Nvidia dropped a detail at its SIGGRAPH 2026 keynote on July 20 that changes the calculus for anyone still deciding between an RTX 4090 and an RTX 5080: DLSS 5 ships this fall, and it runs exclusively on RTX 50-series hardware. No RTX 40-series support has been announced, which means the 4090, Nvidia’s own flagship from 2022, is stuck on DLSS 3 Frame Generation while the cheaper 5080 gets both DLSS 4.5 Multi Frame Generation today and DLSS 5 within months. That single software decision reshapes a comparison that used to be purely about raw horsepower.
The raw horsepower story still favors the 4090. Tom’s Hardware’s GPU Hierarchy testing puts the RTX 4090 roughly 15% ahead of the RTX 5080 in native 4K rasterization, and PCGuide’s 3DMark Port Royal ray-tracing runs show the 4090 opening a 12-16% lead depending on the test pass. But the RTX 5080 launched at $999 against the 4090’s $1,599 debut price, it draws 90 fewer watts, and it’s the only one of the two that will run Nvidia’s newest upscaling and frame-generation stack past this year. As of August 27, 2026, that trade-off is what actually matters for anyone shopping right now, not just the spec sheet.
This piece focuses specifically on gaming performance: native rasterization, ray tracing, and the DLSS frame-generation gap that’s about to widen further. If you’re weighing power draw and efficiency, or VRAM for AI and creative workloads, or the used-market math on a secondhand 4090, tech-insider.org has covered those angles separately, and we’ll link out to them where relevant. Here, the question is narrower and more urgent: which card actually plays games better in August 2026, and does DLSS 5’s RTX 40 snub change the answer for anyone still on the fence?
RTX 4090 vs RTX 5080: Full Specs Comparison
Start with the numbers Nvidia actually publishes. The RTX 4090 is built on the Ada Lovelace architecture and the AD102 die, launched October 12, 2022, at $1,599. The RTX 5080 uses the newer Blackwell architecture on the GB203 die, launched January 30, 2025, at $999. Despite arriving more than two years later and costing $600 less at MSRP, the 5080 doesn’t out-spec the 4090 across the board, and that’s the tension this whole comparison hinges on.
| Spec | RTX 4090 | RTX 5080 |
|---|---|---|
| Architecture | Ada Lovelace (AD102) | Blackwell (GB203) |
| CUDA cores | 16,384 | 10,752 |
| RT cores | 3rd-gen | Blackwell-gen (press-labeled 4th-gen) |
| Tensor cores | 4th-gen | Blackwell-gen (press-labeled 5th-gen) |
| Base clock | 2,235 MHz | 2,295 MHz |
| Boost clock | 2,520 MHz | 2,620 MHz |
| VRAM | 24GB GDDR6X | 16GB GDDR7 |
| Memory bus | 384-bit | 256-bit |
| Memory bandwidth | 1,008 GB/s | 960 GB/s |
| TDP | 450W | 360W |
| Process node | TSMC 4N | TSMC 4N (4nm-class) |
| PCIe interface | PCIe 4.0 x16 | PCIe 5.0 x16 |
| Recommended PSU | 850W | 850W |
| Frame generation | DLSS 3 Frame Gen (single generated frame) | DLSS 4.5 Multi Frame Generation (up to 3 generated frames) |
| Launch date / MSRP | Oct 12, 2022 / $1,599 | Jan 30, 2025 / $999 |
Two lines explain most of the performance gap you’ll read about below. The 4090 packs 53% more CUDA cores and a much wider 384-bit memory bus, so it simply moves more raw data per clock cycle. The 5080 counters with a 100 MHz higher boost clock, a newer process refinement, and a completely different frame-generation pipeline that didn’t exist when the 4090 launched. Neither card wins on paper alone, which is exactly why the benchmarks matter more here than in most GPU comparisons.
Ada Lovelace vs Blackwell: What Actually Changed
The RTX 4090’s AD102 die is still, on raw transistor count, one of the largest consumer GPU dies Nvidia has shipped, packing 76.3 billion transistors into a card that draws up to 450W. It was built for a world where ray tracing was still computationally brutal and DLSS meant “AI upscaling with an optional single generated frame.” Three and a half years later, that architecture is showing its age in one specific area: frame-generation hardware.
Blackwell, the architecture behind the RTX 5080, trades some of that brute-force core count for a redesigned Tensor core pipeline built specifically around generating multiple synthetic frames per rendered frame. Nvidia’s own DLSS documentation describes Multi Frame Generation as a Blackwell-exclusive capability, and outlets tracking the rollout, including Tom’s Hardware’s testing of DLSS 4 MFG on RTX 50-series cards, confirm the feature simply doesn’t run on Ada Lovelace silicon. That’s not a driver limitation Nvidia could patch in later. It’s a hardware decision baked into the flip-metering logic that schedules generated frames, and it’s the single biggest architectural difference separating these two cards despite the 4090 winning almost every raw spec comparison.
The practical result is a strange split. Ask which GPU renders more real, native pixels per second, and the 4090 wins clearly. Ask which GPU can put more total frames on screen, generated or otherwise, in a ray-traced title with Multi Frame Generation switched on, and the 5080 closes most of that gap or occasionally reverses it. Which number matters to you depends entirely on how you play.
RT and Tensor Core Generations, Explained
Nvidia doesn’t publish a clean, side-by-side generation label for RT and Tensor cores the way it does for CUDA core counts, so most of the “4th-gen” and “5th-gen” language you’ll see attached to Blackwell’s ray-tracing and AI hardware comes from architecture briefings and press coverage rather than a single consumer-facing spec sheet. What is well documented is the functional difference. The 4090’s 3rd-gen RT cores handle ray-triangle intersection and bounding-volume-hierarchy traversal largely in fixed-function hardware, the same basic design Nvidia has iterated on since Turing. Blackwell’s RT cores add dedicated hardware for a technique Nvidia calls linear-swept spheres, useful for tracing curved geometry like hair and fur without the triangle-count explosion older methods require.
The Tensor core story matters more for this comparison. Ada Lovelace’s 4th-gen Tensor cores handle DLSS upscaling and the single-frame interpolation DLSS 3 Frame Generation uses just fine. What they can’t do is the flip-metering and multi-frame scheduling Multi Frame Generation depends on, which requires hardware Nvidia only built into Blackwell. That’s the real reason a $1,599 flagship from 2022 can’t run a feature available on a $999 card from 2025: it’s not a marketing decision, it’s a hardware gap that opened up specifically because Nvidia redesigned the Tensor core pipeline around a capability that didn’t exist when the 4090 shipped.
Native Rasterization Benchmarks: 4090 Still Leads
Tom’s Hardware maintains one of the most frequently updated GPU hierarchy charts in the industry, aggregating results across a broad game test suite at three resolutions. Their most recent figures, benchmarked against the RTX 5090 as the 100% baseline, tell a consistent story: the 4090 beats the 5080 at every resolution, and the gap widens as resolution climbs.
| Resolution (Ultra preset) | RTX 4090 (avg FPS) | RTX 5080 (avg FPS) | 4090 lead |
|---|---|---|---|
| 1080p | 183.6 fps | 166.9 fps | +10.0% |
| 1440p | 143.4 fps | 128.3 fps | +11.8% |
| 4K | 89.1 fps | 77.4 fps | +15.1% |
Source: Tom’s Hardware GPU Benchmarks Hierarchy, aggregate results across its standard test suite, 2026.
That widening gap at 4K makes sense once you look back at the specs table. The 4090’s 384-bit bus and 1,008 GB/s of bandwidth matter most exactly where you’re pushing the most pixels per frame. At 1080p, where the bottleneck shifts toward CPU and driver overhead, the two cards sit closer together. At 4K, where memory bandwidth and raw shader throughput dominate, the 4090’s wider architecture pulls ahead by a full 15%. If your monitor is a 4K panel and you play mostly rasterized titles without heavy ray tracing, that’s a real, repeatable performance difference you’ll feel in every session.
Ray Tracing Performance: The Gap Widens Further
Ray tracing is where the 4090’s core count advantage shows up hardest, and it’s also where PCGuide ran dedicated 3DMark Port Royal comparisons directly between the two cards rather than against the RTX 5090. In the first test pass, the RTX 4090 scored 25,132 points against the RTX 5080’s 22,113, a 12% edge for the older card. In a second, more demanding pass, that gap grew to 16%, with the 4090 hitting 18,940 points against the 5080’s 16,069.
Both numbers land above the roughly 15% gap seen in native 4K rasterization, which tracks with how ray-tracing workloads scale. Tracing rays through a scene is memory-bandwidth-hungry work, and the 4090’s 1,008 GB/s pipe simply has more room to move data than the 5080’s 960 GB/s. Add in a full generation more RT cores by raw count, even accounting for Blackwell’s per-core efficiency gains, and the 4090 keeps its lead in every pure ray-tracing benchmark that doesn’t involve frame generation.
This is the part of the story that gets lost when people only look at DLSS marketing slides. Turn off frame generation and just measure rendered rays per second, and Nvidia’s three-and-a-half-year-old flagship still wins. The 5080’s advantage only shows up once you bring its exclusive software stack into the equation, which is exactly what the next section covers.
DLSS 3 Frame Gen vs DLSS 4.5 Multi Frame Generation
Here’s where the comparison flips. The RTX 4090 runs DLSS 3 Frame Generation, the version Nvidia introduced alongside Ada Lovelace back in 2022. It uses a dedicated Optical Flow Accelerator and the card’s 4th-gen Tensor cores to interpolate exactly one AI-generated frame between two rendered frames, roughly doubling the displayed frame rate in supported titles.
The RTX 5080 runs DLSS 4.5 with Multi Frame Generation, and according to Nvidia’s own DLSS technology page, MFG can generate up to three additional AI frames for every one rendered frame, a theoretical 4x multiplier instead of DLSS 3’s 2x ceiling. That capability is exclusive to Blackwell’s redesigned Tensor core and flip-metering pipeline, which is why it doesn’t and can’t run on RTX 40-series cards regardless of driver updates. Tom’s Hardware’s own testing of MFG on RTX 50-series hardware confirms the feature is locked to Blackwell at the silicon level, not gated by software.
What that means in practice: in a path-traced title with everything maxed out, the 5080 renders fewer native frames than the 4090, but it can then multiply what it does render more aggressively. Reviewers testing DLSS 4.5 MFG 6X mode, a further-refined multi-frame setting Nvidia and outlets like Overclockers.ru have profiled across titles including Cyberpunk 2077 and Alan Wake 2, describe displayed frame rates that close most of the native-rendering gap in supported games, even though the underlying rendered frame count still favors the 4090. It’s worth being precise about what’s actually happening here: generated frames improve smoothness and motion clarity, but they don’t reduce input latency the way a genuinely faster GPU would. If you’re playing something twitch-reflex-dependent, the 4090’s higher native frame rate still gives you a more responsive feel even with a lower displayed number.
DLSS 5 Ships This Fall — And Skips the 4090 Entirely
This is the news hook that makes August 2026 the wrong time to treat this comparison as settled. Nvidia used its SIGGRAPH 2026 keynote, held July 19-23 in Los Angeles, to confirm DLSS 5 as exclusive to GeForce RTX 50-series cards at launch. As of this writing, no RTX 40-series support has been announced, and nothing in Nvidia’s SIGGRAPH materials suggests the 4090 will get access later.
The details Nvidia confirmed matter beyond the exclusivity headline. DLSS 5 ships “this fall,” though Nvidia hasn’t given an exact date or minimum driver version as of August 24, 2026. It runs on a single GPU, reversing the dual-GPU configuration shown in earlier 2026 teaser demos, a change that suggests Nvidia scaled back an experimental multi-GPU rendering path before launch. It packages three selectable AI models with per-object masking controls that let developers fine-tune where and how aggressively the upscaling and generation kick in, rather than applying one blanket setting across an entire scene. Fifteen launch titles are already confirmed, including Starfield, Assassin’s Creed Shadows, Hogwarts Legacy, and Oblivion Remastered.
Nvidia first teased DLSS 5 back on March 16, 2026, calling it the company’s most significant breakthrough in computer graphics since real-time ray tracing debuted in 2018. That framing set expectations the July 20 SIGGRAPH keynote had to live up to, and the reaction since has been mixed. Nvidia’s own research directors described the update as closer to a full rendering pipeline than a simple upscaler, but plenty of gamers and reviewers have pushed back on that framing, arguing that generated frames still aren’t the same as genuinely rendered ones no matter how many AI models sit behind them. That debate matters less for this comparison than the hardware fact underneath it: whatever DLSS 5 turns out to be in practice, only RTX 50-series owners get to judge it firsthand this fall.
For anyone weighing a 4090 purchase today, even at a discount, this is the detail that should carry real weight. You’re not just buying yesterday’s flagship. You’re buying a card that Nvidia has already signaled will sit outside its next major upscaling generation, on top of already lacking Multi Frame Generation. The 5080, despite losing the raw rasterization and ray-tracing benchmarks, is the card actually positioned to keep improving through software updates over the next one to two years.
Real-World Examples: Who Feels the Difference
Benchmark percentages only mean so much until you map them onto how people actually use these cards. Here are six scenarios where the 4090-vs-5080 gap plays out differently, drawn from how these builds tend to get used in practice rather than a single controlled test.
Take a streamer running a dual-purpose rig, gaming on the main display while encoding to Twitch or YouTube in the background. That workload leans on both raw shader throughput and NVENC encoding hardware simultaneously, and the extra headroom from the 4090’s wider memory bus helps it absorb the encoding overhead with less frame-time variance. A 5080 handles the same task capably at 1080p or 1440p output, but the margin for error shrinks once a scene gets ray-tracing heavy and the encoder is competing for GPU resources at the same time.
- The competitive 1440p shooter player. Someone running Counter-Strike or Valorant at 1440p on a 240Hz monitor cares about native frame rate and input latency, not generated frames. The 4090’s 11.8% lead at 1440p Ultra translates directly into a smoother, more responsive feel, and DLSS 4.5 MFG offers little benefit in titles this light on GPU load anyway.
- The path-tracing single-player enthusiast. Someone replaying Cyberpunk 2077 in Overdrive mode or Alan Wake 2 with full ray tracing benefits most from the 5080’s Multi Frame Generation, since these titles are exactly what MFG was built for and the smoothness gain is most visible in slower-paced, cinematic gameplay.
- The hybrid creator-gamer. Anyone splitting time between DaVinci Resolve color grading, Blender renders, or local AI image generation and gaming leans toward the 4090’s 24GB of VRAM, which matters far more for those workloads than for gaming frame rates. Tech-insider.org’s separate breakdown of the 24GB vs 16GB VRAM gap for AI workloads covers this angle in more depth.
- The small-form-factor builder. A 360W TDP versus 450W isn’t just a power bill difference. It changes case airflow requirements, cooler sizing, and whether a build fits in a compact ITX case at all. The 5080’s lower draw and smaller reference cooler footprint make it the easier card to fit into a cramped enclosure.
- The long-term upgrader. Someone planning to keep a GPU for three-plus years without upgrading again should weight DLSS 5’s RTX 50-exclusivity heavily. Buying a 4090 today means starting that three-year clock already behind on frame-generation software, with no announced path to catch up.
- The value hunter comparing used vs new. With used RTX 4090 prices swinging wildly across trackers as of August 2026, some buyers can land a secondhand 4090 for close to what a new 5080 costs, which changes the math entirely toward raw rasterization performance per dollar. The pricing section below breaks down exactly how close that gap actually is.
Which Games Benefit Most From Each Card
Not every game stresses these two cards the same way, and knowing which category your library falls into matters more than any single aggregate benchmark. Competitive shooters like Counter-Strike 2 and Valorant rarely tax either GPU heavily even at high refresh rates, since both cards can comfortably clear 240fps at 1440p on medium-to-high settings without needing DLSS at all. In that category, the 4090’s raw clock and core advantage still nudges frame rates higher, but the gap is small enough that most players wouldn’t notice it blind.
The story changes completely in path-traced, single-player AAA titles. Cyberpunk 2077 with Overdrive ray tracing, Alan Wake 2 with full path tracing, and Black Myth: Wukong’s ray-traced global illumination mode are all demanding enough that neither card hits comfortable native 4K frame rates without upscaling help. This is exactly the category DLSS 4.5 Multi Frame Generation was built for, and it’s where the 5080’s software advantage does the most to offset its weaker native rendering. It’s also the category where DLSS 5’s confirmed launch lineup matters most directly, since Starfield, Assassin’s Creed Shadows, Hogwarts Legacy, and Oblivion Remastered are all exactly the kind of open-world, ray-tracing-capable titles that benefit from smarter frame generation.
A third category sits in between: modded games and simulation titles, where VRAM footprint often matters more than raw compute. Heavily modded Skyrim or Cyberpunk 2077 builds running 4K texture packs, or flight and racing sims with dense scenery caching, can push past 16GB of VRAM usage in extreme configurations. That’s a scenario where the 4090’s 24GB ceiling gives it headroom the 5080 simply doesn’t have, regardless of which card renders faster in a stock benchmark.
Should You Buy Now or Wait for DLSS 5?
Given that DLSS 5 is confirmed for this fall but still has no exact release date as of August 24, 2026, plenty of shoppers are asking whether it’s worth waiting rather than buying either card right now. The honest answer depends on which card you’re actually considering.
If you’re looking at a 4090, waiting doesn’t change anything, since DLSS 5 won’t run on it regardless of when it ships. The only reason to delay a 4090 purchase is standard price-watching: used prices have swung wide enough in 2026 that patience alone can save several hundred dollars on a clean unit. If you’re looking at a 5080, the calculus is different. Buying now gets you DLSS 4.5 Multi Frame Generation immediately, and DLSS 5 arrives as a driver-level update to hardware you already own rather than a reason to buy a different card. Nvidia hasn’t tied DLSS 5 to a new GPU SKU. It’s a software layer landing on already-shipping Blackwell silicon, which means a 5080 bought today in August 2026 gets the same DLSS 5 access as one bought the week it launches this fall.
The one group that should genuinely consider waiting is anyone hoping Nvidia extends DLSS 5 support to RTX 40-series cards after launch. Nothing currently rules that out, but nothing supports it either, and betting a purchase decision on an unannounced future update is a real gamble rather than a reasonable plan. Buy based on what’s confirmed today, not what might change later.
Pricing: MSRP, Street Prices, and the Used Market
Pricing is where this comparison gets genuinely complicated, because these two cards no longer occupy separate price tiers the way their $999 versus $1,599 launch MSRPs suggest. The RTX 5080 has never consistently sold at $999 since launch, and the RTX 4090’s used market spans an enormous range depending on condition and where you’re buying.
| Price point | RTX 4090 | RTX 5080 |
|---|---|---|
| Launch MSRP | $1,599 (Oct 2022) | $999 (Jan 2025) |
| New street price, Aug 2026 | Limited new stock remaining | $1,219 avg, per GPU Poet’s August 2026 tracker |
| New street price (retailer high) | — | $1,547.84 at Amazon, per videocardprices.com |
| Used, budget condition | ~$700-$920 (eBay, regional trackers) | N/A (too recent for a mature used market) |
| Used, “fair value” clean cards | $1,975-$2,750, per a July 26, 2026 tracker snapshot | N/A |
| New street price (budget tier) | — | $1,250-$1,300, per buysellram.com’s August 2026 price report |
That spread is wide enough to be genuinely confusing, and it reflects how differently these two cards are tracked. The 5080 is still an active retail product with consistent new-unit pricing across major trackers like buysellram.com’s August 2026 GPU price report, which lists new units running $1,250-$1,300 and used units, where they exist, at $1,167-$1,391. The 4090, on the other hand, has largely shifted to a secondary market as new-in-box stock has thinned out, and pricing varies enormously by condition, from budget eBay listings in the $700-$920 range up to “fair value” estimates north of $2,300 for clean, recently-boxed units.
The practical takeaway: a careful shopper hunting for a used 4090 in good condition, not a bargain-bin unit with unknown history, is very likely paying at or above what a new RTX 5080 costs new. That changes the buying decision from “which is cheaper” to “which set of trade-offs am I actually paying for.” Tech-insider.org’s dedicated look at the used RTX 4090 market and its 57% premium over comparable new options digs deeper into that specific math.
Power Draw, Thermals, and PSU Requirements
The RTX 4090’s 450W TDP and the RTX 5080’s 360W TDP work out to a 90W gap that shows up in every part of a build. Nvidia recommends an 850W PSU baseline for both cards, but the headroom looks different in practice. A 4090 system running near its power ceiling under sustained load leaves less margin on an 850W unit, especially paired with a high-end CPU also drawing 200W-plus under load. A 5080 system has noticeably more breathing room at the same PSU wattage, and reviewers building around 5080s more comfortably recommend staying at 850W even for overclocked configurations, reserving the 1000W step-up for enthusiasts chasing maximum headroom rather than treating it as a necessity.
Thermals follow the same pattern. The 4090’s reference and partner coolers are typically triple-slot designs built to dissipate 450W without excessive noise, which means bigger cards, heavier weight, and more careful case-clearance checks before buying. The 5080’s 360W target allows for meaningfully smaller cooler designs across the board, which matters directly for anyone building in a mid-tower or smaller case, or anyone who’s had to deal with GPU sag from a triple-slot 4090 pulling on a motherboard’s PCIe slot over time.
5 Use-Case Recommendations: Which Card Fits Your Setup
- 4K, ray-tracing-heavy, single-player gaming: lean 5080 if buying new, since DLSS 4.5 MFG (and DLSS 5 later this year) does more to keep frame rates smooth in path-traced titles than the 4090’s native rendering edge alone.
- Competitive, high-refresh, native-frame-rate gaming: the 4090 wins outright here, and its 10-15% native rasterization lead is a real, repeatable advantage that generated frames can’t replicate for latency-sensitive play.
- Mixed AI/creative and gaming workloads: the 4090’s 24GB VRAM matters more than its gaming benchmarks for anyone running local LLMs, Stable Diffusion, or large Blender/Resolve projects alongside gaming.
- Small-form-factor or power-constrained builds: the 5080’s 360W TDP and smaller cooler options make it the easier fit, especially in cases or PSUs that can’t comfortably absorb a 450W card.
- Long-term software support and future-proofing: the 5080 is the safer bet for anyone planning to keep a card for years, given DLSS 5’s confirmed RTX 50-exclusivity at launch and no announced RTX 40 path forward.
Migration Guide: Moving From RTX 4090 to RTX 5080
If you’ve decided the software roadmap and power savings outweigh the raw performance you’d be giving up, here’s how to actually make the swap cleanly.
- Check your PSU’s connector. The 5080 uses a native 12V-2×6 (PCIe Gen 5) connector on ATX 3.1-compliant units. If your current PSU predates that standard, you’ll need either a compatible adapter or a PSU upgrade before the card even physically fits.
- Confirm case clearance and airflow. Even though the 5080 typically ships in smaller coolers than the 4090, always check the specific card’s dimensions against your case, since some partner 5080 models still run triple-fan designs for acoustics reasons.
- Fully uninstall old drivers before swapping hardware. Use Display Driver Uninstaller (DDU) in Safe Mode to remove all traces of your 4090’s drivers before physically installing the 5080. Mixed driver states are a common source of black-screen and crash issues after a GPU swap.
- Install the latest driver through the Nvidia App, not a cached installer, so you get the current Game Ready driver with full Blackwell and DLSS 4.5 support out of the box.
- Re-check per-game DLSS overrides. Any per-title DLSS presets or Frame Generation toggles you set for the 4090 won’t automatically map to Multi Frame Generation settings on the 5080. Revisit them individually, especially in path-traced titles where MFG modes (2x/3x/4x) need to be selected manually per game.
- Benchmark your own system before selling the old card. Run the same test scene or benchmark on both cards in your actual PC, not just published review numbers, since your specific CPU, RAM, and cooling setup can shift the real-world gap in either direction.
A quick way to confirm your driver and DLSS feature support after the swap is checking the Nvidia App’s system information panel, or running a command-line check for driver version if you’re troubleshooting:
nvidia-smi --query-gpu=name,driver_version,memory.total --format=csvThat single line confirms the card is being correctly detected, which driver version is active, and how much VRAM is available to the system, useful both right after a swap and any time a game reports it can’t find DLSS 4.5 or MFG options in its settings menu.
RTX 4090 Pros and Cons
RTX 4090 Pros
- 10-15% faster native rasterization across 1080p, 1440p, and 4K per Tom’s Hardware’s hierarchy testing
- 12-16% faster in pure ray-tracing benchmarks (3DMark Port Royal) per PCGuide’s direct testing
- 24GB of VRAM, a real advantage for AI, creative work, and heavily modded games
- Wider 384-bit memory bus and higher 1,008 GB/s bandwidth
- Mature driver stack with years of game-specific optimization behind it
RTX 4090 Cons
- Capped at DLSS 3 Frame Generation, with no Multi Frame Generation support at the hardware level
- Confirmed excluded from DLSS 5 at its fall 2026 launch, with no announced future support
- 450W TDP requires larger coolers, more case airflow, and leaves less PSU headroom
- Used-market pricing is inconsistent and, for clean units, can meet or exceed a new RTX 5080
- No longer in mainstream new production, making warranty and long-term support less certain
RTX 5080 Pros and Cons
RTX 5080 Pros
- DLSS 4.5 Multi Frame Generation today, and confirmed DLSS 5 access this fall
- 360W TDP, 90W lower than the 4090, with smaller cooler and case-clearance requirements
- Newer PCIe 5.0 x16 interface and GDDR7 memory
- $999 launch MSRP, hundreds less than the 4090’s $1,599 debut price
- Still in active new production with consistent retail availability
RTX 5080 Cons
- 10-15% slower native rasterization across all tested resolutions
- 12-16% slower in pure ray-tracing benchmarks without frame generation active
- Only 16GB of VRAM, a real limitation for AI workloads and some texture-heavy modded games
- Narrower 256-bit memory bus and lower 960 GB/s bandwidth
- Street prices as high as $1,547.84 at some retailers, well above the $999 MSRP
The Verdict: Which GPU Should You Actually Buy
Judge this purely on rendered pixels and the RTX 4090 wins, clearly and repeatedly, across every benchmark source in this piece. A 10% lead at 1080p, 11.8% at 1440p, 15.1% at 4K, and a 12-16% edge in dedicated ray-tracing tests add up to a GPU that’s still, in raw terms, faster than its cheaper, newer replacement. If your priority is native frame rate for competitive or latency-sensitive gaming, or you need the 24GB VRAM for creative and AI work alongside gaming, the 4090 remains the better tool, provided you can find one at a fair used price.
But “raw terms” is doing a lot of work in that sentence. The RTX 5080 costs $600 less at MSRP, draws 90 fewer watts, and, critically, is the only one of the two cards Nvidia has confirmed will run DLSS 5 this fall. Combined with DLSS 4.5 Multi Frame Generation already closing much of the visible smoothness gap in path-traced titles today, the 5080 is the more forward-looking purchase for anyone buying new in August 2026, even though it loses almost every individual benchmark in this article.
The practical split: buy a 4090 only if you’re getting a genuinely clean used unit at a real discount and you value raw native performance or VRAM headroom above software longevity. Buy a 5080 if you’re purchasing new, care about ray-traced single-player games, or want a GPU that keeps receiving Nvidia’s newest upscaling and frame-generation features for the next several years. Given how close used-4090 and new-5080 pricing has drifted as of August 2026, that second scenario now applies to most shoppers.
Frequently Asked Questions
Is the RTX 4090 still worth buying in 2026?
Only if you can find a genuinely clean used unit at a meaningful discount to a new RTX 5080. At parity or a premium, the 5080’s lower power draw and confirmed DLSS 5 support make it the safer long-term buy, even with the 4090’s raw performance lead.
Does the RTX 5080 beat the RTX 4090 in ray tracing?
Not in native, rendered ray-tracing performance. PCGuide’s 3DMark Port Royal testing shows the 4090 ahead by 12-16% depending on the test pass. The 5080 only closes that gap once DLSS 4.5 Multi Frame Generation is switched on, which boosts displayed frame rate without changing the underlying rendered performance.
What is DLSS 4.5 Multi Frame Generation, and does the RTX 4090 support it?
Multi Frame Generation can insert up to three AI-generated frames for every one rendered frame, a 4x multiplier compared to DLSS 3 Frame Generation’s 2x ceiling. It’s exclusive to Blackwell’s Tensor core and flip-metering hardware, so it does not and cannot run on the RTX 4090.
Will the RTX 4090 ever get DLSS 5?
As of August 27, 2026, Nvidia has confirmed DLSS 5 as exclusive to RTX 50-series cards at launch this fall, with no RTX 40-series support announced. Nvidia hasn’t explicitly ruled it out for the future, but nothing in its SIGGRAPH 2026 materials suggests it’s coming.
Is 16GB of VRAM on the RTX 5080 enough for 4K gaming in 2026?
For gaming specifically, yes, in the vast majority of current titles at 4K with typical texture settings. It becomes a real limitation mainly in VRAM-heavy modded games or when running AI/creative workloads alongside gaming, where the 4090’s 24GB has a clear edge.
What PSU do I need for an RTX 5080 vs an RTX 4090?
Nvidia recommends an 850W PSU baseline for both cards. The 5080’s 360W TDP leaves more headroom at that wattage than the 4090’s 450W draw, particularly in systems paired with power-hungry CPUs.
Is a used RTX 4090 cheaper than a new RTX 5080 in August 2026?
It depends heavily on condition. Budget used 4090 listings can run $700-$920, well under the 5080’s roughly $1,219-$1,300 street price. But “fair value” clean units have been tracked as high as $1,975-$2,750, meaning a well-maintained used 4090 often costs as much or more than a new 5080.
Which card is better for ray tracing overall, RTX 4090 or RTX 5080?
The 4090 wins on native, rendered ray-tracing performance by 12-16%. The 5080 wins on perceived smoothness in supported titles once DLSS 4.5 Multi Frame Generation is active, plus it has a confirmed path to DLSS 5 this fall that the 4090 lacks entirely.