Three months wrong about why my 4-node AMD cluster was slow
Credibility score: 75/100 — Mostly Credible. Mixed credibility - some claims are solid, others need verification.
BSmeter analyzed "Three months wrong about why my 4-node AMD cluster was slow" and rated it 75/100 for credibility (a BS score of 25/100 — mostly credible), on 2026-09-22. Its weakest claim — "Boasts about the APU's power and potential as an AI supercomputer." — scored 75/100 and was flagged as no frame. 5 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 5 claims analyzed: 0 scored under 40, 0 between 40 and 69, and 5 at 70 or above.
Claims analyzed
Boasts about the APU's power and potential as an AI supercomputer. — No Frame (75/100)
At 0:00
He's just setting the stage, hyping up the hardware's theoretical power. No lies, just big talk. 😈
Why this score: The speaker is introducing the capabilities of the AMD APUs, specifically their memory and processing power, and then extrapolating their potential when combined. This is a straightforward introduction to the problem he's about to discuss, not a deceptive claim.
Original quote: “More memory available to the graphics card than a 5090 or even the RTX Pro 6000. 128 GB of unified memory on one chip. Still to this day the most powerful X86 APU you can buy. And four of them on one desk should basically be an AI supercomputer in your living room.”
Admits to three months of misdiagnosis for his cluster's performance issues. — No Frame (75/100)
At 0:20
He's confessing his own screw-up, setting up the whole story. Refreshing honesty, for a mortal. 😈
Why this score: The speaker is openly admitting to a personal struggle and a period of incorrect troubleshooting. This is a direct, self-deprecating statement that forms the premise of the video, rather than a claim that requires external verification or is designed to mislead.
Original quote: “it took me over 3 months to get these things actually to work together in a cluster as fast as they finally did. And most of that was just me being wrong about what the problem even was in the”
Setting the stage with a clear goal: beat Minisforum's marketing video. No trickery here. — No Frame (75/100)
At 0:30
He's laying out his challenge, plain and simple. A rare moment of honesty before the inevitable human folly. 😈
Why this score: The speaker is clearly stating his objective for the video: to outperform a specific marketing benchmark. There's no hidden agenda or loaded language; it's a direct, verifiable goal.
Original quote: “about what the problem even was in the first place. Let me back up here. The chip is the AMD Ryzen AI Max Plus 395, aka Strix Halo. If you watch this channel, you've seen me try this chip on mini PCs and laptops from GMK Tech, Beelink, Framework, Asus. And this one is the Minisforum MS-S1 Max. It…”
Claiming 99% memory utilization and 37 billion active parameters for DeepSeek-R1. — No Frame (75/100)
At 1:05
He's giving specific, technical details about his setup and the model. These are numbers, not opinions. 😈
Why this score: The speaker is providing concrete, measurable data points about his cluster's performance and the AI model's size. These are factual statements about the system's state and the model's architecture, which can be verified by technical means.
Original quote: “After a clean reboot, all four nodes are running. Memory utilization is at 99% on all four machines. And it's running. It's actually printing everything out for me. Now, this is actually 37 billion active parameters in DeepSeek-R1.”
Comparing his token generation rate to Minisforum's marketing video, claiming a win. — No Frame (75/100)
At 1:32
He's presenting his results against a direct benchmark. The numbers are right there, a clear comparison. 😈
Why this score: The speaker is directly comparing his measured performance (6.23 tokens/second) against a specific, cited benchmark from Minisforum's own marketing (5.94 tokens/second) under similar conditions. This is a straightforward, verifiable comparison of results.
Original quote: “4,058 tokens generated in 10 minutes. And guess what? Minisforum's own product video, same chip, same four nodes, DeepSeek-R1 quantized to 4 bits. Their numbers, 5.94 tokens per second. Mine, with VLN tensor parallelism, 6.23 tokens per second. Yes.”
See the full analysis with timestamps →