We're keeping an eye out for the lowest-priced GPUs during the Amazon Prime Day 2026 event, and keeping those listed here for you. The sad state of the memory industry, which is besieged by insanely large memory orders for AI data centers and AI GPUs, has led to increasing prices for gaming graphics cards. Nvidia is purportedly allocating memory based on the amount of money it can make per GB of VRAM, which has made the situation dire for some models. To help you navigate the pricing crisis, we have compiled this GPU index to monitor the availability and pricing of recent graphics card releases from the current and previous generations.

We update this guide constantly with the best prices in the U.S. for each SKU of GPU from the most recent two generations of Nvidia, AMD, and Intel cards. We're listing the lowest price for an available graphics card, regardless of the manufacturer, so it could be an Asus, Zotac, MSI, Sapphire, Gigabyte, Powercolor, or ASRock-branded card, but it will be the cheapest.

A word of warning, with this list, sometimes the retailers switch to third-party sellers automatically. You must be super vigilant and check sellers to ensure legitimacy. For older generation graphics cards, it can often be difficult to find these cards brand-new and on sale from first-party retailers like Amazon, Newegg, and Best Buy, and they are more often than not listed by third parties. Check prices against the current generation of GPUs to understand if you are getting a good deal, or if you're in fact better off paying a little extra for the very latest graphics cards.

Also, check out our list of the best graphics cards and our GPU benchmark hierarchy to see evergreen performance data to help you make an informed decision when choosing a new GPU for your system.

Nvidia RTX 50-Series: Lowest Graphics Cards Prices

The latest 50-series graphics cards from Nvidia include the current most powerful consumer GPU - Nvidia's GeForce RTX 5090. The RTX 5090 uses the new Blackwell architecture and comes with a massive 32GB of GDDR7 VRAM, and given the current prioritization for models with lower amounts of VRAM to boost profits, it's a problem getting your hands on one at anything even in the vicinity of a reasonable price. We are tracking all the best GPU deals in the run-up to Amazon Prime Day 2026.

The 50-series range doesn't have the most impressive gen-on-gen performance uplifts on pure rasterization power compared to the 40-series, but it does support new software and tech such as 5th Gen Tensor cores, 4th Gen Ray Tracing cores, DLSS 4.5 with Multi-Frame Generation tech, and Reflex 2. Some of those technologies, like DLSS and frame gen, can help wring the most performance out of those lower-tier cards with less VRAM.

There are 8GB variants of the RTX 5060 and 5060 Ti, available at near MSRP prices, with the higher-tier GPUs like the RTX 5070 Ti and RTX 5080 in poor supply at any reasonable price.

Nvidia RTX 40-Series: Lowest Graphics Cards Prices

Nvidia's 40-series Ada Lovelace architecture graphics cards include the powerful GeForce RTX 4090 and 4080 Super. The 4090 combines a massive 24GB of GDDR6X VRAM and Nvidia's 4th generation of Tensor cores for increased ray-tracing ability.

The RTX 40-series cards also feature support for Nvidia's DLSS 3 software for AI frame generation, which helps to increase frame rates on higher resolutions and settings. In general, Nvidia cards are the best choice for ray tracing, with more features than the competition. You'll also generally pay more for Nvidia GPUs relative to similarly performing AMD GPUs.

Unfortunately, RTX 40-series GPUs are no longer being produced, so stock levels have dried up on many card variants. Any available cards for sale have a high chance of being either second-hand or ex-mining hardware, yet still ask for exorbitant prices. Be extra vigilant when shopping for 40-series graphics cards, and buy from a reputable seller. Prices for 40-series cards are also much higher than they should be. They can be a good choice for a second-hand purchase if you can get one at the right price and accept all associated risks.

AMD Radeon RX 9000-Series: Lowest Graphics Cards Prices

AMD's competition to Nvidia includes the Radeon RX 9060, RX 9070, and RX 9070 XT graphics cards with MSRP prices of $269, $549, and $599, respectively. Finding one of these GPUs for that price, however, could prove to be a struggle as prices of Radeon graphics cards rise due to ever-expanding memory price hikes. These new cards go toe-to-toe with Nvidia's RTX 5070 and RTX 5070 Ti and offer a great alternative to Team Green.

The RX 9070 XT comes with 16GB of GDDR6 VRAM for those texture-heavy games and provides a large generational increase compared to RDNA 3. Improvements to AI and ray tracing tech have made this generation of AMD cards perform much better in games with ray tracing, but they still fall behind Nvidia in overall performance for ray/path tracing.

There are also 8GB and 16GB versions of the RX 9060 XT available, and they are selling for a fair price at the moment. However, 8GB of VRAM means the card isn't terribly futureproofed, as game requirements continually grow regarding shader and texture memory requirements.

AMD Radeon RX 7000-Series: Lowest Graphics Cards Prices

AMD's 7000-series GPUs are still a great option for a graphics card in your gaming rig if you can find one at a good price. Not the best choice if you're looking for pure ray-tracing performance, but if you want raw computational performance, they are a winner. Combined with AMD's FSR software, these 7000-series GPUs can still crank out high frame rates in the latest games.

The RX 7900 XTX in particular represents a great blend of power and performance, with it being the halo-tier card from the 7000-series and AMD not replicating a super-high-end card for the 9000-series. The RX 7900 XTX still ranks highly in our GPU Benchmark Hierarchy charts.

Here are all the lowest prices for the 7000-series models currently available.

Intel Arc GPUs: Lowest Graphics Card Prices

Intel's entry into the GPU market has been a little hit-and-miss. In some titles, these cards perform relatively well, but in older DirectX games, the cards suffer from some poor performance issues.

Frequent driver updates have made many improvements to the performance of Intel's Arc lineup and turned them into a much more viable choice for a budget gaming PC build. The prices of these GPUs have hit a price sweet spot that has been abandoned by Nvidia and AMD, especially for 1080p gaming. The Arc series of GPUs should certainly not be discounted when considering any budget gaming system build.

Intel's most recent launches include the Arc B580 and Arc B570, with the cards having 12GB of VRAM for the B580 and 10GB for the B570. Pricing is very competitive with the Intel cards, but the manufacturer's variety is limited, as is stock.

More Prime Day Tech Deals

Best Tech and PC deals | Best gaming PC deals | Best RAM combo deals | Best 3D printer deals | Best RAM deals | Best gaming laptop deals | Best monitor deals | Best Wi-Fi Router deals | Best GPU deals | Best SSD deals | Best hard drive HDD deals | Best CPU deals | Best gaming chair deals | Best PC building tool deals | Best PC peripherals deals | Best filament and resin deals | Best motherboard deals | Best CPU cooler deals | Best PC case deals | Best Dell and Alienware deals | Best USB charger deals | Best gaming and productivity laptop deals under $1,000 | Best laptop PC deals

Also, you can join the Tom's Hardware deals Discord for up-to-the-minute hardware deals.

A new report claims that DeepSeek has illegally obtained and operated "several thousand" Nvidia Blackwell GPUs in the process of training and developing its newest large language model. According to coverage by The Information, six unnamed sources all claim DeepSeek's involvement in a convoluted smuggling ring based around the use of fake data centers as fronts to move high-powered servers into mainland China, illegally circumventing U.S. sanctions on newer AI GPUs.

Sources close to the matter allege that DeepSeek is involved in a high-complexity smuggling ring focused on getting Blackwell chips into China illegally through the use of fake data centers. Shell companies purchase data centers worth of Nvidia servers somewhere in Southeast Asia, setting up the data center and its hardware entirely to spec. Nvidia's OEM partners send contractors to inspect the installation, confirming successful installs and export compliance.

After this inspection is finished, smugglers reportedly disassemble the entire data center rack by rack, shipping the GPU servers in suitcases across the border into mainland China, where the purchase and use of certain Nvidia chips are restricted by the United States government. According to the report, sources with knowledge of these smuggling operations claim that smugglers and clients prefer 8-GPU rack servers like the HGX B200 over the powerful GB200 NVL72 for this smaller size and ease of covert transportation.

When asked for comment, an Nvidia spokesperson gave the following statement to Tom's Hardware:

We haven't seen any substantiation or received tips of 'phantom datacenters' constructed to deceive us and our OEM partners, then deconstructed, smuggled, and reconstructed somewhere else. While such smuggling seems far-fetched, we pursue any tip we receive.

DeepSeek's Need for Nvidia GPUs

DeepSeek, the most recognizable Chinese AI firm in the United States, thanks to its R1 LLM making worldwide headlines one year ago, has long been connected with Nvidia GPUs. Its sensational R1 model was trained on only 2,048 Nvidia H800s in two months, a number of GPUs far smaller and more efficient than any Western competitor. Since this time, DeepSeek has consistently been linked to the stockpiling and purchase of as many Nvidia GPUs as it can obtain, with reports constantly swirling about DeepSeek somehow bypassing export restrictions and securing huge numbers of the newest Nvidia chips.

Interestingly, DeepSeek's latest internal reports seem to indicate plans to use Nvidia chips for its newest AI models. In a whitepaper released on December 2nd on DeepSeek V3.2, DeepSeek suggests their bottleneck on performance matches that of frontier models like Gemini-3.0-Pro is pre-training compute; "We plan to address this knowledge gap in future iterations by scaling up the pre-training compute." Pre-training compute is a workflow that Nvidia GPUs and CUDA software perform better than most other competitors, suggesting that DeepSeek engineers count on something changing for its access to high-caliber pre-training compute power.

DeepSeek's track record proves that Nvidia's pre-training abilities fill a niche unmatched by domestic Chinese products. Reports in August claimed that Huawei's Ascend GPU servers were unable to run necessary training workloads, prompting a return to Nvidia hardware in the R2 training process. This was despite government intervention and doctrines calling for DeepSeek to turn to domestic Chinese products for its AI workload. While the Huawei Ascend servers were used for inference for the models, the company could not turn anywhere but to Nvidia, much to the chagrin of China.

Nvidia's Future in China

The Trump administration recently announced plans to unrestrict the Nvidia H200 GPU in China, opening up Nvidia's sales in the country. Speculators claim that this policy U-turn from the White House, which has spent much of 2025 toeing a line of complete export isolationism to China, comes as fears of Huawei's CloudMatrix 384 and Ascend 910C systems grow. Reputable claims hold that these servers match the H200 and GB200 NVL72 in certain performance metrics, causing the U.S. government to release the H200 into China.

This new policy is based on a compromise between flooding China with easy-to-access American Nvidia tech and banning it altogether. The hope is to satiate Chinese tech needs and take away motivation for firms like Huawei to develop their own Nvidia competitors. The adoption of this doctrine, oft-touted by Nvidia's lobbying efforts to the White House, marks a major shift in the "Chip War" trade offensive between Beijing and Washington D.C., which has moved from preventing China from any access to next-gen tech to hoping to slow China's tech power that is beginning to threaten Western tech dominance.

While Trump's Commerce Department continues to insist that China will never see Nvidia Blackwell hardware, keeping the export exceptions limited to Hopper-generation hardware like the H200, time will tell if further Nvidia lobbying and fears of the Chinese tech sector will open the doors further. And of course, if DeepSeek truly is involved in conspiracies of phantom data centers, they won't even need the U.S. to allow them access to Blackwell.

Broken PCIe fingers are becoming a frustrating trend among Nvidia's latest flagship graphics cards based on its RTX 5090 Founders Edition design. Computer technician NorthridgeFix on YouTube published a video detailing the story of one of his customers who suffered such bad luck on not just any high-end RTX Blackwell GPU, but Nvidia's crazy-expensive $10,000 RTX Pro 6000 workstation graphics card. According to the video, the card snapped under its own weight during transit after the unfortunate owner (a tech YouTuber with 40 million subscribers, apparently) didn't uninstall it before moving it.

The technician showed off the PCIe connector itself on the now $10,000 paperweight, revealing that it snapped cleanly into two pieces near the middle. Due to the component's complexity, it is impossible to repair the PCIe connector, so a replacement part is required.

If only that were actually feasible — NorthridgeFix revealed that Nvidia astonishingly does not provide any replacement parts for its OEM graphics cards, despite them boasting a design that allows easy repairability of the PCIe connector alone.

NorthridgeFix revealed this is a big problem with Nvidia's latest Founders Edition designs and spoke about a previous encounter he had with a customer who accidentally broke their RTX 5090 Founders Edition while installing a waterblock. After inspecting the damaged RTX 5090, the technician reported that if Nvidia had provided replacement parts, the card could have been saved.

We covered this story and how the user got a free RTX 5090 Founders Edition replacement from Nvidia. Apparently, Nvidia watched NorthridgeFix's rant on YouTube about the lack of replacement parts for their GPUs, and that was enough to get Nvidia to replace the broken card for free, even though it was technically the customer's fault for breaking it with a waterblock in the first place.

Hopefully, the same thing will happen with the customer of the now-dead RTX Pro 6000 workstation card. Nvidia's latest flagship Founders Edition graphics card design has the unique trait of having modularity in its design, something that is very uncommon in the graphics card world. The RTX 5090 Founders Edition and RTX Pro 6000 graphics cards are made up of three primary PCBs, including a PCB that houses all of the important bits like the GPU core, VRAM, and power delivery system, and a secondary PCB that is only responsible for providing PCIe connectivity to the graphics card.

This design, by nature, makes them uniquely well designed for repairability, at least in theory. If the PCIe connector on a traditional graphics card fails, either the whole PCB has to be replaced or the connector has to be repaired. Both require specialized tools and a qualified technician to do the job. With Nvidia's latest Blackwell OEM graphics cards, their design should allow you to simply swap out the PCIe connector component with a replacement part as if you were replacing DRAM on a motherboard. However, as this unfortunate user found out, Nvidia doesn't dish out the parts, rendering the modular and repairable design somewhat obsolete.

Maybe one day Nvidia will provide replacement PCBs for its flagship graphics cards, but for now, if you own one of these cards, NorthridgeFix warns users that these graphics cards are not easily repairable. If the card breaks, there's a good chance you'll have to beg Nvidia customer support for an entirely new GPU replacement.

Jensen Huang is in Taiwan for TSMC's sports day event right now. Before attending, he spoke to local media, giving insight on various ongoing developments surrounding Nvidia and its relationship with AI constituents — including the prospect of selling Blackwell chips to China. President Trump and U.S. Treasury Secretary have already made it clear that Blackwell is reserved for America first, and Huang has just echoed those sentiments by clarifying that "there are no active discussions" surrounding this matter.

A few days ago, while speaking to CBS, Trump said the most advanced AI GPUs will not be sold to China. Following that, America's finance chief Scott Bessent hinted at a potential strategy, highlighting that China can have Blackwell chips after they're outdated by a year or two. Jensen's latest comments don't reveal anything new, but they corroborate the existing stance permeating Washington.

That said, beyond the White House, Huang reminds us that it's ultimately up to China's discretion whether it wants to deal with Nvidia. Beijing has recently banned its firms from dealing with the chipmaker to focus on local, homegrown silicon as the region grows toward AI independence. Blackwell still makes its way into the country, however, through illicit channels.

The previous-gen, Hopper-based H20 is the top-end AI GPU Nvidia sells in China, and there are reports of a Blackwell-based B30A model launching soon. If that comes to fruition and China allows Nvidia to sell it, the chipmaker will have to share 15% of the revenue from all sales with the White House, and that's after export licenses have been granted.

A dense layer of red tape covers this AI race, which is informing semiconductor trade policy at the moment. Meanwhile, Huang has even claimed that China is aggressively tailing the U.S. in artificial intelligence — if not already ahead — and that most of the popular open-source models in the world today are from China. The region also produces some 50% of all AI researchers globally, implying the competition is stiff.

Even though neither Washington nor Beijing wants to engage with each other on cutting-edge AI silicon, both heads of state struck a historic trade truce last month that's said to last a year. This positive development briefly led the markets to think that a chip breakthrough might be on the cards, though it's now clear the two nations don't even want to talk about it.

A couple of days ago, Trump told CBS News that China will not be allowed to have the most advanced chips Nvidia makes, but that they can otherwise deal with each other and sort out their own matters. Today, U.S. Secretary of the Treasury Scott Bessent corroborated that statement, proposing how "there may be a case down the road" when Beijing can eventually get their hands on Blackwell GPUs.

Speaking to CNBC, Bessent said, "If we think about the Blackwell now, they’re the crown jewel... Given the incredible innovation that goes on at Nvidia, where the Blackwell chips may be two, three, four down their chip stack in terms of efficacy, and at that point they could be sold on." This position aligns with the current landscape in the region where Hopper-based H20 GPUs are still the best Nvidia has to offer — so, when Blackwell becomes a generation or two old, it might trickle down as well.

It's important to note that this is not a concrete, official announcement that represents the White House's stance on the matter, though this thinking is what will likely drive (or influence) policy, especially considering how the Treasury head is a crucial part of the President's cabinet. Trump has already made it clear that the latest-and-greatest is reserved for America first, then other countries.

In his interview, Bessent also highlighted how technological evolution outpaces the rate at which negotiations take place, meaning the silicon is what leads the conversation. If there were a deal to be struck here, it might take longer to agree on the terms than the wait for the next-gen product itself, rendering the discussions pointless in the first place.

Next year, Blackwell (Ultra) will officially be succeeded by Vera Rubin, but it will take time to gain prevalence, which lines up with Bessent's comment on "whether it's 12 or 24 months" before Blackwell is truly behind. At which point, Nvidia will be allowed to sell it to China, who, at least right now, doesn't want anything to do with Nvidia GPUs, with CEO Jensen Huang even claiming that revenue from the region has hit zero.

China is already banned from receiving and high-end Nvidia GPUs, which has led to illicit imports and VRAM modding operations at an unprecedent scale in the region. If any companies wish to sell even neutered versions of their flagship products to China, they have to pay Washington 15% of the cut from the sales — which is what Nvidia would need to do when the cutdown China-only B30A GPU (based on Blackwell, funnily enough) is finally launched.

Despite the bans, Beijing still receives Blackwell chips through grey channels, exporting them from other countries, while China is in the midst of pivoting to homegrown silicon for AI independence. At their meeting last month, President Trump and President Xi Jinping reached a historic trade truce but left out any mentions of Blackwell seemingly even in private conversation, further substantiating the red tape surrounding semiconductor trade policy.

Washington and Beijing have been engaged in an escalating trade war ever since the Trump administration took over, with tensions over semiconductors in particular reaching an all-time peak. Despite a year-long trade truce being reached last week, it seems President Trump is still adamant about his position to restrict high-end AI chip access to China. Speaking to CBS last night, Trump made it clear that Nvidia's current-gen, flagship Blackwell AI GPUs will not be allowed to be sold to China, and that they're reserved for America first.

The ongoing AI race largely stands atop the shoulders of chipmaker Nvidia since its AI GPUs are used for training and hosting models for inference around the world. China is already barred from receiving high-end AI GPUs, with Nvidia being forced to produce neutered variants of its popular accelerators that comply with trade restrictions, allowing them to be sold to China. Nvidia CEO Jensen Huang has said its market share in Beijing has dropped to zero, implying that any sort of breakthrough would change the landscape drastically, but it seems that's not on the cards right now.

In the interview, Trump clarified that "we will let [them] deal with Nvidia" but not in terms of the most advanced chips, which are currently Blackwell (B100, B200) and Blackwell Ultra (B300) GPUs. "The most advanced, we will not let anybody have them other than the United States," he said. Recent rumors suggest that a B30A China-exclusive variant is in the works, featuring half the performance and memory of the regular B300, meeting export control requirements, though nothing is set in stone yet.

Trump met President Xi Jinping last week at a summit in South Korea and reached a historic agreement, pausing the trade war between the two nations for a year. This ease in tensions signaled to the markets that Nvidia's re-entry into China was possible once again, with Jensen Huang even saying, "I think it's really good for America and it’s really good for China that Nvidia could participate in the Chinese market."

Nvidia's Blackwell chips were conspicuously left out of the conversation at this meeting, despite prior indications that they would be a talking point. This ever-important convergence of premiers from different nations and billionaire CEOs resulted in Trump saying that the U.S. serves as "sort of the arbitrator or the referee" between China and Nvidia on the latter's "super duper" Blackwell GPUs.

However, the President has now completely shut the door on Nvidia's flagship product entering Beijing, while China itself has banned local firms from purchasing Nvidia GPUs, focusing on homegrown chip development. Nvidia currently sells the outdated, Hopper-based H20 GPU in the region, which performs poorly considering how gaming GPUs with modded VRAMs are thriving in underground markets.

This latest development is symbolic of the tit-for-tat nature of a trade war between two powerhouses, both unwilling to step down despite a temporary truce in place. Current policy indicates that chipmakers like AMD and Nvidia need to share 15% of their revenue from sales to China, which would extend to cover the rumored Blackwell-based, cut-down B30A GPU. Only time will tell if that picks up steam and helps the company's sales increase in the region once again.

AI currently serves as Nvidia's money-printing machine; thus, the chipmaker must ensure that it's properly fed at all times. Although it may not rival the best graphics cards for gaming, the newly introduced RTX Pro 5000 72GB (via TechPowerUp) will cut through AI workloads as if they were butter.

The RTX Pro 5000 72GB represents an upgraded iteration of the standard RTX Pro 5000, distinguished primarily by increased memory capacity. The former boasts an additional 24GB of GDDR7 memory, a substantial 50% increase in capacity. This significant enhancement in the memory subsystem is expected to be accompanied by a proportionate increase in cost.

The sole modification to the RTX Pro 5000 72GB is an increase in capacity, as the memory modules remain 28 Gbps chips operating over a 384-bit memory interface, which yields a maximum memory bandwidth of 1.3 GB/s. The Blackwell AI graphics card comprises 24 memory modules arranged in a clamshell configuration, each with a capacity of 3GB. In terms of capacity, it is only 25% less than the flagship RTX Pro 6000.

Nvidia RTX Pro 5000 72GB Blackwell Specifications

The other specifications of the RTX Pro 5000 72GB remain unchanged. Utilizing the GB202 silicon—also employed in the GeForce RTX 5090 and RTX Pro 6000—the die of the RTX Pro 5000 72GB features 110 SMs enabled out of the maximum 192 SMs. It corresponds to approximately 57% utilization of the GB202 silicon. Given its enhanced capacity, the RTX Pro 5000 72GB effectively positions itself between the RTX Pro 6000 and the RTX Pro 5000.

According to Nvidia's specification sheet, the 50% additional memory didn't alter the RTX Pro 5000's power draw. The RTX Pro 5000 72GB is still a 300W graphics card that pulls power from a 16-pin power connector. The modest power requirements, combined with its dual-slot, blower-style cooler, make it easy to have up to four of these Blackwell AI graphics cards in a modern system.

The RTX Pro 5000 72GB represents a strategic move by Nvidia, with the potential to yield significant dividends. As an intermediary model, the RTX Pro 5000 72GB offers a viable option for consumers seeking greater memory capacity than the RTX Pro 5000, yet wishing to avoid the exorbitant price of the RTX Pro 6000, which can reach up to $10,000.

Nvidia has not announced the release date or pricing details for the RTX Pro 5000 72GB on its website. The pricing of the standard RTX Pro 5000 varies significantly, ranging from $4,400 to $7,500, although availability is infrequent. The increased memory capacity of the RTX Pro 5000 72GB is likely to elevate its price by approximately 20% to 25%.

Earlier this week, Nvidia’s DGX Spark AI mini-PCs were released to the masses. Symbolic of the splintered state of the AI industry, though, Nvidia’s CEO Jensen Huang hand-delivered DGX Spark systems to Elon Musk and Sam Altman — separately. Altman and Musk once worked closely together, for a common cause, as co-founders of OpenAI, where Jensen once hand-delivered the original DGX-1 nine years ago, long before the AI boom began.

The last time we saw such a star-studded DGX photo-op was when Huang hand-delivered the original DGX-1 to Elon Musk, in his role as co-founder of non-profit OpenAI. Once partners seeking to further the development of safe AI for the benefit of humankind, Musk and Altman have become the fiercest of rivals. The rivalry isn’t very sportsmanlike, either, with the pair now regularly trading barbs and public insults, and even engaging in some acrimonious legal tussles. Since 2023, we have also had two distinct competitive AI products addressing the same market, albeit from different angles: ChatGPT and Grok.

Musk received his DGX Spark while wearing his ‘Chief Engineer at SpaceX’ hat. Huang quipped that he was “delivering the smallest supercomputer next to the biggest rocket,” at the Starbase, Texas facility.

We are reminded by the Nvidia blog that the GB10 Grace Blackwell Superchip-powered DGX Spark packs 128GB of unified memory and delivers a petaflop of AI performance. It is claimed to have enough resources and muscle to run models with 200 billion parameters locally. Moreover, the Green Team’s blog highlights that it has been nine years since Huang personally delivered the DGX-1 to Musk at OpenAI.

Sam Altman also reminisced about Huang’s previous little parcel. “Things have come a long way since the delivery of the DGX-1 9 years ago; amazing to see...” mused the OpenAI boss. Altman was commenting on the President and co-founder (another one) of OpenAI, Greg Brockman's, shared photo. The picture shows Huang nestled between Brockman and Altman.

The Jensen-powered delivery service

Huang famously started his career washing dishes, bussing, and waiting at Denny’s. Now he’s back as a server, but on the menu are slices of AI-accelerating silicon, delivered in person to a select few fellow tech industry giants.

Nvidia’s Jensen-powered delivery service isn’t standard, even though the DGX Spark has gotten a third more expensive since it was first announced. The Nvidia first-party DGX Spark MSRP is now $3,999, and it is shipping direct from Nvidia, Micro Center, and a number of partners.

The first batch of DGX Spark systems was also put into the hands of researchers at AI-processing hungry companies like Anaconda, Cadence, ComfyUI, Docker, Google, Hugging Face, JetBrains, LM Studio, Meta, Microsoft, Ollama, and Roboflow.

If you are interested in the DGX Spark, and its integrated Nvidia AI stack with full CUDA library support, the firm’s partners are also cooking up systems, featuring their own special saucy tech - but they all look fairly similar. DGX Spark systems are being made and marketed by Acer, Asus, Dell Technologies, Gigabyte, HPI, Lenovo, and MSI. You definitely won’t get Jensen-powered delivery with these, though, you’d be lucky to get a free MSI dragon plush toy.

If you're a gamer on a tight budget, the past few years have offered slim pickings for entry-level graphics cards. As we sussed out earlier this year, gamers shopping at the $200-to-$250 price point have long been forced to pick among compromised choices that are all a generation or two old, and modern games are really starting to put the hurt on that older hardware.

Even versus checkered company like Intel’s Arc A750 and AMD’s Radeon RX 6600, Nvidia's last entry-level card of note, the GeForce RTX 3050 8GB, put in a particularly weak showing in our tests. The entry-level Ampere card struggled to clear the 60 FPS average we consider a baseline for solid gaming experiences, even at 1080p.

Enter Nvidia's GeForce RTX 5050, powered by the latest Blackwell architecture. Is this $250 card the game changer that this end of the graphics market so desperately needs?

Let’s start with the basics. The RTX 5050 has 2560 Blackwell CUDA cores, a basic setup that's defined this class of product since the RTX 3050’s arrival. Those CUDA cores are now more capable, though, as a Blackwell SM partition can process up to 32 INT or FP32 per clock, versus 16 FP32 instructions and a second 16 FP32 or 16 integer instructions per clock on Ampere.

(The full GB207 chip may include more inactive SMs for reasons of yield or headroom for use in future products, as a "full" implementation encompassing two Blackwell GPCs would total out to 3072 possible CUDA cores.)

On top of Blackwell's basic architectural benefits, the RTX 5050 is clocked much higher than the RTX 3050, at a rated boost clock of 2572 MHz rather than just 1780 MHz on the Ampere card. All those improvements mean this card offers raw compute potential of at least 13 TFLOPS in boost, up 44% from the RTX 3050.

The move to the Blackwell architecture and a more advanced process node also grants the RTX 5050 a massive increase in L2 cache, from just 2MB on the RTX 3050 to at least 24MB on the Blackwell card. Bigger caches are relatively easy wins for boosting performance, and Nvidia says that Blackwell's large L2 benefits ray-tracing workloads especially.

The biggest difference between the RTX 5050 and other Blackwell cards is Nvidia's continuing use of GDDR6 memory. This card gets 8 GB of 20 Gb/s GDDR6 on a 128-bit bus for an effective 320 GB/s of memory bandwidth, up from 224 GB/s on the typical RTX 3050. Even RTX 5050 mobile implementations get GDDR7.

In and of itself, this choice isn't necessarily a bad thing for performance. AMD's Radeon RX 9000-series cards all stick with GDDR6 and still deliver competitive performance with RTX 50-series products at similar price points.

Taken all together, the RTX 5050 sounds like an impressive multi-generational improvement, but only if your frame of reference excludes RTX 4050 mobile chips. Those similarly provisioned GPUs were clocked at least as high as 2370MHz in their most generous power and thermal envelopes.

At its highest thermal and power specs, the mobile RTX 4050 could have compute potential at least as high as 12 TFLOPS, although its performance is likely hampered by just 6GB of 16Gb/s GDDR6 on a 96-bit bus, producing just 192 GB/s of memory bandwidth. Had Nvidia brought the RTX 4050 to the desktop in a more fully fledged 8GB form, the improvements of the RTX 5050 might be considerably less impressive.

Partially because it sticks with GDDR6, the RTX 5050 still rings in at a 130W total board power rating, the same as the RTX 3050. Like most 3050s, most RTX 5050s still employ an eight-pin PCIe power connector to get all of their necessary juice. Builders hoping for a power-sipping, high-performance slot-powered card will still be left wanting.

If supplies of cutting-edge GDDR7 have to go anywhere, Nvidia certainly prefers that they be used on higher-end, higher-margin discrete cards and in notebooks, where tight power and thermal budgets mean that GDDR7's power usage and efficiency benefits yield all sorts of fruit for engineers trying to squeeze out every last second of battery life or to shave off every last gram from a heatsink or chassis. It's much easier to deal with the higher power and heat production of GDDR6 in the relatively unconstrained power and thermal environment of desktop PCs.

If Nvidia saved anything on the RTX 5050's bill of materials by sticking with last-gen GDDR6, it isn't throwing any bones to the lowest end of the market in turn. The RTX 5050 sticks with the same $250 price as the RTX 3050 did about three-and-a-half years ago, even as tariffs and inflation put the squeeze on today's consumers.

You'd think Nvidia could find a few bucks in one of Jensen's jacket pockets to offer some relief to gamers on a budget, given the staggering amounts of money it's making in the data center, but maybe that's too tough a nut even for advanced AI to crack.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

To represent the RTX 5050 in our testing, we picked Gigabyte’s Windforce OC version of this card. Ringing in right at Nvidia’s $249 MSRP, this model offers everything you need and nothing you don’t for the kind of entry-level gaming build that’s most likely to include an RTX 5050.

Typical of cards built to hit MSRPs, the Windforce OC RTX 5050 lacks any fancy extras, like RGB LED lighting or heavy-duty metal shrouds. You get two counter-rotating fans, a plastic fan shroud and backplate with some prominent Gigabyte and GeForce branding, and a metal mounting bracket, and that’s about it. We do appreciate this card’s compact 7.8" x 4.6" by 1.6" (LWD) dimensions (199 x 116 x 40 mm), which should allow it to fit into most any case.

Gigabyte still outfits the Windforce OC card with a proper aluminum fin stack rather than a cast affair, and two beefy copper heat pipes run through it to transfer heat from the GB207 GPU itself into those fins.

We didn’t pull our 5050 apart entirely to examine the base plate of the heatsink, but it appears that this card cools the four GDDR6 memory chips with thermal pads or gel, making contact with an aluminum base plate. Those are all signs that this card should keep its most important components cool under load.

As is common nowadays, Gigabyte cuts a vent in this card’s backplate to allow air from the forward portion of the fin stack to escape directly upwards into the case’s airflow path.

The single 8-pin power connector is located toward the rear of the card, a somewhat annoying trend that we’ve seen several graphics card OEMs adopt recently. PCIe power cables really have to stretch to reach these distant power connectors in many cases. We’d much prefer that the connector be placed toward the front of the card.

Around back, we find a somewhat unusual complement of two DisplayPort 2.1 connectors and two HDMI 2.1 outputs. The mounting bracket features large vents to allow exhaust air to flow out the rear of the card.

Overall, we have no complaints about Gigabyte’s design choices with this card, given that it rings in right at MSRP. Let’s see how it performs.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

We do our best to deliver clean, reliable benchmark numbers. Each of our tests encompasses 60 seconds or more of real-world gameplay, carefully chosen to represent what we expect will be typical gameplay experiences. We sanity-check every result and retest whenever necessary to ensure that outliers don’t muck up our final standings.

For now, we're sticking with native resolution testing at a mix of high and ultra settings at 1080p, as well as ultra settings at 1440p and 4K.

We're weighing whether pure native-res testing will remain the way to go going forward, as pretty much every vendor-independent TAA or upscaler implementation (like Epic's TSR) looks worse than DLSS 4, FSR 4, or XeSS running at the Balanced or Quality preset.

We think that more and more gamers are taking advantage of the performance and image quality benefits of these upscalers without a second thought, and we're weighing whether to enable them by default so our tests are more representative of the real-world performance figures gamers can expect when buying into and using a vendor's entire hardware-software stack.

This is a tough decision, and not one we're making lightly. Any final verdict on the way forward likely won't occur before 2026. For now, we're leaving upscaling and framegen off and sticking with native-res testing exclusively.

Our first draft of a new test suite covers 12 raster-only titles. Five of those include additional RT options that we employ in our tests, while two more require a graphics card with RT support to run at all. Finally, we tested GTA V Enhanced exclusively with RT enabled, as that extra eye candy is the entire reason to run it over GTA V Legacy, in our opinion.

Here's our complete list of tested titles:

• Black Myth: Wukong (+RT)
• Ratchet & Clank: Rift Apart (+RT)
• Marvel's Spider-Man 2 (+RT)
• Cyberpunk 2077 (+RT)
• Alan Wake II (+RT)
• Fortnite
• Marvel Rivals
• Apex Legends
• Counter-Strike 2
• Microsoft Flight Simulator 2024
• Stalker 2
• Doom: The Dark Ages (RT required)
• Indiana Jones and the Great Circle (RT required)
• Grand Theft Auto V Enhanced (RT)
• Clair Obscur: Expedition 33

We've tried to cover a broad mix of game engines, graphics APIs, and game types in this lineup, from popular esports experiences to crushing AAA visual feasts developed both natively for PC and ports from consoles. Our selected games stress every part of a modern graphics card, from pure compute horsepower to VRAM management to RT to driver overhead at high frame rates. If a card rises to the top of our charts after weathering all of these tests, you can be sure that it's a standout product.

When picking titles to test, we considered games' time in market, active player counts, review scores (to see whether a title is likely to become an enduring part of PC gaming), and the ease of conducting a repeatable benchmark, among other factors.

Wherever possible, we use real, live, eyes-on-screen, hands-on-mouse-and-keyboard benchmark runs. We don't think automated, on-rails, hands-off canned benchmarks fully capture the gameplay experience on a given graphics card.

Only by actually playing a game can we account for factors like how input lag affects the experience, and making sure that a title has acceptable input lag is becoming a key consideration when latency-sensitive framegen techniques are becoming more and more common as performance-boosting tools.

Our current test system comprises the following components:

• CPU: AMD Ryzen 7 9800X3D
• Motherboard: ASUS TUF Gaming X670E-Plus Wifi
• Memory: G.Skill Trident Z5 Neo 32GB DDR5-6000
• Storage: Inland Gaming Performance Plus 4TB PCIe 4.0 NVMe SSD
• Power supply: Corsair RM1000x

With all that out of the way, let's dive into our results.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

If all you want is the TL;DR, the RTX 5050 turns in a 58% improvement to average frame rates versus the RTX 3050 at 1080p without RT enabled, basically tying with Intel's $249 Arc B580 and the $299 RTX 4060. Critically, its overall result clears 60 FPS on average across our range of titles, and its 1% lows are good enough for a generally smooth ride.

Click over to our 1440p results, and it's clear that you really want to step up to a Radeon RX 9060 XT 16GB or RTX 5060 Ti 16GB to enjoy a smooth ride across most games at these higher resolutions, even with DLSS or FSR in play.

4K gaming is heavy enough on compute and VRAM that none of these cards are going to be anybody's first choice for a 4K experience without a heavy application of upscaling, assuming you can even keep VRAM usage in check.

If you’re interested in these cards' performance in a particular title, read on for our commentary on individual game results.

Black Myth Wukong

Why it's here: a stunning showcase for PC graphics

What it stresses: Compute, VRAM, RT (when enabled)

Black Myth Wukong is a gorgeous example of what’s possible from Unreal Engine 5, but it places heavy demands on graphics cards in exchange. This title has a free benchmark tool, but to ensure we were getting the benefits of all the updates that Game Science has made to this game since launch, we picked up the full version and devised our own benchmark sequence.

At very high raster settings, the RTX 3050 struggles in this title, but the RTX 5050 at least has enough oomph to potentially benefit from a DLSS boost.

Alan Wake II

Why it's here: One of the richest AAA feasts for the eyes on PC

What it stresses: Compute, VRAM, RT (when enabled)

Alan Wake II is the latest technical tour-de-force from Remedy, and it really puts the hurt on these lower-end cards. As with other AAA heavyweights in our test suite, the RTX 5050 delivers more oomph than the RTX 3050 (as it well should), but you really want DLSS, lower settings, or both to get to a 60 FPS experience at 1080p, and any higher resolutions will need a much more powerful card to achieve playability.

Counter-Strike 2

Why it's here: over a million active players can't be wrong

What it stresses: Compute, driver overhead at high frame rates

Counter-Strike 2 needs no introduction as one of the most popular esports titles out there. The RTX 5050 lands just shy of the 244 FPS average you'd want to drive an ultra-high-refresh-rate monitor at 1080p, but its 1% lows are weirdly low relative to even aging hardware like the RX 7600.

In general, Blackwell cards really seem to struggle with 1% lows in CS2, even as the RX 9060 XT duo posts both the high average frame rates and 1% lows that we want for a smooth gaming experience.

Microsoft Flight Simulator 2024

Why it's here: One of the most demanding games available on PC right now

What it stresses: Compute, VRAM

Flight Simulator 2024 is honestly too much game for any $300-or-less graphics card, at least at ultra settings. Average frame rates are low on everything that's not an RTX 5060 Ti 16GB or RX 9060 XT 16GB, and that pain is compounded by a hunger for VRAM that sends 1% lows far beneath averages that might appear OK on the surface. Frame rates plummet especially hard as you come in for landings, making consistent control of your aircraft a pain in the butt.

The RTX 5050 basically doubles the RTX 3050's performance in this title, but it's not enough to achieve a truly playable frame rate.

Fortnite

Why it's here: It's Fortnite

What it stresses: Compute

Fortnite may have a reputation as a potato game, but make no mistake: this is an Unreal Engine 5 showcase through and through, and it can deliver stunning visuals in between the heated skirmishes of its pop-culture-blending brawls.

The RTX 3050 struggles to hit even a 60 FPS average at 1080p in this title. The RTX 5050 delivers a nice boost, landing midpack at just above an 80 FPS average and laying a great performance foundation for DLSS if you want it.

The RX 9060 XT duo surprises by landing at the top of the charts in this title. AMD's cards deliver slightly higher average frame rates alongside much higher 1% lows than the RTX 5060 Ti duo. If only Epic would implement FSR 4...

Apex Legends

Why it's here: Unique engine, giant player base

What it stresses: Compute

Apex Legends is another wildly popular esports title that needs no introduction. It normally has a 144 FPS cap, but you can disable it with a launch option to reach a 300 FPS ceiling.

Nvidia called out Apex as a particularly good showing for the RTX 5050 in its launch materials, and it's easy to see why. The baby Blackwell card delivers a 75% performance boost over the RTX 3050 at 1080p, even without the use of DLSS or framegen. You could even push a 144Hz 1440p monitor with this card in this title.

Marvel Rivals

Why it's here: another hugely popular free-to-play title

What it stresses: Compute

Marvel Rivals is another popular free-to-play game that might have a reputation as a bit of a potato title. Unlike other games in our suite that share its UE5 foundation, Rivals doesn’t require a superhero of a graphics card to hit high frame rates. The RTX 5050 can nearly hit a 90 FPS average at 1080p, and it can even hit 60 FPS at 1440p.

Clair Obscur: Expedition 33

Why it's here: One of the most popular RPGs of 2025

What it stresses: Compute

Clair Obscur: Expedition 33 has stunning Unreal Engine 5-powered visuals. Since it keeps VRAM usage in check even on 8GB cards, its performance is mostly defined by how much compute horsepower you can throw at it.

At 1080p Epic settings, the RTX 5050 ekes out about 40 FPS on average, which is a nice 50% boost over the RTX 3050 but still a case where you'd really want to dial back to high settings and consider using DLSS to get to a truly smooth experience.

Ratchet & Clank: Rift Apart

Why it's here: a representative older PlayStation port

What it stresses: Compute, VRAM, RT (if enabled)

Ratchet & Clank: Rift Apart is the kind of experience that will take gamers of a certain age on a nostalgic journey back to summer breaks spent in front of CRT TVs with a PlayStation 2 whirring away nearby. This title features incredibly charming art direction and animation that's still delightful to look at even two years after its arrival on PC.

Rift Apart already demands more than 8GB of VRAM with very high settings even at 1080p, which sort of shows in the relatively low 1% lows alongside high average frame rates on the RTX 5050 and its surrounding 8GB competition. Even so, the RTX 5050 delivers a fine experience, blowing the RTX 3050 out of the water with a 60% boost and taking this game from marginally playable to perfectly enjoyable on an entry-level GeForce.

Marvel's Spider-Man 2

Why it's here: A newer PlayStation port

What it stresses: Compute, VRAM, RT (if enabled)

Marvel's Spider-Man 2 is another solid PlayStation port, but like many games in this vein, it really wants both strong compute resources and plenty of VRAM from a GPU to high high frame rates. The RTX 5050 is much faster than the RTX 3050 in this title, but it's not up to the job of delivering a 60 FPS average at 1080p.

Spider-Man 2 is an example of a game where you might want to enable Multi-Frame Generation on Blackwell cards for an extra performance boost, but because it's already spilling out of 8GB of VRAM at 1080p without upscaling enabled, the feature simply didn't work on the RTX 5050 or RTX 5060.

We're sure you could get MFG going on the RTX 5050 withsome further tweaking, but the promise of RT + DLSS + MFG that Nvidia pushes so heavily with Blackwell isn’t really possible on 8GB GPUs. You might only be able to use one of those features at a time.

Stalker 2

Why it's here: A visually stunning revival of a classic PC franchise

What it stresses: Compute, VRAM (to some degree)

Stalker 2 is another beast of a UE5 game, and none of the cards in this entry-level-to-midrange menagerie can even crack 60 FPS on average at 1080p. As we've seen already, the RTX 5050 is a major improvement over the RTX 3050 in this title, but it's still not enough to raise frame rates to a playable level without turning down some settings or enabling DLSS.

Cyberpunk 2077

Why it's here: One of the biggest PC releases of all time

What it stresses: Compute, VRAM, RT (when enabled)

Cyberpunk 2077 may have been a beast of a title when it arrived five years ago, but after a couple of generations of new graphics cards and countless improvements and optimizations to the base game, everything in our suite except the RTX 3050 can run this title at a 60 FPS average or higher at 1080p. Chalk up another boost for the RTX 5050.

The RTX 5060 Ti 16GB turns in weirdly low 1% low frame rates at 1080p in Cyberpunk, and we found this result to be repeatable. We're not entirely sure what the reason is behind it, but we did try to iron it out without success.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Our ray-traced gaming test suite includes five titles we’ve already seen that feature RT as extra eye candy, plus two modern titles that require RT support to run. We’ve also thrown in GTA V Enhanced as an older title that benefits from an RT-powered visual upgrade.

First, the TL;DR: none of the $300-or-less cards we’re testing today are great for RT, even at 1080p. You should really budget for a GeForce RTX 5060 Ti if you’re serious about RT experiences. Its leading performance in our test suite, as well as its 16GB of VRAM, make it ideal for tuning RT performance with upscaling and framegen.

DOOM: The Dark Ages

Why it's here: one of a new crop of games that requires RT to run

What it stresses: Compute, RT, VRAM

Even with mandatory RT support and everything but texture pool size cranked to Ultra Nightmare, DOOM: The Dark Ages isn't terribly hard for our stable of graphics cards to run well. The RTX 5050 comes in just shy of a 60 FPS average at 1080p and should serve as a fine foundation for DLSS-enhanced gaming if you prefer.

Indiana Jones and the Great Circle

Why it's here: another of a new crop of games that requires RT to run

What it stresses: Compute, RT, VRAM

Indiana Jones and the Great Circle is a weird beast. It uses a unique fork of the id Tech engine to deliver its cinematic adventure gameplay, and it doesn't necessarily play well with Nvidia graphics cards with 8GB of VRAM.

If you crank the texture pool size high enough or enable DLSS or MFG at the ragged edge of VRAM usage on GeForce cards, you're likely to cause an annoying hard crash—something we didn't see on any of our Radeon cards.

Even the RX 9060 XT 8GB will let you max out every setting without crashing, even if it might not run that well. We're surprised at how rough the user experience in this game can be on GeForces given that The Great Circle is an Nvidia-sponsored title.

Keeping texture pool size to a minimum and maxing out every other setting (save for path tracing) is totally possible in The Great Circle on these cards, and it only barely affects image quality, so that's how we tested.

The RTX 5050 clears a 60 FPS average in our test sequence, but as with other 8GB GeForce cards, its 1% lows fall lower than we'd like for a truly smooth experience. Still, the point stands: you can have a great time in The Great Circle even on affordable hardware at 1080p.

Alan Wake II

Alan Wake II's high RT preset enables some path-traced effects for extra eye candy, in theory, but they place quite the strain on this selection of hardware. The biggest issue beyond the crushing performance demands of these settings is that these effects don't really make a visible difference in this title, so you could just as soon ignore them and enjoy much higher frame rates.

Cyberpunk 2077

Cyberpunk 2077 is one of the few RT-heavy titles we've seen where enabling the feature creates major differences in image quality, so you might actually care about its performance on these cards. What we’re really looking for here is a solid foundation for upscaling, not a native 60 FPS.

Even by that measure, the RTX 5050 can’t even break 30 FPS on average with RT enabled. If you’re looking to enable RT in this title, you really want an RTX 5060 or RTX 5060 Ti 16GB for the best experience, and only then at 1080p.

Black Myth Wukong

Black Myth Wukong is tough on graphics cards even before we apply RT, and turning on “full ray tracing” takes performance below even the 30 FPS average on all but a handful of cards in our lineup.

RT does make a visible difference in this title, so it’s worth exploring if you’ve got the graphics card to match its considerable demands, but you really need an RTX 5060 or RTX 5060 Ti 16GB or better to even start considering enabling the feature.

Ratchet & Clank: Rift Apart

As an older RT title, Ratchet & Clank: Rift Apart plays OK even on more affordable hardware like the RTX 5050 with RT enabled at 1080p, but stepping up to 1440p increases VRAM pressure to the point that you really want a 16GB card (and more raw horsepower) to really get a good experience.

As with other RT titles in our lineup, though, Rift Apart doesn’t look that much different with RT than without, so unless you’re peeping pixels, you can just leave the feature off and enjoy higher performance.

Marvel’s Spider-Man 2

Enabling RT in Spider-Man 2 puts the hurt on any graphics card that doesn’t have 16GB of VRAM to play with, even at 1080p, and matters only get worse from there. Sorry to sound like a broken record, but if you’re playing this game on a sub-$300 graphics card, enabling RT really isn’t worth the performance hit.

Grand Theft Auto V Enhanced

It’s hard to believe we’re still testing GTA V more than 10 years after its arrival on the PC, but here we are. The latest Enhanced re-release adds appealing RT eye candy to Los Santos, and its demands on hardware are modest enough that even modest hardware can provide a solid enough performance foundation for upscaling.

We didn’t test the raster version of GTA V Enhanced as a baseline, as most any GPU can run it well. With RT maxed out, the RTX 5050 can’t crack 60 FPS on average, but it at least can provide a good baseline before you turn on DLSS (which you’ll want to do for the best image quality anyway).

Roll up all of our RT and raster results into one chart by the power of geomeans, and the RTX 5050 falls just a tiny bit behind the RTX 4060 and Arc B580, but not by enough to change any overall standings. What this analysis does emphasize is that you really want to step up to an RX 9060 XT 8GB, RTX 5060, or ideally an even higher-end card if ray-traced gaming is something you want to explore, as these cheaper cards fall just short of having enough oomph to deliver a solid baseline experience.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Used alongside the FrameView software that we use to capture performance results, Nvidia’s PCAT hardware allows us to capture live power consumption data with every frame, and we can use that data to communicate real-world power usage figures that are more precise than a worst-case total board power rating.

Power consumption taken in isolation doesn't mean much, though. A card can be both highly efficient and still draw a lot of power to turn in a high overall performance level, and it can also draw a modest amount of power and still deliver relatively low performance per watt.

To express power efficiency, we simply divide a card's average frames per second by its average power consumption across all of our tests. We stuck with our 1080p results for this analysis, as these cards' relatively low performance at 1440p and 4K would make efficiency discussions with those results more academic than anything.

By this measure, the RTX 5050 is much more efficient than the Ampere-powered RTX 3050, but it’s disappointingly inefficient for a Blackwell card. The RTX 5060 turns in 20% higher frames per watt at 1080p, and the RTX 5060 Ti duo is even better still. Any generational efficiency gains Nvidia made in the GB207 GPU on the RTX 5050 appear to be offset entirely by sticking with fast GDDR6 memory at the board level. On net, we end up with efficiency even slightly worse than that of the RTX 4060.

For reference, we've also included the geomean of clock speeds and temperatures across all our cards at the tested resolutions. Some data is missing due to driver and/or software hiccups, which we'll correct in future testing.

We don't think there are any surprises in these results. The RTX 5050 clocks much higher than its rated boost speed, a testament to Nvidia's GPU Boost logic and the thermal and electrical headroom that even this modest card apparently boasts.

The Gigabyte card is the hottest-running of this bunch, but it's also got one of the smallest coolers among the cards we tested, and it's still well below any level that would represent cause for concern.

Compared to the three-and-a-half-year-old RTX 3050 8GB, the RTX 5050 is a solid improvement - and it certainly ought to be, given that Nvidia didn't make an RTX 4050 for the desktop in the meantime. The baby Blackwell card is 60% faster on average than its entry-level Ampere predecessor in our raster results at 1080p (all without any help from DLSS or MFG).

If you squint, the 5050 brings RTX 4060 performance to a $50 lower price point, and given that the 4060 is the most popular graphics card on Steam right now, that sounds like a potentially appealing combo on paper.

Stacked up against other Blackwell cards, though, the RTX 5050 looks less impressive as a generational improvement. The RTX 5060 turns out 20% higher frames per watt at 1080p for raster titles despite consuming just 10% more power in our real-world tests, which emphasizes just how much of Blackwell's perf-per-watt improvements come from the use of GDDR7 at the board level.

The RTX 5050 doesn't suck down power by any means, but it's slightly less efficient than even the RTX 4060 despite being a generation newer. Oops.

If the RTX 5050 was priced in proportion to its raw performance gap with the RTX 5060, our results suggest it really should be a $229 card. At that price point, it would make life extremely difficult for the Arc B570 and B580 for 1080p gaming, and it wouldn't face any AMD competition at all in most markets.

For the $249 Nvidia wants for these cards in reality, we think the 5050 is an overpriced, relatively inefficient product that faces tough challengers from AMD and Intel alike, as well as more appealing options within Nvidia's own lineup. It also feels insulting that this card can't consistently deliver the Blackwell architecture's marquee experiences.

DLSS Multi-Frame Generation sometimes just doesn't work on the RTX 5050 in the titles that could really use the boost, because those demanding games already eat up all of the 5050's 8GB of VRAM, leaving no room for the MFG AI model to reside in local memory.

If you end up with an RTX 5050 in a pre-built system or because you received one as a gift, you certainly won't have a bad time with it at 1080p. Outside of Microsoft Flight Simulator 2024, which really, really wants a powerful GPU with 16GB of VRAM to deliver anything approaching consistent performance even at 1080p, we had a perfectly fine time with 1080p gaming on the RTX 5050. The Gigabyte card we tested runs cool and quiet, too, so it's easy to live with. But if you have any say in the matter, you can do much better for yourself by saving up and spending just a bit more money.

The real spoiler for every $300-or-less graphics card available right now is AMD's Radeon RX 9060 XT 8GB, which has consistently been available for $269 on Amazon of late. At that price point—$30 under its $299 MSRP—it delivers 25% higher frame rates on average than the RTX 5050 for just 8% more money. In raster games that aren't limited by VRAM, which describes many of today's most popular titles on the PC at 1080p, the 9060 XT wipes the floor with everything further down the ladder, and it isn't even close.

Try to render games at a native 1440p, and the 9060 XT 8GB does start to show some cracks, as evidenced by its 1% low frame rates. Of course, you can get around those challenges by lowering settings a bit, using FSR 4 upscaling, or both. You can usually work around 8GB of VRAM in general, but you can't add more raw compute horsepower to begin with, and the RX 9060 XT 8GB has the most raw muscle per dollar of anything on the market if you find it on sale.

If you prefer to stay in the green team's corner to keep access to DLSS 4 and MFG, the RTX 5060 makes a strong case for itself at $300. It delivers 27% higher performance than the RTX 5050 at 1080p for 20% more money. This card's high standings in our overall results at 1080p (and 1440p) show that even if the RX 9060 XT 8GB can outrun and outgun it from time to time, the 5060 hits back by maintaining a more consistent level of performance across our test suite, even as the 9060 XT 8GB sometimes stumbles.

Pair that consistently high performance with wickedly good power efficiency, and it's easy to understand why the RTX 5060 is storming up the Steam Hardware Survey charts of late.

Despite its age, Intel's Arc B580 stays in the affordable gaming fight thanks to some recent price decreases. It can finally be had for its $249 MSRP from some Intel board partners on Newegg. It still delivers solid enough gaming performance at both 1080p and 1440p, along with an ample 12GB of VRAM for better game compatibility across a wider range of resolutions and settings.

The challenge for Intel is that Battlemage can exhibit big performance falloffs even in extremely popular games that aren't VRAM-limited (like Fortnite) that sometimes put it behind even the RTX 5050. Those performance cliffs make it tough to generally recommend.

You really need to make sure your favorite titles play well on Arc before buying a B580 (or B570), and you need to be prepared for the possibility that future titles might not run as well as you'd expect. And that's before we touch on Battlemage's power efficiency, which trails everything but the RDNA 3 cards we tested. For the money, though, you might not care.

Putting the spotlight back on the RTX 5050, this card feels like something Nvidia had to make to keep its system integrator partners happy rather than something it really wanted to put on store shelves for enthusiasts.

The Dells, HPs, and Lenovos of the world that need to build cheap gaming PCs for buyers at Wal-Mart and Best Buy now have access to a product that says RTX 50 rather than RTX 30 on the shelf sticker, and for more casual audiences who mostly spend time in wildly popular free-to-play games like Counter-Strike 2, Fortnite, Apex Legends, or Marvel Rivals, this card provides a much-needed performance boost over the ancient RTX 3050. But price-conscious enthusiasts who are waiting for a true no-compromise, game-changing product around the $250 mark are still going to find themselves tapping their feet.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

The Nvidia B30A, which the company is developing as its next-generation replacement for the China-market H20 AI chip, is expected to cost double the price of the earlier model. Reuters reports that the H20 currently sells between $10,000 and $12,000, meaning the B30A will likely be priced from $20,000 to $25,000. Despite this massive increase, many Chinese companies, like TikTok owner ByteDance, are keen on getting their hands on Nvidia’s latest chips, with some sources saying that these chips are considered great deals.

It’s estimated that the B30A will be six times more powerful than the H20, despite being a watered-down version of the full-fat B300 AI chip. Nvidia has reportedly already developed the B30A, but it’s still waiting for approval from the U.S. government to proceed with the marketing and sale of the Blackwell-based GPU. In the meantime, many Chinese tech firms are still buying H20 chips despite Beijing’s guidance to stop buying them.

U.S. President Donald Trump has previously banned the sales of H20 chips, resulting in a $5.5-billion write-off for Nvidia. However, he reversed course in mid-July 2025, allowing the company to resume deliveries to Chinese customers. But instead of a blanket provision allowing it to sell the H20 to anyone, the U.S. instead started issuing export licenses to Nvidia, allowing it to ship its products.

Because of this, there’s currently a massive license backlog at the U.S. Department of Commerce —the worst it has experienced in 30 years — turning the H20 taps into a trickle. The White House is also still ironing out the 15% deal that AMD and Nvidia signed, so that the companies can market their products to China. We’ve also seen reports that Nvidia is asking its suppliers to wind down H20-related production, with the company only telling Tom’s Hardware, “We constantly manage our supply chain to address market conditions.”

So, with all these delays, red tape, and rumors, many of Nvidia’s Chinese customers are getting apprehensive and want assurance that their H20 orders are delivered. This shows that there is still massive demand for these powerful chips, thus turning it into a massive geopolitical tool to be wielded on the negotiating table by both the East and the West.

Early RTX 5090 V2 benchmarks have been published overseas, revealing a surprising trait of Nvidia's latest Chinese flagship GPU. Chinese review site Yesky benchmarked the RTX 5090D and RTX 5090D V2 head-to-head and discovered the latter performs on par with the former in gaming performance, despite the V2 variant featuring 33% less memory bandwidth.

Yesky benchmarked the RTX 5090D and RTX 5090D V2 in several titles, featuring the full gauntlet of 3DMark benchmarks: Steel Nomad, Speed Way, Port Royal, Fire Strike Ultra, Fire Strike (the vanilla version), TimeSpy Extreme, TimeSpy (the vanilla version), and several games Cyberpunk 2077, Hogwarts Legacy, some mysterious PC game Google Translate is dubbing "Marvel Brawl", Borderlands 3 and Far Cry New Dawn.

In all of the benchmark runs Yesky performed, the RTX 5090D V2 maintained virtually identical performance to the RTX 5090D, with any gaps being pretty marginal. In 3DMark Speed Way, the RTX 5090D was 3.5% faster than the RTX 5090D V2, and in Fire Strike, the RTX 5090D was 2.2% quicker. Speed Way and Fire Strike represented some of the best results for the RTX 5090D, with benchmark results from the rest of the 3DMark suite producing even lower margins.

In Cyberpunk 2077, the RTX 5090D reportedly produced 424 FPS with DLSS 4, and in the same test, the RTX 5090D V2 produced a nearly identical 418 FPS. This behavior was present in the rest of Yesky's testing suite. Hogwarts Legacy achieved 300 FPS on the RTX 5090D and 297 FPS on the 5090D V2 with DLSS 4. "Marvel Brawl" produced 577 FPS on the RTX 5090D and 569 FPS on the RTX 5090D V2.

Even without DLSS 4 or any DLSS upscaling, the two GPUs are closely tied. Borderlands 3 at 4k resolution — a resolution that is normally sensitive to memory bandwidth, the RTX 5090D produced 164 FPS, and the RTX 5090D V2 produced 161 FPS. The same behavior was also present in Far Cry: New Dawn.

Yesky's benchmarking results demonstrate that the RTX 5090D V2 suffers virtually no performance loss from its neutered 384-bit 24GB memory configuration compared to the RTX 5090D's 32GB 512-bit configuration. More interestingly, Yesky's benchmarks appear to suggest that Nvidia's Blackwell architecture cannot take advantage of the RTX 5090 and RTX 5090D's 512-bit memory interface for gaming, even at 4k resolution.

This could be the reason Nvidia has not dabbled with a 512-bit memory interface until now with the RTX 5090 series. Past Nvidia flagships, including the RTX 4090, RTX 3090, and RTX 2080 Ti, all featured 384-bit memory interfaces. Even the highly regarded GTX 1080 Ti only came with a 352-bit interface. By contrast, AMD was dabbling with 2048-bit interfaces at the time of Pascal's debut in cards such as the RX Vega 64.

We already know that architectural advancements (like Ada Lovelace's L2 cache) and speedier memory ICs can make up for skinnier bus widths, but it is very interesting to know that Nvidia's architectures might not be capable of taking advantage of 512-bit interfaces for gaming.

The RTX 5090D V2 is Nvidia's third RTX 5090 variant and the second RTX 5090D model tailor-made to comply with U.S. export trade restrictions aimed at China. The RTX 5090D V2 is a neutered version of the RTX 5090D with "just" 24GB of GDDR7 memory and a 384-bit interface, matching the memory configuration of past Nvidia flagships like the RTX 4090 and RTX 3090. Potential buyers will be buoyed by these early performance scores, given that the V2 costs the same as the outgoing 5090D, despite the lowered specs.

Following the ban of its Hopper H20 accelerations in China, Nvidia is reportedly planning on launching new Blackwell-based solutions at a lower price this year, per Reuters. With mass production anticipated by June, we can expect these solutions to be widely available in the Chinese market by Q3 or Q4. While technical details are still emerging, we can already discern some important details and specifications.

Due to stringent U.S. export policies targeting China, the Hopper family has largely been a cat-and-mouse chase between U.S. regulators and Nvidia. Even before their official debut, the flagship H100 and H200 accelerators were already subject to export bans. Nvidia introduced the H800 to circumvent these regulations, which eventually faced a similar fate in October 2023. The cut-down H20 served as Nvidia's primary AI solution for the Chinese market in the interim until its recent ban under the current administration last month, which forced Nvidia to write off $5.5 billion in GPU supply.

Reuters reports that Nvidia's follow-up to the H20 will be based on the Blackwell architecture, more specifically, the RTX Pro 6000D. Further clarification by tipster Jukanlosreve at X, citing a report from China's GF Securities, suggests the RTX Pro 6000D will be dubbed B40 (likely a successor to the Ada Lovelace L40). Reuters classifies this as a server-class GPU which uses traditional GDDR7 memory instead of HBM, and notably avoids the use of TSMC's CoWoS packaging technology, likely signaling at its monolithic nature.

Silicon possibilities and B40 pricing

Nvidia has reportedly broken another AI world record, breaking the 1,000 tokens per second (TPS) barrier per user with Meta's Llama 4 Maverick large language model, according to Artificial Analysis in a post on LinkedIn. This breakthrough was achieved with Nvidia's latest DGX B200 node, which features eight Blackwell GPUs.

Nvidia outperformed the previous record holder, SambaNova, by 31%, achieving 1,038 TPS/user compared to AI chipmaker SambaNova's prior record of 792 TPS/user. According to Artificial Analysis's benchmark report, Nvidia and SambaNova are well ahead of everyone in this performance metric. Amazon and Groq achieved scores just shy of 300 TPS/user — the rest, Fireworks, Lambda Labs, Kluster.ai, CentML, Google Vertex, Together.ai, Deepinfra, Novita, and Azure, all achieved scores below 200 TPS/user.

Intel has announced the unveiling of three new Intel Xeon 6 P-Core CPUs that it says are designed specifically to handle the most advanced GPU-powered AI systems. The new processors will debut in Nvidia's DGX B300 AI systems.

The new processors, replete with Intel's Performance-cores, also feature new Intel Priority Core Turbo (PCT) and Intel Speed Select Technology – Turbo Frequency, which the company claims delivers customizable CPU core frequencies to improve GPU performance for demanding AI workloads.

All three are available now, and the Intel Xeon 6776P also comes integrated in the Nvidia DGX B300, the company's latest AI-accelerated systems.

Intel says that the introduction of PCT and Intel SST-TF as a pairing marks "a significant leap forward in AI system performance." PCT should allow for the dynamic prioritization of high-priority cores, enabling higher turbo frequencies. Meanwhile, lower-priority cores operate at base frequency in parallel to optimize resource distribution. PCT can reportedly run up to eight, high-priority cores at elevated turbo frequencies, according to Intel.

Intel's Xeon 6 CPUs include up to 128 P-cores per CPU and 20% more PCIe lanes than previous-generation Xeon processors, with up to 192 PCIe lanes per 2S server. Intel also claims Xeon 6 offers 30% faster memory speeds compared to the competition (specifically the latest AMD EPYC processors), thanks to Multiplexed Rank DIMMs (MRDIMMs) and Compute Express Link, and up to 2.3x higher memory bandwidth compared to the previous generation.

Intel says its P-Core Xeon 6 processors have 2 DIMMs per channel (2DPC), and says the 2DPC configuration supports up to 8TB of system memory. It also says MRDIMMs boost bandwidth and performance, all while reducing latency. The new CPUs also feature up to 504 MB L3 cache for faster data retrieval.

Intel Xeon 6 processors also feature Intel AMX, which can offload certain tasks to the CPU. Intel confirmed AMX now features support for FP16 precision arithmetic, which enables efficient data pre-processing and critical CPU tasks in AI workloads. Alongside its three new P-Core processors, Intel has also added a B-variant 6716P-B.

There are dozens of solid-state drives with a PCIe 5.0 interface over at Computex 2025, and they no longer surprise us; however, we have seen a drive that features a PCIe 6.0 x4 interface and can potentially boast a 30.25 GB/s sequential read and write speed at the trade show.

Micron’s 9650 Pro SSD with a PCIe Gen6 interface will unlikely launch anytime soon, but for now, the unit is an important test vehicle for companies like Astera Labs, which plays an important role in developing next-gen AI platforms. At the show, Astera Labs used Micron’s PCIe 6.0 SSD to demonstrate its Scorpio PCIe 6.0 4x16 switch, Aries Bandwidth-matching Gearbox software, and Aries 6 timers.

Although Nvidia’s Blackwell GPUs feature PCIe 6.0 x16 connectivity, no CPU platforms formally support PCIe 6.0. Nonetheless, PCIe Gen6 SSDs like the Micron 9650 Pro can be quite useful when utilized with platforms featuring PCIe 6.0 switches. For example, PCIe 6.0 switches can enable peer-to-peer communications between AI GPUs and SSDs, bypassing the CPU. In addition, when paired with Astera’s Gearbox software and hardware, they can reduce the number of PCIe 6.0 lanes required to work with PCIe 5.0 hosts (e.g., to achieve a PCIe 5.0 x8 performance only for PCIe 6.0 lanes are needed) and therefore enable instalation of more drives into a box, which can be critical for some AI systems.

Certification of PCIe 6.0 devices by PCI-SIG was delayed from mid-2024 to the second half of 2025, so while GPUs like Nvidia’s Blackwell support the technology, they have not passed official interoperability tests. As for Micron’s 9650 Pro SSD, the unit used by Astera is marked as EVT3, so this is the third revision of the Engineering Validation Test. Generally, this means that the product has undergone at least two prior engineering validation builds (EVT1 and EVT2), and the third build is undergoing further validation and testing.

Although Nvidia’s GeForce RTX 5060 is massively slower than higher-end Blackwell models, it has one indisputable advantage over its bigger brethren: low power consumption. This opens doors for makers of graphics cards to build GeForce RTX 5060 add-in-boards in a variety of form-factors to address different systems. And Zotac is one of the companies that does not hesitate to build miniature GeForce RTX 5060 cards.

The company is demonstrating a low-profile GeForce RTX 5060 as well as Mini-ITX GeForce RTX 5060 at Computex. Both boards are based on Nvidia’s GB206 graphics processing unit with 3840 CUDA cores and carry 8 GB of GDDR7 memory with a 128-bit interface, just like full-size boards. However, while the Mini-ITX version has four display connectors, the low-profile AIB has only three display outputs.

As for coolers, the Zotac Gaming Mini-ITX GeForce RTX 5060 comes with a rather sophisticated single-fan cooling system with multiple heat pipes, whereas the low-profile product is equipped with a dual-slot triple-fan cooler, which is something to be expected from a graphics board with a 145W total board power. To enhance the compatibility of both cards with inexpensive systems, Zotac equipped them with an eight-pin auxiliary PCIe power connector, which is perfectly fine for such devices.

As both gamers and makers of PC hardware try to squeeze all the juice from parts like graphics cards, the latter became quite huge in the last 10 years as they tend to come equipped with huge coolers no matter how much power they actually consume. To that end, it became close to impossible to find a gaming graphics card in low-profile and Mini-ITX form-factor. Fortunately, there are companies like Zotac, which tend to address rather niche markets with more or less decent offerings for gaming, providing an upgrade path for compact systems.

Zotac plans to start selling its low-profile and Mini-ITX versions of GeForce RTX 5060 graphics cards this summer, though it remains to be seen whether these boards will be available at price points close to GeForce RTX 5060’s MSRP of $299, or the company charges a premium for exclusivity as small graphics adapters are rare these days.

New, possible details about a potential RTX 5080 Super have emerged, thanks to an X posting from well-reputed hardware leaker Kopite, who has a solid track record when it comes to all things Nvidia. The RTX 5080 Super is alleged to address GPU memory limitations with 24GB of fast GDDR7 memory, and a TGP (Total Graphics Power) that's rumored to exceed 400W. The leaker has not shared any details in regards to pricing and availability.

GDDR7 memory modules are currently available in 16Gb (2GB) and 24Gb (3GB) densities, with even higher capacities down the pipeline. The denser option directly translates to a 50% increase in VRAM capacity, even if the memory bus width remains the same.

Despite sharing the same 256-bit bus width as the RTX 5080, the RTX 5090 Mobile achieves its 24GB configuration with denser memory modules. The remainder of the Blackwell family sticks with standard 16Gb modules, similar to GDDR6X. However, this new leak suggests Nvidia might be eying transitioning to denser 24Gb options for the RTX 50 Super refresh.

The RTX 5080 Super reportedly employs the full-fat GB203 die, similar to the RTX 5080, with 10,752 CUDA cores or 84 Streaming Multiprocessors. The 256-bit interface enables eight memory modules, which have been populated with 24Gb (3GB) modules for 24GB of GDDR7 memory, rated at 32 Gbps. This puts the memory bandwidth at an impressive 1 TB/s, or 6.6% faster than the stock RTX 5080. Nvidia is also said to increase the power requirements for the RTX 5080 Super over 400W.

Nvidia CEO Jensen Huang is speaking live from Computex Taipei and the live stream is embedded below! The event takes place at 11 am Taiwan time which is 11 pm ET, 8 pm PT and 4 am BST.

We expect Huang to talk a lot about AI and Nvidia's industry-leading hardware. He'll likely make some announcements and we can imagine many will have to do with datacenter computing, but perhaps there will be some tidbits for desktop users as well.

Watch the feed above to find out.

Despite withholding pre-release RTX 5060 drivers from reviewers, Nvidia has reportedly given select outlets access to its upcoming RTX 5060 driver to create strict "preview" articles about the new mid-range Blackwell GPU. One such outlet, Ascii.jp, published an article benchmarking the RTX 5060 against the RTX 5060 Ti (8GB and 16GB) and other graphics cards in Doom: The Dark Ages, Cyberpunk 2077, and Marvel Rivals.

GameStar Tech claims there are strict Nvidia guidelines for the RTX 5060 previews. According to the German publication (machine translation), "The reason for this unusual situation: Nvidia has decided to only make the driver, which is essential for GPU testing, available in advance under certain conditions and to a very limited number of media outlets worldwide. This includes us."

The news outlet goes on to state, "What's particularly crucial is that we weren't able to freely choose which graphics cards and games we would measure and with which settings for this preview."

Apparently, previews could only compare the RTX 5060 to the RTX 3060 and RTX 2060 Super. The list of allowed titles is reportedly limited to Avowed, Doom: The Dark Ages, Marvel Rivals, Cyberpunk 2077, and Hogwarts Legacy. Meanwhile, the settings must be configured to 1080p resolution, ultra image quality, DLSS in quality mode, and ray tracing.

Nvidia GeForce RTX 5060 Benchmarks

Surprisingly, Ascii's RTX 5060 "preview" only included Doom: The Dark Ages, Cyberpunk 2077, and Marvel Rivals. However, the Japanese outlet did compare the RTX 5060 to several generations of GeForce x060-class GPUs, including the RTX 2060, RTX 3060 (12GB), RTX 4060, RTX 5060, and both variants of the RTX 5060 Ti.

Graphics settings were also limited to running frame generation exclusively, with the RTX 3060 and RTX 2060 running FSR 3 frame generation (since they don't support the DLSS counterpart).

Across all three titles, the new RTX 5060 trails the RTX 5060 Ti by 15% on average using 2x DLSS frame generation. If it wasn't obvious, take these results with a massive grain of salt. Frame generation is not a perfect multiplier of native performance, so there is bound to be a bit of variance between all six GPUs.

However, the RTX 5060 and RTX 5060 Ti comparisons can be taken with more weight since they rely on the same GPU architecture and use identical frame generation techniques. We wouldn't be surprised if the performance difference between the two GPUs is very similar with frame generation turned off.

In Doom: The Dark Ages, the RTX 5060 was 17% faster than the RTX 4060, 35% faster than the RTX 3060, and over three times faster than the RTX 2060.

Focusing on the RTX 5060's performance against the RTX 5060 Ti, the RTX 5060 Ti 8GB was 15% faster than the RTX 5060 (again using 2x frame gen). The 8GB and 16GB variants perform the same in this title. Ultra nightmare settings were used for the graphics settings, with DLSS used on the RTX 4060 and above (FSR 3 was used on the older GPUs).

In Cyberpunk 2077, the RTX 5060 was 25% faster than the RTX 4060, 81% faster than the RTX 3060, and nearly four times faster than the RTX 2060. Against the RTX 5060 Ti, the 8GB variant was 15% faster than the RTX 5060, and the 16GB variant was 17% faster.

RT overdrive was used for the graphics settings with the DLSS 4 transformer model using the Quality preset on the RTX 4060 and above. The RTX 3060 and RTX 2060 exclusively used FSR 3 frame generation, with no notes on whether or not the two GPUs used FSR 3 upscaling.

In Marvel Rivals, the RTX 5060 was 16% faster than the RTX 4060, 48% faster than the RTX 3060, and 85% faster than the RTX 2060. Compared to the RTX 5060 Ti, the Ti variant was 14% faster than the vanilla RTX 5060. Again, the 16GB had worse performance than the 8GB variant of the RTX 5060 Ti in this title so that we can chalk it up to the margin of error.

Image quality was set to best with the DLSS 4 transformer model used on the RTX 4060 and above. This time, DLSS upscaling was used on the RTX 3060 and 2060 in conjunction with FSR 3 frame generation (though there are no notes on what version of DLSS was used on these GPUs).

This is the first time Nvidia has greenlit a "preview" of its future GPU products. Traditionally, reviewers are given one to two weeks (sometimes more) with a pre-release driver to create a review of a future Nvidia GPU that will go live before the GPU launches. For the RTX 5060, Nvidia has opted to stop reviewers from starting their GPU testing until launch day (May 19) when the card releases, forcing reviewers to publish their reviews days after the RTX 5060 release. In the meantime, Nvidia is using "preview" articles to stir hype for its upcoming mid-range GPU.

Two outlets have benchmarked Nvidia's newest RTX Pro 6000 Blackwell GPU, revealing its full capabilities against the RTX 5090 and other GeForce GPUs. Privaterbok on Reddit revealed eight benchmarks on the new workstation GPU, featuring 3DMark and Geekbench 6 benchmark runs.

The Redditor published benchmark runs of the Blackwell-based workstation GPU in TimeSpy, TimeSpy Extreme, Steel Nomad, Port Royal, and Geekbench 6. On average, the RTX Pro 6000 is roughly 10-15% faster than a stock RTX 5090, and even faster with an overclock, according to RTX 5090 data from Videocardz.

For benchmark comparisons, we paired the Redditor's RTX Pro 6000 GPU benchmarking data against the best RTX 5090 results from 3DMark's browser. All of these 3DMark results take advantage of highly overclocked RTX 5090s, giving us an idea of what the workstation GPU is capable of compared to the fastest consumer GPU Nvidia makes currently on full tilt.

The RTX Pro 6000 at stock speeds loses to the overclocked RTX 5090 results, with the RTX 5090 being roughly 5-10% quicker on average. Looking at the overclocked RTX Pro 6000 results, the GPUs are on more equal footing, though the RTX Pro 6000 still loses in Time Spy with the RTX 5090 being 5% quicker. However, in Time Spy Extreme, the workstation GPU gains a slight win with a score 1% better than the RTX 5090.

These are by no means perfect comparisons; each 3DMark RTX 5090 score probably comes from a different person with different setups and different overclocks.

Geekbench 6 is more straightforward; we took the RTX 5090 results straight from the Geekbench website, which are officially validated. In Geekbench 6, the RTX Pro 6000 is anywhere between 9% and 15% faster than the RTX 5090.

Level1Techs also put the RTX Pro 6000 through some gaming-centric benchmarks, with several runs of Cyberpunk 2077. At ultra settings running at 4K with DLSS on auto (i.e., performance mode) with path tracing off, the RTX Pro 6000 achieved 127.02 frames per second, and with path tracing on 92.54 FPS

The RTX Pro 6000 is Nvidia's latest high-end workstation GPU (a Titan-like card) and successor to the RTX Pro 5000 (which was also Blackwell-based). The RTX Pro 6000 has 10% more CUDA cores than the RTX 5090 (24,064) and 96GB of GDDR7 memory. Outgoing pricing for the RTX Pro 6000 is in the $8,000 range.

Right now, AMD's Instinct MI300-series accelerators are aimed at both AI and HPC, which makes them universal but lowers maximum performance for both types of workloads. Starting from its next-generation Instinct MI400-series, AMD will offer distinct processors for AI and supercomputers in a bid to maximize performance for each workload, according to SemiAnalysis. However, there might be a problem with the scalability of these compute GPUs.

AMD plans to offer the Instinct MI450X for AI and the Instinct MI430X for HPC sometime in the second half of 2026. Both processors will rely on subsets of the CDNA Next architecture, but will be tailored for low-precision AI compute (FP4, FP8, BF16) or high-precision HPC compute (FP32, FP64). Such bifurcation of positioning will enable AMD to remove FP32 and FP64 logic from MI450X as well as FP4, FP8, and BF16 logic from MI430X, therefore maximizing die space for respective logic.

In addition to workload optimizations, AMD's Instinct MI400-series accelerators will also feature not only Infinity Fabric but also UALink interconnections, which will make them some of the first AI and HPC GPUs to feature UALink, a technology designed to challenge NVLink. But there is a major problem with UALink.

Nvidia and MediaTek's hotly anticipated AI PC chips may not see meaningful shipment volumes to the mass market until the second half of 2026, according to a new report. As reported by Digitimes, Nvidia and MediaTek are expected to jointly unveil their new 'N1' Arm chips for Windows PCs at Computex.

According to Digitimes, the joint chip will "likely debut under the Nvidia brand," with both N1X and N1 models planned, echoing previous reports. Digitimes says that both companies are well into production ramp-up, however, states "insiders believe meaningful shipment volumes won't emerge until the second half of 2026."

The chips, which we expect to be unveiled at Computex next week, will mark the second offspring of Nvidia and MediaTek's newly-minted partnership, inaugurated by the GB10 chip unveiled earlier this year.

As the report notes, early N1X benchmarks hint at performance that lags behind some Arm-based PC chips, noting "the results have raised industry concerns."

Perhaps more worryingly, Digitimes reiterates reports that there are "unresolved integration issues with endpoint devices." These manufacturing headaches have previously been reported elsewhere, and could explain the hefty lead time of 2H26 on these chips.

An AIB is reportedly preparing a dual-GPU variant of Intel's rumored Arc B580 24GB, totalling 48GB of VRAM on a single board, via VideoCardz. According to the source, this model is slated to be revealed at Computex 2025, which is just days away at this point. Specific details like the AIB, interconnect technology, and price are in the dark, but we can expect more details at the trade show, if the rumors are true.

This is the third leak that references a 24GB edition of the Arc B580, following EEC filings from Maxsun and an earlier slip-up from Sparkle. It's quite surprising that a GPU that's supposed to rival the RTX 4060 is getting a 24GB memory configuration. Many PCs don't even possess that much system memory. These cards are intended as a cost-effective solution for AI/ML developers, where similar capacity cards from AMD and Nvidia carry a steeper price tag. The most affordable Blackwell GPU with 24GB of memory is the RTX Pro 4000, costing over $1,500 based on preliminary listings.

An AIB is apparently doubling down on this approach by building a dual-GPU solution, housing two of these rumored 24GB B580 GPUs on a single PCB. Keep in mind, this product is reportedly a one-time design from the AIB, not a standard reference model from Intel.

Traditionally, multi-GPU setups nowadays, like Nvidia's B200 and Apple's M1 Ultra, rely on their own advanced interconnect solutions like NVLink and UltraFusion. While Intel does have Xe Link, it likely isn't compatible with the B580 and would be too costly for a one-off project. The most probable contender is a PCIe bridge linking the GPU's interfaces, allowing them to communicate through one slot.

A new report claims that Nvidia has recently raised the official prices of nearly all of its products to combat the impact of tariffs and surging manufacturing costs on its business, with gaming graphics cards receiving a 5 to 10% hike while AI GPUs see up to a 15% increase.

As reported by Digitimes Taiwan (translated), Nvidia is facing "multiple crises," including a $5.5 billion hit to its quarterly earnings over export restrictions on AI chips, including a ban on sales of its H20 chips to China.

Digitimes reports that CEO Jensen Huang has been "shuttling back and forth" between the US and China to minimize the impact of tariffs, and that "in order to maintain stable profitability," Nvidia has reportedly recently raised official prices for almost all its products, allowing its partners to increase prices accordingly.

Despite the hikes, Digitimes claims Nvidia's financial report at the end of the month "should be within financial forecasts and deliver excellent profit results," driven by strong demand for AI chips outside of China and the expanding spending from cloud service providers.

The report states that Nvidia has applied official price hikes to numerous products to keep its earnings stable, with partners following suit. As an example, Digitimes cites the RTX 5090, bought at premium prices upon release without hesitation, such that channel pricing "quickly doubled."

The report notes that following the AI chip ban, RTX 5090 prices climbed further still, surging overnight from around NT$90,000 to NT$100,000, with other RTX 50 series cards also increasing by 5-10%. Digitimes notes Nvidia has also raised the price of its H200 and B200 chips, with server vendors increasing prices by up to 15% accordingly.

According to the publication's supply chain sources, price hikes have been exacerbated by the shift of Blackwell chip production to TSMC's US plant, which has driven a significant rise in the price of production, materials, and logistics.

There is some hope that the measures could be temporary; however, following the news that the US and China have agreed on a trade deal that should cut tariffs by 115%, thanks to a 90-day pause on reciprocal tariffs. As with latent price increases being passed on to consumers, however, it could be some time before prices start to fall.

In the annals of the RTX 40 generation, the RTX Titan Ada reigned as the elusive and mysterious Bigfoot of its time. Various sources unearthed prototypes of this GPU, though Nvidia never released this beast for the mainstream. Roman 'Der8auer' Hartung has now managed to secure a prototype of this card, and in a new video, the overclocking expert showcased its capabilities across various games and synthetics, revealing what could have been.

Nvidia's Titan-class GPUs occupied a unique space among enthusiasts, outfitted with the fastest dies and configurations of their respective generation. While being powerhouses for workstations, they also managed to beat their consumer equivalents in gaming as well. The Titan RTX, revealed in 2018, was the last of its kind and these GPUs have since been superseded by Nvidia's 90-class GPUs: RTX 3090, RTX 4090, and RTX 5090.

The prototype carries a massive quad-slot cooler design, shadowing even the RTX 4090 Founders Edition. The card's shroud is stamped with the "Titan" designation. Der8auer defines it as "The biggest GPU I've ever held in my hand." Internally, the PCB has been mounted vertically to the side, as reflected by the I/O ports at the rear. Specifications-wise, we're looking at a fully-enabled AD102 chip with all 144 Streaming Multiprocessors functional, equating to 18,432 CUDA cores, 12.5% more than the RTX 4090. This is complemented by 48GB of G6X memory, achieved by placing 24x 16Gb modules in clamshell mode. We've already seen images of this GPU's PCB previously.

Der8auer says this GPU has a 450W TGP, despite the dual 12V-2x6 design, and that's probably a limitation of the vBIOS used. The driver used dates back to 2023, but the YouTuber didn't provide further details, likely to protect their sources.

Nvidia and MediaTek are expected to introduce their jointly developed Arm-based processors for PCs at Computex 2025, according to ComputerBase.

The upcoming chips — N1X and N1 — are aimed at desktops and laptops and mark Nvidia's deeper entry into the Windows-on-Arm ecosystem. But retail availability may be delayed until 2026 due to unresolved technical hurdles, reports German news site Heise, citing SemiAccurate.

CEOs of both companies — Jensen Huang from Nvidia and Rick Tsai from MediaTek — are scheduled to deliver back-to-back presentations at the Computex trade show in Taipei, Taiwan, on May 19 and May 20, respectively. One likely announcement is a new family of processors combining MediaTek's Arm-based CPU with Nvidia's Blackwell GPU. The collaborative products will rely on the companies' technologies to build the GB10 platform for compact AI workstations.

Through its partnership with MediaTek, Nvidia hopes to address markets that are currently served by AMD's APUs with high-performance Radeon graphics as well as Arm-based Snapdragon X processors. Nvidia's discrete Blackwell GPUs promise to offer higher performance and better compatibility with games than AMD's Radeon and Qualcomm's Adreno, so without any doubt, it will attract the attention of gamers.

According to early reports, the N1X and N1 processors are expected to feature up to 10 Cortex-X925 high-performance cores and up to 10 Cortex-A725 cores, though less powerful CPU configurations will likely be introduced to address markets currently underserved by Qualcomm and its Snapdragon X processors as well as AMD's Ryzen APUs with built-in Radeon graphics processors.

To support the production of an unknown high-volume product, MediaTek has reportedly secured a large amount of packaging capacity for flip chip ball grid array (FCBGA) chips, which implies their use in PCs rather than mobile devices. According to DigiTimes, this capacity reservation occurred around the end of 2024 and was unusually large. However, it remains unclear if the packaging timeline aligns with the initial production of processors jointly designed by MediaTek and Nvidia.

Despite the enthusiasm surrounding processors by MediaTek and Nvidia, the actual launch timeline remains uncertain. Multiple sources reported that development issues could significantly postpone the commercial rollout of systems based on the new processors. Some projections suggest that these delays could push the release into 2026, but for now, there is no confirmation on those rumors.

The official release notes for Nvidia's CUDA 12.9 Toolkit explicitly indicate that the next major release will no longer support Maxwell, Pascal, and Volta-based GPUs. Note that this deprecation is only limited to the compute side, as these GPUs will likely continue receiving normal GeForce drivers for the time being. That being said, this is likely the last SDK version that can be used to develop CUDA applications targeting the aforementioned architectures.

While the previous release hinted at this change, Nvidia's stronger wording now serves as a definitive signal for developers to shift to more modern architectures. CUDA 12.x series (and before) will still allow application development for these GPUs. The deprecation targets offline compilation and library support. Essentially, future CUDA compilers (nvcc) will lack the ability to generate machine code compatible with these GPUs. In the same vein, upcoming versions of CUDA-accelerated libraries like cuBLAS, cuDNN, etc., will not offer support for GPUs built using these architectures.

Nvidia has not specified an exact date for the upcoming major release (likely CUDA 13.x). Similarly, we aren't sure how many interim releases are to follow in the 12.9.x branch. Either way, this is quite a significant change as Nvidia is dropping three major architectures with one swing. Volta's consumer equivalent Turing (RTX 20) is next in line, but it likely has a lot more to offer before it too hits the chopping block.

NVIDIA CEO Jensen Huang warned the United States' House Foreign Affairs Committee that the country faces being left behind by a rapidly-moving surge in AI development, calling on the country to maximize and accelerate the adoption of American AI worldwide.

"We are at an inflection point: The United States needs to decide if it is going to continue to lead the global development and deployment of AI or if we are going to retreat and retrench," Huang told the committee on Thursday, May 1.

"America cannot lead by slowing down. If we step back, others will step in. And the global AI ecosystem will fragment — technologically, economically, and ideologically."

When it comes to long-term prosperity in the high-tech world, it's all about setting standards. Intel once set the standard with x86, PCIe, and USB and now the vast majority of devices use these technologies in one way or another. Nvidia now enjoys its investments in the CUDA ecosystem and is setting the standard in AI compute in general.

To a large degree, Nvidia's efforts made the U.S. industry the leader in AI. But containing AI hardware in the U.S. will provoke rapid development of competing AI ecosystems that can eventually outperform the one developed in America, the company now warns.

Nvidia is everywhere - for now

For now, Nvidia and its CUDA ecosystem set the standards for AI applications across the world, both for training and for inference. With products like GB200/GB300 NVL72 Nvidia can address companies that need on-premise AI deployments for their AI applications. Given the ubiquity of CUDA, such deployments are easy and relatively inexpensive.

Nonetheless, Nvidia has domestic rivals, including traditional competitors like AMD and Intel as well as newcomers like D-Matrix and Tenstorrent. Most of their efforts are aimed at inference though, as Nvidia is the de facto king of AI training thanks to CUDA.

Thanks to the omnipresence of CUDA, Nvidia leads in AI not only in the U.S., but also in Europe and China. The vast majority of China's high-tech giants — Alibaba, ByteDance, Tencent, just to name a few — use Nvidia hardware and virtually all European companies use Nvidia hardware.

Meanwhile, when it comes to China, Nvidia has major rivals both on the hardware and platform sides. On the hardware front, Nvidia has competitors like Biren Technology, InnoSilicon, and Moore Threads.

These companies are quite formidable competitors, even though for now their market share is negligible. All three companies use PowerVR GPU IP developed by the U.K.-based Imagination Technologies and have loads of experience with GPU development, according to Jon Peddie, the head of Jon Peddie Research.

The founder of Moore Threads, Zhang Jianzhong (also known as Zhang Jian Zhong), previously worked at Nvidia: he was the general manager of Nvidia's operations in China.

The founder of Biren Technology, Li Bing, previously worked at Huawei and also had experience at other tech companies. Co-CEO of Biren Technology is Allen Lee (also known as Li Xinrong), who used to be vice president and general manager of AMD's China R&D Center. He joined the startup in 2021 as co-CEO, and now he oversees organizational management and product design.

"Li Bing's background and expertise in the tech industry likely influenced his vision for Biren Technology, which focuses on developing high-performance GPUs for various applications," Peddie told Tom's Hardware.

The importance of Huawei

But while Biren, InnoSilicon, and Moore Threads have rather good hardware, for now they lack an ecosystem that is comparable to Nvidia's CUDA. However, there is a company in China that can compete with Nvidia not only on the hardware side of matters, but also on the platform level: Huawei. Huawei has its Cloud Matrix 384 system, which it claims can outperform Nvidia's GB200 NVL72 rack-scale machine for AI. Perhaps more importantly, the company has its own AI-oriented, heterogeneous Compute Architecture for Neural Networks (CANN) platform designed specifically to use the potential of Huawei's HiSilicon Ascend AI processors.

Just like Nvidia's CUDA, Huawei's CANN offers a complete suite of development resources such as runtime systems, model-building tools, and compilers. It works with both Huawei's MindSpore platform and widely-used AI libraries like TensorFlow and PyTorch, making it flexible for developers. The framework includes a broad range of tuned computational components to speed up model execution and is also compatible with ONNX Runtime, allowing it to run ONNX-based models efficiently on the company's Ascend accelerators for AI.

Although CANN is a key part of Huawei’s AI infrastructure, it has drawn criticism for being difficult to work with, mainly due to unstable performance, inadequate documentation, and reliability issues that complicate model training and deployment. Huawei has acknowledged these problems and is actively working to strengthen the platform and improve its usability. For now, imperfection of Huwei's CANN and significant efforts that are required to port programs from CUDA to CANN (several months and hundreds of developers) limit success of Huawei's hardware.

Nvidia's dominance may not last forever

However, if Nvidia's GPUs will be unavailable for Chinese and European buyers, they will at least consider Huawei, or perhaps Biren, Innosilicon, or Moore Threads hardware. This will not only decrease Nvidia's revenues by tens of billions every year and its market capitalization by hundreds of billions, but could also eventually make its competitors from China as trend setters in the AI segment, the company believes.

"Regardless of how one feels about DeepSeek’s open-source R1 model, it is a clear indication that innovation is moving rapidly around the world, with or without leading U.S. tech," the statement by Nvidia reads. "If U.S. platforms are absent, companies will turn to strategic competitors like Huawei to fill the gap. This is why leadership in AI depends not just on what we restrict — but on what we enable. Ecosystems are essential. It is not just about who can build the biggest data center or train the most advanced model. […]. One of Nvidia's key strengths is our global network of 6 million developers who build on our platforms. If we lose that ecosystem to our competitors, it will be almost impossible to get it back."

AI diffusion rule to be implemented from May 15

The new U.S. export rules for compute GPUs — known as the AI Diffusion Rule — come into effect on May 15. Under the Biden administration's AI Diffusion framework, unrestricted access to high-end AI chips like Nvidia's H100 is reserved for companies in the U.S. and a select group of 18 allied countries classified as 'Tier 1.' Companies in 'Tier 2' nations are subject to an annual limit of approximately 50,000 H100-class GPUs, unless they secure verified end user (VEU) approval. They can still import up to 1,700 units per year without a license, and these do not count toward the national quota. However, countries listed as 'Tier 3' — including China, Russia, and Macau — are essentially blocked from receiving such hardware due to arms embargo restrictions. The Trump administration is now reviewing this tier system to make it more straightforward and enforceable, and is rumored to make limitations for Tier 2 nations even stricter.

Not only will Nvidia cease to be able to sell its GPUs to China, which is one of its largest markets, but its Chinese customers will be forced to either use its GPUs in the cloud, or switch to processors developed in China, such as those designed by Huawei or one of the aforementioned companies. While this will slow down development of China's AI sector in the short term, it will give a strong boost for its AI hardware ecosystem in the mid and long-term future.

Huawei already spends tens of billions of dollars on R&D every year. Once Huawei and others increase sales of their AI hardware, they will be able to invest more in development of their AI ecosystems, which will get more competitive against those developed by Nvidia and other American companies (such as AMD and Intel) than they are today.

Huawei can become a global AI leader

Having China as a fortress and being able to sell its hardware elsewhere, Huawei and other Chinese companies will compete against Nvidia and other American entities for European and Middle-East AI hardware markets. What's more important, they will be able to set standards of the AI market and that will not only reduce Nvidia's influence on such standards, but it will greatly reduce American influence on AI development. AI leadership is about more than market share, it is about strategic control over future governance models.

"Today, the U.S. semiconductor industry is being pushed out of China, the world's largest market," the letter by Nvidia concludes. "On May 15th, if the AI Diffusion Rule comes into effect without significant changes, we will be forced to similarly retreat from the rest of the world."

The U.S. has already seen the consequences of ceding technological leadership, when Huawei gained a dominant foothold in global 5G deployments by offering cheaper and faster-to-deploy infrastructure. This serves as a cautionary example of how losing control over foundational standards can shift both market power and geopolitical influence. Nevertheless, whether the current administration has learnt from similar past mistakes remains to be seen.

Nvidia has yet to officially reveal Blackwell-based graphics boards for professional visualization (ProViz) applications, but such cards are already undergoing testing in the wild by interested parties in Geekbench. An OpenCL benchmark result of Nvidia's upcoming RTX Pro 6000 Blackwell Workstation Edition graphics card has emerged, confirming rumored specifications of the GPU and memory while raising questions about performance.

Nvidia's RTX Pro 6000 Blackwell Workstation Edition add-in-board is based on the GB202 graphics processing unit (GPU) with 24,064 CUDA cores (188 streaming multiprocessors, 128 CUDA cores) allegedly operating at up to 2,617 MHz and carries 96 GB of memory with ECC, according to the Geekbench listing. By contrast, Nvidia's GeForce RTX 5090 — the best graphics card for gaming money can buy these days — features the GB202 GPU with 21,760 CUDA cores operating at up to 2,410 MHz and carries 32 GB of GDDR7 memory.

At first glance, the new RTX Pro 6000 Blackwell Workstation Edition graphics board should easily beat its gaming counterpart, but this is not the case as the unit scores 368,219 points in GB6 6.4.0 OpenCL benchmark, whereas the GeForce RTX 5090 can score around 376,858 in the same benchmark. A 2.3% performance difference is hardly a big deal, but given significant hardware differences between the boards, it is natural to expect the new ProViz card to beat the gaming board.

Huawei's next-generation HiSilicon Ascend 910D AI processor is expected to offer better performance than Nvidia's H100, reports Reuters. The new processor will be slower on a chip vs chip basis compared to Nvidia's Blackwell B200 and Blackwell Ultra B300 GPUs, never mind the next-generation Rubin GPUs slated to launch next year. However, Huawei's approach of building pods with hundreds of processors should allow the Ascend 910D to compete against pods based on Nvidia's current Blackwell and upcoming Rubin GPUs.

Huawei is preparing to start tests of its most advanced artificial intelligence processor, the Ascend 910D, with the performance goal of surpassing Nvidia's H100 and offering a domestic alternative amid U.S. export restrictions. According to sources, Huawei has approached several local companies to assess whether the new Ascend 910D chip meets performance and deployment requirements. Initial samples are expected by late May.

Separately, Huawei plans to start large-scale shipments of its dual-chiplet Ascend 910C AI processors to Chinese customers (and probably full systems based on the chips) as early as next month. The majority of of these processors were reportedly produced by TSMC for a third-party company. It remains to be seen whether the Ascend 910D will be made by China-based SMIC, or whether — nearly five years after the U.S. government restricted Huawei's access to leading-edge semiconductor production capabilities — Huawei will once again find a way to circumvent U.S. sanctions.

Reaching Nvidia H100 performance levels won't be easy for Huawei. The company's latest dual-chiplet Ascend 910C offers around 780 BF16 TFLOPS of performance, whereas Nvidia's H100 can deliver around 2,000 BF16 TFLOPS. In order to achieve H100 performance levels, Huawei will have to redesign the internal architecture of the Ascend 910D and possibly increase the number of compute chiplets.

To stay competitive in the AI industry next year, Huawei will have to achieve performance comparable to that of AI clusters developed in the U.S. This year, the company introduced its CloudMatrix 384 system with 384 Ascend 910C processors. It can reportedly beat Nvidia's GB200 NVL72 in certain workloads, but at the cost of significantly higher power consumption due to dramatically lower performance-per-watt. It also has over five times as many 'AI processors' as an NVL72 rack. Whether the interconnect can scale well to the required number of processors remains to be seen.

Without access to leading-edge process technologies, it will become significantly more difficult for Huawei to maintain competitive positions next year. Nvidia is on-track to introduce its codenamed Rubin GPUs for AI and HPC in 2026. Rubin GPUs are set to be made on TSMC's N3 (or a more advanced) fabrication process, and they should offer even higher performance-per-watt than the current-generation Blackwell GPUs.

Rubin GPUs are slated to offer around 8,300 TFLOPS of FP8 training performance, and presumably half that for BF16 — about twice the performance of the B200. Huawei's Ascend 910D and next-generation CloudMatrix systems with 384 of such processors could theoretically offer competitive AI performance on the rack level. However, it remains to be seen what performance benefits Huawei's Ascend 910D and Nvidia's Rubin GPUs will offer compared to existing offerings. Also, it should be noted that Nvidia will barely be able to sell its high-performance Rubin GPUs in China, so for that market Huawei won't really have a direct competitor.

Regardless of performance or efficiency, Huawei's Ascend 910D processors will likely become China's workhorses when it comes to AI training in the coming years. Given the strategic importance of AI, the power consumption of the Ascend 910D (or any other domestic AI processor) will not be a limiting factor, as the number of deployed units could offset the efficiency of Nvidia's (or AMD, Intel, Broadcom, etc.) AI processors. The main limiting factor for China will be its ability to produce enough processors — either domestically, or overseas using proxy companies.

Nvidia's budget-friendly RTX 5060 GPUs are rumored to hit shelves on May 19th, according to VideoCardz's sources. This release window falls just a day before Computex, and a day after AMD's RX 9060 XT series, allegedly.

Nvidia is reportedly relaying the embargo timelines to AIB partners as we speak, which is likely the origin of this leak. Since embargoes are always subject to change, we should treat this leak with skepticism.

Last week, Nvidia unveiled the RTX 5060 family of Blackwell GPUs, including the RTX 5060 Ti 8GB/16GB ($429/$379) and the RTX 5060 ($299), with the latter slated for an unspecified launch date next month. Contrary to a typical one-day gap between the review and sales embargoes, VideoCardz reports that for the RTX 5060, both will be lifted on the same day: May 19th. Instead of relying on Nvidia's skewed first-party benchmarks, interested customers should wait for independent reviews before jumping the gun.

With the sharp spike in shader count and bandwidth, you could be looking at 20-30% higher performance than the RTX 4060, but that's just a rough estimate. Equipped with the GB206 core, the RTX 5060 features 3,840 CUDA cores, 25% more than the RTX 4060, along with a 128-bit interface for 8GB of memory. Thanks to the adoption of GDDR7, there is a 64% increase in memory bandwidth (from 288 GB/s to 448 GB/s), which will prove useful in memory-intensive tasks.

RTX 5060 Ti 16GB vs RTX 4060 Ti 16GB Introduction

The Nvidia GeForce RTX 5060 Ti 16GB and RTX 4060 Ti 16GB represent the new and old guards of the 16GB realm for Team Green. Both of these are mainstream GPU offerings, ostensibly priced in the $430–$500 range. We weren't particularly impressed with the 4060 Ti 16GB at launch, but our feelings have tempered over time — helped by an unofficial $50 price cut — with both having appeared on our list of the best graphics cards.

The Nvidia Ada Lovelace 4060 Ti 16GB card originally launched over a month after the initial RTX 4060 Ti 8GB card, with a $100 price premium. Throughout the ensuing two years, the 8GB card has been more readily available, though as noted already, street prices on the 16GB card did drop by $50 at times. 

Conversely, the RTX 5060 Ti 16GB and 8GB variants technically launched at the same time on April 16, 2025, with only a $50 price gap. So far, the 16GB models have been far more widespread in the U.S., to the point where we have not yet been able to purchase an 8GB card at retail for testing.

Given the ongoing concerns with having only 8GB of VRAM, we feel it makes the most sense to compare the new and old 16GB xx60 Ti cards. We'll look primarily at performance, as well as power, features, and pricing. In theory, this should be a very easy win for the new Nvidia Blackwell GPU, but let's see how they actually stack up.

RTX 5060 Ti 16GB vs RTX 4060 Ti 16GB Specifications

Specifications for the RTX 5060 Ti 16GB and RTX 4060 Ti 16GB are mostly similar, with a few key upgrades for the newer GPU. While both the GB206 and AD106 dies have a maximum of 36 SMs (Streaming Multiprocessors), the 5060 Ti uses the fully enabled chip while the 4060 Ti disables a pair of SMs. That cascades through the GPU shaders, tensor cores, RT cores, and TMUs, giving the 5060 Ti a theoretical 6% advantage. However, clock speeds also factor into performance, and the 5060 Ti comes with a 2,572 MHz boost clock compared to 2,535 MHz on the 4060 Ti, another 1.5% advantage.

The core counts are a "hard" spec, but the clock speeds are a bit fuzzy. Power limits also come into play, and the 5060 Ti has a 180W TGP compared to 165W on the 4060 Ti 16GB. In practice, across our 21-game test suite, the 5060 Ti 16GB averaged clock speeds of 2,776 MHz compared to 2,725 MHz on the 4060 Ti 16GB — a slightly larger 1.9% advantage in favor of the newer card. Combined with SM and core counts, that gives the 5060 Ti 16GB an 8% advantage in raw compute.

That's pretty negligible in terms of a gen-on-gen increase, but architectural updates can further impact the real-world performance. More pertinent than the raw compute is the memory subsystem. The RTX 5060 Ti 16GB has 28 Gbps GDDR7 memory while the RTX 4060 Ti 16GB has 18 Gbps GDDR6 memory. That means the newer part has a 56% advantage in raw memory bandwidth — though architecture and timings can still play a role.

Taken together, based on the specifications, you should expect to see the newer GPU outperform its predecessor by anywhere from 8% to 56%, depending on whether a particular game or workload relies more on raw compute, memory bandwidth, or a blend of the two. In practice, compute tends to be the more important metric. As such, we'd expect a 15~25 percent lead from the 5060 Ti. Let's see how things go in actual benchmarks.

RTX 5060 Ti 16GB vs RTX 4060 Ti 16GB Gaming Performance

For most people, gaming performance will be the most important factor in choosing a graphics card. We have a test suite of 21 games, 6 of which have ray tracing enabled and 15 with pure rasterization. Upscaling and frame generation are disabled for all these tests, as we view those as "performance enhancers" rather than baseline features that should be compared.

Both the RTX 5060 Ti 16GB and RTX 4060 Ti 16GB support Nvidia's various DLSS features, though the 5060 Ti also offers MFG (Multi Frame Generation) — which can smooth out performance but definitely doesn't improve the actual feel of games that support it as much as the numbers might suggest. The experience of DLSS 3 framegen in our view tends to be inflated by about 50–60 percent (meaning, divide the resulting number by 1.5–1.6 to get a realistic view of what it feels like), while MFG 4X is about a 170–200 percent inflation (divide by 2.7–3.0).

Those are fuzzy approximations, but basically if you get less than a 50% "increase" in framerates via framegen, we would argue that it actually feels like a step backward. Likewise, less than a 170% increase from MFG 4X would be a net loss. Or to put hard numbers on it, if the base performance is 60 FPS, framegen needs to run at 90 FPS or more and MFG 4X needs to run at 160 FPS or more. Some would argue even those figures are too generous, but our main point is that "FPS" inflation via frame generation techniques has been heavily abused as a true indication of performance.

The charts and tables provide all the important numbers, so we're not going to dig into every single result. We're mostly interested in the overall trends, recognizing that individual games can come in above or below the averages. That's why we have an extensive 21 game test suite, with equal weighting given to each result via a geometric mean calculation.

The nice thing about comparing two Nvidia GPUs, even though they're different architectures, is that relative performance ends up being pretty consistent across both rasterization and ray tracing games. AMD and Intel GPUs vary quite a bit more, depending on what specific games are used.

Overall, the 5060 Ti 16GB outperforms the RTX 4060 Ti 16GB by 16% at 1080p medium, 17% in the ray tracing suite, and 16% in the rasterization suite. That means the 5060 Ti and 4060 Ti generally aren't CPU limited, even at 1080p medium. The minimum FPS (1% lows) shows similar consistency. Looking at the individual games, the 5060 Ti 16GB leads by 6% (Assassin's Creed Mirage) to as much as 29% (A Plague Tale: Requiem).

Bumping up to 1080p ultra settings, most of the deltas are virtually unchanged. The 5060 Ti 16GB still leads by 16% overall, 17% in the ray tracing suite, and 16% in the rasterization games. Minimum FPS ends up slightly closer overall at 18%, but with a wider 23% spread in our ray tracing tests and 16% in the rasterization games. Across all 21 games, we measured an 8% (Starfield) to 26% (Plague Tale: Requiem and Space Marine 2) delta in favor of the 5060 Ti.

Given the memory bandwidth advantage, the 5060 Ti 16GB should increase its overall lead at higher resolutions, and it does — just not by much. It's 18% faster overall at 1440p ultra, and 18% in both the rasterization and ray tracing suites. The various games show an advantage of 9% (Starfield) to 29% (Plague Tale and Space Marine 2, again).

Finally, at 4K ultra, we see the biggest lead for the 5060 Ti 16GB. Across the full 21-game suite it's 21% faster than the 4060 Ti 16GB, with a 20% lead in the ray tracing games and a 22% lead in the rasterization tests. The smallest advantage improves to 12% (Avatar), with the biggest lead being 33% (Space Marine 2).

RTX 5060 Ti 16GB vs RTX 4060 Ti 16GB AI and Content Creation Performance

Looking at other potential workloads, we have various disciplines that all loosely fall under the content creation umbrella. AI workloads include Stable Diffusion image generation, LLaMa text generation, image classification, inference, and other tasks. We also have 3D rendering via Blender, video transcoding using SPEC Workstation 4.0, and the Viewport test suite from SPEC Workstation 4.0.

Many people — particularly gamers — may never do any of these tasks on their PC, though it's worth nothing that Nvidia's push for "neural rendering" techniques can also extend into AI-powered NPCs. That's not something we can directly benchmark at present, but there's potential for the AI performance to become more important in the coming years if AI NPCs catch on in some major games.

Of all the content creation tasks, we think the AI tests deserve the most attention. As noted above, such workloads could become a bigger part of how people use their PCs and graphics cards in the future, even within games. Overall, the 5060 Ti 16GB delivers about a 24% lead in AI workloads compared to the RTX 4060 Ti 16GB. It's also worth mentioning that none of these AI tests at present leverage the FP4 support that was added to the tensor cores in the Blackwell architecture, so there's potential for even more of an AI deficit if future applications make use of that number format.

Video transcoding ends up being a virtual tie, indicating there's not a huge difference between the NVENC/NVDEC blocks in Ada and Blackwell. The 5060 Ti takes a 4% lead by virtue of running at slightly higher clock speeds, though Nvidia did supposedly improve the encoding quality — that's not something these tests measure, unfortunately.

3D modeling using Blender basically matches what we saw in gaming performance. Overall, the 5060 Ti 16GB runs 20% faster than the 4060 Ti 16GB, with a spread of 14% to 29% in the three different scenes. SolidWorks agrees with that number, while 3ds Max shows an even larger 31% lead for the 5060 Ti.

Those last two are part of the SPEC Workstation 4.0 Viewport tests, which show a 17% lead overall for the 5060 Ti, with a range of 1% (Creo) to as much as 31% (3ds Max). Most of the SPEC Workstation tests are truly for professionals, however, so we wouldn't put too much weight on them. Most paid professionals will have a workstation equipped with a workstation GPU (e.g. Nvidia RTX 6000 Ada Generation or the new RTX Pro 6000 Blackwell).

RTX 5060 Ti 16GB vs RTX 4060 Ti 16GB Features and Software

The vast majority of Nvidia features and software tools are available on both the RTX 4060 Ti 16GB and the RTX 5060 Ti 16GB. As such, it's easy to think that the two are "tied" in features. However, there are a few noteworthy exceptions.

First, the Blackwell RTX architecture adds support for Multi Frame Generation. We don't love the way the numbers are used to indicate massive performance improvements (i.e. Nvidia CEO Jensen Huang tried to claim that an "RTX 5070 is as far as an RTX 4090" at the launch event), but MFG isn't inherently bad. There are games where it works better, and the additional frame smoothing can make games feel better, subjectively. It's a nuanced discussion, in other words, and not the "fake frames BAD" mentality that often gets tossed around.

Blackwell RTX, and by extension the RTX 5060 Ti 16GB, is also supposed to have better support for so-called neural rendering (i.e. Microsoft's Cooperative Vectors API). We don't yet know precisely how much better it might be compared to older GPUs, but we do know Cooperative Vectors is supposed to work with all RTX GPUs — and that it's supposed to be better on Blackwell RTX.

As noted earlier, the RTX 5060 Ti also has support for FP4 (and FP6) numerical formats in its tensor cores. Making proper use of FP4 presents challenges for AI models, but various algorithms have demonstrated superior performance and reduced memory requirements with a "negligible" drop in overall inference quality. As AI continues to evolve, it's reasonable to expect the use of FP4 to become more commonplace.

We've seen that with FP8 support as well. First introduced with the RTX 40-series GPUs, the use of FP8 modes was relatively rare for most of the past 2.5 years. However, Nvidia's new DLSS Transformers algorithms for upscaling (super resolution) and ray reconstruction appear to leverage native FP8 support. It's why the performance hit for DLSS Transformers tends to be substantially lower on RTX 40- and 50-series GPUs compared to RTX 30- and 20-series GPUs.

But Nvidia giveth and Nvidia also taketh away. Blackwell graphics cards no longer have PhysX support, so the RTX 5060 Ti 16GB simply can't run PhysX in older games that used the feature. (Not many recent games have used GPU PhysX, so this mostly applies to games like the Batman: Arkham series and Mirror's Edge, all of which are at least 10 years old now.) 32-bit CUDA support was likewise retired, which again mostly impacts older apps that are no longer being actively updated.

RTX 5060 Ti 16GB vs RTX 4060 Ti 16GB Pricing

The RTX 4060 Ti 16GB launched in July 2023 with an MSRP of $499, $100 more than the 8GB variant. The RTX 5060 Ti 16GB came out in April 2025 with a $429 MSRP. On paper, that clearly suggests a pricing advantage for the 5060 Ti 16GB. In practice, it's also an advantage when shopping around for new cards, but that's because the 4060 Ti 16GB (and the rest of the RTX 40-series) has been retired.

Since both GPU lines are made using TSMC's 4N process node, there's no reason for Nvidia to continue producing the older chips, and in fact all indications are that wafer orders for Ada Lovelace may have ended in the fall of 2024. Supplies have completely dried up and prices have shot up as a result.

The availability of RTX 50-series GPUs has been relatively poor compared to what we saw with the 40-series launch. Sure, the RTX 4090 sold out for the first few months, but eventually it reached the point where you could expect to find some models selling slightly below the base MSRP, and all of the other 40-series GPUs eventually dropped below the base MSRP by late 2023 and early 2024. So far, the RTX 50-series has rarely even approached MSRP, often selling at 20% to 50% above the minimum suggested price.

At present, the least expensive RTX 5060 Ti 16GB card we can find is an Asus Dual RTX 5060 Ti 16GB for $492.98 at Newegg. That's about 15% above the base MSRP that Nvidia set, but Nvidia can't force its add-in board (AIB) partners to actually sell cards at MSRP. With ongoing GPU shortages, it's little surprise that nearly all 50-series GPUs are "overpriced" relative to MSRP.

As bad as that might seem, the RTX 4060 Ti 16GB is in an even worse state. With production of those GPUs likely halted last year, only limited supplies remain. Mostly, it's third-party marketplace resellers on Newegg, Amazon, and similar storefronts trying to make some serious money. Even the 8GB cards start at $575 (MSRP is supposed to be $400), while the cheapest RTX 4060 Ti 16GB costs $679. Yeah, that's a bad choice unless for some reason you specifically need the older hardware.

RTX 5060 Ti 16GB vs RTX 4060 Ti 16GB Final Verdict

This was pretty much a foregone conclusion, but the RTX 5060 Ti 16GB easily beats the RTX 4060 Ti 16GB in every conceivable metric. There's only one potential advantage to the 4060 Ti, and it's nebulous at best: It supports PhysX and 32-bit CUDA (though both are likely deprecated and will fade away in the coming months).

In contrast, the 5060 Ti 16GB offers about 20% higher gaming performance on average, across nearly all tested resolutions and settings. It's also about 20% faster in AI workloads and 3D rendering, and about 17% faster in a selections of professional applications. That's largely thanks to having 56% more memory bandwidth, courtesy of GDDR7, though other architectural upgrades and slightly increased core counts also factor into things.

The real kicker is the price. Unless you're lucky enough to find an RTX 4060 Ti 16GB on a steep discount, the 5060 Ti 16GB costs quite a bit less at popular online retail sites like Newegg and Amazon. Just say no, in other words. Slower performance, missing new features, and a price that's currently 38% higher? Pass. It's also a poor proposition when you look at total system cost (using $750 for the price of the rest of the PC). The 5060 Ti 16GB delivers 24% more performance per dollar than the RTX 4060 Ti 16GB in that case (using MSRPs for both, which benefits the 4060 Ti).

If you're wondering about power use and efficiency, even though the TGP (Total Graphics Power) on the 5060 Ti 16GB is 15W higher on paper, in practice it averaged lower power use in our testing. Power use was 1W higher at 4K ultra (a setting both GPUs struggle with), 5W lower at 1440p ultra, and 8–9 watts lower at 1080p. With 21% higher performance at 4K, that works out to much improved overall efficiency.

Hopefully it goes without saying, but just because the RTX 5060 Ti 16GB offers more performance and costs less than a new 4060 Ti 16GB, that doesn't mean people who already own the prior generation card should be in a rush to upgrade. A 20% performance uplift combined with a few new features might be nice to have, but we generally only recommend upgrading if you can get at least a 50% performance improvement, and doubling performance is often possible if you're upgrading from a card that's a couple of generations old.

Let's also not forget that there are 8GB variants of both GPUs. Again, just say no to such cards. Unless you can score a particularly impressive deal, the lack of VRAM is increasingly becoming a problem for 8GB cards. Sure, older and lighter games can run just fine on such GPUs, but it's the newer games that need faster hardware, and they're having problems. The 4060 Ti 16GB and 8GB offer nearly identical performance in our test suite at 1080p, but the 16GB card pulls ahead by 7% at 1440p, and by 36% at 4K — with the most demanding newer games exceeding those averages. We anticipate the story will be largely the same with the 5060 Ti 8GB.

So there you have it: While the generational gains of the RTX 5060 Ti 16GB over the RTX 4060 Ti 16GB aren't as large as we've seen with prior architectural upgrades, the lack of progress on process nodes and the relatively similar die sizes and transistor counts mean we shouldn't expect 50% performance improvements at the same price going forward. 20% more performance for an ostensibly 14% lower price is about as good as we're likely to see. That's about 40% more performance per dollar, which beats any of the other RTX 50-series GPUs.

Igor's Lab has discovered a worrying trend among all Nvidia Blackwell RTX 50-series graphics cards, including even faux-budget GPUs like the RTX 5060 Ti. According to a lengthy blog post the outlet published, most if not all RTX 50-series AIB partner cards are allegedly prone to high-temperature hotspots in the power delivery area, which could potentially damage these graphics cards after extended heavy use.

The problem lies in the construction of power delivery systems for the affected graphics cards. Igor's Lab states that several components that make up the power delivery system, such as the FETs, coils, drivers, and traces connecting everything together, are grouped too close together, creating temperatures that can potentially deteriorate the power delivery system over the course of a card's life, potentially killing the card after just a few years of use.

RTX 5070 vs RTX 5060 Ti Introduction

The Nvidia GeForce RTX 5070 and RTX 5060 Ti 16GB are the two most recent GPUs launched by Team Green. Since the beginning of 2025, Nvidia has released five Blackwell RTX 50-series GPUs — six if you want to count the RTX 5060 Ti 8GB as a separate entry.

To say that supplies have been limited and insufficient to keep up with demand would be a gross understatement, but the same applies to any of the best graphics cards right now. In fact, of all the Blackwell RTX GPUs, the 5070 and 5060 Ti are currently the most readily available; just don't expect to find most models at MSRP.

Our GPU benchmarks hierarchy ranks all the graphics cards by performance, and naturally the more expensive cards come out ahead of their less expensive siblings. But how do the RTX 5070 and RTX 5060 Ti 16GB stack up? We'll look at the performance, approximate pricing — because no GPU prices are reliably set in stone right now — as well as other aspects of the cards to help you decide which one might be right for your gaming PC.

Most of the features between the two GPUs will be identical. Both use the Blackwell architecture and support DLSS 4 and MFG, for example. Still, let's start with the specifications to see how the 5070 and 5060 Ti stack up.

RTX 5070 vs RTX 5060 Ti Specifications

As you'd expect, the RTX 5070 offers more raw performance on paper and in practice. It uses a GPU that's 45% larger, with 41.6% more transistors. More importantly perhaps, it has 33% more SMs, tensor cores, RT cores, and related computational units. It also has 50% more memory bandwidth and a 50% wider memory interface, plus 50% more L2 cache.

Memory interfaces don't scale as well as core logic on modern process nodes, and there are other aspects of the Blackwell GPU — the video encoders/decoders, display outputs, etc. — that don't really change, which explains why the chip size and transistor counts don't necessarily scale linearly.

Clock speeds aren't quite the same, at least on paper. The 5060 Ti has a 2572 MHz boost clock and the 5070 has a slightly lower 2512 MHz boost clock. But we're also looking at the 5070 Founders Edition with reference clocks, and a PNY 5060 Ti 16GB OC that has a 2692 MHz boost clock. Plugging in those numbers gives the 5070 30.9 teraflops of FP32 compute and the PNY 5060 Ti has 24.8 TFLOPS. That means the 5070 'only' has 25% more compute on paper — which isn't actually all that accurate.

In our full suite of gaming benchmarks, the 5070 Founders Edition averaged clock speeds of 2826 MHz while the PNY 5060 Ti 16GB OC averaged 2776 MHz. That means despite theoretically 180 MHz lower clocks, in practice Nvidia's 5070 card came out 50 MHz ahead.

That means the 5070 FE offers about 34.7 TFLOPS compared to 25.6 TFLOPS, a 35.5% advantage. Practically speaking, though, the 5070 should be up to 35% faster in games that are compute bound, and up to 50% faster for games that are memory bandwidth bound — and potentially the 5060 Ti could close the gap if VRAM capacity comes into play.

RTX 5070 vs RTX 5060 Ti Gaming Performance

There are multiple facets of performance to discuss. We'll start with gaming performance, as that's what the majority of people looking at the RTX 5070 and RTX 5060 Ti 16GB care about the most. We have an expanded test suite of 21 games, with 15 rasterization benchmarks and six ray tracing benchmarks — slightly more than the 18 games we use in our GPU hierarchy (we're planning to add these into the hierarchy but haven't tested every GPU yet).

The charts group things together by overall performance (geometric mean across all 21 games, where each gets equal weighting), rasterization-only performance, and ray tracing-only performance. We also have separate tables that show the percentage differences below the charts (sorted in reverse order due to the vagaries of Excel).

It's safe to say that neither the RTX 5070 nor the RTX 5060 Ti 16GB are primarily targeted at native 4K ultra gaming. The 1440p result works as a proxy for 4K with DLSS quality mode upscaling, and the 1080p result stands in for 4K with performance mode upscaling, though there would be potential differences. Upscaling does have some overhead, but since we're looking at two Nvidia GPUs we can expect a pretty consistent delta cause by DLSS. CPU bottlenecks would potentially affect performance at lower resolutions as well, which we can see by the results.

At 4K, the RTX 5070 leads the RTX 5060 Ti 16GB by 32% on average, with a performance difference of -5% (Indiana Jones and the Great Circle) to +43% (A Plague Tale: Requiem). Indiana Jones is known to want a lot of VRAM — you can even run the ultra setting on cards with less than 12GB — while various other games appear to want more memory bandwidth.

Dropping to 1440p, the 5070 lead shrinks fractionally to 31% overall, and this time there are no performance deficits. There may be a few other games out there that can push beyond 12GB of VRAM use, but from our test suite — a rather demanding suite overall, we'd say — 12GB remains sufficient in all games for 1440p, and in all but one of the games for 4K.

1080p starts to run into CPU limits, even more so at medium settings, which shrinks the delta between the 5070 and 5060 Ti to 28% at ultra settings and 24% at medium settings. That's still a sizeable gap, with a few of the games showing basically equivalent performance due to the CPU bottleneck (Baldur's Gate 3, Flight Simulator 2020, Spider-Man 2, Spider-Man: Miles Morales, and Stalker 2 end up tied in one or both cases).

Looking at the rasterization and ray tracing performance, overall there's not much difference at the various resolutions and settings. 4K ultra does run into VRAM limits on one of the RT games, so that the gap is slightly narrower there than at 1440p and 1080p, and our RT test suite does end up with lower FPS than the rasterization games at ultra settings. Medium settings with RT enabled actually runs faster on average for our test games, but only Control and Cyberpunk are truly demanding RT games in our title selection.

RTX 5070 vs RTX 5060 Ti Content Creation Performance

Our content creation test suite consists of a variety of test that all loosely fall into the "professional" category. We have multiple AI tests, 3D rendering benchmarks, video transcoding performance courtesy of SPEC Workstation 4.0, and the SPEC Workstation 4.0 Viewport benchmarks (basically the same tests as SPECviewperf).

As expected, the RTX 5070 generally walks away with the content creation crown. It's 45% faster in the Blender 3D rendering benchmarks, 33% faster in MLPerf text generation tokens per second, and 20–43 percent faster in the Procyon and SPEC AI tests that we ran. The 5060 Ti 16GB did have a faster time to first token in MLPerf, though it's possible that's due to updates to the benchmark (we'll recheck the 5070 in the near future, once our test rig isn't busy).

For SPEC Workstation 4.0's Viewport tests, there's another instance of the 5060 Ti 16GB garnering a win — it's 2% faster in Creo. But overall the 5070 still has a 23% lead. Handbrake video transcoding meanwhile ends up as a tie, which is expected as the fixed function encoding and decoding hardware is the same, so the only difference would be GPU clocks during that test.

Neither GPU is specifically marketed as a professional solution, though we anticipate there will be professional variants of both chips (with more VRAM and drivers that enable additional accelerations in some of the SPEC Viewport tests). Mostly, the AI results are interesting as a hobbyist solution, though it's possible future games might leverage the AI hardware more.

RTX 5070 vs RTX 5060 Ti Conclusion

Which is better: RTX 5070 or RTX 5060 Ti 16GB? Price, as we said at the start, will arguably be the biggest factor in deciding between the RTX 5070 and RTX 5060 Ti 16GB. It's also the one thing that we can't really pin down at present. Since the 5060 Ti launched just a few days ago, pricing and availability are more likely to fluctuate in the near term. The 5070 has been out for a month now, and we have a better idea of what to expect.

First, let's talk about MSRPs and performance. On paper, Nvidia gives the 5060 Ti 16GB a $429.99 MSRP and the 5070 has a $549.99 MSRP — so $430 and $550. That means the 5070 is supposed to cost 28% more than the 5060 Ti 16GB. And based on the performance results, that's exactly in line with what you get.

Linear performance scaling generally means you're better off buying the more expensive card — we would normally expect to see diminishing returns. So if you buy an RTX 5060 Ti 16GB at MSRP and get 0.171 FPS/$ across our gaming tests, versus the RTX 5070 with 0.170 FPS/$, we give the win to the 5070. The reason is because the GPU doesn't exist in a vacuum; there's the rest of the PC to consider.

Alternatively, let's talk current street prices. These will change dramatically over time, but we can currently find an RTX 5060 Ti 16GB for $509.98, or an RTX 5070 for $619.98. That's the second pair of tables in the above gallery, which obviously skews in favor of the 5070 at current prices. And that's what you need to look at, first and foremost.

Finally, imagine a PC where the CPU, motherboard, SSD, RAM, PSU, and case together cost $750 for example. Now you're looking at a total cost of $1,180 for the 5060 Ti 16GB compared to $1,300 for the 5070 build (using the base MSRPs for the GPUs). That works out to just 10% more total money for about the same 25~30 percent performance uplift. It's what the last set of tables above show, and again, it favors the 5070 heavily.

Given the 25~30 percent performance advantage of the RTX 5070 over the RTX 5060 Ti 16GB, spending up to 30% more on the RTX 5070 would make sense. It's the clear winner at current online prices, or at the given MSRPs. In fact, we would argue that it's worth spending up to 40% more for the 5070 compared to the 5060 Ti, as we normally expect diminishing returns.

We don't know where prices will end up, in the U.S. or elsewhere, but you can use the above 30–40 percent figure as a guideline. If the RTX 5070 were to cost $550, we wouldn't recommend spending more than $425 on the RTX 5060 Ti 16GB — and $400 would be better. Conversely, if the best price you can find on the RTX 5070 ends up being $700, then the 5060 Ti 16GB would be worth $500 to at most $540. (And you can replace the dollar signs with whatever monetary symbol you choose.)

Long-term, we expect both GPUs will trend toward about a 30% higher price for the RTX 5070 vs RTX 5060 Ti 16GB. Anything less makes the slower card very difficult to justify, even if it does technically offer more VRAM. So unless you specifically need a card with 16GB, perhaps to run a particular LLM, the RTX 5070 offers the better overall value right now, and will probably continue to do so in the future.

AMD is reportedly preparing to debut its RDNA 4 workstation GPU offerings for desktops, presumably under the Radeon Pro W9000 family. As put forwaard by Hoang Anh Phu, who frequently obtains inside scoops, AMD is considering using the Navi 48 XTW die for its top-end SKUs, paired with 32GB of video memory, likely GDDR6. As always, this leak shouldn't be taken as definitive, but there's likely some truth to it given the proximity of Computex next month, followed by AMD's Advancing AI event in June.

Radeon PRO GPUs are aimed at workstation setups, rivaling Nvidia's (former) Quadro or (now incumbent) RTX PRO offerings for prosumers. These graphics cards bridge the gap between consumers and server domains, for applications like AI, HPC, DCC, CGI, CAD, VR/AR, and the list goes on.

It seems that AMD is sticking to more conservative figures for its flagship workstation offerings this generation. That's somewhat expected since Navi 48 (356mm²) is in the same ballpark as GB203 (378mm²), found in the RTX PRO 4500 Blackwell. Nvidia's top-end GB202 at 750mm², is home to the RTX PRO 6000 Blackwell featuring a massive 96GB frame buffer.

Navi 48, with its 256-bit interface, enables either 16GB of memory (via eight 32-bit channels) or a theoretical maximum of 32GB in clamshell mode, which is the exact configuration being reported here.

Jensen Huang, chief executive of Nvidia, said in an interview broadcasted by CCTV that his company will continue supplying competitive products for the Chinese market going forward. 

Huang noted the importance of the Chinese market for Nvidia, emphasizing that it is crucial for the company to continue offering both datacenter and gaming products for the People's Republic. Yet, this is getting harder due to stricter export controls imposed by the U.S. gov't.

"We are going to continue to make significant effort to optimize our products that are compliant with the regulations and continue to serve the Chinese market," said Jensen Huang. 

The Trump administration restricted sales of Nvidia's China-specific HGX H20 GPU for AI earlier this week, which made Nvidia write off $5.5 billion worth of inventory for the first quarter. As a result, Nvidia now must obtain an export license from the U.S. Department of Commerce to ship its H20 for the Chinese market. However, the government is unlikely to grant one as it reviews such license applications with a presumption of denial. 

The U.S. government blamed H20's memory bandwidth and interconnect bandwidth for the restriction as both potentially enable usage of the processor inside supercomputers that can be used to develop weapons. It remains to be seen whether Nvidia develops a variant of H20 with lower memory bandwidth and fewer interconnects before the sweeping U.S. AI Diffusion Rule comes into effect in mid-May, though this is not a likely scenario. 

In fact, it is completely unclear how Nvidia plans to 'optimize' its GPUs for the Chinese market from mid-May and onwards as the new export rules prohibit selling American AI GPUs to adversary countries like China and Russia. Nonetheless, it looks like Huang's company is trying to find a solution. 

Meanwhile, Nvidia has criticized the AI Diffusion Rule saying that it would not stop development of Chinese AI technologies, but will likely encourage local companies like Biren and Huawei to develop their own processors, and, more importantly, standards. 

"The increased restriction has impacted our company significantly," said Huang. "We have grown up in China in fact, and China has watched us grow in the last 30 years. Of course it is a very large market interactions and working and serving the Chinese companies here […] so it has made us both better and so we are going to continue to make significant effort to optimize our products that are compliant with the regulations and continue to serve the Chinese market."

GPU Benchmarks Introduction

Tom's Hardware exhaustively benchmarks every GPU to find out which are worthy of our list of the best graphics cards. Our GPU benchmarks hierarchy ranks current and previous generation graphics cards by performance. Whether it's playing games, running artificial intelligence workloads, or doing professional video editing, your graphics card typically plays the biggest role in determining performance — even the best CPUs for gaming take a secondary role.

Our 2026 GPU Hierarchy testing spans three generations of Nvidia and AMD graphics cards, as well as Intel's Arc B-series GPUs.

Our testing has been made easier by the fact that no truly new gaming GPUs have been introduced in almost a year. If you haven't already upgraded your graphics card after the GeForce RTX 50-series and Radeon RX 9000-series launches in 2025, well, you're still looking at the exact same products now.

AMD did make its formerly China-only Radeon RX 9070 GRE available globally after Computex 2026, but in our review, we found that $549 product to be too expensive given the level of performance it delivers and the compromises made to hit its price point. Check out that coverage for all the details.

Most of the products we recommend remain at elevated prices compared to their MSRPs, but this is just life in mid-2026. It's admittedly cold comfort, but unless you're shopping for an RTX 5090, graphics card prices haven't risen much more than they already did earlier this year.

Compared to the doubling or tripling of prices we've seen for RAM kits and SSDs in 2026 versus last year, a GPU upgrade remains a relatively affordable (and self-contained) option, either as a boost for an existing PC or part of an all-new parts list.

Let's dive into our ranking of GPUs past and present so you can figure out how all those cards stack up.

Prime Day exceptional graphics card deals

These are a few of the standout deals from Amazon's 2026 Prime Day event, which is currently taking place. Be sure to check out our full list of the best Prime Day graphics card deals, too. The GPU Benchmarks Hierarchy continues below.

Our full GPU hierarchy using traditional rendering (aka rasterization) comes first, and below that we have our separate ray tracing hierarchy. We've also mashed up these results into one overarching ranking for the complete pictures.

The results are all collected at native resolution, without enabling DLSS, FSR, or XeSS upscaling or frame generation.

As a brief refresher of the cards in this version of our hierarchy, Nvidia's cutting-edge Blackwell architecture and its DLSS 4 suite of tech upgrades the quality of both upscaling and ray reconstruction on RTX 50-series GPUs. It also adds Multi Frame Generation support. As of mid-2026, MFG can add anywhere from one to five AI-generated frames in between natively rendered ones.

Even if you're not into framegen, DLSS 4.5 upscaling and its transformer neural network architecture can offer a big boost in image quality at the same output resolution compared to earlier DLSS versions. That tech can benefit all GeForce RTX GPUs going back to the RTX 20-series, albeit at a heavier performance cost on older hardware compared to earlier DLSS versions.

Nvidia's Ada Lovelace architecture powers its prior-gen RTX 40-series cards. Ada introduced DLSS Frame Generation, which can double output frame rates in supported games. Ada cards don't benefit from MFG, though.

AMD's Radeon RX 9000 series cards, powered by its latest RDNA 4 architecture, get a big boost to both ray-tracing and AI capabilities with dedicated RT and matrix math accelerators. AMD uses those capabilities to enable its FSR 4 upscaler and its much-improved image quality in a small but growing range of titles, whether through native support or with driver-level overrides.

The FSR Redstone update last year brought ML Frame Generation to RX 9000-series cards. Like FSR 4.x upscaling, ML Frame Generation can be directly integrated in games or enabled through a driver override.

Meanwhile, the last-gen RDNA 3 architecture powers seven RX 7000-series seven desktop cards. Until very recently, RX 7000-series cards couldn't officially run FSR 4 upscaling, but an official version of that tech will be made available for those cards in July 2026.

Intel's Battlemage-powered Arc B580 and Arc B570 offer major improvements in performance and efficiency compared to its Alchemist architecture. Battlemage only serves the entry-level end of the graphics card market today. Intel introduced a larger Battlemage chip in the form of the Arc Pro B70 earlier this year, but that product is intended for AI and professional visualization, and it's priced like it.

On page two, you'll find our 2025-2026 test data for posterity. Page three has our 2024–2022 benchmark suite, which covers previous-generation GPUs running an older test suite and a Core i9-12900K. Page four has an even older 2020–2021 test suite with only raster games, running on a Core i9-9900K testbed. The legacy tables are no longer being actively updated. There's also the "Ancient Legacy GPU hierarchy" (without benchmarks, sorted by theoretical performance) for reference purposes.

The following tables sort everything solely by our performance-based GPU benchmarks, from fastest to slowest. Factors including real-world pricing, graphics card power consumption, overall efficiency, and features aren't factored into the rankings here. The latest results use an AMD Ryzen 7 9800X3D testbed. Here are the tables and benchmark results — rasterization games first, then ray tracing games, and finally the content creation results.

GPU Benchmarks Hierarchy 2026: The tests

For our latest GPU benchmarks, we test every card at a mix of high and ultra settings, depending on the game. We test across three resolutions: 1080p, 1440p, and 4K.

All the scores are expressed as percentages relative to the top-ranking 1080p ultra card, which is of course the GeForce RTX 5090.

Our 2026 test suite comprises the following games. 11 are raster titles, and of those, four have RT support. We test another four games that either require RT to run (DOOM: The Dark Ages and Indiana Jones and the Great Circle) or look best with RT (Assassin's Creed Shadows and Grand Theft Auto V Enhanced.)

• Black Myth Wukong (+RT)
• Alan Wake II (+RT)
• Apex Legends
• Counter-Strike 2
• Fortnite
• Arc Raiders
• Stalker 2
• DOOM: The Dark Ages (RT required)
• Grand Theft Auto V Enhanced (RT only)
• Marvel's Spider-Man 2 (+RT)
• Indiana Jones and the Great Circle (RT required)
• Marvel Rivals
• Assassin's Creed Shadows (RT only)
• Cyberpunk 2077 (+RT)
• Clair Obscur: Expedition 33

GPU Benchmarks Hierarchy 2026: Raster gaming

The FPS score is the geometric mean (equal weighting) of all 11 games. Note that the specifications column links directly to our original review for the various GPUs.

Rasterization GPU Benchmarks, Key Findings

• Unsurprisingly, the RTX 5090 takes the top spot across the board, but prices remain stratospheric in mid-2026. It's also difficult to fully engage the GB202 GPU for pure raster gaming unless you're playing at 4K.
• Among current cards, the RX 9060 XT 8GB and RTX 5060 duke it out for the best performance per dollar at 1080p.
• The Radeon RX 9070 is the 1440p raster value champ. The recently introduced RX 9070 GRE sits just behind, and the RTX 5070 is a somewhat distant third.
• At 4K, the RX 9070 XT is the cheapest way to get into 4K 60 FPS native gaming. But the RX 9070 and RTX 5070 are also strong options among current GPUs if you're willing to enable a dash of upscaling.

Overall, if you're only interested in native raster gaming, Radeons are a strong choice in 2026. But Nvidia offers superior DLSS 4.5 upscaling and Multi Frame Generation support across all RTX 50-series cards, which makes matching your GPU's performance to your monitor's refresh rate a snap.

FSR 4.x upscaling isn't on par with DLSS 4.5 yet, and FSR ML Frame Generation is limited to a 2x framerate boost where it's available.

Rasterization GPU Benchmarks Rankings 2026

Remember that we're not including any upscaling or framegen results in the above table. DLSS, FSR, and XeSS offer differing image quality, and we want to keep things directly comparable.

Don't buy any of the cards at the top of our hierarchy without a high-refresh-rate 2560x1440 or 4K monitor to match. Even one of our best gaming CPUs, like the Ryzen 7 9800X3D, can only do so much when a powerful card becomes CPU-bound.

In general, if you're still running a 60 Hz fixed-refresh-rate monitor, it's entirely possible that you're not seeing all the frames your graphics card can generate. Practically any current-gen graphics card from the RTX 5060 on up is good enough for high-refresh-rate gaming at 1080p in 2026 even without upscaling or framegen, and the RTX 5070 or RX 9070 will give you the same experience at 1440p.

In this era of upscaling and framegen, a high-resolution monitor is no longer an obstacle to the best gaming experience. Upscalers work best with higher-than-1080p output resolutions, and the advent of the DLSS 4.5 upscaling model especially means that you can get both smooth performance and crisp image quality from surprisingly modest graphics cards.

But if you don't have a high-refresh-rate 1440p or 4K monitor to begin with, you can't take full advantage of the free boost to both performance and image quality that AI-powered upscaling offers, nor can you enjoy the full smoothness boost of framegen. Check out our list of the best gaming monitors for a potential upgrade if you're still rocking a 1080p 60 Hz display from the 2010s.

On to our 2026 raster results. Among currently available graphics cards, Blackwell rules the top three spots. The RX 9070 XT ends up fourth, just behind the potent (but much more expensive, as of June 2026) RTX 5070 Ti.

Moving into the midrange, the RX 9070's strong raster performance gives it the edge over the RTX 5070, but it's close. And the 5070 has the full arsenal of DLSS 4.5 features at its disposal in virtually every modern game.

Given the image quality and smoothness advantage of DLSS 4.5 upscaling and MFG, the 5070 got the nod in our most recent round of best graphics card picks after our 2026 retests.

The recently introduced RX 9070 GRE lands between the 5070 and the lower-end 5060 Ti and 9060 XT, a position for which Nvidia has no current-gen answer (and no easy path to one.) But AMD may have priced the GRE too high at $549, and we're already seeing big price drops on that product that might make it uniquely appealing in a market where the RTX 5060 Ti is its direct price competition.

The formerly midrange RTX 5060 Ti 16GB is hopelessly overpriced for the performance it offers in mid-2026, and the formerly somewhat affordable 8GB model is also selling for elevated prices now, making the entire 5060 Ti family impossible to recommend.

The RX 9060 XT 16GB holds down the entry-midrange 16GB position by itself, and its $459 price tag isn't so absurd as to make it un-recommendable. But as RX 9070 GREs start going on sale for $500-ish, the step up is going to look mighty tempting.

As we move further down the stack, the $300-ish Arc B580 represents Intel's best card right now, but it remains very much a budget part in mid-2026, trading blows with the RTX 3060 Ti and RTX 4060. The only reason it lands as high as it does in our results is that its 12GB of VRAM prevents its performance from totally plummeting at 4K, as it does for those older 8GB cards.

The RTX 5060 is 24% faster than the Arc B580 at 1080p and 17% faster at 1440p, and its prices are proportionately higher. The scarce RX 9060 XT 8GB is a whole 30% faster at 1080p and 23% faster at 1440p, for similarly more money.

We don't think the potential longevity afforded by the B580's 12GB of VRAM is worth the massive performance tradeoff versus current-gen 8GB GPUs. Save up a few more bucks and grab an RTX 5060 or RX 9060 8GB instead (and enjoy the more widely adopted DLSS 4.5 or FSR 4 versus XeSS while you're at it).

The $250 Arc B570 is still a good value in mid-2026, and as the only modern, decent graphics card available for (well) under $300, it's still worth a look if you can't stretch to an RTX 5050. But recent software ills that took months to fix have led us to consider it more of a gamble than a product worthy of a recommendation, even though those issues did eventually get corrected.

Ray Tracing GPU Benchmarks 2026

For 2026, we're testing a range of RT titles that present a progressively more difficult workload to the GPUs under test.

Grand Theft Auto V Enhanced, DOOM: The Dark Ages, and Indiana Jones and the Great Circle are all relatively lightweight RT titles that will run on a wide range of RT-ready hardware. Spider-Man 2, Cyberpunk 2077, and Assassin's Creed Shadows represent greater challenges to compute resources, VRAM, or both. And Black Myth Wukong and Alan Wake II still bring even the most powerful graphics cards to their knees at native resolutions.

Our results for those more demanding games represent more of a jumping-off point for upscaling and framegen rather than native performance. But lighter RT titles mean that you can explore the differences in image quality that ray tracing offers without crushing your GPU flat.

In any case, graphics cards at the top of our RT tests are ready for practically any RT game. But as you'll see, that's still quite the high bar to clear in 2026.

Ray Tracing GPU Benchmarks, Key Findings and Notes

• The RTX 5090 really comes into its own with RT enabled, as expected.
• AMD's RX 9070 XT outperforms the RX 7900 XTX in our RT tests, showing that the ray-tracing improvements in RDNA 4 deliver.
• Even with those improvements, the RX 9070 XT can only manage a tie with the RTX 4070 Ti with RT in play, coming in eighth place overall.
• Maintaining 60 FPS in RT titles (as a good foundation for upscaling and framegen enhancements) is a very tall order. Among recent products, you'll want:
  • an RTX 5060 Ti 16GB, RTX 5070, or RX 9070 GRE to handle ~60 FPS RT at 1080p without running out of VRAM
  • An RTX 4070, an RTX 5070 or RX 9070 at 1440p, or an RX 9070 XT or RTX 5070 Ti for the best possible experience
  • An RTX 5080, RTX 4090, or RTX 5090 at 4K

Ray Tracing GPU Benchmarks Rankings 2026

The RTX 5090 and RTX 4090 lead our RT results. But if you want to enjoy high-end RT without a 4090- or 5090-sized bankroll, the RTX 5070 Ti proves itself as the last card with a reasonable price-to-performance ratio before things get crazy.

The 5070 Ti leads the RX 9070 XT by a wide margin in our RT tests, and it's anywhere from 10%-15% behind the absurdly overpriced RTX 5080 across the board. But if you want the absolute best RT performance possible without stepping up to a 4090 or 5090, the extra cash for a 5080 is your best — and only — option.

That said, you should really explore DLSS 4.5 upscaling and MFG on the 5070 Ti before spending big to step up to a 5080.

The RX 9070 XT remains AMD's fastest RT GPU, outpacing the 7900 XTX across all tested resolutions. It leads the RDNA 3 halo card by about 10% across the board, which is an impressive generational leap considering its 33% deficit in CUs versus the XTX.

But the 9070 XT can only trade blows with the RTX 4070 Ti and RTX 5070 with Nvidia competition in the picture. It beats out the 5070 at 1080p and 1440p, but DLSS 4.5 upscaling and MFG give Blackwell gamers more flexible options for boosting performance with only minimal impact to image quality versus both RDNA 4 (and Ada).

Of course, the RX 9070 XT has FSR 4 upscaling and framegen at its disposal in compatible titles, but support for those features isn't as widespread as DLSS.

Meanwhile, the RX 9070 comes in slightly behind the RTX 5070 at 1080p and 1440p for RT. Even though its 16GB of VRAM prevents the 9070's 4K RT performance from completely nosediving, the average frame rate at that resolution isn't high enough to make it a 4K RT choice.

At 1080p, the RX 9070 GRE presents an intriguing RT option for Radeon gamers, delivering a near-60-FPS average baseline that would make a great FSR 4 foundation. But its 12GB of VRAM quickly becomes a liability at 1440p versus the 9070, and 4K is out of the question.

Overall, our demanding 2026 RT test suite shows that the bar for entry to a good RT experience remains high, especially as elevated prices on the RTX 5060 Ti 16GB mean that the first "real" RT-ready Blackwell card at 1080p and up is the RTX 5070. And on the Radeon side, you really want an RX 9070 GRE or RX 9070 for the best experience.

Test Setup and Hardware for GPU Benchmarks

We've used several different PCs for our testing over the years. The latest update switches to an AMD Zen 5 processor: the unparalleled Ryzen 7 9800X3D and its 3D V-Cache-enhanced performance.

Here are the specifications for our latest GPU test PC.

Tom's Hardware 2026 GPU Benchmarks Testbed

CPU: AMD Ryzen 7 9800X3D
CPU Cooler: Thermalright Phantom Spirit 120 SE
Memory: G.Skill Trident Z5 32GB (2x16GB) DDR5-6000 CL30
Motherboard: Asus TUF Gaming X670E-Plus Wifi
SSD: Inland Performance Plus 4TB
PSU: MSI MPG Ai1600TS 1600W

We test across the three most common gaming resolutions, 1080p, 1440p, and 4K, using a mix of high and ultra settings, depending on the title. Where possible, we use 'reference' cards for all of these tests, like Nvidia's Founders Edition models and AMD's reference designs. Most midrange and lower GPUs don't get reference models, however, and in some cases we only have factory-overclocked cards for testing. We do our best to select cards that are close to the reference specs in such cases.

For each graphics card, we follow the same testing procedure. We run one pass of each benchmark to "warm up" the GPU after launching the game, then perform our actual test runs across each resolution.

We carefully review our test data and check for anomalies. For example, we always expect the RTX 5080 to be faster than the RTX 5070 Ti. If it's not, and we're not in a CPU limited situation, we'll recheck both cards to ensure that our standings our accurate. We also check and retest in cases of subtler issues, as when a transient hitch or frame-time spike causes a large dip in 1% low FPS.

Due to the length of time required for testing each GPU, updated drivers and game patches inevitably come out that can impact performance. We periodically retest a few sample cards to verify our results are still valid, and if not, we go through and retest the affected game(s) and GPU(s). We may also add games to our test suite over time, if one comes out that is popular and conducive to testing. See what makes a good game benchmark for our selection criteria.

GPU Benchmarks: Individual Game Charts

The above tables provide a summary of performance, but for those that want to see the individual game charts, for both the standard and ray tracing test suites, we've got those as well. These charts only cover current-gen GPUs for readability.

These charts are up to date as of June 2026.

GPU Benchmarks — 1080p

GPU Benchmarks — 1440p

GPU Benchmarks — 4K

GPU Benchmarks — Power, Clocks, and Temperatures

Most of our discussion has focused on performance, but for those interested in power and other aspects of the GPUs, here are the appropriate charts. We'll run these from highest to lowest settings, as 4K ultra tends to be the most strenuous workload on most of these GPUs.

If you're looking for the legacy GPU hierarchy, head over to page two! We moved it to a separate page to help improve load times in our CMS as well as for the main website. And if you're looking to comment on the GPU benchmarks hierarchy, head over to our forums and join the discussion!

Choosing a Graphics Card

Choosing a Graphics Card

Choosing a Graphics Card

Which graphics card do you need? To help you decide, we created this GPU benchmarks hierarchy, comprising 48 GPUs from three generations of hardware for Nvidia and AMD and also including Intel's Battlemage cards. Not surprisingly, the fastest cards are using the latest GPU architectures, though they're not always a major upgrade over the prior generation.

Of course, it's not just about playing games. Many applications use the GPU for other work. But a good graphics card for gaming will typically do equally well in complex GPU computational workloads. Buy one of the top cards and you can run games at high resolutions and frame rates with the effects turned all the way up, and you'll be able to do content creation work as needed. Drop down to the middle and lower portions of the list and you'll need to start dialing down the settings to get acceptable performance in regular gameplay and GPU benchmarks.

If your main goal is gaming, you can't forget about the CPU. Getting the best possible gaming GPU won't help you much if your CPU is underpowered or many years out of date. So be sure to check out the Best CPUs for gaming page, as well as our CPU Benchmarks Hierarchy to make sure you have the right CPU for the level of gaming performance you're looking to achieve.

And don't forget about your monitor. Be sure to get a display whose refresh rate range matches the average frame rates of the graphics card that you want in the games that you play. Check out our list of the best gaming monitors for a starting point.

GPU Benchmarks from 2022–2024

Our 'legacy' GPU benchmarks used different hardware. Here are the details for the 2022–2024 configuration that used an Alder Lake 12900K CPU.

Tom's Hardware 2022–2024 GPU Testbed

Intel Core i9-12900K
MSI Pro Z690-A WiFi DDR4
Corsair 2x16GB DDR4-3600 CL16
Crucial P5 Plus 2TB
Cooler Master MWE 1250 V2 Gold
Cooler Master PL360 Flux
Cooler Master HAF500
Windows 11 Pro 64-bit

We have upgraded our test system and changed our test suite for 2025, effectively resetting our benchmarking and rankings. However, the old data — collected using a Core i9-12900K PC — remains valid. We aren't testing new GPUs with the old testbed, as that would double the time required for what is already a lengthy process, but we felt it would be helpful to some to keep the final data available.

We also had an interim system that never quite got wrapped up, which had a Core i9-13900K Raptor Lake CPU. Our original CPU was one of the impacted units that failed over time, which explains in retrospect why so much of the data felt a bit questionable. Thankfully, the new Ryzen 7 9800X3D system seems to be running just fine. Here are the 12900K results.

For these GPU benchmarks, we tested nearly every GPU released between 2016 and 2024, plus a few extras. All graphics cards were tested at 1080p medium and 1080p ultra, and we sorted the table by the 1080p ultra results. Where it made sense, we also tested at 1440p ultra and 4K ultra. All the scores are scaled relative to the top-ranking 1080p ultra card of the time, which in our suite is the RTX 4090 — especially at 4K and 1440p.

The above summary charts show the relative performance of the cards we've tested across the past several generations of hardware at 1080p ultra — swipe through the above gallery if you want to see the 1080p medium, 1440p, and 4K ultra images. There are a few missing options (e.g., the GT 1030, RX 550, and several Titan cards), but otherwise it's basically complete. Note that we also have data in the table below for some of the other older GPUs.

The eight games used for our standard GPU benchmarks hierarchy are Borderlands 3 (DX12), Far Cry 6 (DX12), Flight Simulator (DX11 Nvidia, DX12 AMD/Intel), Forza Horizon 5 (DX12), Horizon Zero Dawn (DX12), Red Dead Redemption 2 (Vulkan), Total War Warhammer 3 (DX11), and Watch Dogs Legion (DX12). The fps score is the geometric mean (equal weighting) of the eight games. Note that the specifications column links directly to our original review for the various GPUs.

*: GPU couldn't run all tests, so the overall score is slightly skewed at 1080p ultra.

While the RTX 4090 does technically take first place at 1080p ultra, it's the 1440p and especially 4K numbers that impress. It's less than 2% faster than the RTX 4080 Super at 1080p ultra, but that increases to 9% at 1440p and then 25% at 4K. Also note that the fps numbers in our table incorporate both the average and minimum fps into a single score — with the average given more weight than the 1% low fps.

Again, keep in mind that we're not including any ray tracing or DLSS results in the above table, as we use the same test suite with the same settings on all current and previous generation graphics cards. Since only RTX cards support DLSS (and RTX 40-series if you want DLSS 3), that would drastically limit which cards we could directly compare. You can see DLSS 2/3 and FSR 2 upscaling results in our RTX 4070 review if you want to check out how the various upscaling modes can help.

The RTX 4090 comes at a steep price, though on paper it's not that much worse than the previous generation RTX 3090. In fact, we'd say it's a lot better in some respects, as the 3090 was only a minor improvement in performance compared to the 3080 at the time of launch, but with more than double the VRAM. Nvidia pulled out all the stops with the 4090, increasing the core counts, clock speeds, and power limits to push it beyond all contenders. There are two problems with the 4090, however: It's not available at MSRP any longer, due to demand from the AI sector — it often costs $2,000 or more — and there are still concerns with pulling 450W of power over the 16-pin connector.

Stepping down from the RTX 4090, the RTX 4080 Super and RX 7900 XTX trade blows at higher resolutions, while CPU bottlenecks come into play at 1080p. We'll be switching our testbed in the near future, with the current results from our 13900K testing in the charts at the bottom of the page.

Outside of the latest releases from AMD and Nvidia, the RX 6000- and RTX 30-series chips still perform reasonably well and if you're using such a card, there may not be any need to upgrade at present. Intel's Arc GPUs also fall into this category and are something of a wild card.

We've been testing and retesting GPUs periodically, and the Arc chips running the latest drivers now complete all of our benchmarks without any major anomalies. (Minecraft was previously a problem, though Intel has finally sorted that out.) They're not great on efficiency, but overall performance and pricing for the A750 is quite good.

Turning to the previous generation GPUs, the RTX 20-series and GTX 16-series chips end up scattered throughout the results, along with the RX 5000-series. The general rule of thumb is that you get one or two "model upgrades" with the newer architectures, so for example the RTX 2080 Super comes in just below the RTX 3060 Ti, while the RX 5700 XT basically matches the newer and less expensive RX 6600 XT.

Go back far enough and you can see how modern games at ultra settings severely punish cards that don't have more than 4GB VRAM. We've been saying for a few years now that 4GB was just scraping by, and these days we'd avoid buying anything with less than 8GB of VRAM — 12GB or more is the minimum we'd want with a mainstream GPU, and 16GB or more for high-end and above. Old cards like the GTX 1060 3GB and GTX 1050 actually failed to run some of our tests, which skews their results a bit, even though they do better at 1080p medium.

Now let's switch over to the ray tracing hierarchy.

Ray Tracing GPU Benchmarks 2022–2024

Enabling ray tracing, particularly with demanding games like many of those we're using in our DXR test suite, can cause framerates to drop off a cliff. We're testing with "medium" and "ultra" ray tracing settings. Medium generally means using the medium graphics preset but turning on ray tracing effects (set to "medium" if that's an option; otherwise, "on"), while ultra turns on all of the RT options at more or less maximum quality.

Because ray tracing is so much more demanding, we're sorting these results by the 1080p medium scores. That's also because the RX 6500 XT, RX 6400, and Arc A380 basically can't handle ray tracing even at these settings, and testing at anything more than 1080p medium would be fruitless.

The five ray tracing games used are Bright Memory Infinite, Control Ultimate Edition, Cyberpunk 2077, Metro Exodus Enhanced, and Minecraft — all of these use the DirectX 12 / DX12 Ultimate API. The fps score is the geometric mean (equal weighting) of the five games, and the percentage is scaled relative to the fastest GPU in the list, which again is the GeForce RTX 4090.

If you want to see what the future may hold with ray tracing, check out our Alan Wake 2 benchmarks where the full path tracing barely manages playable performance even with upscaling on non-Nvidia GPUs. However, and this is a critical point, the number of games where RT truly makes a striking difference in visuals is extremely limited. For most games, we still feel running pure rasterization rendering makes more sense.