Dmitry Vlasenko

Programmer 👨‍💻, traveler (18 countries) 🛩

[Dmitry Vlasenko @DimonDeveloper]

@AngelGa66156513 @videotech It’s my free opinion, but VGC probably got money from Microsoft for that article

[Dmitry Vlasenko @DimonDeveloper]

@OutofGalaxyy iPhones get 6–7 years of support, and every supported device gets the new iOS on day one. Within 3–4 months, ~60–70% of users update.
Add another 3–4 years of app support for older iOS versions from devs, and that’s ~9–10 years of support. Android adoption takes much longer.

[Dmitry Vlasenko @DimonDeveloper]

@Sensible_Swag @markgurman No, it won’t. They’re in totally different price categories (the Fold is at least 3-3.5 times more expensive) and cater to bit different customer segments.

[Ahmad @TheAhmadOsman]

Local AI hardware = capacity * bandwidth * software stack

- Capacity tells you what fits
- Bandwidth tells you how hard the box can breathe
- The software stack tells you how much of the spec sheet you can actually cash out.

Hardware by Memory Bandwidth
- Mac Studio M3 Ultra: up to 512GB @ 819 GB/s
- RTX PRO 6000 Blackwell: 96GB @ 1792 GB/s
- RTX 5090: 32GB @ 1792 GB/s
- RTX 4090: 24GB @ 1008 GB/s
- RX 7900 XTX: 24GB @ 960 GB/s
- Radeon PRO W7900: 48GB @ 864 GB/s
- AMD Radeon AI PRO R9700: 32GB @ 640 GB/s
- Intel Arc Pro B65: 32GB @ ~608 GB/s
- Tenstorrent Wormhole n300: 24GB @ 576 GB/s
- Tenstorrent Blackhole p150: 32GB @ 512 GB/s + 800G
- MacBook Pro M5 Max: 460-614 GB/s
- MacBook Pro M5 Pro: 307 GB/s
- DGX Spark: 128GB @ 273 GB/s (coherent + CUDA)
- Mac mini M4 Pro: 273 GB/s
- Ryzen AI Max / Strix Halo: ~256 GB/s (~96GB usable GPU)
- MacBook Air M5: 153 GB/s
- Snapdragon X2 Elite: 152-228 GB/s
- Intel Lunar Lake: 136 GB/s
- Snapdragon X Elite: 135 GB/s
- Mac mini M4: 120 GB/s
- Arc Pro B60: 24GB @ ~456 GB/s

Verdict

- GPUs are still the bandwidth kings

- Apple wins: stupid amounts of memory, don't want to shard across GPUs
- Apple loses: when raw tokens/sec & concurrency matter more

- DGX Spark: coherent memory + NVIDIA stack

- Strix Halo / Ryzen AI Max: first real x86 unified-memory contender

- Tenstorrent: fully OSS stack, excited to see this mature

Fitting != serving

Even if it fits, you still pay for
- bandwidth during decode
- KV cache growth
- dequantization
- batching + concurrency
- scheduler quality
- framework overhead

The only mental model that matters:

1. What must fit?
2. What bandwidth tier do I need?
3. What software stack can actually deliver it?

In short:
- NVIDIA -> fastest raw speed
- Apple Studio M3 Ultra -> biggest one-box memory
- Strix Halo -> first real x86 unified
- DGX Spark -> coherent NVIDIA dev appliance
- AMD / Intel Arc -> rising alternatives
- Tenstorrent -> fully opensource stack

Ask: "which bottleneck am I buying?"

Do not ask: "which hardware is best?"

[Dmitry Vlasenko @DimonDeveloper]

Here’s how the #Geekbench 7 Metal GPU benchmark compares to #Geekbench 6.

• GPU scores are lower across every device I tested.
• The drop is remarkably consistent, around 5–10%, regardless of the SOC.

[Dmitry Vlasenko @DimonDeveloper]

I retested all of my #Apple devices with #Geekbench 7. Here’s CPU benchmarks.

The interesting part:
• Single-core scores are consistently lower across every device.
• But on some devices, multi-core scores are significantly higher (M1 Pro: +14.3%, M2 Pro: +11.5%).

[Dmitry Vlasenko @DimonDeveloper]

@VadimYuryev Why would anyone want a Dynamic Island on the iPad? It would only make the user experience worse.

Apple should simply reduce the bezels as much as the technology allows, and that’s it.

Future iPhones will eventually get rid of the Dynamic Island too.

[Dmitry Vlasenko @DimonDeveloper]

@PlayStation Expected #PS6 specifications.
#PSSR3 combined with the PS6’s massive AI compute upgrade could dramatically improve upscaling quality, delivering sharper visuals, frame generation and more stable image

[Dmitry Vlasenko @DimonDeveloper]

It’s not an iPhone in @forallmankind_ , it’s a Newton. Apple released the first Apple Newton in 1993 but discontinued it in 1998; apparently, in this universe, @Apple continues to make new models.

[Dmitry Vlasenko @DimonDeveloper]

Apple exist in the @forallmankind_ universe. In the new season they’ve already got iPhones but they look kinda weird. Way thicker, and rough design that never existed 😄.
The camera bump looks exactly like the one on the iPhone X. I like polished borders but UI looks terrible.

[Dmitry Vlasenko @DimonDeveloper]

@LeakerApple The 70W charger only costs $20 USD more in the U.S. ($24 CAD in Canada), and it’s free in Australia — what were you thinking when you ordered your MacBook Air?

[Dmitry Vlasenko @DimonDeveloper]

My first Apple display 🤩 — the brightness and contrast are amazing. I think it’s overpriced, but still worth it.

[Dmitry Vlasenko @DimonDeveloper]

@EthanLipnik LG 27GM950B 5K 165Hz, 2304 Local Dimming Zones

[Dmitry Vlasenko @DimonDeveloper]

@GregoryMcFadden It’s actually the worst. I got a Vision Pro with the M5 and used it for work with Mac Virtual Display. The brightness isn’t nearly enough, and the text is blurry too. It’s great for movies and images but not for text.

Plus, it’s really uncomfortable because it’s too heavy.

[Dmitry Vlasenko @DimonDeveloper]

@LeakerApple @VadimYuryev One guy is left, who cares?
Snapdragon X Elite 2 is a big performance leap, so it’s too early to say Windows on ARM is dead.
Software support is the real weakness. More ARM laptops sold means more developer support.
Desktop ARM CPUs would help but needs motherboard manufacturers.

[Dmitry Vlasenko @DimonDeveloper]

@akizukitooru It won’t get faster with more RAM; it’s about GPU architecture and memory bandwidth. My M1 Pro (16-core GPU, 32GB, 200GB/s) generates the image in 220 seconds, while the M4 Pro (16-core GPU, 24GB, 273GB/s) does it in 100 seconds. A better GPU and higher bandwidth increase speed

[Dmitry Vlasenko @DimonDeveloper]

@4nkpaked @GameRoll_ It seems like performance mode isn’t 60 FPS, but only 40

[Dmitry Vlasenko @DimonDeveloper]

@4nkpaked @GameRoll_ It has two graphics options: Fidelity 30 FPS and Performance 60 FPS. Performance mode runs with slightly lower resolution. At least I have this option on the 17 Pro Max.

[Dmitry Vlasenko @DimonDeveloper]

@digitalix Hey! I really enjoy your content. Could you test diffusion models like Flux2, Qwen, and Z-Image in ComfyUI on your setup? I’m curious how the performance scales when using 1, 2, 3, or 4 connected Macs, thank you.