Graphics Cards - Page 13

ASUS's flagship ROG Astral GeForce RTX 5090 OC graphics card has been chilled under LN2 and pushed to its limits by overclocker "Splave" to break the 3DMark Port Royal world record.

The 3DMark Port Royal world record now sits at a huge 47,375 points using the ASUS ROG Astral RTX 5090 OC and overclocker Splave, breaking the previous record set by overclocker OGS who used the GeForce RTX 5090D and 47,196 points: a record held for just a few days until Splave's new record was set.

Splave used LN2 cooling on his ASUS ROG Astral RTX 5090 OC Edition graphics card, with the overclocker hitting 219.33 FPS in Port Royal, a new world record for the RTX 5090. The average score for the RTX 5090 on Port Royal is a little over 36,000 points, so with a score of 47,375 we have a 31% increase over the average score... impressive stuff.

NVIDIA's RTX 5050 desktop graphics card has been keenly awaited for some time now, and it may go on sale a bit sooner than we hoped in the US.

As a quick refresher, NVIDIA revealed the desktop GPU alongside the laptop variant last week, pinning an MSRP of $249 on the former.

We were told that the board would launch in the second half of July - at least outside of Asia (it could arrive sooner in China, perhaps) - so that means we theoretically still have a couple of weeks to wait before we can buy this budget GPU.

Last week, NVIDIA's new DLSS 4 Super Resolution, which utilizes a more advanced and complex Transformer model, exited beta. Debuting alongside the GeForce RTX 50 Series earlier this year (with support for GPUs from the RTX 20, 30, and 40 Series cards), the new DLSS 4 Super Resolution sets a new standard for AI-powered upscaling that delivers incredible image quality that is often sharper and more detailed than native rendering.

It's become a key selling point for the GeForce RTX 50 Series thanks to it already being available in over 125 games and apps. With it officially exiting beta and becoming part of the DLSS Super Resolution and Ray Reconstruction SDK 310.3.0, it will start becoming available in more and more titles, with the older, existing CNN (Convolutional Neural Network) variant set to be phased out.

With an updated version of the DLSS SDK for game developers now available, which includes the official release of DLSS 4's new Transformer model for Super Resolution, we've also learned that NVIDIA was able to reduce the VRAM requirement for running the latest model.

Rumors suggest that NVIDIA will release 'SUPER' refreshes for the GeForce RTX 5070, RTX 5070 Ti, and RTX 5080 at some point in the future. If these refreshed GPUs are similar to the mid-generation upgrade we saw with the GeForce RTX 40 Series, then it's likely we'll get our first look at the lineup with hands-on demonstrations around the time of CES 2026.

With renowned leaker and GeForce insider @kopite7kimi posting the specs for the upcoming GeForce RTX 5070 SUPER earlier today, they've now followed this up with specs for the upcoming GeForce RTX 5070 Ti SUPER.

Like the RTX 5070 SUPER, the RTX 5070 Ti SUPER is receiving a significant 50% increase in VRAM capacity, expanding from 16GB of GDDR6 on a 256-bit interface to 24GB. It appears that for the GeForce RTX 50 SUPER Series refresh, NVIDIA is utilizing 3GB memory modules to expand the VRAM capacity of the lineup for 4K gaming and intensive creator and AI workloads.

NVIDIA launched the GeForce RTX 5070 earlier this year, with the popular mid-range successor to the RTX 4070 delivering around 20% faster performance at 1440p. While the arrival of DLSS 4 helps unlock a new level of performance and image fidelity for the RTX 5070, our initial review highlighted two main criticisms: the 12GB of GDDR7 capacity and the fact that raw performance was only slightly better than the GeForce RTX 4070 SUPER refresh from 2024.

Well, it seems that we're going to be getting a GeForce RTX 5070 SUPER at some point in the future (rumors are pointing to an early 2026 release). The latest leak comes from the renowned insider @kopite7kimi, who has posted specs for the upcoming GPU on social media, detailing the upgrades over the baseline GeForce RTX 5070.

First up, the big difference is that the GeForce RTX 5070 SUPER is set to receive a massive 50% increase in VRAM capacity, featuring 18GB of fast GDDR7 memory on a 192-bit bus. According to reports, NVIDIA is now able to cram more VRAM onto its SUPER series refresh for the GeForce RTX 50 Series thanks to the arrival of 3GB memory modules.

A new prototype external GPU solution has been tested by ETA Prime, featuring NVIDIA's new mid-range GeForce RTX 5060 graphics card with Thunderbolt 5 and OCulink connectivity. Check it out:

The new external RTX 5060 graphics solution uses the desktop RTX 5060 and not the mobile variant, with 3840 CUDA cores (versus the 3328 CUDA cores in the RTX 5060 Laptop GPU) with a higher 150W TDP (compared to the mobile version with 115W through Dynamic Boost), joined by 8GB of GDDR7 memory.

This isn't the final version of the external RTX 5060 graphics solution, but we should expect Thunderbolt 5 connectivity in the final version compared to Thunderbolt 4 in the prototype that ETA Prime tested, which has Thunderbolt 4 (USB4) and OCulink connectivity.

ASUS has introduced a couple of new BTF GPUs into its growing BTF ecosystem, with the new ROG Astral GeForce RTX 5090 BTF OC Edition, which is the first ROG Astral to use the detachable Graphics Card High-Power (GC-HPWR) power connector.

The new GC-HPWR is built on the improvements in version 2.5 of the connector, which has been engineered to supply over 1000W of power into the high-power slot, alongside the 12V-2x6 power connector that's found behind the PCIe connector. The new ASUS ROG Astral RTX 5090 BTF Edition has everything that is built into the non-BTF series card, but the new BTF series version features enhanced power delivery and compatibility.

ASUS's new ROG Astral RTX 5090 BTF Edition has the same (and kick-ass) four-fan configuration, vapor chamber, optimized fin spacing, while the phase-change GPU thermal pad makes sure there's maximum contact with the GB202 GPU die. ASUS says that its quad-fan thermal layout improves airflow and pressure by up to 20%, while additional features include real-time monitoring of temperatures and power draw through the 12V-2x6 power connector which can be tweaked through its in-house GPU Tweak III software.

Yesterday, NVIDIA formally announced the new GeForce RTX 5050 desktop and laptop GPU, positioned as a successor to the GeForce RTX 3050 and an affordable entry-level option for PC gaming, backed by the latest DLSS 4 technology. Performance-wise, NVIDIA claims that the GeForce RTX 5050 is 60% faster than the RTX 3050, the last 50-class GPU. And that's without DLSS 4 or Multi Frame Generation being enabled.

Interestingly, this is one of those rare GeForce RTX GPUs where the desktop and laptop variants share the same core specifications, except for a couple of notable differences. Both feature the same CUDA, Tensor, and RT cores, as well as 8GB of memory capacity; however, the laptop variant's memory is the newer GDDR7, whereas the desktop variant's memory is GDDR6.

This makes the desktop GeForce RTX 5050 the only GPU in the GeForce RTX 50 Series lineup that uses older GDDR6 memory. As for why, NVIDIA's Director of Global PR for GeForce, Ben Berraondo, responded to this very question on social media.

NVIDIA announced the arrival of DLSS 4 alongside its big GeForce RTX 50 Series reveal at CES 2025. One of the most significant advances that DLSS 4 brings is a shift to a new and advanced Transformer model for Super Resolution AI upscaling, replacing the older and existing CNN, or Convolutional Neural Network, model.

At a high level, the new Transformer model can examine every pixel in a rendered frame instead of a partial solution that makes estimates, resulting in a more detailed, sharper, and cleaner image comparable to a natively rendered image. With double the parameters, the Transformer model is also faster than CNN, reducing the latency from 3.25ms to just 1ms on a GeForce RTX 5090.

DLSS 4 has been a game changer for image quality. Using the Performance preset in 4K delivers better image quality than the previous version's Quality preset. DLSS 4's Transformer model is also used in its Ray Reconstruction technology, which dramatically improves the quality of ray-traced effects (as seen above).

NVIDIA formally announced the new desktop GeForce RTX 5050 gaming GPU yesterday, which will launch in the second half of July, starting from $249. As an entry-level RTX Blackwell GPU, it's being positioned as a successor to the GeForce RTX 3050. It offers 60% faster performance on average and supports the latest DLSS 4 suite of technologies, including Multi-Frame Generation.

With no Founder Edition model on the way, we've got our first look at GeForce RTX 5050 cards and models on the way from some of NVIDIA's biggest partners - ASUS, MSI, GIGABYTE, ZOTAC, INNO3D, and COLORFUL.

Kicking things off with ASUS, the company is set to launch four GeForce RTX 5050 models, including PRIME and two DUAL variants. The triple-fan PRIME models include an aluminium fin-stack heatsink and ASUS's Axial-Tech fans. The OC variant pushes the Boost Clock speed to 2707 MHz, a sizable increase over the 2572 MHz reference spec.

NVIDIA has officially announced the GeForce RTX 5050 desktop and laptop GPU. The desktop variant is the first new 50-class card for PC gaming since the GeForce RTX 3050. Availability is expected in the second half of July, and pricing will start from $249, making the entry-level GPU the most affordable option in the GeForce RTX 50 Series lineup.

In its announcement, NVIDIA confirms that "x50-class GeForce GPUs are among the most popular in the world," second only to 60-class GPUs like the RTX 3060. The RTX 5050 is being marketed as an entry-level RTX GPU, an upgrade for older 50-class systems with GTX 1050 or RTX 3050 GPUs, and a solution for home theater PCs.

Although NVIDIA's benchmark and performance charts for the RTX 5050 focus on DLSS 4 with Multi Frame Generation, there is some raw performance data for esports and competitive games like Apex Legends, Counter-Strike 2, and Overwatch 2. NVIDIA confirms that the new GeForce RTX 5050 is 60% faster on average, in raw raster performance, than the GeForce RTX 3050.

NVIDIA's new China-specific GeForce RTX 5090D graphics card is reportedly gearing up for an August launch according to new reports.

In a new article from the folks at Benchlife, the RTX 5090DD is launching in August, where they reported: "by the way, the GB202-240 GeForce RTX 5090DD (tentative) is expected to be released in August". We've already heard rumors of what to expect inside of the RTX 5090DD, with the same CUDA core count as the RTX 5090D, but with a cut-down configuration of GDDR7 memory.

In recent rumors, we heard that NVIDIA is preparing a GeForce RTX 5090DD with the same GPU core count as the RTX 5090D with 21760 CUDA cores and the tweaked GB202-240-K-A1 GPU. The VRAM has been scaled down to 24GB (down from 32GB on the RTX 5090) on a 384-bit memory bus clocked at 28Gbps and the same 575W TDP but leaks tease "and there's a surprise".

New reports indicate that NVIDIA is moving up the launch date for its new entry-level desktop graphics card, the GeForce RTX 5050 8GB, to July 1. Known insider MEGAsizeGPU has taken to social media to confirm that the original launch window of late July has been brought forward, with NVIDIA notifying its partners of the new date.

The GeForce RTX 5050 8GB GPU is set to be positioned as an entry-level option for PC gaming. It is the first desktop 50-Class GPU release since the GeForce RTX 3050 8GB from 2022, and will feature 2560 CUDA Cores, which is 33% fewer than the RTX 5060 and 16% fewer than the RTX 4060. Interestingly, it's the exact CUDA Core count as the GeForce RTX 3050.

Other specs confirmed by multiple sources are that the GPU will ship with 8GB of GDDR6 memory in desktop form, not the newer and faster GDDR7 memory found in the rest of the GeForce RTX 50 Series line-up. The good news is that the GDDR6 memory will be clocked faster, delivering 320 GB/sec of memory bandwidth.

That's not a typo. Some rare GPUs with both GeForce and Radeon branding are making their way into gamers' hands. This looks to be limited to ASUS's TUF Gaming models, as both TUF Gaming Radeon RX 9000 Series and GeForce RTX 50 Series GPUs now share identical shrouds and cooling.

As seen in our reviews of the ASUS TUF Gaming Radeon RX 9070 OC Edition and ASUS TUF Gaming GeForce RTX 5070 OC Edition graphic cards, the only thing visually separating the two is the 'AMD Radeon' and 'GeForce RTX' branding. With that, it looks like a production issue or assembly mix-up is causing Radeon and GeForce RTX-branded TUF Gaming components to be put on the wrong GPUs.

Over on Reddit, user Blood-Wolfe posted a picture of their brand-new ASUS TUF Gaming Radeon GeForce RTX 5070 Ti GPU, stating they would exchange it. The latest case occurred four days ago, with Reddit user Fantastic-Ad8410 posting images of their rare ASUS TUF Gaming GeForce Radeon RX 9070 XT.

In a game where you come across vast open fields or dense forest-like environments full of foliage and trees, these are often created using procedural algorithms that require large amounts of memory. AMD just released a new paper titled 'Real-Time GPU Tree Generation.' It could be a game changer for developers creating large open-world environments to explore.

AMD and Coburg University of Applied Sciences and Arts, Germany's approach shifts everything to the GPU, using Work Graphs with mesh nodes to generate "gigabytes-worth of tree geometry from kilobytes of generation code." What does that mean? Well, a detailed polygonal forest full of trees and complex leaves that could theoretically require 34.8 GB of memory to store all of that data is cut down to a remarkable 51 KB. Yes, that's Kilobytes.

AMD's paper tests its innovative approach to procedural tree generation by testing its performance on a Radeon RX 7900 XTX. It notes that rendering a scene with unique trees, fully animated, takes just 3.13 ms on the GPU.

ASUS's flagship ROG Astral GeForce RTX 5090 graphics card has been shunt-modded by professional overclocker Der8auer, pushing its power limit to a much higher 800W with some interesting results.

The liquid-cooled ASUS ROG Astral RTX 5090 is a beast on its own with a larger 2.5-slot thermal solution, with a 360mm AIO cooler and three fans, as well as the fan on the graphics card itself. In its stock form, the ASUS ROG Astral RTX 5090 has a 600W power limit and can't be changed through software, with the card designed to be fed 600W.

The 12V-2x6 power connector on the ASUS ROG Astral RTX 5090 is designed for 600W of power flowing into it, but that's not the maximum that the cable can handle. This is where Roman "Der8auer" Hartung steps in, shunt-modding the card and allowing a higher 800W power limit.

China has just launched the nation's first 6nm GPU which was meant to rival NVIDIA's mid-range GeForce RTX 4060, but instead, it's only keeping up with the 13-year-old GeForce GTX 660 Ti graphics card.

Lisuan Technology announced the power-on of its new G100 GPU last month, powered by its new in-house TrueGPU architecture, manufactured on the 6nm-grade process node, which is most likely coming from SMIC. The new G100 GPU is meant to have performance on-par with the RTX 4060, but it falls (very) short of those claims.

The new G100 GPU in its Geekbench testing reveals 32 Compute Units, 256GB of VRAM, and a 300MHz GPU clock. The 300MHz GPU clock seems quite odd, which points to how the performance is so low... so much so, that we have performance that has the new China-made GPU sitting at the GTX 660 Ti levels of performance. The same goes for 256GB of VRAM, which is quite a hefty amount of VRAM.

AMD's next-generation UDNA architecture has had some new leaks teasing performance from next-gen Radeon graphics cards, the PlayStation 6, and next-gen Xbox... and it sounds fantastic.

In a new post on the NeoGAF forums, leaker Kepler_L2 teased some new leaks on the next-gen UDNA architecture that will power future Radeon GPUs on the desktop, as well as the next-gen consoles in the PS6 and Xbox.

Kepler said that we should expect around 20% or so additional performance out of the next-gen UDNA GPU and "around" 2x performance uplift in both ray tracing (RT) and AI. Kepler was asked if the home console SoC setup would have 3D V-Cache, to which he simply replied with "no". Lastly, VolticArchangel asked about the GPU side, and whether we'd see 64 CUs with 6 disabled, to which he said "don't know yet".

ASRock has just unveiled its new Radeon AI PRO R9700 Creator graphics card, based on the PRO R9700 GPU that the company launched earlier this week.

The new ASRock AI PRO R9700 Creator graphics card is aimed at both edge AI acceleration and professional visualization markets, and it's based on the Navi 48 GPU on TSMC's new 4nm process node, meaning the GPU has its maxed-out specs including all 64 CUs with 4096 stream processors, 128 AI accelerators, and 64 RT accelerators.

The big difference between the gaming-focused Radeon RX 9070 XT and PRO R9700 is that the workstation GPU features 32GB of GDDR6 memory versus 16GB on the gaming GPU. VRAM helps out with AI workloads, so with double the GDDR6 memory, the new ASRock Radeon AI PRO R9700 Creator is geared and ready to go for higher-end workloads.

AMD's next-generation Radeon GPUs based on the new UDNA (or RDNA 5) GPU architecture will support the upgraded HDMI 2.2 standard with 80Gbps of bandwidth according to the latest leaks.

In a new post on X by leaker @Kepler_L2, we're hearing that UDNA-based Radeon GPUs will support the new HDMI 2.2 standard with 80Gbps of bandwidth, upgraded from the HDMI 2.1 ports on current-gen Radeon GPUs that offer just 48Gbps of bandwidth in comparison.

AMD's next-gen Radeon GPUs will reportedly offer two different HDMI 2.2 modes with 64Gbps and 80Gbps, which is a big upgrade over HDMI 2.1 and its 48Gbps of bandwidth, but less than the 96Gbps ceiling of HDMI 2.2 in its full unleashed mode.