Graphics Cards - Page 4

Discord is one of the most popular chat and community tools among PC gamers, with over 200 million monthly active users. Although Discord is free for everyone, Discord Nitro is the platform's premium subscription that offers expanded upload capabilities, custom icons, emojis, themes, HD video streaming, and Xbox Game Pass in select regions. Plus, monthly discounts on Logitech G and SteelSeries gaming gear.

As part of NVIDIA's GeForce Rewards program, available for those who log into the NVIDIA App, GeForce gamers with a GeForce GTX 10 Series or above PC or laptop are eligible for the Discord Nitro Reward that offers three months of Discord Nitro. There are some terms and conditions, including that it's only valid for those who haven't subscribed to Discord Nitro in the past 12 months, but it's a great offer for those regular Discord users that haven't given Nitro a go.

This Discord Nitro Reward isn't the only thing happening this week, as NVIDIA is also giving away a custom-wrapped CONTROL Resonant GeForce RTX 5080 Founders Edition GPU, alongside codes for Remedy's highly anticipated CONTROL Resonant. All you need to do is go to the social media post (embedded below) for the giveaway, comment, and you're in the running.

The latest AMD Software: Adrenalin Edition 26.7.1 Driver for Radeon gamers is here. In addition to game support and fixing some outstanding issues, it also adds support for the new entry-level Radeon RX 9050 GPU that was quietly announced earlier today. The entry-level 1080p GPU is a cut-down version of the Radeon RX 9060 XT, with half the Compute Units and a power rating of only 92W.

As for new game support, this new driver offers optimizations for the Gears of War: E-Day Open Beta and Early Access weekend that kicks off on August 6 for those who preorder the game and PC Game Pass subscribers. This event will allow gamers to jump into the multiplayer side of the game early, so Radeon owners looking to jump in will definitely want to install the latest driver.

The other new game getting support with AMD Software: Adrenalin Edition 26.7.1 is Out of Control Evolution, a survival PvP game based on Rust and published by Tencent Games. As for the fixes, they're mostly for issues with Radeon RX 7000 Series GPUs, including one that fixes crashing when using AMD FSR Upscaling 4.1 for RDNA 3 on Windows 10 systems. Here's a look at the full Release Notes.

With official driver support and AMD's product page live, the company has quietly launched a new entry-level PC gaming GPU in the form of the Radeon RX 9050. Based on the same RDNA 4 architecture as the rest of the Radeon RX 9000 Series, the Radeon RX 9050 features 16 Compute Units, 1024 Stream Processors, and 8GB of GDDR6 memory on a 128-bit bus. On paper, it's a significantly cut-down version of the Radeon RX 9060 XT, with half the Compute Units and half the memory.

And with that, we've already got details on models from AMD partners ASRock, PowerColor, and SAPPHIRE. As an entry-level SKU, these Radeon RX 9050 GPU variants are all compact dual-slot, dual-fan cards from each partner's more affordable designs. Even so, the compact ASRock Radeon RX 9050 Challenger 8GB still sports a vented, reinforced metal backplate alongside an LED strip that adds some style to the design.

It also sports the company's Striped Axial Fans, a copper base for the GPU, and premium thermal pads for heat transfer. The compact Challenger design is one we're familiar with, with the specs confirming that it ships with the reference Boost Clock speed of 2600 MHz, alongside two DisplayPort 2.1a ports and one HDMI 2.1b port.

In a surprise move, AMD has quietly announced a new entry-level RDNA 4 GPU: the Radeon RX 9050, which will be available with 8GB or 4GB of VRAM. The official product or SKU page for the card is online, and at a high level it presents a significantly cut-down version of the 4nm Navi 44 chip that powers the mainstream Radeon RX 9060 XT, with 16 Compute Units and Ray Accelerators instead of 32.

With AMD noting that it's a 92W GPU, with 8GB of GDDR6 memory on a 128-bit bus delivering memory bandwidth of up to 288 GB/s, it's definitely something aimed at entry-level 1080p gaming, similar to NVIDIA's GeForce RTX 5050. On that note, as an RDNA 4 GPU, it benefits from the advanced AI hardware and video encoding/decoding capabilities of the latest Radeon architecture.

As for pricing and availability, AMD has confirmed a $279 SEP or MSRP with additional reporting by Wccftech's Hassan Mujtaba, stating that it will be arriving in limited markets at first - Asia-Pacific, Japan, and Latin America. As for the 4GB variant, this confusing option (4GB of VRAM in 2026 does feel antiquated) not only lowers the memory capacity but also the overall bandwidth. This model hasn't been officially confirmed, but was spotted in a product listing that has since been taken down.

NVIDIA has released a new WHQL GeForce driver, version 610.88, that adds game-ready support for Halo: Campaign Evolved, Gears of War: E-Day Beta, and seven bug fixes, including two from the previous hotfix driver.

Fixed gaming bugs feature five total fixes, including game stability improvements in Silent Hill F, and a game crash fix for Strange Brigade/Zombie Army 4: Dead War caused when the NVIDIA App in-game overlay is used. The driver also includes fixes for crashing and performance problems in Halo: Campaign Evolved and Path of Exile 2 that were introduced with the 610.82 hotfix driver last week. Under the "Fixed General bugs" patch notes for the 610.88 driver, a fix was introduced for RTX Dynamic Vibrance being incorrectly applied to Wallpaper Engine, and a crash fix for Paint.NET when NVIDIA Surround is enabled.

Only one known issue was listed in the patch notes, where enabling "Prefer Maximum Performance" may not be applied correctly. This problem is also present in the 610.82 hotfix driver.

RTX 50 series prices in China are increasing massively thanks to new price increases in GDDR6 and GDDR7 affecting NVIDIA and AMD graphics cards. WCCFTech reports that RTX 50 series GPU prices from official distributors selling Colorful cards have increased up to a whopping 60% above MSRP. MSI RTX 50 series graphics cards are also selling up to 20% higher than what they were just a week ago.

According to Colorful's distributor sheet dated July 24th, the highest-end RTX 5080 on the list, the Vulcan W OC 16GB, has increased to ¥13,200 or $1949. The highest-end RTX 5070 Ti on the list, the RTX 5070 Ti Vulcan W OC 16GB, is selling for ¥10,000 or $1,476.65. The highest-end RTX 5070 Vulcan X OC 12GB is selling for ¥7,200 or $1,063. The highest-end RTX 5060 Ti, the Ultra W OC 16GB is selling for ¥5,300, or $782. The highest-end 8GB variant, the Advanced OC, is selling for ¥3,950, or $583. The highest-end RTX 5060, the Advanced OC 8GB, is selling for ¥3,200 or $472.

For MSI, an MSI official distributor price sheet shows price increases of 10-20% for all listed RTX 50 series graphics cards in China compared to a week ago. The RTX 5090 D v2 Ventus 3X OC has increased in value from ¥22,999 ($3,396) to ¥25,999 ($3,839). The RTX 5080 Gaming Trio OC has increased in value from ¥10,999 ($1,624) to ¥12,999 ($1919), and the RTX 5070 Ti Gaming Trio OC has increased in price from ¥8,199 ($1,210) to ¥9,699 ($1,432).

Building an x86 system without any components from Intel, AMD, or NVIDIA is starting to become more feasible thanks to growing competition from Chinese CPU and GPU makers. YouTuber GPUSpecs published a video pairing the LX7G100 with a Zhaoxin KX7000 CPU to see how a gaming PC with no Intel, AMD, or NVIDIA components fares in games. Performance was playable, with more demanding AAA games such as Cyberpunk 2077 achieving just above 30 FPS.

The YouTuber initially ran the Zhaoxin KX7000 with 16GB of DDR4 and a Moore Threads MTT S80 GPU with 16GB of GDDR6 memory in a previous video. But the system struggled to play any graphically intensive games such as Red Dead Redemption 2 and Black Myth: Wukong at near 30 FPS due to immature graphics drivers. To fix this, the YouTuber swapped the MTT S80 for a Lisuan LX7G100 featuring more stable and mature graphics drivers.

The much more potent LX7G100 was able to boost frame rates to the point where performance was being bottlenecked by the CPU. Forza Horizon 5 achieved 62.5 FPS average at 1080p using low settings, Cyberpunk 2077 achieved 32.8 FPS average, and Red Dead Redemption 2 achieved 39 FPS average at 1080p medium settings. Shadow of the Tomb Raider in DX11 mode provided unplayable performance around 15 FPS, but in DX12 mode the game ran considerably better, achieving 62 FPS average at the highest settings at 1080p. Resident Evil 3 ran at 1080p, achieving 67.2 FPS average. Black Myth Wukong ran at 30 FPS average at 1080p low settings.

With the memory crisis seemingly taking a turn for the worse, with no end in sight, it looks like GeForce RTX pricing is set to increase. According to a new report over at BenchLife (via VideoCardz), NVIDIA has reportedly contacted its various partners, or AIBs, to inform them that the prices for GeForce RTX GPU kits are set to increase.

Like AMD, when it comes to supplying companies like ASUS, MSI, GIGABYTE, and others, with GPU hardware so they can assemble and ship their custom models, NVIDIA supplies kits that pair the GPU dies for various discrete GeForce RTX cards alongside GDDR7 and GDDR6 memory.

According to the report, the price increase covers both GDDR7 and GDDR6 products, which would refer to the GeForce RTX 50 Series as well as the resurgence of the GeForce RTX 3060, which utilizes the older memory technology. Now, a price increase for kits doesn't necessarily translate to a higher retail price, but with the cost of everything going up, that's looking likely.

Coinciding with AMD's Advancing AI 2026 event, the company has launched its latest AMD Instinct data center GPUs purpose-built for the AI era, with the Instinct MI400 Series led by the flagship AMD Instinct MI455X GPU. Built on a 2nm process, it sports 320 billion transistors, a maximum clock speed of 2.4 GHz, and 432 GB of high-speed HBM4 memory with a peak memory bandwidth of 23.3 TB/sec.

According to the company, the AMD Instinct MI455X GPU is its first GPU to be designed and built for rack-scale AI deployments, which is why it serves as a key part of the new AMD Helios rack, which features 72 Instinct MI455X GPUs for 31TB of HBM4 memory with an impressive 1.7 PB/s memory bandwidth and up to 2.9 exaflops of compute performance.

And with the AMD Instinct MI455X being built for Helios, this means the GPU has been designed with scale-up and scale-out capability with shared-memory capacity, communication bandwidth, and more connectivity compared to previous AMD Instinct GPU generations.

If you know your GPUs, the GTX 750 series only ever came in 1GB and 2GB variants at launch with 4GB models rolled out later by various AIB partners. However, according to YouTuber Budget-Builds Official, China has been repurposing old GTX 750 series graphics cards for the Chinese market, and modding them with 8GB of memory. The YouTuber managed to grab one of these cards for testing purposes and found the makeshift 750 Ti 8GB can technically play modern titles at above 30 FPS thanks to the extra memory capacity.

The card itself is made by a no-name third party, with a generic single-fan cooler and a GM107 Maxwell die transplanted onto a new PCB with eight GDDR5 memory modules running in clamshell mode. Despite being called a GTX 750 Ti, the card more closely aligns with the lower-end GTX 750 featuring 512 cores, and monitoring software identifies the GPU as a vanilla GTX 750. The only aspect of the card that matches the GTX 750 Ti is its clock speed.

Armed with 8GB of VRAM, the card reportedly achieved surprisingly playable frame rates in some of the more demanding games made over the past few years, including Cyberpunk 2077. The YouTuber was also able to crank texture quality to their maximum in the games they tested without sacrificing performance. In Cyberpunk 2077, the makeshift GTX 750 Ti 8GB achieved 34 FPS average with 1% lows of 26 FPS running at 720p and using FSR upscaling set to balanced.

Battlefield 6 just got its latest major update, Season 4, which introduces the biggest map in the game so far. Tsuru Reef is a large and dense island chain that sees a previous luxury resort become the battleground for land, sea, and aerial combat supporting all modes in the game. It also introduces a new wave system for the water that adds a dynamic layer to the series' all-out warfare.

Throw in new vehicles, weapons, and the return of the classic Wake Island map that's set to debut later in the season, and Battlefield 6 continues to evolve as it heads toward its first anniversary. However, all is not well for PC players rocking the latest GeForce RTX 50 Series graphics cards, as several players have been reporting issues with the game crashing when using the latest GeForce Game Ready 610.74 WHQL driver.

This has led the official Battlefield Comms account on social media platform X to issue a statement confirming the "NVIDIA GPU Driver Crashes on Certain Maps." The statement confirms that the issue affects NVIDIA GPUs, "more concentrated among players using GeForce RTX 50 Series cards," but not all PC gamers with GeForce RTX cards are affected.

When it comes to GPU monitoring, one of the biggest stories in recent weeks has been third parties restoring GPU Hotspot temperature readings on GeForce RTX 50 Series cards. Although unofficial, the consensus is that these readings can be beneficial in troubleshooting issues relating to cooling and performance, which is why tools like HWiNFO have now added this feature.

This week we've got another revelation for GeForce RTX monitoring, this time related to GPU memory. Thanks to the Brazilian Paulo Gomes mod team, who discovered a way to read temperature data on individual memory modules by accessing memory-mapped I/O (MMIO), which can then be mapped to physical modules surrounding the GPU die. According to the team, current memory temperature readings always show the current hotspot reading; this new mod expands this to cover individual modules.

And the good news is that this covers GDDR6, GDDR6X, and the new GDDR7 memory found on the GeForce RTX 50 Series. With the new mod, the latest HWiNFO beta includes these readings alongside the Hotspot.dll plugin for MSI Afterburner from Asder00. However, due to the different memory configurations, there is some nuance to the data. For GeForce RTX 5090 GPUs, it has 16 memory packages, but you've only got eight measurement points covering paired GDDR7 modules.

With the GeForce RTX 50 Series, MSI introduced new GPU designs alongside refreshed versions of its classic, tried-and-true models. One of the new design variants is the INSPIRE series, which features a compact form factor with a minimal, understated design best-suited for creators looking for a stylish GPU for their workflows and build. Of course, INSPIRE GPUs are still optimized for PC gaming, but deliver that with a more understated look.

And with that, when MSI introduced the INSPIRE lineup, the GPUs featured a brushed silver-and-gold finish with black fans and accents. Cut to 2026, and MSI is introducing a stealthy black variant (thanks, VideoCardz), with the product page for the new MSI GeForce RTX 5080 16G INSPIRE 3X OC BLACK now live. In addition to the black color, the revamped design also includes etching-like patterns on the front not present in the original design.

With dimensions of 288 x 112 x 50mm, the compact 2.5-slot design is compatible with ATX, M-ATX, and even ITX cases. The ZERO FROZR cooling is powered by MSI's new STORMFORCE fans with enhanced airflow via the custom core pipe and finstack design on the heatsink. It also includes a nickel-plated copper baseplate for the GPU and memory, with a vented metal backplate with MSI and GeForce RTX branding.

Popular YouTuber and enthusiast Der8auer has created a new conversion kit for the $7,000 RTX 5090 Lightning that removes the card's integrated AIO liquid cooling system and replaces it with a waterblock capable of being hooked up to a custom loop while preserving the Lightning's built-in screen and shroud. Der8auer showed off the new waterblock in a YouTube video and found the new watercooling system is capable of running the RTX 5090 GPU well below 60C while drawing 800W of power.

Der8auer showed his process of tearing down the RTX 5090 Lightning and modifying the original waterblock to work with a custom loop. During the teardown, the YouTuber showed off the pump inside the integrated AIO and the integrated waterblock, which Der8auer believes was made as cost-optimized as possible based on the waterblock's design.

The waterblock itself features a copper coldplate attached to an aluminum cover on top. The mod involved replacing the aluminum cover with a stainless steel plate featuring the necessary fittings and terminals to hook the GPU up to a custom loop. Furthermore, making the cover stainless steel was critical to ensuring any coolant a customer might use in the modded waterblock will not corrode.

NVIDIA's professional Quadro graphics cards are not what you would typically think of as gaming graphics cards. But a YouTuber has discovered that NVIDIA's Maxwell-based Quadro M4000 from 15 years ago makes for one of the most potent used bare-minimum entry-level gaming GPU buys in 2026. YouTuber Budget-Builds Official benchmarked the card, with an overclock, and found the GPU regularly outperforms the GTX 970 and RX 580 4GB thanks to its 8GB memory capacity, allowing the card to approach GTX 1060 performance. Best of all, he picked up the card for just $30 on the used market.

The YouTuber benchmarked the M4000 (with a hefty manual overclock) in a variety of games, including GTA V, Counter-Strike 2, Arc Raiders, Oblivion Remastered, Samson, Cyberpunk 2077, Fortnite, PUBG, Baldur's Gate 3, Red Dead Redemption 2, S&Box, BeamNG, Fallout 4: Anniversary, M&B Bannerlord, and Crysis Remastered.

The 15-year-old card couldn't play all of these games at a perfect 60 FPS, but all the aforementioned games were at the very least capable of achieving 30 FPS, including the aforementioned titles running on Unreal Engine 5. Oblivion, a title that demands 8GB of VRAM at a minimum, was able to run at 30 FPS using a 900p resolution.

We have seen cooling mods before, but this one really pushes the boundaries. Australian YouTuber TrashBench, known for pushing GPU cooling mods into genuinely unhinged territory, just tried his most extreme RTX 3080 build yet. In his latest video, he strips the card down and surrounds it with eleven separate fans plus an AIO liquid cooler, all in an attempt to see how far thermal headroom can actually translate into real performance.

The results are, surprisingly, a mixture of both good and bad. Temperatures dropped by a massive 33°C compared to the stock cooler, which is a serious swing. This result is on par with what TrashBench pulled off in his earlier RTX 3080 project, where an Arctic WS360 workstation AIO dropped VRAM temps from 101.6°C to under 50°C. However, the payoff in frames this time was rather underwhelming. Despite this ridiculous setup, gaming performance only rose by around 4 FPS on average.

That gap between thermals and FPS is something to note. The RTX 3080 is a power-limited card, not really a thermal-limited one, once you get it below the low 80s Celsius. NVIDIA already tuned the Ampere boost algorithm to chase clocks within a fairly tight power envelope, so once a cooler gets temperatures comfortably low, extra cooling stops translating into extra clock speed.

The RTX 50 Super series has been swinging between rumoured cancellations and signs of life depending on which leak you believe. The latest update lands somewhere in between. According to VideoCardz, at least one NVIDIA board partner has already received physical RTX 50 Super graphics cards, meaning development is past the prototype stage. The catch is that NVIDIA has reportedly told its partners to hold off on the launch.

The reason is the cost of 3GB GDDR7 memory chips. The entire Super series is built around these higher-density modules, which allow NVIDIA to increase VRAM capacity without changing the number of chips or widening the memory bus. A single 3GB GDDR7 chip currently costs between $60 and $70 according to VideoCardz's sources. A standard 2GB GDDR7 chip costs around $20. That is three times the price for 50% more capacity, and across a card with six or eight modules, the difference amounts to hundreds of dollars in additional memory costs per GPU.

Three Super models are affected: the RTX 5070 Super with 18GB, the RTX 5070 Ti Super with 24GB, and the RTX 5080 Super with 24GB. The RTX 5050 9GB is also on hold for the same reason. Without these 3GB chips, the Super series loses its entire identity, as the RTX 5070 Ti Super and RTX 5080 Super carry the same CUDA core counts as their standard counterparts. Only the RTX 5070 Super gets a minor 4% bump in CUDA cores.

In the past week, we've seen tools like HWMonitor, HWiNFO, AIDA64, and others expand their support for GeForce RTX 50 Series GPU monitoring by adding readings for GPU hotspot temperatures. This was a feature that was actively blocked by NVIDIA, even though it's been a staple of previous generations. It was third-party monitoring tools and engineering that restored access to hotspot readings, and it's opened the door for countless gamers to re-run benchmarks and tests to get a clearer picture of the thermal performance of their GeForce RTX 50 Series cards.

And with that, we're starting to see some concerning results across a wide range of cards. In one case, a user with a COLORFUL iGame GeForce RTX 5080 Vulcan graphics card found that their GPU hotspot temperature hit 98.5 degrees Celsius even though the overall GPU temperature reading was sitting at 68.6 degrees. This is a 30-degree discrepancy and enough of a cause for concern that they got in contact with COLORFUL's support team.

The good news is that COLORFUL's response was detailed and reassuring, noting that sustained temperatures above 90 degrees were abnormal, that if this continued after a sustained 10 minutes of gaming, it could indicate faulty cooling, and that they should contact "after-sales service" for next steps. But first, they suggest ruling out airflow issues and even dust to ensure that the cooling was working as intended.

MSI is now manufacturing some of its RTX 5060s with NVIDIA's GB205 die with the introduction of its new RTX 5060 "V1" graphics cards. Spotted by Uniko's Hardware on X, the new cards feature a different mounting hole layout compared to MSI's existing RTX 5060 lineup that matches RTX 5070 graphics cards. This subtle difference confirms that the new MSI RTX 5060 graphics cards use NVIDIA's larger GB205 die found in the RTX 5070.

The initial version of the RTX 5060 arrived with the GB206 die, which comes with 4,608 CUDA cores in total and powers both the 5060 and 5060 Ti. The GB206 die is physically different from GB205 in that it is designed with a square die and mounting holes matching its shape. The GB205 die is a more powerful Blackwell die responsible for powering the RTX 5070 but comes in a rectangular shape with mounting holes that reflect its shape, and a total of 6,400 CUDA cores.

There is no performance difference between the RTX 5060 versions; NVIDIA disables cores on each version to reach the GPU's designated 3,840 CUDA core count. On the GB205 version of the RTX 5060, the memory bus is also partially disabled to meet the GPU's 128-bit spec. NVIDIA and its partners commonly use multiple die variants for a single GPU model to save money by recycling defective dies that are still good enough for powering lower-end SKUs. For instance, a defective GB205 die with 2,560 defective CUDA cores is still good enough to ship in an RTX 5060.

One of the small but confusing aspects of the GeForce RTX 50 Series and RTX Blackwell launch was the deliberate removal of GPU hotspot readings. GPU hotspot readings are notably higher than the main temperatures you see when monitoring hardware, as they reflect maximum temperatures recorded by sensors at specific contact points. For the GeForce RTX 50 Series, these sensors are still there, but NVIDIA removed public access to that data. Until now.

Thanks to recent third-party discoveries, access to these GPU hotspot readings has been restored for anyone with a GeForce RTX 50 Series graphics card, and popular tools like HWMonitor have brought this feature back for the Blackwell generation. This has led to a flood of PC gamers and enthusiasts benchmarking and stress testing their GPUs to see the difference in temperatures.

When it comes to MSI Afterburner, one of the most popular GPU overclocking, tweaking, and monitoring tools, it seems these newly discovered GPU hotspot readings won't be coming to the app. And the reason comes down to, well, marketing and MSI's close relationship with NVIDIA.