6, DX12U | DLSS 4 Multi-Frame Gen (native) Vulkan 1.4, OpenGL 4.6, DX12U | 
| Game FPS #1 4K Ultra raster geomean (16-game suite) | ~36 FPS | ~64.8 FPS | ~71.1 FPS | 
| Game FPS #2 Cyberpunk 2077, 4K Ultra, RT off | ~30 FPS | ~59 FPS | ~65 FPS | 
| Game FPS #3 Horizon Forbidden West, 4K Ultra | ~40 FPS | ~92 FPS | ~100 FPS | 
| I paid (or could’ve paid) + year/month | €825.86 ; y:2020/7 | €1,109 ; y:2025/5 | €1,297.78 ; y:2025/5 | 
1.3) Workloads I feel & care about
– 3D & video: more cores + faster memory = shorter GPU-bound renders, encodes, and heavy effects. (I care about minutes saved, not just FPS.)
– Games: DLSS 4 Multi-Frame Generation lifts smoothness dramatically and leans on improved AI models (less VRAM overhead, better reconstruction, better culling). Even where raw 5080-vs-4080 gains are modest, DLSS 4 keeps 4K “effortless.”
1.4) Brand & SKU
No need for halo quadruple coolers. I bought the Palit Gamerock (non-OC) simply because it won EUR/FPS in my market. Spend where it changes outcomes, not on logos. I keep the PC free of RGB – useless power usage that only steals focus.
For example, the ROG Astral RTX 5080 OC pulls ahead of my Palit Gamerock RTX 5080 non-OC by ~5 – 7% in 4K average gaming, thanks to a higher boost clock and more aggressive voltage/cooling tuning – when thermals allow. In real use, expect 3 – 5% uplift in select cases. 
The price difference?
– Totally not worth it. Oh, you’re thinking I gave up on cooling? Here’s an OCCT snippet:
2.1) CPU runway
It isn’t the case for a new platform though. My current Z690UD board supports PCI-E 5.0 at full 32GT/s, while the i7-12700K is blasting 1:1 with the RTX 5080. It’s being kept cool with the Arctic Liquid Freezer III 360mm AIO.
If you’ve seen the video above, the CPU/GPU bottleneck is pretty balanced, in a harsh environment (Alan Wake 2 1440p DLSS Quality – Ultra RT/PT, 2x FG).
Neither were overclocked in that test – I’ll have to check the BIOS for the CPU, as the last update reset the settings, and I think I forgot to switch some power saving thing off, as it didn’t apply my mild 5.0GHz all core OC.
If Intel’s next cycles don’t land well, I’ll pivot to Ryzen later – but only after extracting full value from this platform (sane minimalism). Here are the top 2 CPUs from Intel and AMD, compared to the i7-12700K.
| Spec | Ryzen 9 9950X3D | Core Ultra 9 285K | Intel Core i7-12700K | 
|---|---|---|---|
| Process / nm | TSMC 4 nm | TSMC 3 nm | Alder Lake, Intel 10 nm | 
| Cores / Threads | 16 / 32 | 24 / 24 | 12 / 20 | 
| Boost clock (max) | 5.7 GHz | 5.7 GHz | 5.0 GHz | 
| TDP / PL2 | 170 W / 230 W | 125 W / 250 W | 125 W / 230 W | 
| Price in EU | 700-720 € | 700-720 € | 460€ paid in Feb 2022 | 
| Single-core performance | ~ +35% vs 12700K | ~ +27% vs 12700K | Baseline (used for comparison) | 
| Multi-core / parallel throughput | ~ +120% | ~ +93% | Baseline (used for comparison) | 
| Game FPS #1 (Alan Wake 2, 4K Ultra raster, no DLSS/FG) | ~62 fps | ~58 fps | ~32 fps | 
Aggregated from multiple sources.
2.2) Power + thermals
Official TDP ~360 W, with recommended system power 850 W; a single 12VHPWR (or 3×8-pin via adapter). In real gameplay, many reviews log lower average draw. Everyone whines about high power draw, but here is where it’s at, whether we like it or not. Nothing’s for free, and this performance comes at a price.
I’m not a PC build professional, but the RM1000x and tidy cabling make it silent under load. Never cheap out on the PSU, but don’t go wasting money on it. Taking my PC as example, the RM1000x is on the high-end for this purpose. I replaced my old RM750, which didn’t have the ATX 3.1 standard for the current standard 12v-2×6 single cable. It worked by daisy-chaining 3×8-pin, but not without major issues. Keep reading to understand why.
Rushed, amateur build that stays cool.
2.3) Case, clearance and lessons learned
The RTX 5080 is large and heavy (mine: ~3.230 kg; 332 x 150 x 70 mm). My old Thermaltake View 28 couldn’t fit it without forcing a dangerous bend on the 12V-2×6. I moved to the Endorfy Arx 700 Air – big and breathable – so the problem disappeared.
Early on, my RM750 worked via a daisy-chain 3×8-pin – 12V-2×6 adapter, but the bend was so severe that the new sense pin cut power as protection. Nasty scares until I understood what was tripping it. Lesson learned: 12V-2×6 must have a good large bend, and no other cables to sit on it. Better to route it from above so gravity doesn’t pull it down. Plan the bend radius coming out of the connector (no hard turns, no side panel pressure). Keep the run relaxed and supported. This is why, in the PC overview picture above, I prioritized the 12V-2×6 cable over anything else, even if it’s not pretty.
If you want to use a GPU riser, get a good one and make sure your case supports it and the GPU size before buying.
Drivers & BIOS maturity: early-gen drivers settle fast – update cadence matters in month one. My biggest gremlin was actually motherboard firmware: the Gigabyte Z690 UD BIOS was a mess until F30 (June 19). I chased thousands of WHEA errors, even forced the slot down to PCIe 4.0 (fewer errors, but were still present). After flashing F30: zero errors at any PCIe speed. Lesson learned – don’t jump to blame the GPU; firmware first. Here are some photos from that era:
My old RTX 3070 paranoia-cleaning broke the PCIe lock-pin – but with the 5080 snug in, bracket screws tightened (care here as well, never overtighten stuff, it might create a big pull force), and stabilizers in place, the slot holds steady and runs safe.
Supports & paranoia: I use two GPU supports to reduce slot stress and sag. Remember: never push the GPU upwards – just adjust until the card gently rests on them. Check the above caption as well, never overtighten screws on a PC build. To calm my paranoia, I even placed a stabilizing bubble level near the PCIe slot to check alignment. Don’t worry if the card doesn’t sit perfectly level at the front end – that’s normal due to uneven weight distribution.
No sag achieved, hopefully.
Extra cooling lesson: if you buy aftermarket heatsinks for NVMe SSDs that sit under/behind the GPU, remember they add thickness. A chunky heatsink can prevent the GPU from seating in the PCIe slot at all. Either use low-profile pads/heatsinks or plan your slot layout first.
Voltage stability: also, a nice touch – get one of those AVR relay voltage stabilizers, especially if you live in a zone with frequent brownouts or drops. Get one that allows you to change its fuse if it fries.
If you’ve got the budget, step up to a UPS to save work before shutting down. For me, the AVR is enough: it cuts instantly but cleanly, without “dirty” surges – and I frenquently save work, muscle memory from previous losses, I guess.
I know that modern PSUs are well fitted with several protection levels, but hey, better to fry the fuse on the wallsocket, then the AVR and finally, the PC parts.
👉 Clarification: an AVR cleanly protects hardware, but your PC will still shut off instantly if the power drops – the risk is only data loss or file corruption if you’re mid-write. A UPS adds a few minutes of runtime for a safe exit, but costs more. Hardware-wise, AVR alone is fine.
In conclusion, the RTX 5080 is my buy and forget card. It was the rational step over a 4080 in my market, and it turns watt-hours into saved minutes: faster exports, better gaming, and enough headroom to skip a generation or two without fiddling with settings.
Yes, it’s big and thirsty, but in a roomy case with a native 12V-2×6 and quiet airflow, it disappears until needed. A GPU I plan to keep for years. I just wish the 80-class shipped with 20-24 GB VRAM.
Heavy and light. That’s the shape of this whole page – and, honestly, my way of building systems now. The tower gives me time back; the Air gives me freedom back. I don’t need them to impress anyone. I need them to vanish while I work or relax.
I’ve learned a few things on this run:
On nova, it seems that we’ve settled on “Muscle at the desk, Silence in the bag.” That’s the way it should be. Thanks for traveling through the heavy and the light with me. More field notes, tools, and small rebellions against bloat are already in the pipe. Stay curious, stay kind, and keep your tools aligned with your life.
Signing out,
–Theo