Now that the CES hype has faded, it’s time to look at the real numbers for the 50 series GPUs before deciding whether to pick one up. Is the performance improvement worth the cost? Does the card support the features I need? Are there better alternatives? Should I wait for the next generation? These questions should dominate your quest for a GPU upgrade.
Finding your answers involves gathering information about the Blackwell lineup and reviewing a great deal of comparisons against your current GPU (whichever brand that may be). I’ll provide these details, explain scenarios where upgrading or sticking to the current card makes sense, and provide insights on alternative options.
Overview of the RTX 50 Series
To understand NVIDIA’s 50 series product line, you need the basics about Blackwell, the architecture it’s built on. Blackwell is the successor to Ada Lovelace, the platform for the RTX 40 series. NVIDIA claims the new architecture is built on a special TSMC process node because it fuses two chips to make a larger die (GB202) for the RTX 5090. Apart from that, the rest of the lineup is built on the same 4nm node as 40 series chips.
The prominent upgrades in Blackwell are achieved through AI, so much so that NVIDIA called it the GeForce Blackwell Neural Rendering Architecture.
One of these headline upgrades is gamers’ ability to use neural shaders, which was impossible until now. That means geometric properties that adjust detail in 3D models can now be processed by the GPU’s tensor cores. This ability saves resources and helps developers implement better graphics and gaming details.
While AI is a prominent part of every NVIDIA RTX 50 series card, Blackwell also improves a lot of non-AI stuff. Per NVIDIA, the architecture focuses on memory footprint reduction, energy efficiency, and quality of service. These are the main highlights outside the platform’s AI capabilities:
- 4th Gen RT (ray tracing) cores that offer double the ray triangle intersection rate over the last generation. They also utilize 75% less memory resources.
Ray triangle intersection rates refer to how fast RT cores can determine whether a ray of light intersects with a triangle in a 3D environment. Developers and designers use millions or thousands of small triangles to build most 3D models.
- About 30% more CUDA cores, thanks to the new TSMC 4NP (4nm NVIDIA performance) process, which allows more cores to be integrated onto the 5090 die.
- DisplayPort 2.1 UHBR20 support, which supports 4K resolutions at 240Hz for people with the right monitors.
- GDDR7 memory, which is about 2x faster and more efficient than GDDR6X.
Here’s a quick look at how the current architecture compares to previous generations.
| Features |
Blackwell |
Ada Lovelace |
Ampere |
| RTX Series |
GeForce RTX 50 |
GeForce RTX 40 |
GeForce RTX 30 |
| Process node |
4nm |
4nm |
8nm |
| Memory generation |
GDDR7 |
GDDR6X/GDDR6 |
GDDR6X/GDDR6 |
| CUDA capability |
12.8 |
8.9 |
8.6 |
| DLSS |
DLSS 4 |
DLSS 3 |
DLSs 2 |
| Tensor cores (AI) |
Gen 5 |
Gen 4 |
Gen 3 |
| Ray tracing cores |
Gen 4 |
Gen 3 |
Gen 2 |
| PCIe |
Gen 5 |
Gen 4 |
Gen 4 |
| NVENC (NVIDIA Encoder) |
Gen 9 |
Gen 8 |
Gen 7 |
| NVDEC (NVIDIA Decoder) |
Gen 6 |
Gen 5 |
Gen 5 |
| Display support |
DisplayPort 2.1b with UHBR20
4K at 480Hz or 8K at 165Hz with DSC
|
DisplayPort 1.4a
4K at 240Hz or 8K at 60Hz with DSC, HDR
|
DisplayPort 1.4a
4K at 240Hz or 8K at 60Hz
|
NVIDIA GeForce RTX 50 Series GPU Specifications
Each RTX series features different GPU models with varying specifications. For example, rumors indicate that the RTX 5050 will feature GDDR6 memory rather than GDDR7. The 5090 is the only 50 series graphics card to sport a 512 memory bus, and GPUs in the lineup will have different CUDA, tensor, and ray tracing core counts. Memory sizes will also vary across models.
| Specifications |
5090 |
4090 |
5080 |
4080 Super |
5070 Ti |
4070 Ti Super |
5070 |
4070 Super |
| VRAM |
32GB |
24GB |
16GB |
16GB |
16GB |
16GB |
16GB |
12GB |
| Memory bus width |
512-bit |
384-bit |
256-bit |
256-bit |
256-bit |
256-bit |
192-bit |
|
| CUDA cores |
21760 |
16384 |
10752 |
10240 |
8960 |
8448 |
6144 |
7168 |
| Tensor cores |
680 units (5th Gen)
3352 AI TOPS |
512 units (4th Gen)
1321 AI TOPS |
336 units (5th Gen)
1801 AI TOPS |
320 units (4th Gen)
836 AI TOPS |
280 units (5th Gen)
1406 AI TOPS |
264 units (4th Gen)
706 AI TOPS |
192 units (5th Gen)
988 AI TOPS |
224 (4th Gen)
568 AI TOPS |
| Ray tracing cores |
170 (4th Gen)
318 TFLOPS
|
128 (3rd Gen)
191 TFLOPS
|
84 (4th Gen)
171 TFLOPS
|
80 (3rd Gen)
121 TFLOPS
|
70 (4th Gen)
133 TFLOPS
|
66 (3rd Gen)
102 TFLOPS
|
48 (4th Gen)
94 TFLOPS |
56 (3rd Gen)
67 TFLOPS
|
| Base/boost clock |
2.01/2.41 |
2.23/2.52 |
2.30/2.62 |
2.29/2.55 |
2.30/2.45 |
2.34/2.61 |
2.33/2.51 |
1.98/2.48 |
| Total graphics power (watts) |
575 |
450 |
360 |
320 |
300 |
285 |
250 |
220 |
| Required system power (watts) |
1000 |
850 |
850 |
750 |
750 |
700 |
650 |
650 |
| Maximum GPU temperature in Celsius (C) |
90 |
90 |
88 |
90 |
88 |
90 |
85 |
90 |
Let’s explain the key specifications and why they matter.
These are the total CUDA, tensor, and ray tracing cores available on each GPU. CUDA cores are processing units designed for parallel computing. They execute instructions from NVIDIA’s CUDA software package. The more cores your GPU has, the more compute operations it can handle.
Tensor cores are specialized processors used for AI acceleration and deep learning tasks. They determine how much AI workload your GPU can process, whether it’s for in-game Frame Generation or running local large language models. They are measured in TOPS (trillion operations per second).
Ray tracing cores are also measured in TFLOPS (tera (one trillion) floating-point operations per second) and are used to simulate how light behaves in the real world.
Also read: CPU Cores Explained: Does the Number of Cores Affect CPU Performance
Clock speeds measure (in gigahertz) how fast your GPU cores operate. While faster clock speeds indicate faster GPUs, other specs like core counts and memory speeds also play huge roles.
You’ll notice that the RTX 5090 has a slightly lower boost clock than the RTX 4090 (2.41 GHz vs 2.52 GHz). That’s because it has more CUDA cores. You’ll find the same system implemented across other generations. For example, the 4070 Ti has higher base and boost clocks than the 4090.
Memory size and bus width
Your GPU’s VRAM is a specialized type of RAM that stores graphics data (shaders, textures, and other render instructions) to speed up rendering tasks. It is baked onto your GPU instead of your motherboard and determines how fast and well your games run. VRAM also influences your graphics card’s performance when it runs other tasks like video editing and AI acceleration.
AI and machine learning, 4K gameplay, 3D modeling, and data mining are examples of tasks that require high amounts of VRAM.
Memory bus width, on the other hand, is the size of the physical lane where data travels from the RAM to the GPU. It is measured by how many bits can travel through a lane at any given time. For example, the RTX 5090, with its 512-bit interface, features the highest bus width on any card. The higher the bus width, the more data moves between your GPU and memory at any given time.
That said, the speed at which data moves from the memory to the GPU is determined by your video card’s memory generation. GDDR7 is currently the fastest generation and is deployed across Blackwell GPUs.
Also read: DDR4 vs. DDR5: Is DDR5 Worth It? Here’s What You Need to Know
DLSS 4 & Multi Frame Generation
DLSS 4 is one of NVIDIA’s biggest upgrades to its AI image enhancement suite. The package uses a new transformer model that offers reduced artifacts and ghosting, smoother edges, better scaling, and improved in-motion detail. It represents a serious step up over the CNN (convolutional neural network) model.
Multi Frame Generation (MFG) is another massive upgrade over its predecessor (Frame Generation). The previous offering used AI (assisted by hardware acceleration) to generate one additional frame for every native frame. However, this new version can produce up to three AI-generated frames (without hardware acceleration) instead of one per native frame. According to benchmarks, the performance difference is up to 4x better.
While NVIDIA claims MFG will reduce latency, don’t expect it to eliminate every Frame Gen issue, such as artifacts and image distortion (which are not present when using only DLSS), from past years. While your FPS numbers will dramatically increase, brace yourself for high amounts of visual anomalies.
NVIDIA revealed that DLSS 4 is now available in 100 games (including Cyberpunk, God of War: Ragnarok, Stalker 2, and Alan Wake II). Also, DirectX will start supporting neural shading in April 2025, thanks to the company’s collaboration with Microsoft. These updates are billed to actualize the promise of DLSS 4 and Multi Frame Gen. We’ll see how it all works out.
Ray tracing is the tech responsible for simulating real-world light behavior. It has always been an option that can be turned on or off. However, it is becoming less of a choice, as games like Indiana Jones now add it to their minimum system requirements.
As already mentioned, NVIDIA brought significant upgrades to ray tracing in every 50 series graphics card, thanks to the 4th generation RT cores that detect ray triangle intersections faster and use less memory footprint. But do those mean a giant leap in real-world performance? We’ll cover that later.
Thermal Design Power, or Total Graphics Power, indicates how much heat a GPU can generate. It is measured in watts and helps users determine how much power and cooling is required to run the GPU optimally.
For example, the GeForce RTX 5090 has a TGP of 575 watts, more than 25% above the 4090. That means you need the PSU and power cables to match that wattage. You also need a cooling unit that can keep the GPU from overheating.
The PCIe standard enables point-to-point interactions between components connected to a motherboard. So, it determines how fast your GPU can speak to your CPU, SSDs, and other internal hardware. PCIe 5.0 doubles the data transfer rate over the 4.0 generation to 32GT/s (giga transfers per second) from 16 GT/s.
Download Auslogics BoostSpeed
Your one-stop PC maintenance and optimization tool, this program will clean, speed up, repair and tweak your system to ensure maximum speed and efficiency.
No one believed Jensen Huang’s claim that the RTX 5070 will produce RTX 4090 performance, but it made some of us hopeful. However, review benchmarks have shown us a different (and disappointing) reality.
Generally, you should measure your GPU’s performance by its price. If a card is going to cost a lot more than what you intend to upgrade from, its performance improvement should reflect that cost.
Also read: Start Gaming Right: Expert Picks for the Best Affordable Gaming PCs for Beginners
Native gaming performance numbers across the NVIDIA RTX 50 series models are pretty underwhelming when you factor in price and compare like-for-like upgrades.
Let’s cover real-world benchmarks to paint a clearer picture.
Reviewers post different results for the same game at the same settings because computer configurations (think CPU, RAM, storage, and even motherboard specs) vary.
RTX 5090 gaming benchmarks
The 5090 might be the fastest GPU on the planet, but it only beats the 4090 by negligible margins at native gameplay. At 1440p (ultra settings), the new card edges out the RTX 4090 by 16.8% on average across games tested by Linus Tech Tips.
Benchmark credit: Linus Tech Tips
The 5090 does better in some games versus others. Average FPS metrics from Hardware Unboxed gave it a 19% lead in Cyberpunk 2077: Phantom Liberty and a 24% uplift in Dying Light 2 Stay Human over the 4090.
Benchmark credit: Hardware Unboxed
Real differences start to show when you go to 4K and max out the settings in games that favor the new card. Tom’s Hardware’s average difference between the 5090 and 4090 is 22.3% across 16 games.
The 5090 also outpaces the RTX 3090 Ti by about 68%.
Benchmark credit: Linus Tech Tips
The difference also hovers around the 20 to 28% mark when you turn on ray tracing. The RTX 5090 commands a 26% lead in Cyberpunk 2077 at 4K (ultra settings) and a 27% edge in Resident Evil 4 at the same resolution.
Benchmark credit: PC World
Benchmark credit: Gamers Nexus
DLSS 4 and Multi Frame Gen performance
The 5090 flexes its AI muscles when you turn on DLSS 4. Numbers can double or even quadruple when you use MFG in games that support it.
Benchmark credit: Hardware Unboxed
RTX 5080 benchmarks
The 5080 edges out the RTX 4080 Super (basically the same as the RTX 4080) by only 6 FPS (12%) in Black Myth Wukong at 4K (high settings), according to tests from Gamers Nexus.
Benchmark credit: Gamers Nexus
Linus Tech Tips puts the overall average uplift of the 5080 over its predecessor at about 15%. Their geomean benchmark across all games at 4K showed the 5080 leading the 4080 Super by an average of 18%.
Benchmark credit: Linus Tech Tips
There are also instances where the card shines over the 4080 Super by reasonable margins. These instances include its 20% lead in Cyberpunk 2077 at 1440p and 28% at 4K.
Benchmark credit: PC World
Benchmark credit: PC World
But that’s about where the performance boosts seem arguably impressive for gaming. The numbers never quite go higher across other games at different settings. It delivers poor gains (11%) over the 4080 Super with ray tracing turned on at 1440p in Metro Exodus. The results are similar in Alan Wake II (basically the same FPS output), according to tests from TechSpot.
Benchmark credit: TechSpot
The story is the same with Multi Frame Gen on the 5080 as it is with the 5090. You can double or quadruple your generated frames with the feature turned on.
Benchmark credit: Linus Tech Tips
RTX 5070 Ti benchmarks
The RTX 5070 Ti is considered the successor to the 4070 Ti Super (they feature the same amount of VRAM and a slight difference in CUDA cores). But it follows the same disappointing trend as the higher cards in its generation by not doing very well against the GPU it replaces. The newer video card only pulls ahead in most benchmarks with tiny gains that do not justify an upgrade.
Benchmark credit: Tom’s hardware
However, the story is different if you currently use a 30 series card, as the 5070 Ti is quite better than even the RTX 3090 and 3080, according to Hardware Unboxed’s 1440p 16-game average benchmark.
Benchmark credits: Hardware Unboxed
That said, if you don’t notice the differences between AI-boosted FPS numbers and native frames, you could be impressed by the 5070 Ti.
RTX 5070 benchmarks
The card that raised eyebrows at CES 2025 did not quite hit the home run as promised. The RTX 4090 is about 63% faster than the RTX 5070 on average at native rendering. But the 5070 climbs ahead when you dope it up with NVIDIA’s cocktail of new AI features.
Benchmark credit: TechRadar
If you think the story will change when compared against its predecessor, it does not. It posts middling performance gains over the 4070 Super and sports only 12GB of VRAM.
Benchmark credit: Hardware Unboxed
You may see things favorably if you compare the 5070 Ti directly with the RTX 4070. But the 20 to 25% performance uplift still makes it a dull upgrade.
Productivity and AI benchmarks
Things start to look up for the 50 series GPUs when you view benchmarks for content creation and AI. It shouldn’t come as a surprise since these cards have more tensor cores and higher AI TOPS against their predecessors.
According to NVIDIA, its new Blackwell architecture supports 4:2:2 encoding and decoding. This upgrade makes it easier to work with high-end video files, especially when editing multi-camera footage. Apart from speeding up the encoding process, these encoders also offer higher quality at reduced file sizes.
The RTX 5090 sees a 15.4% jump over the RTX 4090 in the PugetBench test for DaVinci Resolve and about 27.6% in the Blender rendering benchmark.
Benchmark credit: Puget Systems
As for AI, the trend remains consistent, with the 5090 beating the 4090, its closest competitor, by 20 to 35% across different benchmarks.
Benchmark credit: PC Gamer
The RTX 5080 falls on its face again when you compare it against the RTX 4080 Super but does a little better than its direct predecessor. That’s according to scores from Puget Systems.
Benchmark credit: Puget Systems
However, we’ve seen some benchmarks where the RTX 5080 does better than even the 4090 in some productivity tests.
Benchmark credit: Linus Tech Tips
In AI, there’s a 20% gap over the RTX 4080 in text generation and about a 16% improvement in stable diffusion image generation benchmarks.
Benchmark credit: Linus Tech Tips
The 5070 Ti provides a so-so % performance uplift (about 5%) over the RTX 4070 Ti Super in Premiere Pro. But it does much better against its direct older sibling, the RTX 4070 Ti (9%).
Benchmark credit: Linus Tech Tips
Its AI benchmark results are better, as it scores 23% higher than the 4070 Ti Super in text generation and 28% in image generation per tests from Linus Tech Tips.
Benchmark credit: Linus Tech Tips
As you probably surmised, the RTX 5070 also showcases Blackwell’s AI upgrades by beating both the 4070 Super and 4070 Ti in AI benchmarks.
However, according to tests from TechRadar, the 5070 only surpasses the 4070 Super by an average of 3% in productivity tests.
Benchmark credit: TechRadar
Download Auslogics BoostSpeed
Your one-stop PC maintenance and optimization tool, this program will clean, speed up, repair and tweak your system to ensure maximum speed and efficiency.
Best NVIDIA RTX 50 Series for Different Use Cases
Gaming
RTX 50 series cards are primarily gaming GPUs, even if they are quite competent for other use cases. Your choice of GPU model should rely on the type of game you want to play and your preferred graphics settings.
I should point out that you can play at any setting (even 4K) with any card from the 5070 and up if you turn on DLSS 4 and Multi Frame Gen. It all depends on how well you notice artifacts and other image distortions.
But for this section, we’ll stick to which cards perform best at native resolution without any upscaling.
- 4K at high settings: RTX 5090 and RTX 5080
People with the budget to purchase the 5090 and 5080 want to play at the highest resolutions. They have the monitors and CPUs to match 4K at high settings and even throw in ray tracing.
If you want the highest frames at any setting, even if you use DLSS 4, the RTX 5090 is the best choice. However, you can go for the less powerful 5080 if your budget precludes you from purchasing the flagship card.
Still, the 5070 Ti, with its 16GB of VRAM (same as the 5080), can take a crack at 4K.
- 1440p and ray tracing: RTX 5070 and higher
The RTX 5080 is excellent for 1440p gaming, and the 5070 Ti is more than adequate. While these cards may struggle to push out 60 FPS at 4K, they’re quite impressive at 1440p (when you’re not comparing them with last-gen video cards).
If your budget is on the low side, you can eke out respectable frames at 1440p from an RTX 5070 card when you reduce the graphics settings.
- 1080p: RTX 5070 and higher
Every card in the 50 series lineup can push out excellent frames at 1080p. However, we do not recommend using cards like the 5080 and 5090 if you’ll only be gaming at 1080p for the foreseeable future, as they are quite overkill.
The 5070 Ti is the sweet spot if you want to use ultra settings and turn on ray tracing.
Content creation
The RTX 5090 is the most powerful option for video editing, but the 5080 offers the best value. That is not to say other cards won’t get the job done. It all depends on your budget and work needs.
As mentioned, the Blackwell architecture comes with great video editing upgrades. As such, any card in the lineup can pick up video editing tasks. However, if you’re handling a lot of effects and layers, high-end 4K video files, and numerous workloads, you’re better off investing in a card at the top end of the spectrum.
AI
50 series cards are quite capable of AI tasks, thanks to the Blackwell architecture. Your choice here should depend on your budget, as the 5090 remains the fastest, followed by the 5080, at least for now.
When it comes to AI workloads, you need a lot of memory. So, the higher the VRAM capacity, the better the performance. That is why professional cards like the RTX A6000 (with 48GB VRAM) and the upcoming RTX Pro 6000 Blackwell GPU are considered the best for this sort of thing.
Pricing and Availability: Can You Get One?
These are the official price tags of the released cards:
| GPU |
Price |
| RTX 5090 |
$1,999 |
| RTX 5080 |
$999 |
| RTX 5070 Ti |
$749 |
| RTX 5070 |
$549 |
Graphics card launches are a hot mess right now. Finding any 50 series GPU is a tall order unless you’re willing to pay way above MSRP, just like this RTX 5070 listing on Newegg.
Even expensive OC options sell out after launch. It means you currently can’t get an RTX 5070 Ti at $1,000 ($750 MSRP).
The solution? Wait a few more months and hope NVIDIA solves the availability issue. Or better still, grab an old-generation card, like the 4070 Ti, that’s still available on secondary markets.
Reports suggest NVIDIA wound down production of most mid-range to high-end RTX 40 series cards last year. That leaves you with only markets like eBay where you can purchase pre-owned cards or old stock if you want to get the Ada Lovelace generation.
What is NVIDIA saying?
NVIDIA told the gaming community in a recent press call that it has shipped more Blackwell GPUs than the 40-series at this point since launch. The company also released some visual information to back up its claim.
However, that claim seems misleading when considering that the RTX 4090 was the only card released for the first 5 weeks of the 40 series launch. Meanwhile, this time, NVIDIA released the 5090 and 5080 almost simultaneously.
User-Reported Issues with the RTX 50 Series
Gamers Nexus calls the RTX 50 series launch one of the worst in recent history. That’s because there have been many user-reported problems apart from the availability fiasco.
1. Lack of PhysX support
PhysX is a physics engine or middleware SDK created by a company later acquired by NVIDIA. The tech was introduced to help game developers implement high-quality, realistic physical and particle effects in games.
These effects were all-round successful in older games and still work better on NVIDIA cards today.
But PhysX support, at least the 32-bit implementation, is gone on Blackwell GPUs. That’s because NVIDIA withdrew 32-bit CUDA applications from the RTX 50 series graphics cards.
The new development may only affect older games running the 32-bit PhysX API, but some of these titles are still popular. These include Mafia II, Borderlands II, and Batman: Arkham City.
While the games will still run fine, don’t expect the same level of physical and particle effects you experienced on older cards. It’s legitimate reason to skip any 50 series graphics card if you still play a lot of old games that depend on the API.
A ResetEra user compiled a list of games that run the 32-bit PhysX implementation. You can check it out to confirm whether your favorite games are affected.
2. Missing ROPs
ROP is short for raster operations pipeline or render output unit. It is the hardware component within a GPU that processes pixel data and outputs the final rendered image to the screen. As such, ROPs are considered the finalizers of the graphics rendering process.
TechPowerUp first reported that some NVIDIA GPUs were missing some ROPs. NVIDIA validated this discovery by releasing a statement to The Verge claiming the manufacturing defect only affected 0.5% of 5070 Ti and 5090/5090D cards.
But it wasn’t until after a Reddit user reported the same issue with their 5080 card, which was amplified by The Verge, that NVIDIA amended its statement. The company admitted “an early production build” of the 5080s also had missing ROPs.
“Upon further investigation, we’ve identified that an early production build of GeForce RTX 5080 GPUs were also affected by the same issue. Affected consumers can contact the board manufacturer for a replacement…”
Gamers Nexus tested a 5080 that was missing 8 ROPs to verify NVIDIA’s claim about the 4% performance impact. Tests reveal differences across games, with CPU-dependent titles showing little to no difference between the defective and healthy units. However, there were performance swings on games like Total War: Warhammer 3, Dying Light 2 Stay Human, and Dragon’s Dogma 2.
Benchmark credit: Gamers Nexus
The problem with this kind of defect is that you can only detect it after purchasing your graphics card, installing it, and running programs like GPU-Z. On top of that, you have to clean install your NVIDIA graphics drivers to get the correct test results through the app.
3. Driver issues
Many RTX 50 series owners have bemoaned a number of issues affecting their video cards. These problems include games crashing after enabling specific features, system freezing issues after launching or exiting games like Final Fantasy XVI, BSOD errors with HDR enabled, and reduced performance after overclocking.
NVIDIA acknowledged these issues in one of its feedback threads. You can find information about later driver updates that may have fixed some bugs in the left pane.
4. Black screen problems
The black screen issue is one of the main glitches that have been plaguing people lucky enough to buy the GeForce 50 series cards, especially the 5090. A Reddit user (zpinto1234) detailed their experience with the card and sort of confirmed that it only happened with their MSI RTX 5090 Gaming Trio. PC Gamer’s Dave James also reported the same issue on their MSI RTX 5090 Suprim.
Tom’s Hardware’s Jarred Walton confirmed they didn’t see the issue with their ASUS ROG RTX 5090 Astral OC after replacing the MSI RTX 5090 Suprim.
However, workarounds found for the problem included underclocking the affected GPU and reducing the monitor refresh rate to 60Hz. Users can also download and install the hotfixes released by NVIDIA to resolve the issue.
5. Melting GPUs
In an apparent echo from the past, a Reddit user reported that their RTX 5090 cables melted on both ends. The issue is reminiscent of the melting problem faced by some 4090 users shortly after launch and then a year later. Tom’s Hardware also recently shared a case of a 4090 melting for a while. The molten cable debacle led to users criticizing NVIDIA for its decision to use 12VHPWR 16-pin power connectors.
In the case of the 5090, the Reddit user (ivan6953) claimed they perceived a burning smell while playing Battlefield V, and YouTuber Roman “der8auer” Hartung confirmed this claim after assessing the card physically.
Image credit: Reddit (ivan6953)
Per the user’s submission, the cable was fastened and clicked. Hartung also validated this claim and indicated that the issue was due to overheating.
Thankfully, the user told Laptop Mag that NVIDIA and ASUS contacted them to offer RMAs. It seems NVIDIA is still honoring its pledge that it would honor warranty claims for melted cables on the 4090.
That said, not fastening the cable on both ends can cause the melting problem. Ensure your cables are correctly plugged in to reduce the risk of melting.
Download Auslogics BoostSpeed
Your one-stop PC maintenance and optimization tool, this program will clean, speed up, repair and tweak your system to ensure maximum speed and efficiency.
Should You Upgrade to the RTX 50 Series?
As I mentioned earlier, the 50 series has its market. However, these cards are not right for everyone. Here’s an overview of my recommendations depending on your current GPU and use cases:
| Current GPU |
Use case |
Recommended RTX 50 series GPU |
Reason to upgrade/Reasons to skip |
| Mid to lower-end RTX 30 series and below |
1440p gaming and content creation |
RTX 5070 or 5070 Ti |
- The RTX 5070 Ti has significant performance improvements over mid-tier 30 series cards and lower
- If you don’t plan to upgrade your 1080p monitor anytime soon, you can stick with your current card as long as it pumps decent 1080p numbers
- Availability issues
|
| Lower-end RTX 40 series |
1440p/4K gaming, content creation, and AI |
RTX 5070 Ti or 5080 |
- You will gain massive performance bumps if you switch from an RTX 4060 Ti and lower to a 5070 Ti and higher
- You’re still comfortable with 1080p
- You have a lower budget
|
| RTX 4070 and 4070 Super |
4K gaming at high settings, high-end video editing, AI workloads |
RTX 5080 |
- The RTX 5080 does 4K and heavy video editing better than the 4070 cards
- You have the budget
- You want performance bumps in double percentage figures
|
| RTX 4070 Ti Super to 4080 Super |
Enthusiast-level gaming (4K at ultra presets with ray tracing), 4K video editing and cinematography, heavy AI processing |
RTX 5090 |
- You want to game at 4K with ultra presets and ray-tracing
- You don’t mind using DLSS 4 and Multi Frame Gen during gaming
- You run a lot of machine learning and neural processing tasks locally
- You have a high-end 4K monitor
- You want more significant performance improvements
- You consider Multi Frame Gen FPS to be fake
- You have a lower-end CPU that may suffer bottlenecking
|
The following are other factors to consider:
- PCIe Gen 5 advantage: You can go for a 50 series card if you have a PCIe Gen 5 motherboard and want the bandwidth advantage.
- Inflated prices due to supply issues and high demand: The RTX 50 series prices make them a bad value proposition for now. If you must get one, wait until supply issues subside and cards are sold at close-to-reasonable prices.
- You use an RTX 3090 Ti, 4070 Super, or better: Like-for-like successors between the 40 series and 50 series have marginal performance differences that don’t justify an upgrade, especially when you factor pricing. Even the 3090 Ti and 3080 Ti are quite adequate for 1080p and 1440p gaming for the foreseeable future, at least till the 60 series launch.
- You have a 1440p monitor: Older cards can consistently push out excellent frames at 1440p even at the highest graphics setting. If you don’t plan to upgrade your monitor soon, there’s no reason to pick up a 50 series GPU yet.
- PhysX support: If you regularly play older games that use the 32-bit PhysX SDK, it may be a good idea to hold on to your current card or upgrade to a higher 40 series tier.
- Budget constraints for other upgrades: Most CPUs cannot keep up with the RTX 5090 in most cases, even the legendary Ryzen 7 7800X3D. So, to get the best out of the card, you need a CPU upgrade. You may also need to upgrade your cooling unit and PSU to match the GPU’s cooling needs and 575-watt power demand.
Also read: AMD vs. Intel: Which CPU Is Right for You? Here’s a Detailed Processor Comparison
Alternative Options
The 50 series’ underwhelming performance gains, a flurry of driver and hardware problems, and general unavailability are enough to warrant looking elsewhere. The following are options worth considering if you want a new GPU.
1. AMD RX 9070 and 9070 XT
The RTX 90 series has closed some gaps with NVIDIA GPUs in ray tracing, AI upscaling, and productivity. FSR4 is a lot better, and ray tracing has significantly improved.
More importantly, the 9070 XT offers similar gaming performance to the RTX 5070 Ti, which costs $150 when comparing MSRPs. The RX 9070 even beats the RTX 5070 by a staggering 23% in The Last of Us Part 1 at 4K ultra resolution, thanks to its higher memory size.
Benchmark credit: Linus Tech Tips
AMD’s new cards also sold out at launch and could be just as difficult to get close to MSRP.
Even if you want to wait for prices to drop, AMD’s 9070 XT is worth considering over the 5070 Ti if you’re managing a tight budget or just want a better price-to-value option.
2. Higher-end RTX 40 series GPUs
Any card from the RTX 4070 Super and higher will be a better pick when you consider the following:
- 40 series cards are more available at cheaper rates, especially when considering secondary markets.
- The Ada Lovelace GPUs have had time to mature for driver optimization, which means you’re assured stable performance and fewer bugs.
- You’re not leaving too much FPS on the table since the 40 series cards provide similar numbers to their 50 series counterparts.
- You won’t have to replace some components like your PSU to get the best out of 40 series cards, unlike the high-end Blackwell lineup.
- Going from an RTX 4060 to the RTX 4080 Super is a performance upgrade of about 100 to 200%.
Benchmark credit: Gamers Nexus
3. High-end AMD RX 7000 GPUs
You can purchase an AMD RX 7900 XTX right now for about $1,200 to $1,400. You can also purchase it for a lot cheaper if you go to used markets like eBay. That’s great value, considering you cannot purchase the RTX 5080 right now.
You can also game comfortably at 1440p and play some level of 4K with the RX 7900 XT and 7800 XT, which cost significantly less than their NVIDIA counterparts.
However, these cards will only be great deals if ray tracing is not a must and you don’t do a lot of productivity work.
What About RTX 50 Series Laptops?
While real-world reviews for GeForce 50 series laptops are yet to surface, we’re already gearing up for the same level of disappointment judging from released specifications. Here are the specs comparisons between RTX 40 series laptop GPUs and the upcoming graphics cards.
|
RTX 5090 |
RTX 4090 |
RTX 5080 |
RTX 4080 |
RTX 5070 Ti |
RTX 5070 |
RTX 4070 |
| Memory generation |
GDDR7 |
GDDR6 |
GDDR7 |
GDDR6 |
GDDR7 |
GDDR7 |
GDDR6 |
| VRAM (GB) |
24 |
16 |
16 |
12 |
12 |
8 |
8 |
| Memory bus width |
256-bit |
256-bit |
256-bit |
192-bit |
192-bit |
128-bit |
128-bit |
| CUDA cores |
10496 |
9728 |
7680 |
7424 |
5888 |
4608 |
4608 |
| Tensor cores (AI TOPS) |
1824 |
686 |
1334 |
542 |
992 |
798 |
321 |
| GPU power (watts) |
90–150 |
80–150 |
80–150 |
60–150 |
60–115 |
50–100 |
35–115 |
As you’ll find, there are minimal improvements apart from the higher AI TOPS. Also, GPU power remains similar, so you should expect FPS numbers to be in the same ballpark across both generations in native rendering.
There have also been reports from Heise Online and Hardwareluxx, both German publications, that GeForce 50 series laptop GPUs are missing render units (ROPs). NVIDIA categorically debunked those reports to The Verge, but given earlier incidents with the 5090, 5070 Ti, and 5080, it’s difficult to take the company’s word for it.
Should you buy a 50 series laptop?
I recommend against pre-ordering RTX 50 series laptops until you see performance reviews and benchmarks. While 50 series-powered notebooks will join the ranks of the best gaming laptops, seeing how they perform in the real world can help you decide which one to buy.
Download Auslogics BoostSpeed
Your one-stop PC maintenance and optimization tool, this program will clean, speed up, repair and tweak your system to ensure maximum speed and efficiency.
Is the 50 Series Worth the Hassle, or Should You Skip It?
In general, NVIDIA didn’t revolutionize raw GPU performance with Blackwell. The 50 series cards are faster than their direct predecessors by laughable percentage increases only because they draw more power and have a slight bump in core counts.
We could blame the lack of a node shrink in this generation’s fabrication process for the underwhelming improvement.
However, the poor pricing and driver problems are all on NVIDIA.
In the end, I would only recommend upgrading to the 50 series if you’re coming from the Ampere (30 series) generation or earlier. Upgrading from a lower-tiered 40 series GPU to an RTX 5070 Ti and above also makes sense. Additionally, people who do a lot of AI-dependent work and video editing can assess performance differences before switching.
Head to the comments section to tell us what you think and where you’re leaning!
FAQ
No. The RTX 4080 Super beats the RTX 5070 Ti in every benchmark I’ve seen running games at native resolution. However, the 5070 Ti only pulls ahead when DLSS 4 with Multi Frame Generation is turned on.
In some cases, yes. If you own a 1440p or 1080p monitor, there’s no need to buy an RTX 5090. Cards like the RTX 5070 Ti, 4080, and even 4070 Ti and better are excellent for 1440p and 1080p gaming. However, the GPU is adequate for 4K gaming at high graphics settings, especially with ray tracing turned on.
Not by a long shot. The RTX 4090 will be relevant for many years to come, as it is still better than every other GPU except the RTX 5090.
