Nvidia, You’re Late. World’s First 128GB LLM Mini Is Here!

Credibility score: 73/100 — Mostly Credible. Mixed credibility - some claims are solid, others need verification.

Claims analyzed

Nvidia's DJX Spark is still weeks away, always 'coming soon' — Opinion (50/100)

Sassy shade on the eternal 'coming soon' — fair roast on delays 😏

GMKtec Evo X2 has 128GB unified RAM for LLMs, half price of DJX Spark — Solid (85/100)

👌💯

Dual PCIe Gen4 SSDs up to 16TB, Wi-Fi 7, shareable memory via static partitioning — Solid (75/100)

Checks out 👌

Testing Asus ROG Flow Z13 with 128GB chip — Solid (80/100)

👌

LM Studio offloads all 64/64 layers of Q4 Quinn 32B to GPU — Solid (75/100)

✅

Static partitioning: copies from system to GPU memory, unlike Apple's unified — Verified (90/100)

Spot on 👌

Quinn 32B Q4 is 18.4GB, gets 10.58 tokens/sec — OK (65/100)

Plausible benchmark ⚠️

Offloading LLM layers to CPU makes it very slow — Solid (85/100)

👌✅

M4 Mac Mini at 31W, 19.94 tokens/sec — OK (60/100)

Specific benchmark shown — no contradiction but no public match either ⚠️

Apple M4 chip has 120 GB/s memory bandwidth — Verified (95/100)

Spot on number ✓

Chat LLM Teams sponsors with model routing — Sponsored (50/100)

Full-on ad read for Chat LLM Teams — $10/mo dashboard 🚀

Measured 96 GB/s triad on M4 STREAM benchmark — Personal Story (70/100)

Plausible real-world measurement ⚠️

DeepSeek R1-Distill-Qwen-7B Q4 is ~4 GB — Solid (80/100)

👌

Chat LLM features PDF chat, humanizer, agents — Sponsored (50/100)

Continuing the pitch — agents, humanizer, all-in-one AI suite 💼

Llama CPP compiles for ROCm and Vulkan — Solid (85/100)

✅

ROCm not available yet for this setup — Dubious (45/100)

ROCm 'not available'? Tell that to AMD's 2026 docs 🚩

Sources: Compatibility matrix - ROCm Documentation - AMD, Limitations and recommended settings — Use ROCm on Radeon GPUs, Installation prerequisites — ROCm installation (Linux) - ROCm Documentation - AMD

Vulkan Llama.cpp: 935 PP, 46 TG tokens/sec — OK (65/100)

Plausible benchmark numbers for Vulkan on beefy AMD hardware ⚠️

M4 Mac Mini gets 228 PP512, 21 tokens/sec on Llama Bench — Personal Story (70/100)

Dude's own benchmarks — fair play 👌

M4: 120GB/s, M4 Pro: 273GB/s, M4 Max: 540GB/s, M3 Ultra: 819GB/s memory bandwidth — Verified (100/100)

✅

Nvidia DGX Spark has 279 GB/s memory bandwidth — Solid (80/100)

👌

This machine's bandwidth at 279 GB/s, higher than M4 Pro — Solid (80/100)

👌

M4: 96/120 GB/s Stream; M4 Pro: 209/273 GB/s — Solid (85/100)

✅

This machine's Stream benchmark: 120 GB/s vs expected 256 — OK (65/100)

Low real-world vs spec — common but underwhelming ⚠️

Gemma3-1B: ~160 t/s; Llama3.2-1B: 130 t/s on this machine — OK (60/100)

Self-reported benchmarks — impressive if real 📈

Gemma 3 12B: 23.9 t/s, Qwen 2.5 32B: 10.8 t/s, Llama 3 70B Q4: 5 t/s — OK (65/100)

Benchmark numbers look real but Llama 3 '370B' is sus — it's 70B 😂⚠️

Knuckbox Evo X2 desktop beats Flow Z13 laptop in most LLM benchmarks — Dubious (45/100)

'Knuckbox' or 'Nugbox'? And no hits on that exact desktop model 🚩🤔

M4 Max: 49 t/s Gemma 3 12B; Nvidia Spark Blackwell has 278 bandwidth — Solid (80/100)

👌 Numbers align with unified memory advantages

Static memory partitioning is stable but wastes memory and lacks flexibility for modern PCs — Solid (85/100)

✅

AMD at Computex promised better ROCm support for LLM chips like this one — Solid (80/100)

👌

Michael Larabel got ROCm working on this chip but unstable with seg faults — OK (65/100)

Specific tester named but sounds like fragile hack job ⚠️

See the full analysis with sources and timestamps →