Concept illustration (AI-generated) of a future Radeon graphics card. AMD has not shown or announced any RDNA 5 product — this is an artist's impression, not a photo of a real product.

AMD RDNA 5 Graphics Cards Leaked: Everything We Know About the Radeon RX 10000 Series (2027–2028)

AMD RDNA 5 — the architecture expected to power the next generation of Radeon graphics cards, reportedly named the Radeon RX 10000 series — has become the most talked-about rumour in PC hardware right now. And there is a reason the speculation is louder than usual: both AMD and Nvidia are going quiet at the same time. Nvidia has confirmed nothing new for 2026, the RTX 5090 is selling for five times its list price in some markets, and AMD's flagship RX 9070 XT remains the top gaming card the company offers. Into that vacuum have poured leaked roadmaps, codenames, specs, and prices. This guide separates what is known from what is guessed, explains the four leaked Radeon RX 10000 SKUs in detail, and — most importantly — tells you what a GPU buyer should actually do while waiting for cards that may not arrive until late 2027 or 2028.

One essential note before we start: AMD has confirmed absolutely nothing about RDNA 5 or the RX 10000 series. Everything in this article beyond today's officially available products is a rumour, attributed to its source. Wherever this article states a fact as certain, it is about hardware and events that are already on the record.

Why the RDNA 5 Rumours Matter More Than Usual

Normally, a graphics architecture rumour is just enthusiasts trading numbers on forums. This time it matters to real buyers, because the entire GPU market is stuck. Nvidia's RTX 50 series launched in early 2025 and — according to leaker Kopite7kimi's reporting, widely covered in September 2026 — the next architecture (often discussed as RTX 60 / Rubin) may not arrive until 2028. That would make 2026 the first calendar year without a new GeForce generation since the GeForce 256 launched in 1999. At the same time, the memory shortage has sent high-end card prices into orbit: the RTX 5090, with a $1,999 list price, is being listed at $8,500 to $10,000 on some street and resale markets, even as Nvidia SVP Jeff Fisher publicly denied in Taipei on 16 October 2026 that the card has been discontinued. For a full breakdown of that situation, see our report on why the RTX 5090 has become a $10,000 ghost, and our broader look at why graphics card prices are so high in October 2026.

Against that backdrop, AMD's next move is genuinely consequential. The RX 9070 XT, launched in March 2025, is still AMD's flagship gaming card — and it remains a strong value at the high end of the mainstream segment. If RDNA 5 delivers the ray-tracing leap the rumours describe, the competitive picture could look very different by 2028. The leaks also matter because they suggest AMD is skipping the 9000 series successor numbering entirely, jumping from RX 9000 straight to RX 10000 — a naming reset last seen in industry transitions where a company wants a clean break with the past.

What AMD Has Actually Confirmed: Nothing

This section is short by design. AMD has made no official announcement about RDNA 5, the Radeon RX 10000 series, any codename, any date, any price, or any specification. There is no keynote slide, no press release, and no product page. The company's current public gaming lineup is the RDNA 4-based Radeon RX 9000 series, led by the RX 9070 XT, and its official product information lives on AMD's graphics product page.

This matters because the hardware press runs on sourcing, and every "specification" in this article traces back to a small number of leaks. Treating them as confirmed would be exactly how misinformation spreads. From here on, every leaked claim is labelled as such and tied to the outlet or leaker that reported it.

Where the Leaks Came From

The current wave of RDNA 5 information has three main sources, and it is worth knowing which is which because they have different track records.

1. The GameGPU roadmap article (25 October 2026). The site GameGPU published a roadmap built on an archived Broadcom/VMware presentation that was edited and updated with currently circulating leaks. This is the source of the four-SKU RX 10000 lineup, the codenames, the core counts, the memory configurations, and the price ranges discussed below. It is the most detailed single document available — and also the one that should be treated with the most caution, because it blends an old corporate presentation with unverified leak data. Support for the generation is said to run through 2033, which is a normal driver-support horizon, not a surprise.

2. The NeoGAF "Alpha Triton" leak, covered by Notebookcheck. A leak posted on the NeoGAF forums named the RDNA 5 GPU family Alpha Triton, codenamed AT — so the four dies are AT0, AT1, AT2, AT3 and AT4. Notebookcheck's analysis of the leak added performance context: which Nvidia cards each rumoured SKU would compete with, and the striking claim that RDNA 5's ray-tracing gains could substantially outpace its raster (traditional rendering) gains.

3. Kepler_L2 via hardware press (October 2026). The hardware leaker Kepler_L2, posting in the AnandTech forums, claimed AMD is unlikely to announce RDNA 5 before Computex 2027 (the show runs 1–4 June 2027) — and that a reveal later in 2027 is more likely. The claim was reported by Guru3D, TweakTown, and Notebookcheck around 12 October 2026. The same leak suggests only the AT2 die arrives in 2027, with the rest of the family slipping into 2028.

The Leaked Lineup at a Glance

According to the GameGPU roadmap, AMD's RDNA 5 generation would launch as the Radeon RX 10000 series, built on TSMC's N3P process node — the third-generation 3-nanometre-class node, a step ahead of the nodes used by today's gaming GPUs. The roadmap lists four desktop SKUs, spread across two memory technologies and a wide price band. Here is the full leaked specification table. Every number in this table is a rumour, not a specification.

Leaked SKU	Chip (Alpha Triton)	Leaked launch window	Compute units	Leaked clock	Leaked memory	Memory bus	Leaked bandwidth	Leaked price
RX 10050 XT	AT4	Q4 2027	24 CU	3.5 GHz	12–16 GB LPDDR5X	128-bit	160 GB/s	$300–$350 (up to $600–$700 if memory shortage continues)
RX 10060 XT	AT3	Q4 2027	48 CU	3.4 GHz	16–24 GB LPDDR5X	256-bit	320 GB/s	$450–$550 (up to $900–$1,100 if shortage)
RX 10070 XT	AT2	Q3/Q4 2027	68 CU	3.2 GHz	16–24 GB GDDR7	192-bit	864 GB/s	$700–$800 (up to $1,400–$1,600 if shortage)
RX 10090 XT	AT0	Q3/Q4 2027	96 CU	3.0 GHz	24–36 GB GDDR7	384-bit	~1.7 TB/s	$1,000–$1,200 (up to $2,000–$2,400 if shortage)

A few things jump out immediately. The clocks are the headline: 3.0 to 3.5 GHz would be a major step up from anything in today's GPU market, where boost clocks in the high 2 GHz range are typical. The compute-unit ladder (24 / 48 / 68 / 96) covers everything from mainstream to flagship. And the memory split — LPDDR5X on the two cheaper cards, GDDR7 on the two expensive ones — is the most argued-over detail in the enthusiast press, for reasons we will dig into below. The price ranges are also presented in two versions: a "normal market" price and a dramatically higher "if the memory shortage continues" price, which tells you how seriously the roadmap's author takes the ongoing DRAM squeeze — memory makers have warned it could last through 2028.

RX 10050 XT (AT4): The Mainstream Entry

The leaked RX 10050 XT, built on the smallest Alpha Triton die (AT4), is the entry point of the rumoured lineup: 24 compute units at 3.5 GHz, with 12–16 GB of LPDDR5X memory on a 128-bit bus delivering 160 GB/s, aimed at a $300–$350 launch price in Q4 2027 — though the roadmap warns it could cost $600–$700 if the memory shortage is still raging.

The most important leaked claim about this card is its rumoured performance target: raster performance roughly on par with Nvidia's RTX 3070. Notebookcheck's coverage of the NeoGAF leak puts the 10050 XT at that level, which would make it a very curious product in 2027: a brand-new card matching a GPU that launched in 2020. On paper that sounds disappointing — until you remember the price. An RTX 3070-class card at $300–$350 in a 2027 market would be a genuine budget option, particularly if, as the leak suggests, its ray-tracing performance is proportionally better than its raster performance.

The real question is the memory. 160 GB/s of bandwidth is low for a desktop gaming card in 2027, and the leaked 12 GB minimum capacity would be a talking point — even the RX 9070 XT generation treats 16 GB as the mainstream baseline. If the rumoured 16 GB variant materialises at $350, it becomes far more interesting; if only the 12 GB version ships, its longevity at 1440p would be a legitimate concern. The LPDDR5X choice also raises eyebrows (see the dedicated section below) because it is memory designed for power efficiency rather than raw throughput. As a rumour, though, this is exactly the kind of detail that can change between a leaked roadmap and shipping silicon.

RX 10060 XT (AT3): The Mid-Range Workhorse

The leaked RX 10060 XT on the AT3 die doubles the AT4's compute units to 48 CU at 3.4 GHz, with 16–24 GB of LPDDR5X on a 256-bit bus delivering 320 GB/s, at a leaked $450–$550 price — or $900–$1,100 under continued shortage conditions.

This is the card that looks most like today's RX 9070-class positioning: a 1440p and entry-4K gaming card with generous VRAM. The leaked 16–24 GB capacity range is notable — if the shipping card lands at 24 GB for around $500, that would be a memory-per-dollar figure well above anything Nvidia currently offers in this price band. The 320 GB/s bandwidth, however, is modest for a card in this class: today's mid-range cards typically exceed 600 GB/s. That gap is where the rumoured clock speeds (3.4 GHz) and architectural efficiency improvements would have to do heavy lifting.

Kepler_L2's leak also casts a shadow on this SKU's timing. While the GameGPU roadmap shows Q4 2027, the leaker's claim is that only the AT2 die arrives in 2027, with AT3 — along with AT0 and AT4 — potentially slipping into 2028. If that is right, the 10060 XT could be a 2028 product, nearly three years after the RX 9070 series launched. For a buyer shopping in 2027, the practical takeaway is that this particular tier may simply not exist when the first RDNA 5 cards appear.

RX 10070 XT (AT2): The Rumoured Star of the Launch

Every source agrees on one thing: the AT2 die is the centrepiece of the RDNA 5 rollout. The leaked RX 10070 XT carries 68 compute units at 3.2 GHz, 16–24 GB of GDDR7 on a 192-bit bus with 864 GB/s of bandwidth, a leaked $700–$800 price (or $1,400–$1,600 if the shortage persists), and a Q3/Q4 2027 window.

The rumoured performance target is rasterisation roughly on par with the GeForce RTX 5080 — Nvidia's current second-tier flagship, a card with a $999 list price. If a $700–$800 card matches it in 2027, that would be a strong generational story. Kepler_L2 adds detail: AT2 is rumoured at 68–70 CU and 3.1–3.4 GHz, up to ~25% faster than the RX 9070 XT. That 25% figure is the number to watch — it is a plausible generational uplift for a new architecture on a new process node, and it is notably smaller than the headline claims attached to the flagship SKU below.

The AT2 die also has a second life that may explain its priority. The same die family is reportedly destined for the PlayStation 6 and Xbox's Project Helix — the Magnus SoC for the next Xbox is said to use a standard-derived AT2 chiplet. Console contracts run on fixed timelines, which gives AMD a structural reason to bring AT2 to silicon first. That makes the leaker claim — AT2 in 2027, everything else later — feel at least plausible from a business standpoint, not just a schedule slip.

RX 10090 XT (AT0): The Flagship — With a Catch

The leaked RX 10090 XT is the halo product: 96 compute units at 3.0 GHz, 24–36 GB of GDDR7 on a 384-bit bus with roughly 1.7 TB/s of bandwidth, at $1,000–$1,200 — or $2,000–$2,400 if the memory shortage continues. Leaked launch window: Q3/Q4 2027.

The rumoured performance claim is the boldest in the whole leak set: approaching RTX 5090-level performance. The RTX 5090 is the fastest gaming GPU on the market, and a rumoured AMD card matching it at a $1,000–$1,200 list price would be a genuine market disruption — on paper. But this SKU carries two major caveats from the leaks themselves. First, Notebookcheck notes that the full AT0 die is likely reserved for AI products, with only cut-down AT1/AT2 dies going to Radeon gaming cards — meaning the 10090 XT, if it exists, could be a heavily cut-down part rather than the full die. Second, Kepler_L2's timeline puts AT0 in 2028 at the earliest, not 2027.

The price caveat deserves its own emphasis: the leaked "shortage" price for this card is $2,000–$2,400 — double the base price. The roadmap's author is explicitly modelling a world where the DRAM squeeze does not end. Memory manufacturers have warned the shortage could last through 2028, and if that holds, the flagship's real-world price could land anywhere in that enormous band. A rumour with a 100% price uncertainty range is a rumour telling you, honestly, that it does not know.

TSMC N3P and the 3 GHz+ Clock Question

The leaks consistently place RDNA 5 on TSMC's N3P process — the performance-enhanced third iteration of TSMC's 3-nanometre-class node family. A node move is the single most reliable predictor of a generational leap: smaller, more efficient transistors let a designer spend the power budget on clocks, on more units, or on both. The leaked clocks — 3.5 GHz on the smallest die down to 3.0 GHz on the largest — are the most aggressive GPU clocks ever rumoured for a shipping product, and they follow the expected pattern: smaller dies clock higher because there is less silicon to feed and cool.

The unanswered question, flagged by Notebookcheck, is power consumption. Clocks this high on a 3nm-class node could still produce enormous total board power on the 96-CU flagship — or they could be efficient if the architecture is redesigned around them. The leaks say nothing credible about wattage, and without it, the clock numbers are only half a story. A 3.0 GHz card drawing 600 watts is a very different product from a 3.0 GHz card drawing 350 watts, and buyers should treat any efficiency claim about RDNA 5 as pure speculation until AMD publishes real figures.

Illustration (AI-generated): memory choice is one of the most debated details in the RDNA 5 leaks. This is concept art, not a photo of a real component.

The LPDDR5X Debate: Low-Power Memory on Desktop GPUs

The detail that "raises eyebrows" — Notebookcheck's phrase — is the choice of LPDDR5X for the two cheaper SKUs (the 10050 XT and 10060 XT), while the two expensive cards get GDDR7. LPDDR5X is low-power DDR memory: it is the memory inside phones and thin-and-light laptops, optimised for power efficiency rather than maximum throughput. Putting it on a desktop graphics card is, to put it mildly, unusual.

There are two ways to read it, and both are worth holding in mind. The charitable reading: AMD may be designing RDNA 5's lower tiers for efficiency-first segments — compact desktops, and crucially, laptops and handhelds, where LPDDR5X's power savings are a genuine advantage. Notebook memory is often soldered LPDDR already, so an AT3/AT4 die designed to pair with LPDDR5X would slide naturally into mobile designs. The desktop "XT" cards in the leak might then be reusing mobile-optimised silicon, which would explain the odd memory choice. The sceptical reading: LPDDR5X's bandwidth ceiling is simply too low for desktop gaming at 1440p and above — the leaked 160 GB/s for the 10050 XT and 320 GB/s for the 10060 XT are both well below what current-generation cards in those price bands deliver. If those numbers ship unchanged, the cards could be bandwidth-starved regardless of how fast the compute units are.

The honest position is that this detail is exactly what a leak would get wrong or simplify. Memory configurations are among the last things finalised on a GPU — capacities in particular (note the leaked ranges: 12–16 GB, 16–24 GB) suggest AMD itself had not decided. It is also possible the roadmap conflates mobile and desktop variants of the same dies. Treat the LPDDR5X detail as interesting but provisional: it is the single most likely thing to change before launch.

Ray Tracing and AI: Where the Biggest Gains May Hide

Perhaps the most significant leaked claim is not a number but a shape: Notebookcheck's analysis says RDNA 5 may deliver substantially larger ray-tracing gains than raster gains. If true, that would address AMD's longest-standing weakness against Nvidia. For several generations, AMD's cards have offered competitive traditional (raster) performance per dollar while trailing badly in ray-traced games and in the AI upscaling features (DLSS versus FSR) that have become central to modern PC gaming.

Why does this matter more than raw raster numbers? Because the games of 2027–2028 will not be the games of 2020. Ray tracing has moved from a luxury toggle to a baseline expectation in AAA releases, and path-traced lighting — the heaviest variant — is already the defining visual feature of the most demanding titles. A card that is 25% faster in raster but 60–80% faster in ray tracing (purely illustrative numbers — the leaks give no such figures) would feel like a much bigger upgrade than the raster figure suggests. The leaks do not quantify the ray-tracing improvement, so the only responsible claim is directional: AMD appears to be prioritising the area where it is weakest.

There is a related AI angle. The full AT0 die is reportedly reserved for AI products, which fits the industry's broader direction: every GPU vendor is now designing dies that serve both gaming and AI inference markets. Stronger AI hardware on the die would also feed better upscaling — the successor to FSR — which matters because upscaling quality is now a first-order buying criterion for many gamers. None of this is confirmed, but it is consistent with where AMD has been investing: the company's public messaging around its current products emphasises AI acceleration alongside gaming, and a unified die strategy would explain why the flagship gaming card might only get a cut-down AT0.

RDNA 5 vs RDNA 4: What Would Actually Change

Since RDNA 5 is unannounced, the only honest comparison is between what is confirmed about today's lineup and what the leaks claim about tomorrow's. AMD's current flagship gaming card, the RX 9070 XT, launched in March 2025 on the RDNA 4 architecture — and against that baseline, the rumoured RDNA 5 changes cluster in four areas.

Process node. The leaks place RDNA 5 on TSMC N3P, a newer and denser node family than the nodes behind today's gaming GPUs. A full node jump is historically the largest single contributor to a generational leap — it is what makes the rumoured 3.0–3.5 GHz clocks thinkable at all.

Memory technology. Today's high-end AMD cards use GDDR6-class memory; the leaks split RDNA 5 between GDDR7 (the two expensive SKUs) and LPDDR5X (the two cheaper ones). GDDR7's higher per-pin speeds are what enable the leaked ~1.7 TB/s figure on the flagship — a number that would have been unthinkable on GDDR6 without an impractically wide bus.

Ray tracing priority. RDNA 4 already narrowed the gap with Nvidia in ray tracing compared with earlier generations; the leaks suggest RDNA 5 makes it the headline improvement rather than an afterthought. If the "ray-tracing gains larger than raster gains" claim holds, it would be the first AMD generation designed around closing the company's most visible deficit.

Naming and positioning. The jump from RX 9000 to RX 10000 skips an entire numbering generation — a signal, if the leak is right, that AMD wants buyers to read this as a clean break rather than an incremental refresh. Naming resets are marketing, but they also set expectations: a 10000-series badge promises more than a 9000 refresh would.

What would not change, on current evidence: AMD's value positioning. The leaked price ladder ($300 to $1,200 before shortage adjustments) mirrors the company's long-standing strategy of undercutting Nvidia's list prices at each tier rather than chasing the absolute performance crown — with the possible exception of the 10090 XT, whose rumoured RTX 5090-level target is the one claim that breaks the pattern.

The Timeline Problem: 2027, 2028, or Later?

The leaks disagree with each other on timing, and the disagreement is itself informative. Here is the conflict, laid out plainly:

The GameGPU roadmap shows all four desktop SKUs launching in the second half of 2027 — the 10070 XT and 10090 XT in Q3/Q4, the 10050 XT and 10060 XT in Q4 — under the banner of desktop sales beginning in fall 2027.

Kepler_L2's leak, reported around 12 October 2026, says AMD is unlikely to even announce RDNA 5 before Computex 2027 (1–4 June 2027), that a later-2027 reveal is more likely, and that only the AT2 die is expected in 2027 — with AT0, AT3 and AT4 possibly following in 2028.

Reconciling them: the two leaks are not necessarily contradictory. A Computex 2027 announcement with an AT2-based RX 10070 XT shipping in Q3/Q4 2027, followed by the remaining three SKUs through 2028, fits both accounts reasonably well. What it rules out is the optimistic reading — a full four-card stack on shelves for the 2027 holiday season. The realistic base case from the leaks is one card in 2027, the rest in 2028.

There is also a structural reason to believe the slower timeline. The console angle — PlayStation 6 and Xbox Project Helix reportedly using AT2-derived silicon — means AMD's engineering resources and TSMC allocation will be split between console SoCs and discrete Radeon cards. Console launches are immovable dates with contractual penalties; discrete GPU launches are flexible. If allocation gets tight, the discrete cards wait. Add the memory shortage — memory makers warning of a squeeze through 2028 — and a 2028-heavy rollout starts to look less like pessimism and more like planning.

What the Delay Means for Buyers Right Now

Here is the practical bottom line, and it is the most important section of this article: RDNA 5 is, at best, a year away — and for most price tiers, closer to two years away. You cannot buy a rumour. Every GPU purchase decision you make between now and late 2027 has to be made with the cards that actually exist.

The RX 9000 series remains the buy. The RX 9070 XT (March 2025) is AMD's current flagship gaming card, and the rumoured ~25% uplift of the 10070 XT over it is a normal single-generation improvement — not a revolution that makes today's cards obsolete. If you need a GPU now for 1440p or 4K gaming, buying an RX 9070 XT or its siblings at a sane price is the rational move; waiting 12–24 months to save a theoretical 25% is only sensible if your current card still does the job. The 9070 XT will also benefit from years of driver maturity and FSR improvements that a launch-day RDNA 5 card will not have.

The waiting math is simple. A GPU bought today will serve you through the entire RDNA 5 launch window. By the time the full RX 10000 stack is on shelves — plausibly late 2028 — a card bought in late 2026 will be two years old and ready for a normal upgrade cycle anyway. Waiting only makes sense in one scenario: your current GPU is adequate, and you want to see whether the rumoured ray-tracing leap and the sub-$400 tier materialise before spending. In every other scenario, buy for today's games at today's prices.

Watch the memory market, not just the GPU market. The leaked prices carry their own warning: the roadmap prices every SKU twice, with the "shortage continues" price roughly double the base price. If the DRAM squeeze persists through 2028 as memory makers warn, RDNA 5's street prices could look nothing like its launch prices. A buyer in 2027 should budget against the shortage scenario, not the optimistic one — and today's buyers should know that current cards, bought at current prices, are a hedge against that uncertainty.

Mistakes to Avoid While Waiting for RDNA 5

1. Treating leaked prices as real prices. The $300–$350 RX 10050 XT is a rumour of a rumour — and even the roadmap hedges it up to $600–$700. Never plan a build budget around leaked pricing. Budgets are built on products you can add to a cart.

2. Selling your current GPU in anticipation. Selling now to "wait for RDNA 5" leaves you GPU-less for a year or more if the timeline slips to 2028 — and used GPU prices fall fastest exactly when everyone else is selling ahead of a launch. Upgrade when the new card is reviewed and in stock, not when the rumour appears.

3. Ignoring the cut-down caveat on the flagship. The rumoured "RTX 5090-level" 10090 XT may be a cut-down AT0 with the full die reserved for AI products. Flagship rumours are systematically the most inflated numbers in any leak cycle. Anchor your expectations to the 10070 XT's rumoured ~25% uplift over the 9070 XT, which is the more grounded figure.

4. Assuming the LPDDR5X SKUs will look like the leak. Memory configuration is the most changeable part of a GPU design. If you are specifically waiting for a sub-$400 RDNA 5 card, wait for the announcement — not the roadmap — before getting excited about the price or dismissing the bandwidth.

5. Overpaying for a stopgap. If you need something now and plan to upgrade to RDNA 5 later, buy the cheapest card that meets your needs today — ideally used or previous-generation at a discount. The worst financial move is paying a premium for a high-end card now and replacing it in eighteen months.

6. Forgetting the console factor. If PlayStation 6 and the next Xbox really do use AT2-derived silicon, expect AMD's marketing and engineering attention to tilt toward consoles during 2027. Discrete Radeon launches have historically slipped when console ramps consume allocation. Build schedule risk into your expectations.

FAQ: AMD RDNA 5 and the Radeon RX 10000 Series

1. Has AMD confirmed RDNA 5 or the RX 10000 series?
No. As of 31 October 2026, AMD has announced nothing — no architecture name, no product names, no dates, no prices, no specifications. Everything in this article beyond the current RX 9000 lineup is an unconfirmed rumour from leaks and should be treated as such.

2. When will RDNA 5 graphics cards actually launch?
The leaks point to fall 2027 at the earliest, with leaker Kepler_L2 suggesting an announcement no earlier than Computex 2027 (June 2027) and only the AT2-based card shipping in 2027. The remaining SKUs may follow in 2028. Given the console contracts and the memory shortage, a 2028-heavy rollout is the realistic base case.

3. How many RX 10000 cards are rumoured, and what are they?
Four desktop SKUs appear in the GameGPU roadmap leak: the RX 10050 XT (24 CU, LPDDR5X, ~$300–$350), the RX 10060 XT (48 CU, LPDDR5X, ~$450–$550), the RX 10070 XT (68 CU, GDDR7, ~$700–$800), and the RX 10090 XT (96 CU, GDDR7, ~$1,000–$1,200). All figures are rumoured, and each carries a much higher "if the memory shortage continues" price in the leak.

4. Will the RX 10090 XT really match the RTX 5090?
That is the leak's boldest claim, and it comes with two big caveats: the full AT0 die is reportedly reserved for AI products, so the gaming card may be a cut-down part, and its launch may slip to 2028. Treat "RTX 5090-level" as a best-case rumour, not a target you should plan around.

5. Why would AMD use LPDDR5X memory on desktop graphics cards?
Nobody outside AMD knows whether it actually will — this is one of the most debated and most changeable details in the leaks. The plausible explanations are efficiency-focused designs shared with laptops and handhelds (where LPDDR5X is standard), or a roadmap that conflates mobile and desktop variants. The leaked bandwidth figures (160–320 GB/s) look low for desktop gaming, which is why scepticism is warranted.

6. Is RDNA 5 also coming to consoles?
According to the leaks, yes: PlayStation 6 and Xbox's Project Helix (Magnus SoC) are reported to use AT2-derived silicon. This is consistent with AMD's history of supplying both console makers, and it helps explain why the AT2 die is rumoured to come first — console timelines are contractually fixed.

7. Should I wait for RDNA 5 or buy a GPU now?
If you need a GPU now, buy now. The RX 9070 XT remains a strong card, and the rumoured generational uplift (~25% over the 9070 XT for the 10070 XT) does not make today's cards obsolete. Only wait if your current GPU is adequate and you specifically want to see whether the rumoured ray-tracing improvements and the sub-$400 tier materialise. Never sell your current card in anticipation of a launch that is a year or more away.

8. How will the memory shortage affect RDNA 5 prices?
Severely, if it persists. The leaked roadmap prices every SKU twice — for example, the 10070 XT at $700–$800 normally versus $1,400–$1,600 under continued shortage. Memory makers have warned the DRAM squeeze could last through 2028, so buyers should budget against the higher scenario and treat the base prices as optimistic.

9. What does TSMC N3P mean for these cards?
N3P is the performance-enhanced third iteration of TSMC's 3-nanometre-class process. A node jump is the most reliable driver of generational gains — it is what makes the rumoured 3.0–3.5 GHz clocks plausible. But clocks without power-consumption figures are only half the story, and the leaks say nothing credible about wattage.

10. Where can I follow official news when AMD actually announces something?
AMD's announcements will appear first on AMD's official graphics page and through the company's press channels. Until then, treat every specification, price and date as a rumour — including the ones in this article.

Conclusion: Excitement, Tempered by the Calendar

The RDNA 5 leaks paint an appealing picture: a four-card Radeon RX 10000 stack on TSMC N3P, clocks above 3 GHz, GDDR7 on the high end, a possible answer to Nvidia's ray-tracing lead, and a return to the sub-$400 segment. If even half of it ships as described, 2027–2028 will be an interesting time to buy a graphics card.

But the calendar is the story. One card in late 2027, the rest possibly in 2028, prices hostage to a memory shortage that may run through 2028, and a flagship whose full die may never reach gamers — that is the realistic composite of the leaks. AMD has confirmed nothing, and the gap between a leaked roadmap and shipping silicon has swallowed many promising architectures before.

The smart play has not changed: buy the GPU you need when you need it, from the products that exist. The RX 9000 series is here, reviewed, and priced in real money. RDNA 5 is a rumour with a two-year horizon. Keep an eye on Computex 2027 — that is the earliest date any of this becomes official — and until then, enjoy the games your current card can already play.

Sources

• GameGPU — roadmap article on the Radeon RX 10000 and GeForce RTX 60 releases (25 October 2026), source of the four-SKU lineup, codenames, clocks, memory and price ranges.
• Notebookcheck — analysis of the NeoGAF "Alpha Triton" leak, including rumoured performance targets versus RTX 5080/5090/3070 and the LPDDR5X eyebrow-raiser.
• Guru3D — report on Kepler_L2's claim that AMD is unlikely to announce RDNA 5 before Computex 2027.
• Notebookcheck — coverage of the late-2027 reveal rumour and the claim that only the AT2 die arrives in 2027.
• TweakTown — report on the Kepler_L2 timeline leak via the AnandTech forums.
• AMD — official graphics product page, for the current confirmed Radeon lineup.