Graphics Cards - Page 17

The new NVIDIA RTX PRO 5000 72GB Blackwell GPU, which the company says is now available, offers a memory-focused configuration that will sit alongside the existing NVIDIA RTX PRO 5000 Blackwell GPU with 48GB of VRAM. The new variant is targeted at AI developers, scientists, and creatives who require hardware for "modern, memory-hungry workflows."

With generative AI and the rise of complex, multimodal agentic AI, NVIDIA notes that today's large language models (LLMs) and AI agents are growing in both complexity and size - necessitating the release of a beefier NVIDIA Blackwell GPU. With 72GB of ultrafast GDDR7 memory, 50% more than the current 48GB model, and a memory bandwidth of 1,344 GB/s, you're looking at 2,142 TOPS of AI performance.

Naturally, having access to 72GB of GDDR7 memory on a 512-bit interface will allow AI developers to train, run, and tweak models locally. This adds an extra layer of privacy and reduces latency compared to a cloud-based solution.

With the product page now live, AMD has quietly launched a brand-new variant of its mainstream RDNA 4 GPU, the Radeon RX 9060 XT. The AMD Radeon RX 9060 XT LP graphics card is a low-power variant that lowers the TDP to 140W, 20W lower than the standard Radeon RX 9060 XT.

The Radeon RX 9060 XT LP is a 16GB model, so it includes 16GB of GDDR6 memory on a 128-bit interface with a total memory bandwidth of 320 GB/s - the same as the current 160W model. The GPU configuration is also identical with 32 RDNA 4 Compute Units and Ray Accelerators, plus 64 AI Accelerators. It also sports the same Stream Processor count of 2048 with 64 ROPs and 128 Texture Units.

The Radeon RX 9060 XT LP also features DisplayPort .1a and HDMI 2.1b connectivity with hardware-accelerated H264, H265, and AV1 video encoding and decoding. The only thing missing from AMD's product listing is the Boost Clock speed of the new Radeon RX 9060 XT LP variant, so this might be where the difference lies.

NVIDIA is reportedly planning on adjusting the production capacity for GeForce RTX 50-series GPUs in the first half of 2026.

The information has come from Board Channels and backed up by Benchlife, which reported the supply for the GeForce RTX 50-series GPUs may be cut by 30 to 40% compared to what it was in the first half of 2025. According to the report, the reason for the production change by NVIDIA isn't just GPU VRAM, but also memory used in the DIY channel, including motherboard-related memory.

Benchlife backed up the rumor by writing it has been informed by several AIC partners and component suppliers that NVIDIA will first adjust the supply of the GeForce RTX 5070 Ti and GeForce RTX 5060 Ti 16GB GDDR7. It should be noted the Board Channels report states the information isn't confirmed, and that it's rumored.

Colorful has launched new GeForce RTX 50 Series graphics cards specifically for Mini-ITX builds and compact rigs. The new Colorful iGame GeForce RTX 50 Series Mini lineup includes three models: an iGame GeForce RTX 5060 Ti Mini OC 8GB, an iGame GeForce RTX 5060 Ti Mini OC 16GB, and an iGame GeForce RTX 5070 Mini OC 12GB.

All three sport the same compact dimensions of 180 x 123 x 39.8mm, making the GeForce RTX 5070 variant one of the smallest we've seen to date, especially for an OC model with a Boost Clock of 2557 MHz, a small but notable bump from the 2512 MHz reference spec. And with a GPU length of only 180mm, all three models feature a cooling solution with a single 95mm fan.

Both GeForce RTX 5060 Ti variants are also OC models, sporting a Boost Clock speed of 2632 MHz, which is a notable increase over the 2572 MHz reference spec. The all-black design of Colorful's new Mini range of GeForce RTX 50 Series graphics cards also includes minor RGB lighting on the side and a reinforced metal backplate.

We knew it was coming, but no one knows how much and how bad it'll be... but Japanese gamers have noticed that the GeForce RTX 5070 Ti has gone up in price by $100 this week alone.

Most people are worried about the out-of-control DRAM shortage skyrocketing DDR5 memory prices, but the DRAM shortage also hurts anything with DRAM chips -- GDDR6 and GDDR7 memory included -- with multiple sources noticing it impacting the price of NVIDIA's GeForce RTX 5070 Ti and RTX 5070 graphics cards in Japan.

Gazlog reports that the RTX 5070 series is sharply rising in price across Japan, where it normally sells below MSRP. NVIDIA launched the GeForce RTX 5070 Ti at 148,800 Yen, but started selling for 126,000 Yen after a couple of months on the market. However, at 126,000 Yen, it makes the RTX 5070 Ti more expensive compared to the MSRP in the US. The RTX 5070 dropped to 82,000 Yen after its launch, after hitting the market in Japan for 108,000 Yen.

GIGABYTE has officially released its AORUS RTX 5060 Ti AI BOX featuring a fast Thunderbolt 5 port, and desktop RTX 5060 Ti 16GB GPU inside.

The main feature with the new AORUS RTX 5060 Ti AI BOX is the Thunderbolt 5 port, with GIGABYTE including support for Thunderbolt 5 daisy chaining, with the same cable used for charging with USB PD 3.0 rated for up to 100W to the connected laptop.

GIGABYTE is marketing the AORUS RTX 5060 Ti AI BOX as more than just an external GPU enclosure, with the Thunderbolt 5 port joined by 3 x USB 3.2 Gen 2 ports as well as Ethernet.

ASUS has quietly revamped the design of its Dual Series graphics cards for the GeForce RTX 5060 Series. Covering RTX 5060 and RTX 5060 Ti models, the new 'EVO' models offer a slimmer 2.1-slot design for small-form-factor builds, alongside a simpler dual-fan look that does away with the two-tone transparent finish of earlier GeForce RTX 50 Series DUAL designs.

The new models are available in four models, covering OC and non-OC variants, and they are: the ASUS Dual GeForce RTX 5060 EVO 8GB, ASUS Dual GeForce RTX 5060 EVO OC Edition 8GB, ASUS Dual GeForce RTX 5060 Ti EVO 16GB, and ASUS Dual GeForce RTX 5060 Ti EVO OC Edition 16GB. Interestingly, these new EVO designs are skipping the GeForce RTX 5060 Ti 8GB variants, which makes sense as the RTX 5060 Ti 8GB are mostly being sold as part of pre-built systems.

Another notable design change with these new EVO graphics cards is that the single 8-pin power connector for all four models is now located on the left side of the GPU, close to the stainless steel bracket. This is an unusual choice and not something that you usually find.

NVIDIA's RTX 3060 is finally going to sell through entirely likely later this month, after Team Green pulled the plug on the graphics card some time ago.

VideoCardz highlighted a post on the Board Channels, a source of a lot of hardware rumors in China - but one that should be regarded with some caution - insists that time has been called on the RTX 3060.

We're told that board manufacturers including ASUS, Colorful, Galax, GIGABYTE and MSI cleared their remaining RTX 3060 supply in September and October, with Zotac being the only player with a meaningful level of stock left after that - and that went dry in November.

When the mainstream and entry-level Intel Arc B580 launched in December 2024, we were all expecting to see the flagship mid-range Intel Arc B770 sometime this year. There have been several rumors pointing to a potential 'Big Battlemage' launch throughout this year, and just when we were about to lose hope, Intel updated its VTune Profiler performance analysis tool, adding support for the BMG-G31 GPU, which is the chip that will power the Intel Arc B770 or an AI-focused workstation equivalent.

Throw in shipping manifests detailing brackets and parts for a new desktop graphics card, similar to what we saw before the Intel Arc B580's launch, and you could say that the chances of getting an imminent announcement or a CES 2026 reveal are pretty good. Especially now that the official Intel Gaming account on the social media platform X responded to a fan who got excited about upcoming Intel architectures and technologies.

The post, which has since been deleted, basically name-dropped the B770, Panther Lake, and Nova Lake as exciting technologies on the horizon. In response, Intel Gaming opened with, "We're thrilled to hear you're pumped about the B770." Which would be a strange thing to say if there was no plan to release an Intel Arc B770 GPU with the BMG-G31 chip.

NVIDIA's monster TITAN Ada prototype GPU has been given the teardown treatment by overclocker "Der8auer", and it looks damn right beastly in the flesh. Check it out:

Der8auer has previously benchmarked the elusive NVIDIA TITAN Ada prototype GPU in the past, comparing it against the GeForce RTX 5090 and RTX 4090 graphics cards, then had a gander at the custom power adapter from NVIDIA. The power adapter converts dual 8-pin PCIe connectors into dual 12VHPWR connectors and is used for the huge 900W power design of the TITAN Ada prototype GPU.

The teardown reveals NVIDIA's huge quad-slot design and full metal exterior, comparing what he saw with the RTX 4090 Ti prototype GPU teardown from GamersNexus, noting that the backplate assembly on the TITAN Ada uses additional heat pipes and a different surface finish than the other prototype card.

AMD's FSR 'Redstone' update is now available for RDNA 4 owners to play with, improving image fidelity and performance across a wide range of games, including Call of Duty: Black Ops 7, which also features the brand-new FSR Ray Regeneration technology that improves ray-tracing lighting effects. Alongside the debut of FSR 'Redstone,' AMD has also released the AMD FSR SDK 2.1, now available on GPUOpen for all game developers.

AMD FSR SDK 2.1 includes the full FSR 'Redstone' suite of neural rendering technologies: AMD FSR Upscaling 4.0.3, AMD FSR Frame Generation 4.0, AMD FSR Ray Regeneration 1.0, and AMD FSR Radiance Caching (Preview).

As introduced with AMD FidelityFX SDK 2.0 (AMD has renamed all FidelityFX tech to the simple AMD FSR), once a game adds new FSR 'Redstone' technologies, future AMD Software: Adrenalin Edition driver releases for Radeon GPUs can automatically update the version of ML or AI-based technologies used.

AMD Software: Adrenalin Edition 25.12.1 is here, and it adds official support for FSR 'Redstone,' which launched earlier today. For a breakdown on everything you need to know about FSR 'Redstone,' check out our main story on the technology, which includes the new AI-powered FSR Upscaling, FSR Frame Generation, FSR Ray Regeneration, and FSR Radiance Caching - which now fall under the AMD FSR banner.

As we've seen with the launch of FSR 4 earlier this year, which has been renamed to FSR upscaling (ML), the FSR 'Redstone' suite is exclusive to the latest RDNA 4 and Radeon RX 9000 Series graphics cards. With over 200 games set to add FSR 'Redstone' by the end of the year, AMD FSR just got its most significant update since the technology first debuted.

The new AMD Software: Adrenalin Edition 25.12.1 drivers for Radeon GPUs are available for all RDNA gamers, as they also include some fixes for existing issues, alongside adding support for new workstation GPUs, the Radeon AI PRO R9600D and Radeon AI PRO R9700S. Here are the full Release Notes.

NVIDIA is celebrating the 5th anniversary of Cyberpunk 2077 by giving away an ultra-rare GeForce RTX 2080 Ti Cyberpunk 2077 Edition graphics card personally signed by NVIDIA CEO Jensen Huang.

This is all part of NVIDIA's huge GeForce Holiday 2025 giveaway event, where it has listed not one but two custom GeForce RTX 2080 Ti Cyberpunk 2077 Edition graphics cards, with one of them signed by Jensen. The entries are done through NVIDIA's various GeForce social channels, with NVIDIA running anniversary and CES-themed prompts.

CD PROJEKT RED launched the PC version of Cyberpunk 2077 five years ago now if you can believe it, being one of the key showcases of NVIDIA RTX technologies including 4 ray-traced effects including shadows, reflections, ambient occlusion and diffuse illumination, as well as DLSS 2.0 upscaling at the time.

AMD lifted the lid and officially launched its new suite of AMD FSR 'Redstone' technologies today, and with that, it has decided to simplify the naming. FSR originally stood for FidelityFX Super Resolution, an upscaling technology created in response to NVIDIA's AI-powered DLSS Super Resolution. With FSR 'Redstone' introducing new AI-powered technologies and features, AMD FSR now refers to all FSR technologies - past, present, and future.

And when it comes to Super Resolution, FSR 3 and FSR 4 now fall under the same branding - FSR Upscaling. Simple. Well, not exactly, as the new AI-powered FSR 'Redstone' technologies are exclusive to the latest RDNA 4 generation of GPUs - the Radeon RX 9060 XT, RX 9070, and flagship RX 9070 XT. This means there are two versions of FSR Upscaling: the newer RDNA 4-exclusive FSR Upscaling (ML) and the older FSR Upscaling (Analytical).

It's a change that, on paper, makes sense because AMD's Adrenalin drivers and software, or a game with native FSR support, will automatically choose the version depending on the Radeon GPU you have. In a way, it's reminiscent of how NVIDIA has handled its new upgraded Transformer model for DLSS Super Resolution, except the Transformer and older CNN model work across all GeForce RTX graphics cards.

AMD's FSR has entered a new era with the arrival of FSR 'Redstone' for the RDNA 4 generation of graphics cards, led by the flagship Radeon RX 9070 XT. By leveraging machine learning (ML) and AI, FSR 'Redstone' is a suite of four new technologies designed to enhance the performance and fidelity of PC gaming, beginning with the arrival of FSR 4's impressive AI-powered Super Resolution (FSR Upscaling) earlier this year.

And with that, FSR 4 is now called AMD FSR Upscaling (ML), powered by a new ML-based algorithm that delivers a dramatic improvement in image quality when upscaling from lower resolutions to 1080p, 1440p, or 4K. We've already covered this extensively in our reviews of various RDNA 4 GPUs like the Radeon RX 9060 XT, RX 9070, and RX 9070 XT, and it's a game-changer when compared to previous FSR versions like FSR 3 and FSR 4.

As part of AMD's big FSR 'Redstone' reveal, the company has confirmed that FSR Upscaling (ML) is exclusive to FSR 4, and that previous Radeon generations, including RDNA 2 and RDNA 3, will use the older non-AI version, now called FSR Upscaling (Analytical). This brings us to the next FSR 'Redstone' technology making its debut with today's announcement - AMD FSR Frame Generation (ML).

The Intel Arc B770, also referred to as Big Battlemage, has been quietly rumored to be on the way for almost a year now. Powered by the BMG-G31 chip with 32 Xe2 cores and 16GB of GDDR6 memory, it's expected that the Arc B770 would be a mid-range alternative to something like the GeForce RTX 4070 or RTX 5070 in much the same way the Arc B580 is a mainstream alternative to the GeForce RTX 4060.

In addition to Intel confirming the BMG-G31 chip in a recent software update, rumors about the Intel Arc B770 are ramping up, with new custom listings suggesting the mid-range Battlemage GPU could feature a TDP as high as 300W. This would be surprising, as the Intel Arc A770 featured a 225W TDP, and the Intel Arc B770's GPU would use a more efficient 5nm process.

Now, the custom listings refer to a shipment of GPU brackets, with similar codes and naming that were spotted before the launch of the Intel Arc B580. So even though "N38341-001 TASDK 300W GPU BRKT 0.8" doesn't specifically mention the Arc B770 or BMG-G31 GPU, it does match the previous naming utilized for desktop Arc products.

The Intel Arc B770, powered by the BMG-31 GPU, is commonly referred to as "Big Battlemage" and seen as the flagship mid-range desktop companion to the mainstream-focused Intel Arc B580 and Arc B570. Initially rumored to launch earlier this year, we've finally got some indication from the company that 'BMG-G31' is finally ready to launch.

As part of the latest update to the company's Intel VTune Profiler performance analysis tool for both Windows and Linux, Intel confirms support for new hardware - specifically "Intel Arc Battlemage (BMG-G31) and Intel Core Ultra 3 Processors (Panther Lake)." BMG-G31 is the larger desktop GPU variant of the BMG-G21, which powers the Intel Arc B580 and Arc B570. But even so, Intel has yet to announce or confirm the Intel Arc B770 formally.

As the flagship desktop variant of the 'Battlemage' lineup and architecture, BMG-G31 features up to 32 Xe2 cores and 16GB of GDDR6 memory on a 256-bit bus, with up to four SKUs reportedly in the works. Although it's great to see the BMG-G31 GPU pop up like this, the current DRAM shortage would make it a strange time to release a new 16GB GPU for PC gaming and AI.

When NVIDIA launched the RTX Blackwell-powered GeForce RTX 50 Series earlier this year, it was discovered that the company quietly dropped 32-bit support for CUDA. This meant that older titles that used the company's PhysX acceleration for in-game physics offloaded the physics calculations to the CPU, significantly impacting performance.

At the time, some users noted that playing a game like Borderlands 2 with PhysX enabled saw performance drop to below 60 FPS in 4K on the GeForce RTX 5090, when the RTX 4090 could easily push 120 FPS. Well, there's some good news on this front: NVIDIA has reinstated 32-bit GPU-accelerated PhysX support for select games on GeForce RTX 50 Series GPUs with the latest GeForce Game Ready 591.44 WHQL driver release.

"We heard the feedback from the community, and with the launch of our new driver today, we are adding custom support for GeForce gamers' most played PhysX-accelerated games, enabling full performance on GeForce RTX 50 Series GPUs, in line with our existing PhysX support on prior-generation GPUs," NVIDIA writes, confirming that it's reinstating support one game at time, with a focus on the most enduring titles from the 32-bit PhysX era.

GeForce Game Ready Driver 591.44 WHQL is here, adding new game support for the Battlefield 6: Winter Offensive update and Call of Duty: Black Ops 7, which both feature DLSS 4 with Multi Frame Generation. Although Call of Duty: Black Ops 7 support was included in the previous driver release, this update improves the fidelity of DLSS Ray Reconstruction when ray-traced lighting is enabled.

But even if you're not playing either of these games, it's a critical driver release for Windows 11 users as it fixes a Microsoft bug that tanks performance in some games by up to 50%. This issue was recently resolved in a hot fix driver release and has now been included in the official WHQL branch available for download from NVIDIA's main GeForce driver hub or via the NVIDIA App.

Per the full release notes, this new driver fixes the issue where "Users running R580 branch drivers (58x.xx) or newer may observe lower performance in some games after updating to Windows 11 October 2025 KB5066835." The new driver fixes several issues affecting game performance and reinstates 32-bit GPU-accelerated PhysX support for the new GeForce RTX 50 Series.

According to a new report from Tom's Hardware, citing an industry source, AMD has increased the prices of its Radeon graphics cards for the US market. AMD's AIB partners are now spending $10 more per 8GB of VRAM for GPU and memory kits, with these costs expected to be passed on to consumers.

For AMD's current RDNA 4-powered Radeon RX 9000 Series, this could see the MSRP for most models increase by $20. With this price increase, the flagship Radeon RX 9070 XT 16GB will see its $599 MSRP increase to $619, the Radeon RX 9070 16GB will see its $549 price increase to $569, the Radeon RX 9060 XT 16GB will see its $349 price increase to $369, and the Radeon RX 9060 XT 8GB will see its $299 price increase to $309.

AMD's Radeon graphics cards all use GDDR6 memory, and with the current crisis, the cost of obtaining modules for its consumer-focused GPUs has increased dramatically. Although the report doesn't specify a reason for the price increase, at this point it's almost self-evident.