RTX 5080's '220× Ray Tracing' Claim, Decoded (2026)

Your feed probably showed you a headline like "RTX 5080 gets 220× FPS boost with new DirectX ray tracing, 3.4× faster than RX 9070 XT" in the first week of November 2026. Two hundred twenty times. If that number were real for games, every GPU comparison on the internet would be obsolete by Friday. It isn't real for games — but the underlying numbers are real, verifiable, and interesting. The trick is that the headline multiplies one very specific, very weird measurement by nothing at all. Let's take it apart.

TL;DR

• The "220×" comes from Microsoft's D3D12RaytracingClusteredGeometry sample, run with Clustered Geometry on vs. traditional BLAS rebuild: 1.4 FPS → 310 FPS on an RTX 5080, as tested by hobbyist Osvaldo Pinali Doederlein and reported by Wccftech (relayed to Korean readers via QuasarZone on Nov 5, 2026).
• The speedup is almost entirely acceleration-structure rebuild elimination, not ray-tracing throughput: rebuild time dropped 720 ms → 1.0 ms, while ray-dispatch time actually got slightly worse (1.8 ms → 2.0 ms).
• The feature behind it is real and vendor-neutral: DXR Clustered Geometry + PTLAS (Partitioned Top-Level Acceleration Structures), shipped in DirectX 12 Agility SDK 1.722 Preview — Microsoft's largest DXR update since 2018.
• The RX 9070 XT also benefits (1.7 → 91 FPS in the same sample), but the feature-off results flip the story: 9070 XT 1.7 FPS vs. RTX 5080 1.4 FPS. No new "generational NVIDIA advantage" here — just a new code path NVIDIA's drivers currently execute better.
• No real game shows anything close to 220×. The sample has no LOD system and no shading/presentation load of a normal frame — it's a rebuild-cost stress rig by design.

The claim, as actually relayed

The Korean hardware board QuasarZone relayed a Wccftech piece ("NVIDIA RTX 5080 Sees Over 220X FPS Boost With New DirectX Raytracing Technology, Outpaces RX 9070 XT By 3.4X") on November 5, 2026. The relayed test results for the RTX 5080:

Metric (RTX 5080)	BLAS rebuild (traditional)	Clustered Geometry on	Change
Frame rate	1.4 FPS	310 FPS	~221×
Acceleration-structure rebuild	720 ms	1.0 ms	~720×
Ray dispatch	1.8 ms	2.0 ms	+0.2 ms
Acceleration-structure memory	4,716 MB	267 MB	−94.3%

(vs. RX 9070 XT at 91 FPS with the feature on; 1.7 FPS with it off. All figures as reported by Wccftech per the QuasarZone relay — the raw screenshots come from the tester himself.)

Notice what's already suspicious in that table: 310 ÷ 1.4 = 221, but 720 ms ÷ 1.0 ms = 720×, and dispatch regressed. So the "220× FPS" is a downstream side effect of a 720× faster rebuild inside a frame whose only meaningful work was the rebuild. Hold that thought.

What DirectX actually shipped

This is the part that's confirmed from official sources. Microsoft's DirectX engineering blog published "Clustered Geometry, Partitioned TLAS, Indirect Acceleration Structure Builds" as a DirectX 12 Agility SDK 1.722 Preview (with Shader Model 6.10). Three things in one package:

1. Clustered Geometry (CLAS) — triangles are grouped into small, reusable clusters of up to 256 triangles / 256 vertices, each with its own acceleration structure. Animate or edit a mesh, and most clusters are reused untouched; you only rebuild what changed. Compression: vertex positions are stored as distances to a cluster anchor point instead of full float triples, cutting memory.
2. PTLAS — a partitioned top-level structure so large scenes update region-by-region instead of rebuilding the whole TLAS.
3. GPU-driven (indirect) acceleration-structure builds — the GPU schedules its own build work without CPU round-trips.

If the concept sounds familiar, it should: this is essentially DirectX standardizing what NVIDIA has been shipping as RTX Mega Geometry (shown in the Witcher 4 "nanite foliage" demos). How directly the standard maps onto NVIDIA's proprietary tech is documented in an NVIDIA whitepaper per the Wccftech report — I can't independently confirm that paper's contents from NVIDIA's own pages, so treat the paper's specifics as [unverified] until NVIDIA's site links it. The Microsoft side, though, is solid: sample, spec, version, all public.

The workload: what the "220×" was measured on

D3D12RaytracingClusteredGeometry is one of Microsoft's public DirectX Graphics Samples — a self-described "cluster sandbox." Its entire point is to isolate and measure acceleration-structure construction cost: it tracks GPU build/update time, rendering time, and AS memory, so you can run matched workloads on different GPUs.

The tester (Osvaldo Pinali Doederlein, a longtime graphics enthusiast known in the community, did the AMD test first) fixed the animation to the same frame for both runs to make the comparison apples-to-apples. The heavy configuration in the AMD-side test contained roughly 91 million triangles (his reference point: Alan Wake II's most complex scenes can push ~200M triangles before culling — though he explicitly warns this is not a direct workload equivalence).

So the measured thing is: a frame that spends ~99.7% of its time rebuilding one BVH, before and after you make that rebuild basically free. A quick frame-time sanity check:

frame_time_ms = 1000 / FPS

RTX 5080, traditional rebuild:
  1000 / 1.4  ≈ 714 ms per frame
  720 ms rebuild + 1.8 ms dispatch = 721.8 ms  → matches ~1.4 FPS

RTX 5080, Clustered Geometry:
  1000 / 310  ≈ 3.2 ms per frame
  1.0 ms rebuild + 2.0 ms dispatch = 3.0 ms   → consistent (~310–330 FPS)

RX 9070 XT, Clustered Geometry on:
  1000 / 91   ≈ 11.0 ms per frame

The numbers reconcile perfectly. That's why the figure is huge without being fraudulent: make 720 ms vanish and the frame rate explodes. It just has no reference to a game frame, where shading, texture filtering, post-processing, and presentation each own a slice that Clustered Geometry doesn't touch. If rebuild is 5% of a typical game frame, the realistic win is eventually a few percent — not triple digits.

The RX 9070 XT leg of the story — and the footnote that deflates it

The same tester ran the identical sample on AMD's card a day earlier (RX 9070 XT, 1600×920, geometry-heavy config, AMD's Agility SDK preview driver 26.20.12.01 for RDNA 4): 1.7 → 91 FPS, a ~54× jump, with AS memory down nearly 90%. Microsoft's own earlier demo cited 1.2 → 79 FPS on an RX 9070. Even the feature-off baseline told a story: at 100 BLAS, clustering was ~2× faster; scale to 10,000 BLAS and it hit the 50× headline. And scaling the resolution to 4K shrank the multiple to 1.6 → 28 FPS (~17.5×), because at 4K actual ray work per pixel grows while the rebuilt structure stays the same size — the dispatch part (which got up to 40% slower per his observation) starts to matter.

Now the RTX 5080 vs. RX 9070 XT comparison, both states:

Setting	RTX 5080	RX 9070 XT
Clustered Geometry off	1.4 FPS	1.7 FPS
Clustered Geometry on	310 FPS	91 FPS

The 3.4× headline is the "on" row. The off row shows AMD slightly ahead. What that tells you: the two cards are competitive when they both pay full build costs, and the big gap appears only on the new code path — where NVIDIA's BVH compression lineage (mature since Turing, per the tester's explanation in the report) and driver implementation currently pull away. Hardware RT throughput alone doesn't explain a result that flips direction between modes. It's one synthetic sample and one tester, so treat the 3.4× itself as a capability signal, not a product ranking [unverified outside this test].

The buyer-decoder table

Here is the table that turns this news into a purchase decision tool. Read it left to right: marketing phrase vs. what it actually describes.

What the number claims	What it actually describes
"RTX 5080 is 220× faster with new DirectX RT"	One Microsoft sample (D3D12RaytracingClusteredGeometry), feature ON vs. a deliberately rebuild-bound OFF baseline. The OFF frame is ~99.7% acceleration-structure rebuild (720 ms of 721.8 ms) — the worst possible denominator
"New DirectX ray-tracing technology"	An API capability, not a product: Clustered Geometry + PTLAS + indirect build APIs in Agility SDK 1.722 Preview (SM 6.10). Vendor-neutral — AMD and Intel GPUs support it
"3.4× faster than RX 9070 XT"	Both cards on the same new path in one sandbox, one tester, no published test-system details (CPU, driver versions, run repeats). Feature off, AMD was ~21% ahead
"94% less ray-tracing memory"	AS memory only in this sample (4,716 → 267 MB). Console- and texture-level VRAM savings will be a fraction of that
"NVIDIA standardized RTX Mega Geometry" (the whitepaper angle)	NVIDIA's closed tech and Microsoft's open spec tallied up well in a paper per Wccftech [unverified] — the standard thesis is plausible, the details are paper-only so far
"Your games will run hundreds of times faster"	Nobody claimed that. The tester explicitly warned against it: no LOD in the sample, shipping path-traced games do far more optimization

Which numbers to actually wait for

Write this list down; it works for every "N×" headline of this kind, not just this one:

1. Independent reproductions. The sample is public and free — other outlets running both cards in the same session, same driver, with frame-time breakdowns, would confirm the reported figures within weeks. Until then it's one person's system.
2. A real game path. The first title or engine that ships DXR Clustered Geometry (beyond Microsoft's own demo scenes) and a "Mega Geometry ON/OFF in-game" comparison. Rebuild-bound scenes like dynamic foliage or destructibles will show gains; shaded cinematic scenes won't.
3. Out-of-preview status. This is Agility SDK 1.722 Preview, with a separate AMD developer preview driver and per the relay an NVIDIA Game Ready driver (621.01) already carrying support [unverified]. Preview APIs get renamed, re-broken, and re-tuned. Benchmarks of previews are promises.
4. The dispatch-time column. The healthiest sign that an implementation is balanced will be ray dispatch staying flat as rebuild time falls. The RTX 5080's 1.8 → 2.0 ms and the 9070 XT's up-to-40%-slower dispatch both hint the new path still has claws that scenes closer to real budgets will expose.

One more thing worth the wait: nobody should buy an RTX 5080 or RX 9070 XT because of this story. At a $999 list price for the 5080 (per NVIDIA's listing), a synthetic capability demo changes nothing about what either card runs in stores today. It changes what developers might build for 2027 — which is exactly the right frame's worth of excitement.

Wrap-up

• The 220× is arithmetically real inside D3D12RaytracingClusteredGeometry (1.4 → 310 FPS) and mechanically explained: DXR Clustered Geometry in Agility SDK 1.722 Preview removes a 720 ms rebuild that dominated 99.7% of the test frame. Blend it into a normal game frame and the same gain shrinks to whatever percent of the frame that rebuild occupied.
• RTX 5080's 3.4× over the RX 9070 XT holds only with the feature enabled — with it off, the Radeon led (1.7 vs. 1.4 FPS). One tester of the hands-on kind, one sample, zero test-rig documentation: interesting, but that wide should not be quoted as a product ranking.
• The genuinely important news is vendor-neutral: DirectX is standardizing the RTX Mega Geometry approach (CLAS + PTLAS) since 2018's biggest DXR overhaul. The buyer-relevant signal is "pay attention when a shipping game or engine adopts it," not any single-hardware multiplier.

Sources

• DirectX Developer Blog — Clustered Geometry, Partitioned TLAS, Indirect Acceleration Structure Builds (Agility SDK 1.722 Preview)
• Wccftech — NVIDIA RTX 5080 Sees Over 220X FPS Boost With New DirectX Raytracing Technology, Outpaces RX 9070 XT By 3.4X
• Wccftech — AMD's RX 9070 XT Sees Over 50x FPS Boost and Nearly 90% Lower Ray Tracing Memory Usage With New DirectX Technology
• Wccftech — Microsoft Unveils Biggest DirectX Raytracing Upgrade Since 2018
• VideoCardz — Microsoft introduces major DirectX Raytracing update, showing higher FPS and 90% lower memory usage
• Jon Peddie TechWatch — Microsoft advances ray tracing with clustered geometry
• QuasarZone — RTX 5080 220배 향상 기사 릴레이 (2026-11-05, Korean relay of the Wccftech piece)
• Microsoft DirectX-Graphics-Samples — D3D12RaytracingClusteredGeometry (the tested workload)