NVIDIA may be preparing to swap the GeForce RTX 5070’s smaller GB205 processor for the larger GB203 die that currently powers the RTX 5070 Ti, according to reports circulating this week from outlets including Windows Report and Technetbook. The move, still unconfirmed by NVIDIA itself, would mark one of the more unusual mid-cycle changes in the Blackwell desktop lineup and comes as GDDR7 memory pricing and AMD’s next-generation RDNA5 architecture both bear down on NVIDIA’s mainstream GPU strategy.
The RTX 5070 launched in 2025 at a $549 MSRP built around the GB205 die, 6,144 CUDA cores, 12GB of GDDR7 memory on a 192-bit bus, and a 250W power target, according to NVIDIA’s official GeForce specifications and coverage from TechSpot. Its sibling, the RTX 5070 Ti, launched at $749 with the larger GB203-300 die, 8,960 CUDA cores, 16GB of GDDR7 on a 256-bit bus, and roughly 896 GB/s of bandwidth. A GB203-based RTX 5070 would close much of that architectural gap, and it would do so at a moment when every extra gigabyte of VRAM and every additional row of silicon carries a heavier price tag than it did a year ago.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the RTX 5070 GB203 Reports Are Actually Claiming
The core claim making the rounds is straightforward: NVIDIA is weighing a version of the RTX 5070 that ditches the GB205 processor in favor of a cut-down GB203 die, the same silicon family used in the RTX 5070 Ti and RTX 5080. Reports frame this less as a brand-new SKU and more as a continuation of the “Super” refresh pattern NVIDIA has run in prior GPU generations, where mid-cycle cards get bumped VRAM, wider memory buses, and higher core counts without a full architectural overhaul.
Leak coverage from Technetbook earlier this year detailed a rumored RTX 5070 Super built on a GB203-350-A1 die with 6,400 CUDA cores, 18GB of GDDR7 on a 192-bit bus, and a power target near 275W — up from the current RTX 5070’s 12GB and 250W, a configuration also referenced in PCGamesN’s RTX 50 Super guide. A companion RTX 5070 Ti Super rumor, also detailed in spec breakdowns from PC Guide, pointed to the same GB203-350-A1 die with 8,960 CUDA cores, 24GB of GDDR7 on a 256-bit bus, and around 350W of power draw. Pricing chatter has placed the Ti Super near $749, effectively sliding into the current RTX 5070 Ti’s price slot while pushing the standard Ti upward.
None of this has been confirmed by NVIDIA, and the company has not issued any statement addressing a GB203-based RTX 5070 or a broader Super refresh timeline. Readers should treat every spec figure below the official RTX 5070 and RTX 5070 Ti baseline as leak-stage reporting, not a locked product roadmap.
RTX 5070 vs RTX 5070 Ti: The Specs NVIDIA Would Be Blurring
To understand why a GB203-based RTX 5070 would be significant, it helps to see how distinct the two current cards are on paper. The RTX 5070 and RTX 5070 Ti already sit $200 apart at MSRP, and most of that gap traces directly back to the die size and memory configuration difference between GB205 and GB203. Folding the RTX 5070 onto GB203 silicon — even a heavily cut-down version — would shrink that separation considerably, unless NVIDIA introduces a new price tier to keep the lineup spread out.
| Spec | RTX 5070 (current, GB205) | RTX 5070 Ti (current, GB203) | RTX 5070 Super (rumored, GB203) |
|---|---|---|---|
| GPU die | GB205-300 | GB203-300 | GB203-350 (rumored) |
| CUDA cores | 6,144 | 8,960 | 6,400 (rumored) |
| Memory | 12GB GDDR7 | 16GB GDDR7 | 18GB GDDR7 (rumored) |
| Memory bus | 192-bit | 256-bit | 192-bit (rumored) |
| Memory bandwidth | ~672 GB/s | ~896 GB/s | ~672 GB/s (rumored) |
| Power (TGP) | 250W | 300W | ~275W (rumored) |
| Launch MSRP | $549 | $749 | Unconfirmed |
The most telling column is the middle ground: a rumored RTX 5070 Super with more CUDA cores than the standard RTX 5070 but fewer than the RTX 5070 Ti, more VRAM than either non-Super card but on the same narrower 192-bit bus as today’s base RTX 5070. That configuration reads like NVIDIA trying to solve the 12GB VRAM complaint that has followed the RTX 5070 since launch without simply handing buyers a discounted RTX 5070 Ti.
Why NVIDIA Would Move the RTX 5070 to GB203 Now
Three forces converge to make this rumor plausible even without official confirmation. First, the RTX 5070’s 12GB VRAM allotment has been a recurring criticism since the card launched, echoing the same complaint that dogged the RTX 4070 a generation earlier. Second, GDDR7 and DRAM pricing broadly has climbed sharply through 2026, changing the calculus around which die supports which memory configuration. Third, AMD’s next-generation RDNA5 architecture is widely expected to target the same $500–$700 segment where the RTX 5070 competes, and NVIDIA has historically responded to AMD pressure in that band with Super refreshes rather than all-new architectures.
Using the GB203 die across more of the lineup also gives NVIDIA a manufacturing advantage: binning the same large die into multiple SKUs — RTX 5070, RTX 5070 Ti, and RTX 5070 Ti Super, in the rumored scenario — lets the company absorb yield variation more efficiently than maintaining a separate GB205 production line purely for the entry card. That kind of consolidation becomes more attractive precisely when memory and wafer costs are rising, since it reduces the number of distinct die designs NVIDIA has to keep in volume production.
The GDDR7 and DRAM Pricing Backdrop
This rumor doesn’t exist in a vacuum. Memory pricing has been one of the defining hardware stories of 2026, with mainstream DDR5 kits and GDDR7 modules both climbing well beyond their 2025 levels as AI data center demand competes with consumer GPU and PC production for the same fabs. Tech Insider previously reported that 32GB DDR5 kits climbed roughly 485% in 2026, a spike driven largely by memory makers prioritizing AI accelerator contracts over consumer SKUs. GDDR7, the memory standard used across the entire RTX 50 series, has faced comparable upward pressure even though it serves a narrower market than general-purpose DRAM.
Against that backdrop, a VRAM bump on the RTX 5070 — from 12GB to a rumored 18GB — is not a trivial cost decision. Every additional gigabyte of GDDR7 NVIDIA commits to a mainstream card is now measurably more expensive to source than it was when the RTX 5070 originally launched. That’s part of why the leaked RTX 5070 Super configuration keeps the memory bus at 192-bit rather than jumping to 256-bit: NVIDIA appears to be trying to add capacity without also adding the cost of a wider bus and the extra memory chips that requires.
AMD’s RDNA5 Threat and Market Segmentation
NVIDIA’s mainstream GPU tier has not gone unchallenged. AMD’s current RDNA4-based RX 9070 series already trades blows with the RTX 5070 and RTX 5070 Ti on price-to-performance, and early leaks around AMD’s next-generation RDNA5 architecture point to significantly higher clock targets, with some reports citing GPU clocks approaching 3.5 GHz, a substantial jump from current-generation cards. If AMD arrives with RDNA5 parts that undercut NVIDIA on VRAM-per-dollar in the same way RDNA4 has, NVIDIA has a clear incentive to pre-empt that pressure with a beefed-up RTX 5070 rather than waiting for a full next-generation architecture.
Market segmentation is the other half of the equation. NVIDIA has used the Super naming convention in previous generations specifically to insert a mid-cycle refresh between existing tiers without renumbering the entire stack. A GB203-based RTX 5070 Super slotting between the standard RTX 5070 and RTX 5070 Ti would let NVIDIA capture buyers who feel priced out of the Ti tier but underserved by the base card’s VRAM, all without cannibalizing the higher-margin RTX 5080.
NVIDIA’s Super Refresh Playbook: Historical Context
Mid-cycle Super refreshes are not new territory for NVIDIA. The company has used the naming scheme across multiple GPU generations to extend a lineup’s shelf life, respond to competitive pressure, or correct a VRAM configuration that drew criticism at launch. Each prior refresh followed a similar pattern: keep the core architecture unchanged, adjust die selection or memory configuration, and reposition pricing across the existing tiers rather than launching an entirely new segment.
| Refresh cycle | What changed | Primary driver reported |
|---|---|---|
| RTX 20 → RTX 20 Super | More cores and memory at similar price points | Competitive response to AMD Navi pricing |
| RTX 30 → RTX 30 Ti variants | Higher core counts, faster memory on select tiers | Filling gaps in the stack ahead of next-gen launch |
| RTX 40 → RTX 40 Super | Adjusted core counts and, in some tiers, more VRAM | Pricing repositioning against AMD RDNA3 |
| RTX 50 → RTX 50 Super (rumored) | GB203 die reportedly extended down into the 5070 tier | VRAM criticism, GDDR7 cost pressure, anticipated RDNA5 competition |
Seen against that history, a GB203-based RTX 5070 Super would be consistent with how NVIDIA typically manages its mainstream lineup roughly a year into a generation. What would be new is the degree to which the refresh appears aimed squarely at the VRAM criticism that has followed the RTX 5070 since its debut, rather than purely a response to AMD’s roadmap.
Regional Pricing Signals Already Point Upward
Even without a confirmed refresh, RTX 5070 pricing has already been drifting away from its $549 launch MSRP through 2026. Reports on regional pricing this year have described RTX 50-series cards, including the RTX 5070, climbing in markets such as South Korea as component costs rise. In Europe, NVIDIA and retail partners took the opposite approach earlier in the RTX 50 series’ life, trimming prices on select cards by around 10% to keep the lineup competitive — a reminder that pricing on this generation has already moved in both directions depending on region and timing, well before any GB203-based refresh reaches the market.
That volatility matters for anyone deciding whether to buy an RTX 5070 today or wait. If NVIDIA does bring a GB203-based Super variant to market, the standard RTX 5070 could either get discounted to clear inventory or simply get discontinued in favor of the new configuration — both scenarios NVIDIA has followed at different points in past Super transitions.
Competitive Comparison: Where a GB203 RTX 5070 Would Land
A GB203-based RTX 5070 with a rumored 18GB of GDDR7 would compete more directly against AMD’s higher-VRAM Radeon cards in the same price bracket, narrowing one of the clearer advantages AMD has held in recent generations. Tech Insider’s testing has shown AMD’s RX 9070 trading blows with the standard RTX 5070 at the same $549 price point, with AMD often ahead by single-digit to mid-teens percentages depending on the title. A higher-VRAM RTX 5070 Super would blunt one of AMD’s most repeated talking points against NVIDIA’s mainstream tier, even if raw rasterization performance gaps persist.
It would also reshape the internal NVIDIA comparison. The current RTX 5070 Ti’s clearest selling point over the standard RTX 5070 is its larger 16GB frame buffer and wider memory bus. A GB203-based RTX 5070 Super with 18GB, even on a narrower 192-bit bus, would match or exceed the Ti’s VRAM capacity while undercutting it on price — forcing NVIDIA to either drop the standard 5070 Ti’s price or push a Ti Super variant to keep the tiers distinct, as the leaked 24GB Ti Super configuration suggests it may already be planning to do.
What Board Partners and Retailers Are Watching
Add-in board partners like ASUS, MSI, and Gigabyte have continued shipping and refreshing RTX 5070 and RTX 5070 Ti cards throughout 2026, with new custom-cooled variants appearing as recently as this past summer according to product coverage from Technetbook and WCCFTech’s GPU news desk. None of that partner-level activity has referenced a GB203-based 5070 die swap directly, which is one reason the current reports remain in rumor territory rather than confirmed roadmap territory. Board partners typically only receive final specifications close to a product’s public unveiling, so the absence of partner leaks doesn’t rule out a refresh — it simply means NVIDIA hasn’t looped its manufacturing partners in yet, or those details haven’t surfaced publicly.
Retailers, meanwhile, are dealing with a more immediate problem: GPU inventory planning in a market where memory costs are a moving target. A sudden RTX 5070 refresh would force partners to requalify cooler designs and PCB layouts for the new die, a process that historically takes several weeks to a few months once specifications are finalized — one reason any GB203-based RTX 5070 is more likely to land in late 2026 or early 2027 than in the immediate weeks following these initial reports.
How This Fits NVIDIA’s Broader 2026 GPU Strategy
NVIDIA’s desktop GPU roadmap in 2026 has been shaped as much by AI infrastructure demand as by gaming competition. The company’s data center GPU business has continued to dominate its revenue mix, and NVIDIA’s own RTX Spark AI-focused hardware line launched with configurations carrying up to 128GB of unified memory, underscoring how much of the company’s engineering attention has shifted toward high-memory, AI-capable products. A GB203-based RTX 5070 fits into that same broader theme: more memory per card, positioned to serve both gamers and the growing population of enthusiasts running local AI models on consumer GPUs.
That dual-purpose positioning — gaming card and entry-level local AI workstation — has become increasingly common across NVIDIA’s consumer lineup as more software vendors optimize for VRAM-hungry local inference. A jump from 12GB to 18GB on the RTX 5070 tier wouldn’t just help with games; it would meaningfully expand which AI models the card can run without offloading to system memory.
Predictions: What Happens Next
- A formal Super refresh announcement is more likely than a silent GB205-to-GB203 swap. NVIDIA’s history suggests it will brand any GB203-based 5070 as a distinct “Super” product rather than quietly replacing the existing card’s internals.
- Pricing will likely land above the current $549 RTX 5070 MSRP. Given 2026’s memory cost environment, a Super variant with 18GB of GDDR7 would be difficult for NVIDIA to price at parity with the outgoing 12GB card.
- The standard RTX 5070 probably survives as a budget option rather than disappearing immediately. NVIDIA has kept non-Super cards in market alongside Super refreshes in prior generations, at least until inventory clears.
- AMD’s RDNA5 launch timing will influence NVIDIA’s final specs and pricing. If AMD’s next-generation cards slip, NVIDIA has less urgency to rush a GB203-based 5070 to market.
- Official confirmation is unlikely before early 2027 given typical AIB partner lead times. The gap between initial leak-stage reporting and a confirmed NVIDIA announcement has historically run several months in past Super transitions.
What Buyers Should Do Right Now
For anyone shopping in the $500-$750 GPU range today, the practical advice hasn’t changed much despite the rumor. The current RTX 5070 and RTX 5070 Ti are shipping, tested products with known performance characteristics, while any GB203-based Super variant remains unconfirmed and, per these reports, likely months away at best. Buyers whose workloads are VRAM-sensitive — large local AI models, high-resolution texture mods, or multi-monitor content creation — have the clearest reason to wait, since a confirmed 18GB configuration would directly address the limitation driving that hesitation. Buyers focused purely on current-generation gaming performance at 1440p have less reason to delay, since the rumored changes center on memory capacity rather than a fundamental rasterization or ray-tracing performance leap.
It’s also worth watching how NVIDIA’s competitors respond in the meantime. If AMD’s RDNA5 launch details firm up before NVIDIA confirms anything about a GB203-based RTX 5070, that competitive pressure could accelerate NVIDIA’s timeline — or push the company to adjust the rumored specs further before any official unveiling.
Frequently Asked Questions
Is NVIDIA officially confirming a GB203-based RTX 5070?
No. As of this report, NVIDIA has not issued any official statement about moving the RTX 5070 to the GB203 die or launching an RTX 5070 Super. The information circulating comes from industry reports and leak coverage, including from Windows Report and Technetbook, not from NVIDIA directly.
What GPU die does the current RTX 5070 use?
The current, officially shipping RTX 5070 uses the GB205 die with 6,144 CUDA cores, 12GB of GDDR7 memory on a 192-bit bus, and a 250W power target, per NVIDIA’s official specifications.
How is the RTX 5070 Ti different from the RTX 5070?
The RTX 5070 Ti uses the larger GB203-300 die with 8,960 CUDA cores, 16GB of GDDR7 on a 256-bit bus, and a 300W power target, compared to the RTX 5070’s GB205 die, 6,144 CUDA cores, and 12GB on a 192-bit bus. The RTX 5070 Ti launched at $749 versus $549 for the standard RTX 5070.
Why would NVIDIA put a bigger GPU die in a cheaper card?
Using a shared die like GB203 across multiple tiers lets NVIDIA bin chips with minor manufacturing defects into lower-tier products instead of discarding them, which can be more efficient than maintaining a completely separate die design solely for entry-level cards. It also lets NVIDIA respond to VRAM criticism and competitive pressure from AMD without designing an entirely new chip.
When could a GB203-based RTX 5070 actually launch?
No official date exists. Based on typical add-in board partner lead times and the pattern of past NVIDIA Super refreshes, a confirmed launch is more plausible in late 2026 or early 2027 than in the immediate weeks following these initial reports.
Will the current RTX 5070 get cheaper if a Super version launches?
That’s possible but unconfirmed. In past Super transitions, NVIDIA has sometimes discounted outgoing non-Super cards to clear inventory, though pricing decisions ultimately depend on retailer stock levels and memory cost trends at the time of any refresh.
Is this related to the RTX 5070 Ti Super rumors?
Yes. Leak coverage describes both a rumored RTX 5070 Super and RTX 5070 Ti Super, both reportedly built on GB203-350-A1 dies with different core counts and VRAM configurations — 18GB for the 5070 Super and 24GB for the Ti Super, according to reports.
How does GDDR7 pricing affect this rumor?
GDDR7 memory costs have risen alongside broader DRAM price increases throughout 2026, driven largely by AI data center demand competing for the same manufacturing capacity. That cost pressure is one of the market factors analysts point to when explaining why any VRAM increase on the RTX 5070 tier carries real cost implications for NVIDIA rather than being a simple software or binning change.
Related Coverage
- Nvidia Brings DLSS 5 to RTX 40, Skips Full Control [2026]
- RTX Spark N1X: 6144 vs 5120 Cores, 128GB vs 32GB [2026]
- Megaspeed's Malaysia Unit Bought $2B in Nvidia Chips [2026]
- Inspur Used Aivres to Buy $5.6B in Nvidia Chips [2026]
- Cramer: Nvidia's $13B Deal Locks Out AMD, Broadcom [2026]
