RTX 4090 vs RTX 5080: 45% Price Premium Flip [2026]

The RTX 4090 was supposed to be old news by now. NVIDIA stopped building it in October 2024, more than a year before the Blackwell generation fully took over retail shelves. Yet in August 2026, a used RTX 4090 still sells for more than a brand-new RTX 5080, and that single fact is scrambling the usual “just buy the newer card” advice. Between GDDR7 shortages, a 45-55% markup over the RTX 5080’s $999 launch price, and a 24GB-vs-16GB VRAM gap that matters a lot more for local AI workloads than it did for gaming, this is not a simple generational upgrade decision. It is a genuine trade-off, and the right answer depends heavily on what you actually do with the card.

This comparison pulls current pricing, specs, and benchmark data from TechPowerUp, Tom’s Hardware, and independent GPU-review outlet WattWisePC to settle the RTX 4090 vs RTX 5080 question with real numbers rather than launch-day marketing slides.

It also matters who is asking the question. A gamer chasing the highest possible 4K frame rate in a ray-traced title, a hobbyist trying to run a large language model locally without renting cloud GPU time, and a small studio building a render farm on a fixed power budget will all land on different answers even after looking at the exact same benchmark charts. This guide breaks the decision down by workload, not just by spec sheet, so you can match the card to what you are actually going to do with it rather than to whichever number looks bigger in a headline.

·

Google · Preferred Sources

Don't miss new tech stories on Google

Add Tech Insider once in the Google app and our stories appear in your news suggestions.

Add Now

RTX 4090 vs RTX 5080: Full Specs Comparison

Before getting into benchmarks, it helps to see the raw silicon side by side. The RTX 4090 is built on NVIDIA’s Ada Lovelace architecture using the massive AD102 die, while the RTX 5080 uses the newer Blackwell architecture on the smaller GB203 die. That die-size difference explains almost everything else in this comparison, including why the RTX 4090 still wins on raw core count while the RTX 5080 wins on efficiency and newer feature support.

Specification	RTX 4090	RTX 5080
Architecture	Ada Lovelace (AD102)	Blackwell (GB203)
Process node	TSMC 4N (custom 5nm-class)	TSMC 4NP
CUDA cores	16,384	10,752
Base clock	2,235 MHz	2,295 MHz
Boost clock	2,520 MHz	2,617 MHz
Memory capacity	24GB GDDR6X	16GB GDDR7
Memory bus width	384-bit	256-bit
Memory bandwidth	1,008 GB/s	960 GB/s
RT core generation	3rd-gen	4th-gen
Tensor core generation	4th-gen	5th-gen
DLSS support	DLSS 3 (single-frame generation)	DLSS 4 (Multi Frame Generation, up to 4x)
TGP / TDP	450W	360W
Power connector	16-pin (450W-rated adapter)	16-pin (600W-rated adapter)
Launch date	October 12, 2022	January 30, 2025
Launch MSRP	$1,599	$999

The headline gap is CUDA core count: the RTX 4090 packs 16,384 cores against the RTX 5080’s 10,752, a 52% advantage on paper. But raw core count has never told the full story with NVIDIA silicon. Full specs for both cards are published on TechPowerUp’s RTX 4090 database entry and the RTX 5080 database entry, and NVIDIA’s own product page confirms the RTX 5080’s official specifications including the DLSS 4 feature set.

RTX 4090 vs RTX 5080 Pricing in August 2026

Pricing is where this comparison gets genuinely strange. Because the RTX 4090 was discontinued in October 2024, there is effectively no new-stock market for it left, only leftover inventory at inflated prices and a fragmented used market. Meanwhile the RTX 5080, still in active production, has seen its own street price climb well past its $999 launch MSRP thanks to the broader 2026 GDDR7 memory shortage that has also pushed up RAM and SSD prices across the PC hardware industry.

Price point	RTX 4090	RTX 5080
Launch MSRP	$1,599 (Oct 2022)	$999 (Jan 2025)
New retail, Aug 2026	Scarce; leftover units often $2,500+	~$1,450-$1,550 typical, deals from ~$1,200
Used market, Aug 2026	$2,200-$2,700 (eBay sold listings, multiple trackers)	~$1,200 (early resales, limited supply)
Markup vs MSRP	Roughly 40-70% above original $1,599	Roughly 45-55% above $999
Production status	Discontinued October 2024	In active production

Used RTX 4090 trackers converge on a similar range as of early-to-mid August 2026: getpcparts.com puts sold eBay listings at roughly $2,200, Clustermelt lists $2,299-$2,630, and RigPrice shows a current price near $2,700. On the RTX 5080 side, Tom’s Hardware and Videocardz market reports put median new pricing around $1,460-$1,500, roughly 46-50% above the card’s $999 launch MSRP, driven by the same GDDR7 supply crunch documented in Tom’s Hardware’s coverage of the RTX 50-series price spike. The practical upshot: a new RTX 5080 today typically costs $700-$1,200 less than a used RTX 4090 in comparable condition, which is the opposite of what most buyers expect from a “previous flagship vs current mid-flagship” matchup.

Architecture: Ada Lovelace vs Blackwell

The RTX 4090’s AD102 die is enormous by consumer GPU standards, at 76.3 billion transistors, and NVIDIA never released a bigger consumer chip on the Ada Lovelace architecture. The RTX 5080’s GB203 die is a smaller, more efficient part, positioned a tier below where the AD102 sat in the Ada lineup. That is the core reason the RTX 5080 was never meant to be a straight RTX 4090 successor in raw specs; NVIDIA’s actual RTX 4090 successor, the RTX 5090, uses the much larger GB202 die with far more cores and a $1,999 MSRP.

CUDA Cores and Compute Throughput

On paper, the RTX 4090’s 16,384 CUDA cores dwarf the RTX 5080’s 10,752, and in pure shader-bound compute workloads that gap shows up directly. But Blackwell’s architectural changes, including a redesigned streaming multiprocessor and 5th-generation Tensor Cores, close a meaningful chunk of that gap in real-world rendering, especially once ray tracing and AI-upscaling workloads enter the picture rather than pure rasterization.

Memory Bandwidth and the VRAM Trade-off

This is the most consequential spec difference for real buyers. The RTX 4090 has 24GB of GDDR6X across a 384-bit bus for 1,008 GB/s of bandwidth. The RTX 5080 has 16GB of newer, faster GDDR7 across a narrower 256-bit bus, landing at roughly 960 GB/s, close to the 4090’s number despite the smaller bus thanks to GDDR7’s higher per-pin transfer rate. For most current games, even at 4K with ray tracing, 16GB is enough. For local AI inference, large Blender or Octane scenes, and multi-layer video timelines in DaVinci Resolve, the RTX 4090’s 8GB capacity advantage can be the difference between a workload fitting in VRAM and one that spills over into slower system memory.

4K Gaming Benchmarks: Cyberpunk 2077, Alan Wake 2, and Black Myth: Wukong

Gaming is still where most buyers make the RTX 4090 vs RTX 5080 decision, and the results split depending on whether DLSS 4’s Multi Frame Generation is switched on. In native 4K rasterization, with no ray tracing and no upscaling, the RTX 4090’s extra CUDA cores keep it ahead. Testing referenced by GPU review site WattWisePC, drawing on TechPowerUp and Tom’s Hardware benchmark data, shows the RTX 5080 trailing the RTX 4090 by roughly 3-8% in native 4K raster performance depending on the title.

Game / Setting	RTX 5080 Result	vs RTX 4090	Source
Native 4K raster, no RT/DLSS (avg across titles)	Baseline	3-8% slower	TechPowerUp / Tom’s Hardware, via WattWisePC review
Cyberpunk 2077, Overdrive RT + DLSS Quality, 4K	~145 FPS avg	Within ±5%	Tom’s Hardware, via WattWisePC review
Alan Wake 2, 4K Ultra + DLSS Quality	~118 FPS avg	~8% faster with Multi Frame Gen active	TechPowerUp, via WattWisePC review
Heavy path-traced title, Multi Frame Gen 4x (TechPowerUp FE review)	120 FPS (from 35 FPS native)	N/A, frame-gen scaling test	TechPowerUp Founders Edition review

TechPowerUp’s own RTX 5080 Founders Edition review documents frame-generation scaling in detail: a demanding, heavily ray-traced scene runs at 35 FPS natively with frame generation off, climbing to 64 FPS at 2x mode, 91 FPS at 3x mode, and 120 FPS at the full 4x Multi Frame Generation setting, with roughly 13.6GB of VRAM in use. That last figure matters: it confirms the RTX 5080’s 16GB buffer has meaningful headroom left even in a demanding path-traced scenario, which undercuts the argument that 16GB is an automatic bottleneck for current-generation gaming. In VRAM-hungry titles like Black Myth: Wukong, both cards lean on DLSS upscaling to hold 4K frame rates above 60 FPS with ray tracing enabled, and the gap between them narrows to low single digits once upscaling is active on both sides.

Ray Tracing and Path Tracing Performance

Ray tracing is the one area where Blackwell’s architectural generation gap actually helps the RTX 5080 punch above its CUDA core count. The RTX 5080 ships with 4th-generation RT cores, one generation newer than the RTX 4090’s 3rd-generation units, and NVIDIA rebalanced the ray-triangle intersection and BVH traversal hardware between generations specifically to improve path tracing throughput per core.

The practical result, visible in the Cyberpunk 2077 Overdrive RT numbers above, is that the RTX 5080 closes almost all of its raster-mode deficit once heavy ray tracing is switched on, landing within about 5% of the RTX 4090 rather than the 3-8% gap seen in native rasterization. In fully path-traced scenes, where every light bounce is computed rather than approximated, that gap can effectively disappear, and the newer RT core design plus DLSS 4’s improved ray reconstruction model start to matter more than raw core count. For buyers who specifically want the best possible path-traced experience in titles like Cyberpunk 2077 or Alan Wake 2, the RTX 5080 is a legitimate answer to the RTX 4090 despite giving up a third of its CUDA cores.

DLSS 4 vs DLSS 3: What the Frame Generation Gap Means

This is the single biggest software-side differentiator in the RTX 4090 vs RTX 5080 comparison, and it is entirely a matter of hardware generation rather than a driver toggle. The RTX 4090 supports DLSS 3, which includes single-frame generation: for every real rendered frame, the card can insert one AI-generated interpolated frame. The RTX 5080 supports DLSS 4, which adds Multi Frame Generation, capable of inserting up to three AI-generated frames for every one real frame, a 4x total multiplier instead of 2x.

NVIDIA has not backported Multi Frame Generation to Ada Lovelace cards, so the RTX 4090 is locked out of DLSS 4’s headline feature regardless of driver updates. This is why the RTX 5080 can post visible-FPS numbers, like the 120 FPS 4x mode result in TechPowerUp’s testing, that a stock RTX 4090 simply cannot match on the frame-generation side, even though the underlying rendered frame rate is lower on the RTX 5080. The trade-off worth understanding: Multi Frame Generation adds input latency relative to native rendering, so competitive shooter players and anyone sensitive to input lag should treat the visible FPS counter with some skepticism and prioritize the “Framegen Off” baseline numbers instead, where the RTX 4090 still leads.

Power Draw and Efficiency

The RTX 4090’s 450W TGP against the RTX 5080’s 360W TGP is a 16GB memory buffer, with any power‑draw gap versus the RTX 4090 varying by specific model and comparison rather than being a universal 20% reduction in rated board power. At US average residential electricity rates, running a 450W card for a few hours a day adds up meaningfully over a year compared to a 360W card, especially for anyone doing sustained AI training or rendering workloads rather than a couple of gaming sessions per week.

Lower TGP also means the RTX 5080 is a friendlier fit for small-form-factor builds and systems with 750W or 850W power supplies, whereas NVIDIA’s own guidance and most partner-card documentation recommend at least 850W for a stable RTX 4090 build once CPU headroom is factored in. Both cards use a 16-pin power connector, but the RTX 5080’s bundled adapter is rated for 600W versus the RTX 4090’s 450W-rated adapter, giving 5080 builders more headroom if they later add extra fans, RGB, or storage draw to the same PSU rail.

Price-to-Performance: Dollars per Frame in August 2026

Once you combine the pricing and benchmark data above, the price-to-performance conversation flips almost entirely in the RTX 5080’s favor. At a typical $1,460-$1,500 new street price against a used RTX 4090 averaging $2,200-$2,700, the RTX 5080 costs roughly 35-45% less while delivering somewhere between a 3-8% deficit in native raster 4K and a slight lead in ray-traced, frame-generation-enabled scenarios.

That math would have looked very different a year ago, when RTX 5080 pricing sat closer to its $999 MSRP and a used RTX 4090 could still be found for $1,000-$1,200. The 2026 GDDR7 shortage has pushed 50-series pricing up across the board, but it has hit the RTX 4090’s scarce, discontinued supply even harder in percentage terms. For anyone shopping today rather than comparing launch-day MSRPs, the RTX 5080 is simply the better dollar-for-frame purchase in nearly every gaming scenario, and the gap only closes for buyers with VRAM-bound workloads where the RTX 4090’s 24GB actually changes what is possible rather than just how fast it runs.

AI, Machine Learning, and Content Creation Workloads

Gaming benchmarks undersell how different these two cards are for professional and semi-professional workloads. The RTX 4090’s 24GB of VRAM remains one of the most common single-GPU configurations recommended in local LLM and Stable Diffusion communities specifically because it can hold larger quantized models, bigger batch sizes, and higher-resolution training data without offloading to system RAM, which tanks throughput on both cards equally.

A 16GB card like the RTX 5080 handles most consumer AI inference workloads fine, quantized 7B-13B parameter models, most image generation pipelines, and everyday video editing, but it hits a hard wall on larger local LLMs (30B+ parameters at higher precision) and the biggest Stable Diffusion or Flux training runs where the RTX 4090 keeps working. In Blender Cycles, Octane, and Redshift rendering, VRAM capacity directly determines whether a complex scene renders on the GPU at all or falls back to a much slower CPU/GPU hybrid path, which is why professional 3D artists have been slower to jump from RTX 4090 to RTX 5080 than gamers have. The RTX 5080’s faster GDDR7 bandwidth and 5th-generation Tensor Cores do help with inference speed once a workload fits in memory, but capacity, not bandwidth, is usually the limiting factor for these use cases.

The calculus shifts again for anyone running multi-GPU workstations rather than a single card. Two RTX 5080s at 360W each draw less combined power than a single RTX 4090 plus a modest CPU, and 32GB of combined VRAM across two 5080s can match or beat a single RTX 4090’s 24GB for workloads that support model or tensor parallelism, such as larger LLM inference split across GPUs with frameworks like vLLM or text-generation-webui. That approach adds complexity, PCIe lane and motherboard requirements, and NVLink is no longer available on either of these consumer cards, so the two GPUs communicate over PCIe rather than a dedicated high-bandwidth bridge. For single-card simplicity, though, the RTX 4090’s 24GB remains the easier, lower-friction option.

Is the RTX 4090 Discontinued? Availability in 2026

Yes, and it has been for nearly two years. NVIDIA ended RTX 4090 production in October 2024, and by mid-2026, market trackers describe new retail stock as “almost not circulating,” with the few remaining new units priced well above the used market, an inversion of the usual new-costs-more-but-you-get-a-warranty logic. Practically every RTX 4090 changing hands in August 2026 is a used card, which means buyers need to factor in warranty risk, unknown mining or heavy AI-training history, and cooler wear that a brand-new RTX 5080 simply does not carry.

This scarcity is also why used RTX 4090 pricing has been so inconsistent across trackers, ranging from roughly $700-$1,050 on some peer-to-peer marketplace snapshots up to $2,200-$2,700 on eBay sold-listing aggregators in the same general timeframe. Anyone shopping the used RTX 4090 market should treat listings well below $2,000 with extra scrutiny, since heavily used mining or AI-training cards, and outright scams, are more common at the low end of that range.

Buyers considering a used RTX 4090 should also budget for verification steps that a new RTX 5080 purchase skips entirely: running an extended stress test with a tool like FurMark or 3DMark’s stress test mode to check for thermal throttling or instability, inspecting the cooler for bent fins or dried thermal paste that suggests high mileage, and asking the seller directly about mining or 24/7 AI-training use, since sustained high-temperature operation shortens the lifespan of VRAM modules and power delivery components faster than typical gaming use does. None of that risk exists with a new RTX 5080, which is one more reason its lower purchase price understates the real gap in total cost of ownership between the two cards right now.

RTX 5080 Super: How the Rumored Variant Changes the Math

Leaked specifications for an RTX 5080 Super variant point to 24GB of GDDR7 (up from 16GB), a 415W TGP (up from 360W), and the same 32 Gbps memory speed for close to 1 TB/s of bandwidth. Reports through mid-2026 indicate the Super variant’s release has slipped, with current leaks pointing to a 2027 launch window rather than the earlier-rumored late-2026 timing. That timeline is not yet confirmed by NVIDIA, so treat it as a leak rather than a purchasing plan.

If the RTX 5080 Super does land with 24GB, it would directly erase the RTX 4090’s VRAM advantage while keeping the newer Blackwell architecture and DLSS 4 support, which would make the current RTX 4090 vs RTX 5080 trade-off obsolete for anyone willing to wait. For buyers who specifically need 24GB today, whether for local AI work or professional rendering, that is a real reason to consider a used RTX 4090 as a bridge purchase rather than paying a premium for a 16GB RTX 5080 that a Super variant could outclass on the one spec that matters most to them.

Encoding, Streaming, and Display Outputs

Streamers and video editors care about a different set of specs than gamers, and here the two cards are closer than the CUDA core count suggests. Both the RTX 4090 and RTX 5080 include NVIDIA’s 8th-generation NVENC encoder with AV1 support, so live-streaming quality and export speed in OBS, DaVinci Resolve, or Premiere Pro are effectively identical between the two cards for most workflows. Where they diverge is dual-encoder throughput: the RTX 4090 ships with two NVENC encoder chips, letting it run two simultaneous 4K encode streams (useful for multi-platform streaming or fast dual-format exports) at full speed. The RTX 5080 also carries dual encoders, but pairs them with lower total board power, which matters for anyone running a streaming and gaming rig from a single compact PSU.

Display output configurations are identical on both cards’ Founders Edition models: three DisplayPort 2.1 outputs and one HDMI 2.1b port, both capable of driving 4K at high refresh rates or multiple monitors from a single card. Neither card has an advantage here; the differentiator for multi-monitor and high-refresh setups comes down to which resolution and refresh combination you are actually driving, and whether DLSS 4’s frame generation, exclusive to the RTX 5080, helps you hit that target without dropping settings.

Overclocking and Thermal Headroom

Overclocking headroom favors the RTX 5080 in relative terms, even though the RTX 4090 has more absolute power budget to work with. Because the RTX 5080 ships at a 360W TGP with a 600W-rated power adapter included, many partner cards leave 15-20% of thermal and power headroom on the table at stock settings, giving overclockers room to push core clocks and memory speeds up without immediately hitting the power limit. The RTX 4090, already running close to its 450W ceiling at stock in most factory-overclocked partner cards, has comparatively less room before hitting thermal throttling, though its larger cooler designs (many RTX 4090 partner cards ship with triple-slot, triple-fan coolers) do help it sustain boost clocks longer under extended workloads.

In practice, most reviewers find the RTX 5080 gains roughly 5-8% additional performance from a manual power-limit increase and memory overclock, while the RTX 4090’s factory-tuned partner cards typically have only 2-4% left on the table since AIB partners already push close to the silicon’s practical ceiling out of the box. Neither card requires exotic cooling for a mild overclock, but sustained overclocked 4090 operation does push case temperatures and fan noise noticeably higher than a similarly overclocked 5080, a real consideration for anyone building in a small or poorly ventilated case.

5 Real-World Use Cases: Who Should Buy Which Card

Specs and benchmarks only matter in the context of what you are actually going to do with the card. Here is how the trade-off plays out across five common buyer profiles.

• 4K/120Hz single-player gamer with ray tracing as a priority: the RTX 5080 wins here. DLSS 4’s Multi Frame Generation and the newer RT cores put it within a few percent of the RTX 4090 in Cyberpunk 2077-style ray-traced titles, at a lower price and lower power draw.
• Freelance 3D artist rendering in Blender Cycles or Octane: the RTX 4090’s 24GB VRAM avoids out-of-memory failures on complex scenes that a 16GB RTX 5080 can choke on, making it worth the used-market premium for this specific workload.
• Local LLM hobbyist running 30B+ parameter models via Ollama or vLLM: the RTX 4090’s extra 8GB of VRAM is often the difference between a model fitting entirely on-GPU versus spilling into much slower CPU offload, so the 4090 remains the better single-GPU pick here.
• Small studio building a render farm or multi-GPU AI training rig: the RTX 5080’s 360W TGP versus the RTX 4090’s 450W lets more cards run per circuit and per PSU, which can matter more than per-card VRAM once you’re scaling horizontally.
• Competitive esports player on 1440p or lower, prioritizing frame rate over visuals: neither card is really necessary at this resolution, and the money is better spent on a high-refresh monitor; if forced to choose, the RTX 5080’s lower price and power draw make it the more sensible pick.
• Livestreamer running simultaneous 4K capture and dual-platform export: both cards handle this well thanks to dual NVENC AV1 encoders, but the RTX 5080’s lower power draw leaves more thermal and PSU headroom for a combined gaming-and-streaming rig running from one machine.

Migration Guide: Upgrading From RTX 4090 to RTX 5080

For current RTX 4090 owners considering a lateral move to a new RTX 5080, mainly to shed power draw and gain DLSS 4, the process is straightforward but worth doing in the right order to avoid driver conflicts and instability.

1. Back up any custom driver profiles, DLSS overrides, and undervolt/overclock settings you rely on, since these do not carry over between architectures.
2. Fully uninstall the existing NVIDIA driver using Display Driver Uninstaller (DDU) in Safe Mode to remove Ada-specific driver components before installing Blackwell drivers.
3. Power down, unplug the PSU, and physically swap the card, noting the RTX 5080’s shorter length and lower cooling demands may allow different case routing than the RTX 4090 required.
4. Connect the bundled 16-pin adapter to your PSU’s 8-pin PCIe rails, checking the RTX 5080 only needs a 600W-rated adapter path versus the RTX 4090’s higher-draw configuration.
5. Install the latest Game Ready or Studio driver directly from NVIDIA, then reboot before launching any games or creative applications.
6. Enable Resizable BAR in your motherboard BIOS if it was not already active, since both cards benefit but Blackwell’s driver stack leans on it more heavily for DLSS 4 performance.
7. Update in-game DLSS profiles to DLSS 4 presets where available, and manually enable Multi Frame Generation in supported titles, since it is off by default in most games.
8. Re-run your usual benchmark suite (3DMark Time Spy Extreme or Port Royal, plus one or two real games you play) to confirm stability before selling the old card.

A quick way to confirm the new card and driver are both recognized correctly after the swap is to check the output of NVIDIA’s system management interface:

nvidia-smi --query-gpu=name,driver_version,memory.total,power.limit --format=csv

That command should return “NVIDIA GeForce RTX 5080,” the installed driver version, 16384 MiB of memory, and a power limit near 360W, confirming the swap completed cleanly before you sell or repurpose the old RTX 4090.

RTX 4090 Pros and Cons

• Pro: 24GB of VRAM handles large local LLMs, big Blender/Octane scenes, and heavy Stable Diffusion training that a 16GB card cannot.
• Pro: Still leads in native 4K raster performance without upscaling, by roughly 3-8% depending on title.
• Pro: Mature, well-documented driver stack with years of game-specific optimization behind it.
• Con: Discontinued since October 2024; every purchase today is a used, unwarrantied card with unknown history.
• Con: 450W TGP demands a beefier PSU and generates more heat and noise under sustained load.
• Con: Locked out of DLSS 4 Multi Frame Generation permanently, regardless of driver updates.
• Con: Used pricing ($2,200-$2,700) currently exceeds a brand-new RTX 5080’s street price.

RTX 5080 Pros and Cons

• Pro: Brand-new with full warranty coverage, unlike the used-only RTX 4090 market.
• Pro: DLSS 4 Multi Frame Generation delivers substantially higher visible frame rates in supported titles.
• Pro: 360W TGP is 20% lower than the RTX 4090, easing PSU and cooling requirements.
• Pro: Currently costs $700-$1,200 less than a used RTX 4090 in equivalent condition.
• Con: 16GB VRAM is a hard ceiling for the largest local AI models and professional rendering scenes.
• Con: Trails the RTX 4090 by 3-8% in native raster 4K performance without DLSS.
• Con: Street price has climbed 45-55% above its $999 MSRP amid the 2026 GDDR7 shortage.

Resale Value and Long-Term Investment

The RTX 4090’s resale story is unusual enough to be worth its own section. Most GPUs depreciate steadily after launch, but the RTX 4090 has effectively appreciated in inflation-adjusted terms over parts of 2025 and 2026, driven by its discontinued status, its continued relevance for AI workloads, and broader GDDR-market supply constraints. A card bought used for $1,000-$1,200 in early 2025 could realistically resell for $2,000 or more in August 2026, an outcome almost no other consumer GPU in recent memory has matched. That makes the RTX 4090 one of the rare hardware purchases that has functioned more like a scarce collectible than a depreciating asset, though buyers should not count on that trend continuing indefinitely once Blackwell Super variants or the next Rubin-based generation arrive and further reduce demand for Ada Lovelace silicon.

The RTX 5080’s long-term value picture is more conventional. As an actively produced card, its price should gradually normalize toward MSRP as the 2026 GDDR7 shortage eases, meaning early buyers at today’s $1,450-$1,550 street prices may see the card’s value dip before any scarcity premium develops, if one ever does. For buyers thinking purely in resale terms rather than immediate use, that makes the RTX 4090 the more speculative but potentially higher-upside hold, while the RTX 5080 behaves like a normal depreciating component you buy to use, not to flip.

The Verdict: RTX 4090 vs RTX 5080

For the overwhelming majority of buyers in August 2026, the RTX 5080 is the better purchase, and the pricing data makes the case more than the benchmarks do. A new RTX 5080 at roughly $1,450-$1,550 with a full warranty beats a used, unwarrantied RTX 4090 at $2,200-$2,700 on nearly every practical measure: it costs less, draws 20% less power, and gains DLSS 4’s Multi Frame Generation, closing most of its raw-performance deficit in ray-traced, upscaled scenarios where most 4K gaming actually happens now.

The exception is narrow but real: if your workload is VRAM-bound rather than compute-bound, running large local LLMs, training Stable Diffusion or Flux models, or rendering complex scenes in Blender Cycles or Octane, the RTX 4090’s 24GB is not just faster, it is the difference between the job running at all or failing outright. For that specific group, paying the used-market premium for a RTX 4090 remains defensible, at least until the rumored 24GB RTX 5080 Super materializes, which current leaks put no earlier than 2027. Everyone else should buy the RTX 5080.

Frequently Asked Questions

Is the RTX 4090 still worth buying in 2026?

Only for specific VRAM-bound workloads like large local LLMs or professional 3D rendering. For gaming, a new RTX 5080 currently costs less than a used RTX 4090 while offering a full warranty and DLSS 4 support.

Is the RTX 5080 faster than the RTX 4090?

It depends on the setting. In native 4K rasterization without upscaling, the RTX 4090 leads by roughly 3-8%. With ray tracing and DLSS 4’s Multi Frame Generation enabled, the RTX 5080 can match or slightly exceed the RTX 4090’s visible frame rate, per TechPowerUp and Tom’s Hardware benchmark data.

How much VRAM do I actually need in 2026?

16GB, as found on the RTX 5080, covers the vast majority of current games at 4K, even with ray tracing enabled. Only large local AI models, big professional rendering scenes, and heavy AI training workloads consistently need more than 16GB, which is where the RTX 4090’s 24GB becomes relevant.

Is the RTX 4090 discontinued?

Yes. NVIDIA ended RTX 4090 production in October 2024. As of August 2026, new retail stock is nearly gone, and the used market, priced $2,200-$2,700 by most trackers, is effectively the only way to buy one.

Should I wait for the RTX 5080 Super?

If you specifically need more than 16GB of VRAM, it may be worth waiting. Leaked specs point to 24GB of GDDR7 and a 415W TGP, but current reports put the launch no earlier than 2027, and NVIDIA has not confirmed the variant or its timing.

Does the RTX 4090 support DLSS 4?

No. DLSS 4’s Multi Frame Generation is exclusive to Blackwell-generation cards like the RTX 5080. The RTX 4090 remains on DLSS 3, which supports single-frame generation but not the 4x Multi Frame Generation mode.

Which card is better for AI and local LLMs?

The RTX 4090, primarily because of its 24GB VRAM capacity. Larger quantized models and bigger batch sizes fit entirely on-GPU on the RTX 4090 where a 16GB RTX 5080 may need to offload to slower system memory.

What power supply do I need for each card?

NVIDIA and partner documentation generally recommend at least 850W for a stable RTX 4090 build accounting for CPU headroom, versus roughly 750W for a comparable RTX 5080 build given its lower 360W TGP.

Related Coverage

• RTX 5080 vs RX 9070 XT: 25% RT Gap, $400 Price Divide [2026]
• RTX 5070 Ti vs RX 9070 XT: 15% RT Gap, $250 Price Split [2026]
• How to Install RAM: 12 Steps, 30 Min [2026]
• How to Overclock a CPU: 12 Steps, 90 Min [2026]
• Ryzen 9800X3D vs 9900X3D: 1-2 FPS Gap, $120 Price Split [2026]
• Nvidia Vera CPU Takes On Intel, AMD With $20B Bet [2026]