NVIDIA may be preparing a second version of the GeForce RTX 5070, this time built on the larger GB203 die instead of the GB205 chip that ships in the card today. The claim comes from hardware leaker hongxing2020 and has since been picked up by Guru3D, TechPowerUp, WCCFtech, Lowyat.NET, VideoCardz, Windows Report and Technetbook.
Nothing here comes from NVIDIA itself. No driver database entry, retail listing or board-partner slide has surfaced to back it up. But the timing lines up with a new GeForce driver that hongxing2020 says lands September 9, 2026, and that alone has pushed the rumor to the top of PC hardware forums this week. Here’s what the reporting actually supports, what it doesn’t, and why a chip swap that changes zero listed specs still matters to anyone shopping for a midrange GPU in September 2026.
What the New RTX 5070 Rumor Actually Claims
The core claim is narrow. Hongxing2020 posted that “the 5070 uses the GB203 core” and added that a new driver was arriving the following day. That’s the entire tip. Every other detail, including device ID, board design, cooler changes, and retail timing, remains unknown, and the outlets covering the leak have said so directly.
Today’s RTX 5070 runs on the GB205 GPU, a 263 mm² die with 31.1 billion transistors. The rumored variant would instead use GB203, a 378 mm² die with 45.6 billion transistors, currently reserved for the RTX 5070 Ti and RTX 5080. Swapping in a bigger, more expensive chip for a cheaper card sounds backwards until you consider why manufacturers do this in the first place: not every GB203 die that comes off the line is good enough for a 5070 Ti or 5080, and NVIDIA has an obvious incentive to sell the leftovers rather than scrap them.
Who’s Reporting the GB203 Swap, and How Solid Is It
Leak credibility in the GPU space usually comes down to two things: the source’s track record and how many independent outlets pick up the claim without contradicting each other. On both counts, this rumor sits in familiar territory. Hongxing2020 has posted GPU-configuration leaks before, and the claim spread within a day to at least seven outlets across four countries, from Malaysia’s Lowyat.NET to Guru3D’s Netherlands-based desk.
None of that makes it confirmed. Guru3D was explicit about the gap, noting there are no retail listings, driver database entries, product photos, or board-partner announcements to independently verify the change. That’s the standard NVIDIA rumor lifecycle: a single leaker starts the thread, tech press amplifies it within 24 to 48 hours, and actual confirmation (or a quiet debunking) shows up weeks later once cards start appearing in driver INF files or store listings.
GB203 vs GB205: Comparing the Silicon
The two chips at the center of this story sit at very different points on NVIDIA’s Blackwell lineup. GB205 is the mid-tier die that currently powers the vanilla RTX 5070. GB203 is one tier up, shared today by the RTX 5070 Ti and RTX 5080. Putting a GB203 die into the cheapest card in the family would be unusual, but it wouldn’t be unprecedented for NVIDIA, which has reshuffled die assignments before when yields or demand shifted mid-cycle.
Die Size and Transistor Count Compared
GB203 measures 378 mm² and packs 45.6 billion transistors. GB205 measures 263 mm² and packs 31.1 billion. That’s a 44% larger die area and roughly 47% more transistors for GB203, a gap wide enough that the two chips aren’t remotely interchangeable at the manufacturing level. A GB203-based RTX 5070 would, by definition, use meaningfully more silicon per card than the GB205 version does today, even if a large chunk of that silicon sits disabled.
Why the Reported Spec Sheet Doesn’t Change
Here’s the part that trips people up: outlets covering the leak report that the configuration NVIDIA is testing keeps the same 6,144 CUDA cores, 12GB of GDDR7 memory, 192-bit memory bus, and 250W board power that the current GB205-based RTX 5070 already ships with. In other words, a card built on a much bigger chip would perform identically to the one it replaces, at least on paper. That’s the detail that turns this from a “secret performance upgrade” story into a supply-chain story.
The Binning Theory: Why Recycle a Bigger Die
Chip binning is standard practice across the semiconductor industry. Every wafer produces dies with slightly different defect rates and clock ceilings, and manufacturers sort (“bin”) them into product tiers based on which ones pass which tests. A GB203 die that fails to meet RTX 5070 Ti or RTX 5080 thresholds, whether due to a defective memory controller segment, a damaged cluster of cores, or power-delivery inconsistencies, doesn’t have to go in the scrap bin. If enough of the die still works, NVIDIA can disable the broken sections and sell what’s left at a lower tier.
That’s the theory Guru3D and other outlets have floated for why a GB203-based RTX 5070 would make business sense even without a spec bump. It stretches yield further, cuts waste, and gives NVIDIA a second source of RTX 5070 supply if GB205 output tightens for any reason, whether from fab capacity constraints or from GB205 being prioritized for other SKUs.
Precedent: NVIDIA Already Pulled This Move on the RTX 5060
A GB203-based RTX 5070 wouldn’t be NVIDIA’s first mid-cycle die swap in this generation. The RTX 5060 launched on the GB206 chip, but board partners later released RTX 5060 cards built on a cut-down GB205 die instead. MSI shipped several of these as “V1” models, and Yeston released its own RTX 5060 variant using GB205-200 silicon. Buyers who didn’t check the fine print on the box got a card with a different underlying chip than the one reviewed at launch, even though the marketed specs stayed the same.
That precedent is the strongest evidence that this rumor is plausible, independent of whether hongxing2020’s specific tip pans out. NVIDIA has a documented pattern this generation of swapping the die underneath a given model name once yields and market conditions call for it.
The September 9 Driver Signal
The detail giving this rumor extra weight this week is timing. Hongxing2020’s post ties the GB203 claim to a driver release dated September 9, 2026, the same day this story is breaking. New GeForce drivers routinely contain device ID entries for unreleased hardware months before a card ships, which is exactly how past NVIDIA SKUs have leaked. If a GB203-based RTX 5070 device ID shows up in that driver’s INF file, it would be the first hard evidence beyond a single leaker’s claim.
In the days since, that question looks answered. Reporting on the September 9 driver, released as version 616.92, says its INF files added a second RTX 5070 entry carrying the PCI device ID 10DE:2C09, which sits in the GB203 range next to the RTX 5070 Ti’s 2C05 rather than alongside the current 5070’s original 2F04. That is the first driver-level evidence beyond hongxing2020’s single tip, and it matches how past NVIDIA SKUs have leaked through INF files before launch. It is still not official confirmation: NVIDIA has announced nothing, and no retail listing, board-partner slide, or shipping card carrying that ID has surfaced, so for now the GB203 RTX 5070 exists as a driver entry rather than a product on a shelf.
Current RTX 50-Series Street Prices Set the Stakes
Part of why this rumor is landing hard with buyers is that RTX 50-series pricing has drifted well above launch MSRP across most of the stack. The RTX 5070 launched at $549, a price NVIDIA’s own product pages and WCCFtech’s coverage of the launch both confirm, and one that briefly returned this year when partner cards dipped back to or below that mark. Price-tracking site GPUPriceHistory put the RTX 5070 at roughly $1,045 in the US as of September 6, 2026, nearly double its launch price. The RTX 5070 Ti and RTX 5080 show similar patterns: TechSpot’s Q3 2026 GPU pricing roundup listed the 5070 Ti moving from a $749 MSRP to around $1,050 street, and the 5080 moving from $999 to roughly $1,550, a jump of about 55%.
Any change to how a $549-MSRP card gets built, even one that leaves the spec sheet untouched, matters more in a market where actual shelf prices are running 90% to 100% above that number.
RTX 5070 Family Specs: Confirmed vs Rumored
| Card | GPU Die | Die Size | Transistors | CUDA Cores | Memory | Status |
|---|---|---|---|---|---|---|
| RTX 5070 (current) | GB205 | 263 mm² | 31.1 billion | 6,144 | 12GB GDDR7, 192-bit | Confirmed, shipping |
| RTX 5070 Ti | GB203 | 378 mm² | 45.6 billion | Not disclosed in this report | Not disclosed in this report | Confirmed, shipping |
| RTX 5080 | GB203 | 378 mm² | 45.6 billion | Not disclosed in this report | Not disclosed in this report | Confirmed, shipping |
| RTX 5070 “GB203 variant” | GB203 (reported) | 378 mm² | 45.6 billion | 6,144 (reported unchanged) | 12GB GDDR7, 192-bit (reported unchanged) | Unconfirmed rumor |
Market Impact: What It Means for Buyers and Board Partners
For everyday buyers, a die swap with an unchanged spec sheet is close to invisible at checkout. A GB203-based RTX 5070 would still list 6,144 CUDA cores, 12GB of memory, and a 250W power draw, so a benchmark chart wouldn’t flag which chip sits under the cooler. The only way most buyers would ever know is by checking the GPU-Z tool or a similar utility that reads the die identifier directly off the card, exactly how enthusiasts caught the RTX 5060’s GB205 swap earlier this generation.
Board partners face a bigger question: cooler and PCB design. GB203 and GB205 don’t necessarily share a footprint, power-delivery layout, or thermal profile, so a straight die swap inside an existing cooler design isn’t guaranteed to be simple. If NVIDIA moves forward, expect partner SKUs to carry a small revision marker (a “V2” or similar suffix, as MSI used on its GB205-based RTX 5060) rather than a full model refresh, and expect retailers to be slow or inconsistent about disclosing which chip ships in which box.
Competitive Landscape: How AMD and Intel Fit In
NVIDIA isn’t reshuffling silicon in a vacuum. AMD and Intel both compete directly against the RTX 5070 tier with their own midrange cards, and NVIDIA’s ability to keep RTX 5070 supply flowing, even via a costlier die, protects its share of the segment that sells in the highest volume. A card built on excess GB203 supply also gives NVIDIA a cushion if GB205 output gets squeezed by the same memory and fab capacity constraints that have already pushed RTX 50-series street prices well above MSRP across the board this year.
The bigger competitive risk sits with cost, not performance. Manufacturing a $549 card on a 378 mm² die instead of a 263 mm² one raises NVIDIA’s per-unit cost even before yield savings are factored in. If NVIDIA absorbs that gap to protect margin on the flagship-tier chips, RTX 5070 street pricing has little reason to fall back toward $549 anytime soon, keeping the segment favorable for whichever AMD or Intel card can undercut it on price without sacrificing rasterization performance.
Historical Context: NVIDIA’s Silicon Reshuffling Playbook
NVIDIA has swapped dies underneath an existing model name more than once across recent generations, almost always driven by yield economics rather than a desire to quietly upgrade a product. The RTX 5060’s shift from GB206 to a cut-down GB205 earlier this year is the closest analog, and it followed the same pattern being reported now: a leak first, tech press coverage within days, and confirmation only once retail units and driver entries caught up with the claim.
What makes the RTX 5070 case notable is direction. The RTX 5060 swap moved to a smaller, cheaper die (GB205 replacing GB206’s role in some SKUs), which is the more common and more easily justified version of this story. A rumored RTX 5070 swap moving the other way, onto a bigger and pricier die, is a less common pattern and one industry watchers will scrutinize harder before treating it as confirmed.
Sept. 2026 GPU Street Pricing Snapshot
| Card | Launch MSRP | US Street Price, Sept. 2026 | Approx. Change vs MSRP | Source |
|---|---|---|---|---|
| RTX 5070 | $549 | ~$1,045 (Sept. 6, 2026) | ~90% above MSRP | GPUPriceHistory, WCCFtech |
| RTX 5070 Ti | $749 | ~$1,050 | ~40% above MSRP | TechSpot Q3 2026 pricing report |
| RTX 5080 | $999 | ~$1,550 | ~55% above MSRP | TechSpot Q3 2026 pricing report |
What Would Actually Confirm This Rumor
Three things would move this from rumor to fact. First, a device ID for a GB203-based RTX 5070 turning up inside the September 9 driver package, the same way past NVIDIA SKUs have leaked through INF files before launch. Second, a retail listing or FCC filing from a board partner referencing a new RTX 5070 revision. Third, a teardown or GPU-Z reading from an actual retail card showing GB203 silicon under the cooler, which is exactly how the RTX 5060’s GB205 swap got caught earlier this year.
Until one of those three shows up, every outlet covering this, including this one, is repeating a single leaker’s claim. That doesn’t make it wrong. Hongxing2020’s driver-timing tip lines up with a real driver release date, which is more corroboration than most single-source GPU leaks get on day one.
Predictions: Where the GB203 RTX 5070 Story Goes Next
- A device ID surfaces within a week. If the September 9 driver contains new hardware entries, expect GPU-focused outlets like VideoCardz and TechPowerUp to extract and publish them within days, not weeks.
- NVIDIA stays silent through Q4 2026. Companies rarely comment on unannounced SKU changes, and the RTX 5060 precedent suggests NVIDIA will let board partners quietly ship the new variant rather than issue a formal statement.
- Board partners label it discreetly, if at all. Expect a “V2” suffix or similarly subtle model-number change on partner product pages, mirroring how MSI handled its GB205-based RTX 5060.
- Street prices don’t fall because of this change alone. A costlier die per unit gives NVIDIA less room to cut RTX 5070 pricing back toward $549, especially with memory and fab capacity already tight across the RTX 50 lineup.
- Enthusiast tools become the verification method of record. With the reported spec sheet unchanged, GPU-Z and similar die-identification utilities will be how buyers and reviewers confirm which chip actually shipped in any given card, just as they did with the RTX 5060 swap.
For readers deciding whether to buy an RTX 5070 right now, the honest answer is that nothing about today’s rumor changes performance, price, or availability yet. It’s a manufacturing story before it’s a buying story, and it will likely stay that way until a device ID, an FCC filing, or a teardown turns speculation into a documented product change.
Frequently Asked Questions
Is the GB203-based RTX 5070 officially confirmed by NVIDIA?
No. The claim originates from leaker hongxing2020 and has been picked up by outlets including Guru3D, TechPowerUp, WCCFtech and Lowyat.NET, but NVIDIA has not issued a statement, and no retail listing or driver entry has independently verified it as of this writing.
Would a GB203-based RTX 5070 perform better than the current model?
Reports indicate the configuration keeps the same 6,144 CUDA cores, 12GB of GDDR7 memory, 192-bit bus, and 250W power draw as the existing GB205-based card, so performance should be effectively unchanged even though the underlying die is larger.
Why would NVIDIA use a bigger, pricier chip for a cheaper card?
The leading theory is binning: GB203 dies that don’t meet the quality bar for an RTX 5070 Ti or RTX 5080 can still work at RTX 5070 specifications once the failing sections are disabled, letting NVIDIA recover otherwise-wasted silicon.
How can I check which GPU die is actually in my RTX 5070?
Utilities such as GPU-Z read the die identifier directly from the card’s hardware and will report GB205 or GB203 regardless of what the box or listing says. This is the same method enthusiasts used to catch the earlier RTX 5060 die swap.
Has NVIDIA done this kind of die swap before?
Yes. The RTX 5060 launched on GB206, and later board-partner models, including several MSI cards and a Yeston model, shipped with a cut-down GB205 die instead, while keeping the same marketed specs.
When will we know if this rumor is true?
The clearest near-term signal is the GeForce driver reportedly releasing September 9, 2026. If a new device ID for a GB203-based RTX 5070 appears in that driver’s files, outlets that regularly extract driver data are likely to report it within days.
Does this affect current RTX 5070 owners?
No. Cards already sold and in use aren’t affected by a future manufacturing change. This would only apply to new units built and shipped after any such switch takes effect.
Is now a good time to buy an RTX 5070?
That depends on price rather than this rumor. Street pricing for the RTX 5070 was running around $1,045 in early September 2026, according to GPUPriceHistory, nearly double the $549 launch MSRP, and a die swap with unchanged specs isn’t likely to move that price on its own.
