RTX 6090 specs and rumors: what's actually leaked (and what isn't)

Search interest in the RTX 6090 is already real, so here's the honest version: a set of next-generation NVIDIA specs has been circulating, it's specific enough to look like a leak, and almost none of it is verified. This page lays out exactly what's rumored, who said it, and which parts are speculation dressed up as fact. If you want a card you can actually buy today, jump to our best graphics cards of 2026 instead.

Where these "specs" come from

This matters more than the numbers. The detailed RTX 60-series specs that spread across the web in early 2026 all trace to one source: YouTube leaker Red Gaming Tech (RGT). Overclock3D wrote it up (their headline calls them "alleged"), forums and aggregators re-posted it, and even TechPowerUp's news desk covered the claim. But those aren't five independent leaks — they're five re-reports of the same RGT video. As one forum reader put it bluntly: "Not a leak, just speculation."

The precise-looking core counts you'll see quoted elsewhere (24,576 CUDA cores, 768 TMUs, 192 ROPs) come from a single place: a placeholder entry in TechPowerUp's GPU database. That entry is flagged noindex, uses a generic placeholder image, and describes the card with "expected to" language — it's the database's own projection, not a leaked figure. We don't repeat those numbers as if they were real.

The same care applies to the biggest number floating around — a ~28,000-CUDA-core / "256-SM" figure. It comes from a die-shot analysis (High Yield) of "Rubin CPX," which leaker kopite7kimi says is GR212, a data-center part — explicitly not the gaming die. (The die-shot itself showed 192 SM and up to 256 ROPs; the 256-"SM" / 28,000-core figure is a downstream projection for a hypothetical gaming variant.) The gaming flagship is a different die, GR202, so those numbers are almost certainly from the wrong chip. A separate Wccftech snippet listing 8,192 CUDA and 48 GB HBM2e is the same confusion — data-center Rubin specs, not a GeForce card.

The rumored RTX 6090, vs cards that actually exist

Here's the rumored 6090 next to two GPUs whose specs are confirmed (because they shipped). The 4090 and 5090 columns are NVIDIA's published specs (and if you're weighing today's flagships, see our RTX 5090 vs 5080); the 6090 column is rumor, source-linked in the sections below.

* No reported leak states a CUDA-core count for the 6090. The only figure in circulation (24,576) is a TechPowerUp database placeholder estimate, not a leaked spec.

The rumored RTX 60 series: 6090, 6080 and 6070

People search the 6080 and 6070 too. The same RGT leak maps the stack to the GR20x dies — all rumored, none confirmed:

All single-source (RGT), re-reported by Overclock3D and HotHardware. GR206/GR207 dies are expected below these but unspecced. For the full stack and what's actually known at each tier, see our RTX 60 series hub.

The architecture: "Rubin" on a 3 nm node

The one claim every source agrees on is the family: the RTX 60-series is rumored to use NVIDIA's "Rubin" architecture — the same generation name NVIDIA uses for its next data-center AI GPUs — with the flagship built on a "GR202" die on a TSMC 3 nm process (per RGT, via Overclock3D and TechPowerUp). RGT's leak also mentions 6th-generation Tensor cores and 5th-generation RT cores, and clock speeds "between high 2 GHz and low 3 GHz" — a modest bump over Blackwell, with the real gains said to come from architecture rather than raw core count.

On the memory side, the consistent rumor is 32 GB of GDDR7 on a 512-bit bus — matching the 5090's memory width. Memory speed is where sources diverge: one (@TeksEdge, summarizing RGT) lists 28 Gbps for about 1.79 TB/s of bandwidth; a separate YouTube analysis claims 32–36 Gbps for "over 2 TB/s." Neither is official.

On the software side, the same coverage flags a next-generation DLSS 5 as part of the Rubin feature set. NVIDIA's upscaling has been its single biggest edge over AMD, so a new DLSS generation would matter as much as any silicon change — but it's unconfirmed, like everything here.

Performance: a path-tracing play, not a raster leap

The headline rumor is ray tracing. Per RGT (via Overclock3D), NVIDIA is said to be targeting roughly 2× path-tracing performance over the 50-series — which is what TechPowerUp's coverage meant by "massive RT performance gains." Traditional raster performance is rumored to rise a much smaller ~30–35%, because the uplift is said to come from architectural changes, not a big jump in SM count (the 6090 is rumored at "up to 192 SM," versus the 5090's 170 — under 13% more).

Treat any single big number with suspicion: one YouTube video claims "40–50% real-world or up to 75% on paper" versus the 5090, but that figure isn't in the RGT leak the rest of the reporting is based on, so we flag it as uncorroborated.

Why path tracing, not raster? The rumored priority tracks the market. The next consoles — PlayStation 6 and Xbox "Project Helix" — are shaping up as the first heavy ray-tracing machines (both on AMD RDNA 5, rumored for 2027–2028). Once games are built ray-tracing-first for consoles, path-tracing headroom is where a PC flagship earns its price — exactly the case NVIDIA would be making with a 2×-PT Rubin card.

Nobody has benchmarked an RTX 6090 — no silicon has been tested. These are the rumored targets, all tracing to the same Red Gaming Tech leak, not measured results:

The through-line: the 6090 is rumored as a ray-tracing play, not a raster leap. When a real card ships, we'll replace this with measured numbers — the way our GPU FPS benchmarks cover today's cards — from the same value framework as our 2026 GPU rankings.

Price and release date

Release window — now less certain than it was. Through early 2026 the rumors converged on the second half of 2027 (per Tom's Hardware and @TeksEdge). That timeline is now contested. In mid-2026, reporting from The Information — re-reported by TechPowerUp and PCGamer — said NVIDIA pushed the RTX 60 series back beyond 2027, with some coverage floating a 2028 launch. (The nearer-term RTX 50 "Super" refresh, once reported shelved, has since resurfaced: as of mid-July 2026 the Super cards are reportedly already at board partners, with the launch on hold purely over the price of the new 3 GB GDDR7 modules — built and waiting, not cancelled.) The reason isn't engineering: a global DRAM/memory shortage is pushing NVIDIA to steer scarce GDDR7 toward far-more-profitable AI accelerators — the same squeeze we track in our GPU pricing explainer, and why high-VRAM cards for local AI command a premium right now. So the honest answer to "when": rumored H2 2027, but a 2028 slip is now on the table, and NVIDIA has confirmed nothing.

Price: there is no official or leaked MSRP — NVIDIA has announced nothing. What exists are estimates, and they've drifted upward as the memory crisis has worsened: the most-repeated figures now cluster at a $2,000–$2,500 starting point (PCGuide and community leak roundups), with fringe claims of $3,000+ tied to the same DRAM shortage. For grounding, the 5090 launched at a $1,999 MSRP and has sold above it, so a next flagship plausibly starts higher — but "plausibly," from third-party estimates, is the operative caveat. Treat any 6090 price as a guess, not a leak.

The release-date timeline: why it keeps slipping later

Worth stating plainly: no genuinely new RTX 6090 spec leak has landed since the March 2026 Red Gaming Tech video. What has actually moved all year is the timeline — and it has drifted later, not closer. The leakers don't agree:

Why it's slipping: the 2026 memory crunch. The delay isn't engineering — it's a global DRAM/GDDR7 shortage. AI-datacenter demand has booked memory production out through 2027 (SK hynix has warned of "chipflation"; China's CXMT is reportedly sold out into 2027), so NVIDIA is steering scarce GDDR7 toward far-more-profitable AI accelerators. Hard signals reported across 2026: RTX 50 supply cut roughly 20%, street prices up 15–19% since late 2025, and The Information reporting NVIDIA plans no new gaming GPUs in 2026 at all — the Super refresh included. It's the same squeeze we track in our 2026 GPU pricing explainer. The freshest reporting leans 2028 / late-2027; if you're timing a purchase, plan around "not soon."

What is not confirmed (the honest list)

• Anything from NVIDIA. No announcement, no teaser, no official spec.
• The CUDA-core count. No leak states one; 24,576 is a database placeholder.
• Board power (TGP) and price. Only third-party estimates exist.
• SM count. "Up to 192" (RGT) vs a one-off "256" (a YouTube video) — unresolved.
• Memory speed. 28 Gbps vs 32–36 Gbps, depending on who's summarizing.
• VRAM capacity. 32 GB is the rumored figure; the "48 GB" you'll see is an editorial guess layered on top, not a leak.
• That a leak even happened. VideoCardz argues the "specs" never actually leaked — they're projections.

We'll update this page if NVIDIA confirms anything or a genuinely new, multi-source leak appears. Until then, it's rumor.

Frequently asked questions

When will the NVIDIA RTX 6090 be released?

NVIDIA has not announced a release date. Rumors pointed to the second half of 2027, but mid-2026 reporting (The Information, via TechPowerUp and PCGamer) says NVIDIA pushed the RTX 60 series back beyond 2027 — some coverage says 2028 — because a global DRAM shortage is diverting memory to AI chips. So: rumored H2 2027, with a 2028 slip now on the table. Nothing is confirmed.

What are the rumored RTX 6080 and RTX 6070 specs?

The same single-source RGT leak maps them to the GR20x dies: RTX 6080 on a GR203 die with 20 GB of GDDR7 on a 320-bit bus; RTX 6070 on GR205 with 16 GB on 256-bit. GR206/GR207 dies are expected below these but unspecced. All rumored and re-reported (Overclock3D, HotHardware), none confirmed by NVIDIA.

Does the RTX 6090 really have 28,000 CUDA cores?

Almost certainly not. That figure comes from a die-shot analysis of "Rubin CPX," which leaker kopite7kimi identifies as GR212 -- a data-center part, not the gaming die. The gaming flagship is a different die, GR202, rumored at "up to 192 SM." No leak actually states a CUDA-core count for the gaming 6090; the widely-quoted 24,576 is a TechPowerUp database placeholder.

How much faster will the RTX 6090 be than the RTX 5090?

Rumored, not confirmed: roughly 2x path-tracing performance and about a 30–35% raster improvement, with the gains said to come from the new "Rubin" architecture rather than a big core-count increase (up to 192 SM vs the 5090's 170). Treat these as speculation until NVIDIA or independent testing confirms them.

Will the RTX 60 series bring DLSS 5?

The rumor coverage flags a next-generation "DLSS 5" as part of the Rubin feature set. NVIDIA's upscaling has been its biggest advantage, so a new DLSS generation would be significant -- but like every RTX 60 detail, it is unconfirmed.

Will AMD's RDNA 5 beat the RTX 6090, and which launches first?

Both are 2027-ish rumors, so "faster" is unknowable until real silicon exists. On timing, some reporting suggests AMD could get RDNA 5 (the Radeon RX 10900 XT) to market first, in mid-2027, ahead of NVIDIA's RTX 60. See our RDNA 5 rumor tracker for the AMD side.

Will the RTX 50 Super launch before the RTX 6090?

Almost certainly. The RTX 50 Super refresh is reportedly already built and at board partners, with launch on hold over the price of new 3 GB GDDR7 memory modules -- while the RTX 6090 is a 2027-2028 rumor. If you're deciding what to buy, the Super refresh (and current-gen cards) is the real near-term choice, not the 6090.

Is NVIDIA releasing any new gaming GPU in 2026?

Reportedly not -- 2026 coverage says NVIDIA is holding new GeForce launches (including the RTX 50 Super) over the memory shortage, which would break a long run of near-annual releases. The RTX 60 series is rumored for 2027 at the earliest.

How much will the RTX 6090 cost?

There is no leaked or official price. Third-party estimates cluster at a $2,000–$2,500+ starting point and have drifted upward with the memory crisis, with fringe claims above $3,000. For reference, the RTX 5090 launched at a $1,999 MSRP and currently sells far above it. Treat any 6090 price as a guess, not a leak.

Should I wait for the RTX 6090 or buy a GPU now?

If you need a graphics card today, don't wait on a card rumored for late 2027 with no confirmed specs or price — buy from the current 2026 lineup. Waiting only makes sense if you already run an RTX 4090 or 5090 and care solely about maximum ray tracing, and even then the timeline is a rumor, not a plan.

Is the RTX 6090 confirmed by NVIDIA?

No. As of this update NVIDIA has made no announcement, teaser, or official specification. Every figure in circulation traces to leakers and projections, not to NVIDIA.

Why is the RTX 6090 delayed?

The consensus reason is the 2026 memory crisis, not engineering. A global DRAM/GDDR7 shortage -- driven by AI-datacenter demand booking memory production out through 2027 -- is pushing NVIDIA to route scarce GDDR7 to far-more-profitable AI chips. Reporting across 2026 (The Information, via TechPowerUp and PCGamer) says NVIDIA plans no new gaming GPUs in 2026 and pushed the RTX 60 series back, with a 2028 launch now on the table. So the delay is about memory supply and margins, not a technical problem with the chip.

How much VRAM will the RTX 6090 have -- 32GB or 48GB?

The actual leak points to 32 GB of GDDR7 on a 512-bit bus -- the same capacity and bus width as the RTX 5090. The '48 GB' figure you'll see is not part of the leak; it's an editorial guess (TweakTown speculated the card would 'most likely' land somewhere around 48 GB). Treat 32 GB as the rumored number and 48 GB as speculation layered on top. NVIDIA has confirmed neither.

What power supply (PSU) will the RTX 6090 need?

There is no confirmed board power -- NVIDIA has announced nothing. Third-party estimates float 550-600 W, in the same ballpark as the RTX 5090's 575 W TDP. If those hold, you'd want a quality 1000 W+ ATX 3.1 unit with a native 12V-2x6 connector -- the same class we recommend for a 5090 build in our best PSU for gaming guide. But the wattage is a guess, not a leak.

Is the 'Rubin CPX' the RTX 6090?

No -- that's a common mix-up. Rubin CPX is an NVIDIA data-center AI GPU, and a die-shot of it showing rendering blocks is what sparked the '6090 with 28,000 cores' speculation. But leaker kopite7kimi identifies it as GR212, a data-center part, while the gaming flagship is a different die, GR202. Leakers now say the CPX is not the consumer 6090's base die.

Will the RTX 6090 use a 2nm (TSMC N2) node?

Rumor says no. The Red Gaming Tech leak points to a custom TSMC 3 nm FinFET variant, and coverage (TweakTown) specifically notes it 'puts an end to' the sub-2 nm speculation. So the rumored node is 3 nm-class, not N2 -- unconfirmed, like everything else here.

Will NVIDIA announce the RTX 6090 at CES 2027?

Unknown. CES 2027 (January 6-9) is NVIDIA's next big stage, but the more likely reveal there is the RTX 50 'Super' refresh (rumored for CES 2027 with new 3 GB GDDR7 modules), possibly with a Rubin tease. A full RTX 6090 launch is rumored for later -- 2027 at the earliest, with 2028 now floated. Treat any CES-2027 6090 reveal as speculation.

How good will the RTX 6090 be for AI and local LLMs?

If the 32 GB GDDR7 rumor holds, it would match the RTX 5090's capacity -- enough for sizeable local models, but not a leap in VRAM over today's flagship, which is what local-AI users most want. The rumored gains are concentrated in ray tracing and new Tensor cores, not memory capacity. For running LLMs locally right now, see our best GPU for AI 2026 guide -- capacity, not the next-gen rumor, is what decides it.

In the meantime: GPUs you can actually buy

A 2027 rumor doesn't help you build a PC today. If you're shopping now, our best graphics cards of 2026 covers the cards that are real, in stock, and ranked by value — from the RTX 5090 down to budget picks — with current pricing and what each one is actually good at. Here are three worth a look across the range:

Prices move — the links above go to the live Amazon listing so you see current pricing and stock. See all five picks, including creator and AMD options, in the full 2026 GPU roundup.

Want the whole next-gen lineup, not just the flagship? See our RTX 60 series hub for the rumored 6080, 6070 and below, each rated for how much is actually known.

Curious how the competition stacks up? AMD is rumored to answer with a high-end RDNA 5 flagship (the Radeon RX 10900 XT) — and some reporting suggests AMD could get RDNA 5 to market first, in mid-2027, ahead of NVIDIA. The nearer-term RTX 50 Super refresh is the more immediate NVIDIA rumor — both tracked here with the same honest framing. On the processor side, the next-gen AMD Zen 6 vs Intel Nova Lake race is shaping up for the same CES 2027 window.

Related buyer's guides

All reviews →

Buyer's guide / PC builds

Best gaming PC under $1,500 in 2026

The honest $1,500 build: a 16GB RX 9060 XT, Ryzen 5 and 32GB DDR5 for strong 1080p-ultra and entry-1440p gaming — with the real shortage-adjusted total, where every dollar goes, and whether a prebuilt beats it this year.

See the build →

Buyer's guide / PC builds

Best gaming PC under $2,000 in 2026

The value sweet spot: near-flagship 1440p gaming built around the RX 9070 XT and an X3D CPU — with the honest 2026 total, budget allocation, and exactly how to duck under a hard $2,000.

See the build →

Buyer's guide / PC builds

Best gaming PC under $3,000 in 2026

Where 4K becomes real: the fastest gaming CPU paired with an RTX 5080 for 1440p-max and 4K-high with DLSS 4 — the honest shortage-adjusted total, where every dollar goes, and a matched monitor.

See the build →

Buyer's guide / PC builds

Best gaming PC under $5,000 in 2026

The honest high-end build: a fully-maxed RTX 5080 creator-gaming machine with a 16-core 9950X3D, 64GB and 4TB — plus the truth about why $5,000 does not reliably buy a 5090 at 2026 street prices.

See the build →

Buyer's guide / AI PCs

LLM VRAM calculator

Enter a model size, quantization and context length to estimate the VRAM or unified memory an LLM needs to run locally — and instantly see which 2026 machines can hold it.

Use the calculator →

Buyer's guide / AI PCs

How to run LLMs locally: the 2026 hardware guide

How much VRAM or unified memory you need to run 7B-120B models locally, the capacity-vs-bandwidth tradeoff, and the 2026 hardware that hits each point on the curve — GB10, Ryzen AI Max, Apple Silicon and RTX GPUs.

Read the guide →