Video summary

DO NOT BUY: AMD Ryzen 9 9950X3D2 CPU Review & Benchmarks | 24 Charts in 24 Hours

Main summary

Key takeaways

Product Review

Product reviewed

AMD Ryzen 9 9950X3D2 (aka 9950X3D II / X3D2) — a $900 16-core, 32-thread Zen 5 CPU for AM5 with stacked 3D V-Cache on both CCDs (total 192MB L3), plus a separate IO die. Claimed specs: 200W TDP, 4.3GHz base, up to 5.6GHz max boost.

Main features highlighted

  • Dual-CCD V-Cache everywhere: both CCDs use stacked 3D cache, totaling 192MB L3 (vs the earlier 9950X3D having 128MB and only one cache stack).
  • Simplified scheduling/core parking: the video argues this makes CPU scheduling easier (less complexity around “which CCD is faster/cached”).
  • Real-world framing: testing is presented “as consumers” using public parts/software, rather than AMD vendor-tuned guidance.
  • Power draw (Blender all-core worst case):
    • about 294W peak
    • about 285W nominal at the EPS 12V rails
    • software readings around ~250W

What the review found (performance)

Gaming performance (core message: gains are tiny vs price)

  • Roughly 0% to 6% faster vs Ryzen 9 9950X3D I.
  • Often within error bars / not visibly noticeable (“you could actually observe with your eyes”).
  • Examples mentioned:
    • Baldur’s Gate 3: 178 FPS, about +5% vs 9950X3D; price up about ~33%.
    • Outer Worlds 2: essentially tied (no meaningful benefit).
    • Kingdom Come: Deliverance II: about +2.6% vs 9950X3D at 1080p; still small differences at 1440p.
    • Dragon’s Dogma II: about +4.8% vs 9950X3D, but still framed as not worth it for gaming.
    • Cyberpunk 2077 (medium): within ~1%, functionally the same as 9950X3D.
    • F1 25: tied (363 vs 364 average).
  • Multiple charts are described as the CPU being “top-ish” but not moving the needle.

Productivity / “experimental” workstation benchmarks

  • Mostly 0% to 6% uplift over 9950X3D I in the video’s production/general tests.
  • Main exception: OpenFOAM (experimental)
    • about +34% vs 9950X3D I
    • presented as cache-sensitive, but explicitly labeled experimental and potentially less reliable.
  • Other SPEC Workstation items (experimental charts referenced):
    • LAMMPS: 2.09 (vs ~270K plus 2.04; vs 9950X3D I 1.97) → about ~6% vs X3D I
    • Convolution: 3.07 vs 2.92 (X3D I) → about ~5% better; Intel 270K plus at 2.44
    • NAMD: 2.4 vs 2.28 (X3D I), but Intel 270K plus scored 2.87 and is framed as better on value
  • Production benchmarks (non-SPEC-style):
    • Blender rendering: effectively the same (5.8 min vs 5.9 min)
    • Chromium code compile: improvement shown
      • 9950X3D2: 88.1 min
      • 9950X3D I: 95 min → about ~7.2% faster
    • 7-Zip compression: about +9% with the extra cache
      • ~215,000 MIPS vs a lower number on X3D I
    • 7-Zip decompression: no meaningful benefit
      • reruns sometimes tie/near-tie
      • cache advantage doesn’t consistently help

Overall takeaway from charts

  • Most workloads: 2–5% improvement (often less), sometimes literally within variance.
  • Gaming: improvements are described as not worth noticing at the $900 price.
  • Central flaw: value—the uplift doesn’t justify the cost.

Pricing / value comparisons (strongly emphasized)

  • $900 launch price: described as “astronomical” / high-end.
  • vs AMD 9950X3D I: about $657 listed (video notes roughly ~37% more expensive).
  • vs AMD 9800X3D: around $464; framed as better value for gaming.
  • vs AMD 7800X3D: around $389; described as a very strong value path.
  • Competition: Intel Core Ultra 270K Plus
    • around $300 at launch, ~$350 at retailers in the video
    • in several non-gaming tests, it’s described as punching close to AMD’s $500–$520 9950X tier and offering strong productivity pricing.
  • Threadripper: mentioned as faster for some workstation cases, but “priced out” of direct value comparison.

Pros mentioned

  • Top performance in select production tasks, e.g. 7-Zip compression (near top of chart) and Chromium compile gains.
  • Excellent gaming leadership at the top end (often near-tied for highest FPS), but still not worth it.
  • OpenFOAM outlier: potentially very meaningful cache sensitivity (~+34%) if it generalizes to your workload.
  • Boost behavior matches claims (mostly): max boost commonly ~5580–5625 MHz vs stated 5.6GHz.

Cons mentioned

  • Primary con: not worth $900
    • “Almost 0% gain, sometimes 5%” is framed as insufficient.
    • Extra cost doesn’t translate into noticeable gaming improvements.
  • Gaming value is poor: better options exist at about half the price or less.
  • Production value is poor: the regular 9950X (non-3D) is framed as delivering most of what you’d want at $500–$520.
  • OpenFOAM caveat: big improvement is on an experimental workload → uncertain real-world applicability.
  • Architecture intent doesn’t pay off broadly: the cache/scheduling design doesn’t yield a wide enough uplift.

User experience / practical guidance

  • For gaming: don’t buy it expecting a “best gaming CPU.” Performance is described as frequently close or tied, and not worth the extra spend.
  • For production: only certain workflows show meaningful uplift; otherwise it behaves similarly to cheaper alternatives.
  • Recommendation: a direct “Do not buy” for general consumers based on the charted results and value gap.

Unique points about test methodology / presentation

  • Reviewer mentions the CPU arrived quickly and was used overnight to produce rapid multi-chart output.
  • Notes an aesthetic issue: the sample shipped appeared upside down (due to source orientation), claimed to be non-performance-affecting—it “just looks different.”

Concise verdict

Overall recommendation: Do NOT buy the AMD Ryzen 9 9950X3D2 at $900.

The video repeatedly concludes that improvements are typically 0–6% vs the cheaper 9950X3D I and often within variance, especially for gaming, while much better value options exist—especially 9800X3D / 7800X3D for gaming and 270K Plus or 9950X for productivity value.

The major exception—OpenFOAM (~+34%)—is on an experimental chart, so it isn’t enough to justify the premium unless that workload matches your needs.

Speakers / contributions (grouped)

  • Main reviewer (likely Dave / team-host voice): delivers the “don’t buy” conclusion, pricing/value framing, and most benchmark interpretation.
  • Methodology/verification voice (within the same narration style):
    • clock validation (boost checks around 5.6GHz)
    • confidence framing (experimental vs non-experimental benchmarks)
    • measurement context (power, variance, reruns)
  • Other referenced contributors:
    • Steve (Hardware Unboxed): credited for shipping the CPU
    • Arctic: mentioned as sponsor for MX-7 thermal paste, including counterfeit-check context (not a CPU feature)

Original video