Hands-On with QSSR - AI Upscaling For Standard PS5 - Image Quality, PSSR Comparisons + More

Summary of the video (technological concepts + findings)

Sony’s QSSR (Quick Spectral Super Resolution) is presented as a new machine-learning based upscaler now available on standard PS5 via a patch for Marvel’s Wolverine and Ghost of Yote. The video argues QSSR can function as a “drop-in replacement” for games that already support PS5 Pro’s PSSR, delivering notably better image quality even though base PS5 lacks the Pro’s dedicated ML acceleration.

---

What QSSR is (and how it relates to PSSR)

• QSSR is described as a companion technology to PS5 Pro’s PSSR.
• On base PS5, QSSR is adapted/streamlined from PSSR:
  • It is effectively a refactoring of PSSR to work with FP16 (which base PS5 supports).
  • This adaptation is needed because base PS5 does not (as far as is known) have dedicated ML silicon / “300 TOPS”-class acceleration like PS5 Pro.
• Sony’s engineer (principal engineer, graphics R&D) characterizes it as:
  • a drop-in replacement for titles that already integrate PSSR.
  • potential minor follow-ups may be required (e.g., higher quality mips/assets).
• Oliver notes the technical approach is surprising but plausible:
  • moving from INT8/low-precision formats (fast on ML-accelerating hardware) to FP16 (more general, potentially slower).
  • despite this, results appear shippable rather than proof-of-concept.

---

Image quality findings (Marvel’s Wolverine: QSSR on vs off)

The hosts highlight real-time in-game toggles (menu QSSR on/off switch) and emphasize that differences are visible immediately.

Visual improvements attributed to QSSR

• Sharper, cleaner detail in foliage and surfaces.
• Reduced temporal instability (less flicker / crawling).
• Remedies for artifacts that appear without QSSR, including:
  • Subpixel shimmering and instability on embroidery/textures.
  • Cleaner edges with less “pixel popping/crawl.”
• Improved handling of key content types:
  • Water: less ghosting; sometimes described as slightly “painterly,” but clearer.
  • Hair: smoother silhouettes, less breakup/noise.
  • RT reflections: smoother and more temporally consistent.

Comparison against another upscaler (IGTI)

Compared to Insomniac’s referenced stock method (called IGTI in the transcript), QSSR is repeatedly described as producing fewer artifacts, especially in:

• specular flicker,
• fence/railing subpixel breakup,
• ghosting-heavy elements like water.

---

PSSR (Pro) vs QSSR (base PS5) comparisons

The video positions QSSR as broadly similar to PSSR in what it improves:

• Reported as generally similar in edge construction and reflection-resolution behavior.
• A notable difference: hair behavior may vary, possibly due to:
  • higher internal resolution on PS5 Pro,
  • different hair rendering tuning across performance modes.

Resolution/tier observations in Wolverine

Three “quality tiers” are discussed:

1. Base PS5 performance mode
2. Base PS5 with QSSR
3. PS5 Pro with PSSR

Overall takeaway: QSSR provides a meaningful quality bump, even if it doesn’t reach Pro-level “pristine” quality.

---

Performance analysis (base PS5 computational cost)

Main takeaway

QSSR costs frame time, but the hosts argue it’s worth it in the settings where developers most need it.

Test methodology described

• Use cutscenes and VRR for repeatable measurements.
• Attempt “like-for-like” resolution matching (QSSR on vs off), even if internal resolutions differ slightly.

Reported performance costs

• Example cutscene (VRR, ~864p internal baseline described):
  • FPS drops (headline: ~8–10 fps in the example).
  • More importantly, QSSR adds about 1.5–1.8 ms of frame-time cost.
• Higher output case (~2160p/4K output described):
  • QSSR cost can rise to about ~4.4 ms when FPS differences are larger.

Conclusion

• QSSR is most useful for 60fps performance modes targeting around 1440p output from lower base resolutions.
• If games use dynamic resolution to regain performance headroom, the quality benefits may be retained at the target frame rate.

---

Ghost of Yote: performance behavior differs from Wolverine

Time constraints limit coverage, but the video reports:

• Performance impact is slight / near-imperceptible compared to Wolverine.
• The hosts speculate QSSR may be operating at a similar computational cost to FSR3 in this scenario.
• They caution that more verification is needed across:
  • additional games,
  • different dynamic-resolution behavior.

---

Broader implications / expectations for adoption

Why adoption could be easier

• QSSR looks especially promising because:
  • if a game already supports PSSR, QSSR can likely be retrofitted with relatively low effort.
• Sony claims retrofits are relatively painless (as described in the transcript).

What the hosts want to see next

More testing in third-party titles, particularly those using FSR in 60fps modes, where image quality has been weaker.

Examples they specifically want to see improved with QSSR:

• Alan Wake 2
• Avatar
• Assassin’s Creed Shadows
• other large third-party “heavy hitters,” plus possible upgrades to existing PSSR-equipped titles

---

Closing sentiment

The video frames QSSR as a surprisingly successful ML upscaler port to base PS5 (despite lacking dedicated ML acceleration). Sony is praised for making a machine-learning upscaler workable on hardware not originally designed for ML acceleration.

Overall: a promising debut, though the dataset so far is limited (primarily Wolverine + a smaller look at Ghost of Yote).

---

Main speakers / sources

• Rich — primary host; leads demonstrations and test explanations
• Oliver McKenzie — co-host; contributes analysis, test setup, and comparisons
• Sony source: Daniel Craig — Principal Engineer, Graphics R&D, SIE (Sony Interactive Entertainment) (as referenced in the transcript)