A used RTX 4090 now sells for more than a brand-new RTX 5080. That single fact, confirmed across half a dozen pricing trackers heading into September 2026, has scrambled the usual upgrade math for anyone shopping the high end of Nvidia’s lineup. Multiple eBay-tracking sites, including GetPCParts, which measured a used RTX 4090 24GB averaging $2,541 in sold listings through August 29, 2026, and ResalePrices, which pegged the September 2026 market average at $2,500, now put the current fair value of a used RTX 4090 24GB at roughly $2,500 to $2,600, while new RTX 5080 cards are streeting around $1,150 to $1,400 against a $999 launch MSRP. The gap didn’t happen because the RTX 4090 got better. It happened because the GDDR7 memory that powers the RTX 50 series is in short supply, and Nvidia’s planned RTX 5080 Super refresh, which was supposed to fix the 5080’s 16GB ceiling, never shipped.
This comparison breaks down what you actually get for the money on each card in August 2026: full specs, independently sourced benchmark deltas from TechPowerUp, Tom’s Hardware, and Guru3D, current street and used pricing, real VRAM-bound workloads where the gap matters, and a migration path if you’re moving between the two. We’ll also walk through where the RTX 5080 Super disappeared to, why GDDR7 now eats more than 80% of a high-end GPU’s bill of materials, and which of these two cards is the rational buy depending on what you actually do with a GPU.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
RTX 4090 vs RTX 5080: Full Specs Comparison
The RTX 4090 launched in October 2022 on the Ada Lovelace architecture at $1,599. The RTX 5080 launched in January 2025 on the newer Blackwell architecture at $999. On paper, that looks like a straightforward “newer and cheaper” story. In practice, the two chips take very different approaches: the 4090 is a bigger, hungrier die with a wider memory bus and more VRAM, while the 5080 is a smaller die that leans on faster GDDR7 memory and DLSS 4’s frame generation to close the gap. Here’s how the specs actually stack up.
| Specification | RTX 4090 | RTX 5080 |
|---|---|---|
| Architecture | Ada Lovelace (AD102) | Blackwell (GB203) |
| Launch date / launch MSRP | October 2022 / $1,599 | January 2025 / $999 |
| CUDA cores | 16,384 | 10,752 |
| Boost clock | 2.52 GHz | 2.617 GHz (FE) |
| Memory capacity | 24GB GDDR6X | 16GB GDDR7 |
| Memory bus width | 384-bit | 256-bit |
| Memory bandwidth | 1,008 GB/s | Up to 960 GB/s (30 Gbps modules) |
| Ray tracing cores | 3rd-gen RT cores | 4th-gen RT cores |
| Tensor cores | 4th-gen Tensor cores | 5th-gen Tensor cores |
| Total board power (TBP) | 450W | 360W |
| Production status (Aug 2026) | Discontinued, ~October 2024 | In production, no Super/Ti variant shipped |
| Typical August 2026 price | $2,200–$2,500 (used) | $1,150–$1,400 (new, street) |
The headline numbers favor the RTX 4090 on raw silicon: 50% more VRAM, a wider memory bus, and 52% more CUDA cores. The RTX 5080 answers with a newer architecture, faster per-pin memory speed on GDDR7, and a 90W lower power draw. Nvidia’s own spec pages for the RTX 4090 and RTX 5080 confirm both configurations, and neither has changed since launch.
Why RTX 5080 Never Got Cheaper: The GDDR7 Shortage Explained
The RTX 5080 launched at $999 in January 2025. Eighteen months later, it’s still not reliably available at that price. The reason isn’t scalpers or retailer markups in the way GPU shortages usually get blamed. It’s memory. According to a July 31, 2026 report from Phandroid, GDDR7 memory now accounts for more than 80% of the bill-of-materials cost on a high-end Nvidia GPU, and that figure has kept climbing through 2026 as AI datacenter demand competes with consumer graphics cards for the same production lines.
Nvidia has cut supply of RTX 50 series cards to its board partners by an estimated 20% to 40% during the first half of 2026, according to reporting from Tech Times and separate analysis from RoIC.ai. That supply cut, combined with rising per-chip GDDR7 costs, pushed street prices for the RTX 5080 up roughly 31% to 35% by early August 2026, based on a Tech Times report tracking South Korean retail data that found no regional market was insulated from the increase. A separate April 2026 memory-economics breakdown from SoftwareSeni put the RTX 5080’s per-gigabyte VRAM cost at roughly $62.44, nearly identical to the flagship RTX 5090’s $62.47 per gigabyte, despite the two cards selling in completely different price tiers.
The RTX 5090 has it worse. Street prices for Nvidia’s flagship have been reported anywhere from $3,000 to over $6,000 against a $1,999–$2,199 MSRP, a premium exceeding 150% in some secondary-market listings. One market tracker cited by LocalAIMaster found the RTX 5090’s street price rose 79% in three months during mid-2026, compared to a 35% rise for the RTX 5080 over the same window. Neither card is anywhere close to its launch price, and industry analysts quoted in a July 2026 GPU drought report from DropReference don’t expect meaningful relief until at least Q4 2027, since GDDR7 and broader DRAM shortages are tied to the same AI-driven memory demand that’s also inflating RAM and SSD prices across the PC industry.
Regional Pricing: The Shortage Hits Every Market
The GDDR7 shortage isn’t a US-only phenomenon, and the used RTX 4090 pricing data confirms it, even if the ride there was bumpier than today’s numbers suggest. PCPrice.Watch’s own tracker actually clocked the used RTX 4090 as low as $1,462 in June 2026, 9% below its original $1,599 MSRP, and Bandurart’s citation of BestValueGPU data put it even lower, around $1,099 on eBay in April 2026, at a time when leftover “new” units were still fetching roughly $2,755 on Amazon. That dip didn’t last: PCPrice.Watch’s eBay market trackers show a used RTX 4090 selling for a median of $2,099 in the United States market as of August 17, 2026, with most sales falling between $2,012 and $2,185. The same tracker’s Germany listing, last updated August 14, 2026, shows a nearly identical median of €2,098, with a typical range of €2,010 to €2,165. A UK snapshot from July 10, 2026 put the median at £2,101, with most sales between £1,998 and £2,181. New RTX 50 series cards show the same pattern from the supply side: the Tech Times report on South Korean retail found RTX 5070 Ti and RTX 5080 pricing up 31% to 35% locally, explicitly noting that “no market was insulated” from the increase. Whichever currency or region you’re shopping in, the underlying economics are the same: GDDR7 scarcity, not local retailer behavior, is setting the floor on price.
The Stranded RTX 5080 Super: Why Nvidia’s VRAM Fix Never Shipped
Nvidia’s answer to the RTX 5080’s 16GB VRAM limit was supposed to be a Super refresh. Leaked plans described an RTX 5070 Super with 18GB, an RTX 5070 Ti Super with 24GB, and an RTX 5080 Super with 24GB, all built around 3GB-per-chip GDDR7 modules instead of the 2GB modules used in the current RTX 50 lineup. As of August 18, 2026, none of these cards have launched. A July 15, 2026 industry report from DropReference described the refresh bluntly: “It’s dead.” The problem isn’t marketing strategy, it’s physics and allocation. The higher-density 3GB GDDR7 modules needed for a 24GB RTX 5080 Super are the same modules in shortest supply, and datacenter AI accelerators are absorbing that capacity at prices consumer GPU makers can’t match.
This matters directly for anyone comparing the RTX 4090 and RTX 5080 today, because it means the VRAM gap between the two cards isn’t closing anytime soon. A 24GB RTX 5080 Super would have neutralized the RTX 4090’s biggest advantage for creator and AI workloads. Instead, buyers in August 2026 are choosing between a 16GB card that’s readily available and a 24GB card that’s been out of production for nearly two years and is only obtainable secondhand. Craftrigs’ GDDR7 shortage analysis, last updated in July 2026, forecasts real supply normalization no earlier than Q4 2027, meaning the earliest realistic window for a 24GB Blackwell refresh at anything close to MSRP is 2027 at best.
Used RTX 4090 Pricing in August 2026: The eBay Data
Nvidia stopped manufacturing the RTX 4090 around October 2024, according to supply-chain reporting from The FPS Review that circulated at the time. Every RTX 4090 sold in 2026 is either leftover retail inventory or a used card changing hands on the secondary market, and the pricing data shows just how much demand has kept those cards expensive. Multiple trackers pulling from verified eBay sold listings converge on a similar range for August 2026.
| Source | Date | Used RTX 4090 Price |
|---|---|---|
| GetPCParts (verified eBay sold listings) | Aug 29, 2026 | $2,541 average |
| ResalePrices (10-listing sample) | Sep 2026 | $2,500 market average |
| DealHunter (daily-updated active listings) | Aug 2026 | $2,570 median asking price |
| RigPrice (65 active listings) | Aug 2026 | $2,700 average ($2,600–$2,999 range) |
| Clustermelt secondary-market range | Aug 1, 2026 | $2,299–$2,630 |
| Silicon Comps fair-value model | Jul 26, 2026 | $2,325 fair value, $1,975–$2,750 typical band |
| LocalAIMaster (AI workload buying guide) | Aug 12, 2026 | $2,100–$2,350 |
Every recent tracker, including GetPCParts’ verified eBay sold-listing tracker, which measured a $2,541 average through August 29, 2026, DealHunter’s daily-updated median of about $2,570 for August 2026, and RigPrice’s 65-listing sample putting the going rate at $2,700, with most sales between $2,600 and $2,999, that same month, now clusters in a $2,300 to $2,800 window, a meaningfully tighter band than you’d expect from a two-year-old discontinued card, echoed by Alibaba Electronics’ report of wholesale units trading between $2,200 and $2,800 in June 2026, a 40% to 75% markup over the original $1,599 MSRP. That consistency is the clearest evidence that used RTX 4090 pricing in 2026 isn’t being driven by scarcity marketing or a handful of outlier listings, it’s a real, sustained secondary market where the 24GB of GDDR6X still commands a premium over anything Nvidia currently sells new below the RTX 5090. Compare that to the RTX 5080’s $1,150–$1,400 street range, and a shopper looking at both cards side by side in late August 2026 is paying roughly 64% to well over 100% more for the older, out-of-warranty card, depending on which listings you compare.
Benchmark Showdown: 4K Gaming, Ray Tracing, and Rasterization
Independent lab data tells a consistent story: the RTX 4090 is still the faster gaming card in pure rasterization and ray tracing, even against a newer architecture. TechPowerUp’s RTX 5080 Founders Edition review measured the 5080 beating the RTX 4080 Super by 16% and the vanilla RTX 4080 by 19% at 4K resolution in pure rasterization, without ray tracing or DLSS enabled. Tom’s Hardware’s separate 4K rasterization testing found a smaller gap, with the RTX 5080 beating the RTX 4080 Super by roughly 9% at 4K ultra settings, and Tom’s reviewers noted directly that “outside of multi-frame generation, it’s not significantly faster than the 4080 Super in most of our tests.”
The RTX 4090, by comparison, established a 25% to 35% lead over the RTX 4080/4080 Super in 4K rasterization back when both cards launched, a gap that’s well documented across TechPowerUp’s and Tom’s Hardware’s historical review databases. Cross-referencing the RTX 5080’s measured gains over the 4080 Super with the RTX 4090’s established lead over that same baseline puts the RTX 4090 roughly 10% to 20% ahead of the RTX 5080 in average 4K rasterized frame rates, and the gap widens to an estimated 15% to 25% in heavy ray-traced titles, where the 4090’s larger die and wider memory bus carry more weight. Guru3D’s Founders Edition review lists the RTX 5080 at 10,752 CUDA cores against a 2.617 GHz boost clock, confirming the spec-level ceiling behind those numbers.
Driver maturity is a live variable here too. Tom’s Hardware’s RTX 5080 review, updated as recently as mid-2026, flagged “driver issues in some games and apps” that were holding the card below its expected performance ceiling relative to the RTX 4080 Super in specific titles. That means some of the RTX 5080’s underperformance versus its own hardware potential is a software problem Nvidia can still fix with future Game Ready driver releases, not a hard architectural limit. As of August 2026, though, no driver update has closed the gap enough to flip the overall hierarchy: the RTX 4090 remains the faster card for pure gaming performance, particularly at 4K with ray tracing enabled.
| Benchmark Source | Test Condition | Result |
|---|---|---|
| TechPowerUp (RTX 5080 FE review) | 4K raster, no RT/DLSS, vs RTX 4080 Super | RTX 5080 +16% |
| Tom’s Hardware (RTX 5080 review) | 4K ultra raster, vs RTX 4080 Super | RTX 5080 +9% |
| Combined estimate vs RTX 4090 | 4K raster (historical 4090 vs 4080 Super baseline) | RTX 4090 ~10–20% faster than RTX 5080 |
| Combined estimate vs RTX 4090 | 4K ray tracing | RTX 4090 ~15–25% faster than RTX 5080 |
| Guru3D (RTX 5080 FE review) | Total board power under gaming load | 360W (vs 450W on RTX 4090) |
VRAM Matters: Path Tracing, AI/LLM Inference, and Creator Workloads
Raw frame rate isn’t where the RTX 4090’s 24GB of VRAM pays off. It’s in the workloads that fill VRAM before they run out of compute. Full path-traced games at 4K with maxed-out texture settings, like Cyberpunk 2077’s RT Overdrive mode, can push VRAM usage close to or past 16GB, which forces the RTX 5080 into texture-quality compromises or streaming stutters that the RTX 4090’s extra 8GB simply avoids.
The same pattern shows up in AI and creative work. Local LLM inference and Stable Diffusion generation both scale directly with available VRAM: larger models, higher batch sizes, and longer context windows all need headroom that a 16GB card runs out of faster than a 24GB card. LocalAIMaster’s August 2026 buying guide, which specifically compares the RTX 5070 Ti against the RTX 4090 for AI workloads, frames the extra 8GB as the deciding factor for anyone running 13B-parameter-plus models locally without quantization tricks. Blender and other GPU path-tracing renderers behave the same way: once a scene’s geometry, textures, and BVH data exceed available VRAM, most renderers fall back to system RAM or the CPU, which can turn a render that takes minutes on a 24GB card into one that takes hours, or fails outright, on a 16GB card. DaVinci Resolve users editing standard 4K timelines generally won’t hit this wall on the RTX 5080, but anyone working with 8K footage, heavy Fusion compositing, or multiple layered high-bit-depth effects will find the RTX 4090’s headroom noticeably safer.
It’s worth being precise about what “hitting the VRAM wall” actually looks like in practice, because it isn’t always a crash. In games, it usually shows up first as texture pop-in or a brief hitch when the engine has to evict and reload assets mid-scene, before eventually degrading to a hard stutter if the shortfall is severe enough. In Blender or other GPU renderers, the failure mode is blunter: once a scene exceeds available VRAM, the render either falls back to a much slower CPU/system-RAM path or throws an out-of-memory error and stops outright, which is why professionals treat VRAM headroom as a hard requirement rather than a nice-to-have. For AI inference, the symptom is usually an explicit CUDA out-of-memory error rather than a graceful slowdown, forcing the user to drop quantization levels, reduce batch size, or abandon the model entirely. None of these failure modes show up in a standard FPS benchmark chart, which is exactly why the RTX 4090’s 24GB matters more in practice than the raw specs sheet suggests for anyone running these specific workloads.
Power Consumption and Efficiency: 450W vs 360W
The RTX 4090 carries a 450W total board power rating from Nvidia’s own launch specifications. The RTX 5080, according to Guru3D’s Founders Edition testing, runs at 360W under full gaming load, a 90W difference that adds up over a year of regular use and matters for anyone building around a smaller power supply or a case with tighter thermal headroom. On pure performance-per-watt, the RTX 5080’s newer Blackwell architecture and GDDR7 memory clearly do more with less power, even though it’s giving up outright performance to the RTX 4090’s larger die. For a PC builder choosing a power supply, that gap can be the difference between needing an 850W unit for the 4090 versus a comfortable 750W for the 5080, especially when paired with a modern high-end CPU.
The efficiency gap also compounds over time in ways that are easy to overlook at purchase time. Running either card for four to six hours a day of gaming or rendering, the 90W difference adds up to real electricity cost over a year, and it also means the RTX 5080 build runs cooler and quieter under sustained load, since less heat has to be dissipated through the same case airflow. For anyone building a small-form-factor PC or working in a warm climate without strong air conditioning, that efficiency advantage can matter as much as the price difference, since a 450W card in a cramped case is far more likely to hit thermal throttling under a long render or extended gaming session than a 360W card in the same enclosure.
Pricing Table: MSRP, Street Price, and Used Market in August 2026
Nothing about either card’s price has stayed anchored to its launch number. Here’s the full pricing picture as of August 18, 2026, combining official MSRP, current street pricing for new cards, and used-market data for the discontinued RTX 4090.
| Card | Launch MSRP | August 2026 Price | Change vs MSRP |
|---|---|---|---|
| RTX 4090 (used, since discontinued) | $1,599 (2022) | $2,200–$2,500 | +38% to +56% |
| RTX 5080 (new, street) | $999 (2025) | $1,150–$1,400 | +15% to +40% |
| RTX 5090 (new, street) | $1,999–$2,199 (2025) | $3,000–$6,000+ | +50% to +150%+ |
| RTX 5070 Ti (new, street, for reference) | $749 (2025) | ~$900 | +20% |
Every card on this table costs more today than it did at launch, but the RTX 4090’s used-market premium stands out because it’s fighting two forces at once: general GPU price inflation from the GDDR7 shortage, and scarcity value from being out of production entirely. A buyer choosing between a used RTX 4090 at GetPCParts’ late-August 2026 average of $2,541, or as high as $2,700 per RigPrice’s 65-listing sample, and a new RTX 5080 at roughly $1,400 is paying an $1,100-plus premium, plus giving up a manufacturer warranty, for 8GB of extra VRAM and a 10-20% raw performance edge. Whether that trade makes sense depends entirely on what the card is doing eight hours a day.
Real-World Buying Scenarios in August 2026
The specs and benchmarks only mean something in context. Here’s how the RTX 4090 vs RTX 5080 decision plays out for five different types of buyers shopping right now.
- The 4K competitive gamer on a budget: A player upgrading from an RTX 3080 for 4K/120fps competitive titles gets better value from a new RTX 5080 at $1,400 with a full warranty than a used RTX 4090 at $2,200 with none, even accounting for the 4090’s raw speed edge.
- The freelance 3D artist rendering in Blender: Someone rendering architectural visualization scenes with 8K textures routinely hits VRAM ceilings on 16GB cards. For this buyer, the RTX 4090’s 24GB justifies the used-market premium outright, since a failed or CPU-fallback render costs more in time than the price difference.
- The local AI hobbyist running open-weight LLMs: Running 13B to 30B parameter models locally without heavy quantization needs VRAM headroom the RTX 5080 doesn’t have. This buyer is one of the few groups where paying $2,200+ for a used, out-of-warranty RTX 4090 is the economically rational choice over a new 5080.
- The video editor cutting standard 4K social content: Typical 4K timelines in Premiere Pro or DaVinci Resolve don’t come close to exhausting 16GB of VRAM. This buyer should take the RTX 5080’s lower price, lower power draw, and warranty coverage without hesitation.
- The small AI startup outfitting a handful of workstations: Buying five or six used RTX 4090s at $2,200 each means paying roughly $13,200 for cards with no warranty and unknown prior usage history (mining, rendering farms, or otherwise). Several startups in this exact position are instead opting for fewer, warrantied RTX 5080 or professional-tier cards and offloading heavier training to cloud GPU rental instead.
- The PC builder assembling a new rig from scratch in August 2026: Anyone building fresh has to budget for GDDR7-driven price hikes across RAM and SSDs too, not just the GPU. For a from-scratch build, the RTX 5080’s lower TBP also reduces the required PSU wattage, trimming a further $30–$50 off the total build cost.
Upgrade Path: Migrating From RTX 4090 to RTX 5080 (or the Reverse)
Moving between these two cards, in either direction, isn’t a simple swap. Follow this sequence to avoid the most common mistakes.
- Check your PSU wattage first. An RTX 4090 needs a PSU with comfortable 850W+ headroom given its 450W TBP; dropping to a 360W RTX 5080 means a 750W unit is typically sufficient, but don’t assume, check your specific CPU and case airflow too.
- Confirm your power connector. Both cards use the 12VHPWR/12V-2×6 connector standard, but always verify your PSU cable and adapter are rated for the card’s full draw before connecting.
- Uninstall old drivers with DDU. Use Display Driver Uninstaller in Safe Mode before installing a new card’s drivers, regardless of direction, to avoid conflicts between Ada Lovelace and Blackwell driver branches.
- Reassess your workload’s VRAM ceiling before selling. If you’re moving from a 4090 to a 5080 to cash in on used pricing, audit your actual VRAM usage in Task Manager or GPU-Z across a week of normal use first, so you’re not surprised by stutters in path-traced games or large AI models afterward.
- Time your resale around the eBay pricing data. Given the $2,200–$2,500 range multiple trackers report for August 2026, selling a used RTX 4090 now, while GDDR6X demand remains elevated, nets meaningfully more than waiting for the RTX 50 Super refresh to eventually arrive and potentially cool secondary-market demand.
- Rebuild your case airflow around the new TBP. A 90W drop in heat output changes fan curve requirements; retune your case fan profile after installing an RTX 5080 rather than leaving 4090-era settings in place.
- Reinstall DLSS/frame-gen profiles per game. Blackwell’s DLSS 4 multi-frame generation isn’t available on Ada Lovelace, so games you previously ran with older frame generation settings need their upscaling profiles reconfigured after moving to an RTX 5080.
RTX 4090 Pros and Cons in 2026
Pros: 24GB of VRAM handles path tracing, large AI models, and heavy creator workloads without compromise; 10-20% faster raw rasterization performance than the RTX 5080; wider 384-bit memory bus benefits bandwidth-heavy workloads; four years of driver maturity and game compatibility behind it.
Cons: Discontinued since October 2024, so every unit is used and out of manufacturer warranty; used pricing at $2,200–$2,500 now exceeds a new RTX 5080 by hundreds of dollars; 450W power draw demands a larger PSU and generates more heat; no access to Blackwell-generation features like DLSS 4 multi-frame generation or 5th-gen Tensor cores; unknown usage history on any given used unit (mining, 24/7 render farms, etc.).
RTX 5080 Pros and Cons in 2026
Pros: New, in-production, and covered by manufacturer warranty; 360W TBP lowers PSU and cooling requirements; newer Blackwell architecture with 4th-gen RT cores, 5th-gen Tensor cores, and DLSS 4 multi-frame generation; lower total cost than a used RTX 4090 even at inflated August 2026 street pricing.
Cons: 16GB VRAM ceiling limits path tracing, large local AI models, and heavy creator workloads; still selling 15-40% above its $999 MSRP due to the GDDR7 shortage; the planned 24GB Super refresh that would have fixed the VRAM gap has not shipped and isn’t expected before 2027; some titles still show driver-related underperformance relative to its hardware ceiling as of mid-2026.
5 Use-Case Recommendations
- Pure 4K/1440p gaming, no AI or creator work: Buy the RTX 5080 new. The warranty and lower power draw outweigh a 10-20% frame rate deficit for most gamers.
- Local LLM inference above 13B parameters: Buy the used RTX 4090. The 24GB VRAM ceiling is the deciding factor, and no amount of driver optimization fixes a hard memory limit.
- 3D rendering and VFX with large scene files: Buy the used RTX 4090, but budget for the lack of warranty by setting aside repair or replacement funds.
- Budget-constrained builds or first-time high-end upgrades: Buy the RTX 5080. It’s the only one of the two cards still available new at a price under $1,500.
- Multi-GPU AI workstations for a small team: Consider skipping both and evaluating cloud GPU rental for training workloads, since five or six used RTX 4090s at current pricing can exceed the cost of a cloud contract for equivalent compute.
Alternatives Worth Considering Before You Buy
Neither the RTX 4090 nor the RTX 5080 is the only option on the table in August 2026, and the same GDDR7 shortage warping their prices is warping the rest of the lineup too. The RTX 5070 Ti, Nvidia’s next tier down, launched at $749 and is streeting around $900, a smaller absolute jump that makes it worth a look for buyers who don’t strictly need flagship-tier performance. LocalAIMaster’s AI-workload comparison specifically frames the RTX 5070 Ti against the RTX 4090, arguing that for many local AI use cases the extra 8GB on the 4090 is the only thing that actually decides the trade, not raw compute.
On the high end, the RTX 5090 delivers a genuine generational leap in raw performance, but its street pricing of $3,000 to over $6,000 puts it out of reach for anyone not treating a GPU purchase as a business expense. If you’re comparing across brands rather than sticking with Nvidia, AMD’s RX 9070 XT is worth cross-shopping against the RTX 5080, since tech-insider.org’s own RTX 5080 vs RX 9070 XT comparison found a 25% ray tracing gap and a roughly $400 price divide between the two, numbers that shift the value calculation depending on whether ray tracing or price is your priority. And if none of the current options fit your budget or your patience, the honest alternative is simply waiting: Craftrigs’ shortage analysis puts real relief no earlier than Q4 2027, so “wait it out” is a legitimate strategy only if your current GPU still gets the job done in the meantime.
Verdict: Which Card Makes Sense in the 2026 GPU Market
For most buyers in August 2026, the RTX 5080 is the more rational purchase, and the data backs that up cleanly: it costs $800 to $1,100 less than a used RTX 4090, comes with a manufacturer warranty, draws 90W less power, and runs a newer feature set including DLSS 4 multi-frame generation. The RTX 4090 only wins the trade for buyers whose workloads are hard-capped by VRAM: local LLM inference above roughly 13B parameters, uncompressed path tracing at 4K and beyond, and large-scene 3D rendering where running out of memory means a failed job rather than a slower one.
The larger story here isn’t really about which card wins a benchmark chart, it’s about what the GDDR7 shortage has done to the entire high-end GPU market. Nvidia’s own fix for the RTX 5080’s VRAM ceiling, the 24GB Super variant, is stranded by the same memory shortage that’s inflating every card’s price, and no analyst tracked in this piece expects that to change before Q4 2027 at the earliest. Until then, the RTX 4090 vs RTX 5080 decision isn’t a normal generational upgrade choice, it’s a bet on which shortage-driven price you’re more comfortable paying.
If you take one number away from this comparison, make it this one: an $800 gap between a warrantied new card and a warranty-free used one is not how GPU generations are supposed to behave. Normally, the newer, cheaper card is the easy call. In August 2026, the RTX 5080 is still the right default for most buyers, but it’s the right default for reasons that have almost nothing to do with it being newer, and everything to do with a memory shortage that’s made the alternative unusually, and probably temporarily, expensive.
Frequently Asked Questions
Is the RTX 4090 still worth buying used in 2026?
Only if your workload is capped by VRAM. At $2,500-$2,700 used, per multiple late-2026 eBay-based trackers, including GetPCParts’ $2,541 average through August 29, 2026, ResalePrices’ $2,500 September 2026 market average, and RigPrice’s $2,700 median across 65 active listings, the RTX 4090 costs far more than a new RTX 5080 while carrying no warranty. It’s worth it for local LLM inference, heavy 3D rendering, and uncompressed path tracing where 24GB of VRAM prevents workflow failures. For general 4K gaming, the RTX 5080 is the better value.
Why is the RTX 5080 still above its $999 MSRP in 2026?
A GDDR7 memory shortage, driven by AI datacenter demand competing for the same production capacity, has pushed GDDR7 to more than 80% of a high-end GPU’s bill-of-materials cost, according to a July 2026 Phandroid report. Nvidia has also cut RTX 50 series supply to board partners by an estimated 20-40% during the first half of 2026, tightening availability further.
Did Nvidia release an RTX 5080 Super or Ti in 2026?
No. As of August 18, 2026, no RTX 5080 Super or Ti variant has shipped. Leaked plans described a 24GB RTX 5080 Super, but the higher-density 3GB GDDR7 modules it requires are the same modules in shortest supply, and industry reporting from DropReference describes the refresh as effectively stranded, with no clear launch window before 2027.
How much faster is the RTX 4090 than the RTX 5080 in gaming?
Based on combined data from TechPowerUp and Tom’s Hardware reviews, the RTX 4090 runs roughly 10% to 20% faster than the RTX 5080 in average 4K rasterized frame rates, and the gap widens to an estimated 15% to 25% in heavy ray-traced titles. Driver maturity on the RTX 5080 continues to close small pieces of that gap in specific titles as of mid-2026.
Which card uses less power, RTX 4090 or RTX 5080?
The RTX 5080 draws significantly less power. Guru3D’s Founders Edition testing measured a 360W total board power rating for the RTX 5080, compared to Nvidia’s official 450W rating for the RTX 4090, a 90W difference.
Is 16GB of VRAM enough for gaming in 2026?
For the vast majority of games at 4K with standard ray tracing settings, yes. The 16GB ceiling only becomes a real limitation in full path-traced titles with maximum texture settings, such as Cyberpunk 2077’s RT Overdrive mode, or in non-gaming workloads like large local AI models and heavy 3D rendering.
Should I wait for the RTX 50 Super series instead of buying now?
Only if you can wait a year or more. Craftrigs’ GDDR7 shortage analysis, last updated July 2026, doesn’t project meaningful memory supply normalization before Q4 2027, and even then there’s no guarantee Super-series cards launch at their original planned MSRP given how much GDDR7 costs have already climbed.
Can I still buy a new RTX 4090 in 2026?
Realistically, no. Nvidia’s manufacturing partners stopped RTX 4090 production around October 2024, based on supply-chain reporting from The FPS Review. Any card marketed as “new” in 2026 is most likely leftover retail inventory rather than freshly manufactured stock, and buyers should verify seller and box condition carefully before treating a listing as genuinely unused.
