AMD Instinct MI350P First Look: 144GB HBM3E in a 600W PCIe Card
Comments on “AMD Instinct MI350P First Look: 144GB HBM3E in a 600W PCIe Card”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
StorageReview's first look at AMD's Instinct MI350P, a dual-slot PCIe card with 144GB of HBM3E and 4TB/s of bandwidth, tested it at 450W against NVIDIA's RTX PRO 6000 (600W) and H200 NVL. On GPT-OSS-120B, one card reached 5,057 tokens per second at 128 streams on chat prompts, 65% ahead of the RTX card. Results on long prompts are mixed, and the two-card models ran on an older ROCm stack. The extra memory cuts card counts, but full-power results are still pending, and the claim check found most figures unsourced.
Why it ranked
No. 12 of 139 with a score of 0.70. The panel rated it highest on new information (0.94) and technical depth (0.93), lowest on audience breadth (0.58). Placed No. 12, No. 17 by score alone: the placement terms moved other stories past it.
- AI judgeJudge 2claude-haiku-5-5
- Detailed hardware benchmark of AMD MI350P versus NVIDIA cards with reproducible setup and stack versions; strong technical depth and substantiation, a product launch with real measured data, moderate audience breadth beyond AI infrastructure buyers.
- AI judgeJudge 3gpt-5.6-luna
- A detailed, benchmark-heavy first look at a new AMD accelerator, with reproducible test configurations, extensive model results, caveats about power and software stacks, and clear implications for AI infrastructure buyers.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.70total under rubric v2.3
- Significance0.65
- Reader interest0.81
- Audience breadth0.58
- Practical relevance0.73
- Substantiated0.85
- Technical depth0.93
- New information0.94
- Counts against: Promotional0.08
- Counts against: Overclaim0.23
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.68 | 0.50 | 0.78 | 0.65 | 0.28 | +0.24 | +0.157 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.69 | 0.85 | 0.90 | 0.81 | 0.21 | +0.26 | +0.211 |
| Audience breadthHow many people who build technology does it concern? | 0.54 | 0.50 | 0.70 | 0.58 | 0.20 | +0.13 | +0.075 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.64 | 0.70 | 0.86 | 0.73 | 0.22 | +0.22 | +0.161 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.84 | 0.80 | 0.91 | 0.85 | 0.11 | +0.05 | +0.043 |
| Technical depthHow technically substantive is it? | 0.98 | 0.85 | 0.94 | 0.93 | 0.14 | +0.05 | +0.046 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.88 | 0.95 | 0.98 | 0.94 | 0.10 | +0.05 | +0.047 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.07 | 0.10 | 0.08 | 0.08 | 0.03 | −0.19 | −0.016 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.34 | 0.20 | 0.16 | 0.23 | 0.18 | −0.11 | −0.026 |
| Sum of contributions | +0.699 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.699 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-11-05/12/"><img src="https://rubric.hrih.si/badge/2026-11-05/12.svg" alt="Ranked #12 on Rubric, 5 Nov 2026" width="238" height="28"></a>Claim check
Claim check: 7 no source given · 1 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
No source givenFirst, only or fastest
The AMD Instinct MI350P is AMD's first new PCIe Instinct GPU in four years, following the MI210 in 2022.
The article gives no link or citation for it.
The article says “The AMD Instinct MI350P is AMD’s first new PCIe Instinct GPU in four years, following the MI210 in 2022”
The source says No source to compare with.
Couldn’t checkNumber
The MI350P offers up to 4.6 PFLOPS of MXFP4 performance, 144GB of HBM3E and 4TB/s of memory bandwidth in a dual-slot, 600W air-cooled card.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “with up to 4.6 PFLOPS of MXFP4 performance, 144GB of HBM3E, and 4TB/s of memory bandwidth”
The source says No passage could be compared.
Source cited by the article: amd.com/en/products/accelerators/instinct/mi350/mi350p.html
No source givenFirst, only or fastest
The MI350P carries the most memory of any current PCIe accelerator, 3GB above the H200 NVL and 48GB above the RTX PRO 6000 Blackwell.
The article gives no link or citation for it.
The article says “The MI350P carries the most memory of any current PCIe accelerator, 3GB above the H200 NVL and 48GB above the RTX PRO 6000 Blackwell”
The source says No source to compare with.
No source givenBenchmark result
At 450W, the MI350P measured 580 TFLOPS in BF16, 1,120 in FP8 and 2,228 in FP4 in MAMF, 36 to 38 percent above the 600W RTX PRO 6000 at every precision.
The article gives no link or citation for it.
The article says “At 450W, one MI350P reaches 580 TFLOPS in BF16, 1,120 in FP8, and 2,228 in FP4”
The source says No source to compare with.
No source givenBenchmark result
GPT-OSS-120B on one MI350P reached 5,057 output tokens per second at 128 streams on the chat workload, 65 percent ahead of the RTX PRO 6000 and 50 percent ahead of the H200 NVL.
The article gives no link or citation for it.
The article says “GPT-OSS-120B reached 5,057 output tokens per second at 128 streams on the chat workload, 65 percent ahead of the RTX PRO 6000 and 50 percent ahead of the H200 NVL”
The source says No source to compare with.
No source givenBenchmark result
Eight MI350Ps in a PowerEdge XE7745 produced 48,971 tokens per second offline and 40,868 in the server scenario on GPT-OSS-120B in the MLPerf Inference v6.1 submission of September 16.
The article gives no link or citation for it.
The article says “eight MI350Ps in a PowerEdge XE7745 produced 48,971 tokens per second offline and 40,868 in the server scenario on GPT-OSS-120B”
The source says No source to compare with.
No source givenNumber
One MI350P reads its HBM3E at 3,291GB/s and writes at 3,204GB/s, about 80 percent of the 4TB/s rating, about 2.1x the RTX PRO 6000's usable bandwidth.
The article gives no link or citation for it.
The article says “One MI350P reads its HBM3E at 3,291GB/s and writes at 3,204GB/s, about 80 percent of the 4TB/s rating”
The source says No source to compare with.
No source givenQuote
AMD's estimate for the extra 150W at 600W is about 10 percent more performance.
The article gives no link or citation for it.
The article says “We hear that AMD’s estimate for the extra 150W is about 10 percent more performance”
The source says No source to compare with.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 4 pts, not on front page
Recorded for context only, checked 5 Nov 2026, 05:46 UTC. It played no part in the ranking.