Video Cards & GPUs News - Page 1

NVIDIA's next-generation GeForce RTX 50 series "Blackwell" GPU configurations have been leaked: we now have the skinny on the GB202, GB203, GB205, GB206, and GB207 GPUs.

In a new post on X by leaker "kopite7kimi" we're learning that the flagship GB202 GPU -- the one that will power the GeForce RTX 5090 flagship graphics card -- will reportedly have a massive 512-bit memory bus, powered by ultra-fast, next-generation GDDR7 memory. GB202 will reportedly have 24,576 CUDA cores, which is a 33% increase of CUDA cores inside of AD102, the GPU that powers the RTX 4090.

AMD didn't talk about its next-generation RDNA 4 GPU architecture at Computex 2024 last week, but we know it's coming... and now we've got some Navi 48 XTX graphics cards that have been spotted in a new shipping manifest.

NVIDIA will have its powerful next-gen Blackwell-based GeForce RTX 5090 and GeForce RTX 5080 graphics cards later this year, which will be the new flagship gaming GPUs to buy... but the market under that, is what AMD is aiming at this generation. RDNA 5 will hopefully mark the return to the high-end GPU business for AMD, for now... Navi 44 and Navi 48 are coming.

Sure, we didn't get any official confirmation on when we can expect to see the GeForce RTX 5080 and RTX 5090, but word around Computex was that NVIDIA's next-gen GPU hardware was on track for an end-of-2024 launch. This is worth remembering because even though we got to see two very cool and unique GPU designs for the GeForce RTX 4080 SUPER and RTX 4090 from MSI, odds are these designs are for next-gen cards.

First, there's the new GeForce RTX 4080 SUPER EXPERT FUZION, which looks like the brand-new Founder Edition-style GPU we reviewed earlier this year when NVIDIA launched the RTX 4080 SUPER. It's a fantastic all-metal and minimal design. In EXPERT FUZION form, it's a little thicker (not by much) as it features an integrated 'Liquid-cooled Hybrid System' with a low-profile AIO-style pump.

Paired with the push and pull design of the new EXPERT Series from MSI, aluminum shell, custom fans, and copper baseplate, it's all about offering the benefit of liquid cooling without the need for a waterblock and external radiator. It's an impressive design that we assume will become part of the GeForce RTX 50 Series.

The switch to the new 12VHPWR or 12V-2X6 power connector standard for modern GPUs (specifically NVIDIA's GeForce RTX 40 Series) was a rocky transition, to say the least. For the high-end flagship GeForce RTX 4090, which draws up to 450W (that can be pushed higher), a steady stream of reported melted power connectors fusing with a GPU's PCB led to multiple investigations and even a redesign of the connector for the SUPER Series.

The cable melting issue might not be in the headlines every single day like it was back when the GeForce RTX 40 Series launched, but it's still an issue and something to be concerned about.

This makes the new WireVew PRO device we saw at Computex 2024 one of the most innovative little gadgets (especially for enthusiasts, reviewers, and overclockers). It's a simple monitoring device designed to mitigate issues relating to 12VHPWR or 12V-2X6 power connectors.

AMD might not have unveiled its next-gen RDNA 4 GPU architecture at Computex 2024 last week, but Radeon AIB partner PowerColor had an awesome concept Radeon graphics card with a built-in AI NPU that reduces power consumption by up to 20%.

The NPU (Neural Processing Unit) is something we've seen on Intel's current-gen Core Ultra 100 series "Meteor Lake" CPUs, a growing amount of AMD Ryzen CPUs -- and their upcoming Zen 5-based CPUs and APUs -- but now we're seeing graphics cards that will eventually have an NPU inside using AI to do their magic.

PowerColor was using a custom AI NPU chip on a concept graphics card at Computex 2024, separate to the GPU itself, which is capable of adjusting power consumption and fan profiles on-the-fly, reducing power consumption by up to (an impressive) 20% without much of a performance hit (not that you'll notice it at super high FPS, anyway).

Those desperate to witness the launch of next-gen graphics cards from AMD and NVIDIA may not want to hear the latest chatter from the GPU grapevine, as it points to delays with both Blackwell and RDNA 4.

As VideoCardz noticed, Tweakers has posted two reports on both next-gen GPU ranges sharing what's been gleaned from talking to connected sources at Computex, though we should add you must grab handfuls of seasoning to go with these assertions.

The first report was about AMD, and also Intel's, next-gen GPUs, and claims that both RDNA 4 and Battlemage won't be out until 2025, or this is the most likely case. Both ranges should debut at CES 2025, or that's the strongest possibility, according to Tweakers - though an announcement could be made late in 2024.

NVIDIA has been dominant for years, but in a new report from analyst firm Jon Peddie Research (JPR) we're seeing the AIB GPU market is absolutely dominated by NVIDIA with 88% market share, AMD with just 12%, leaving Intel with exactly 0% market share.

JPR's report states that in Q1 2024, 8.7 million units were shipped, down from 9.5 million units in Q4 2023. The market share for desktop discrete GPU suppliers "shifted" in the quarter, with NVIDIA's market share growing since last quarter by 8%, while AMD's share decreased by 7%.

Intel entered the AIB market in Q3 2022, close to two years ago now, with its Arc A770 and Arc A750 graphics cards, but they "remained flat" says JPR as the company "has yet to gain significant traction in the add-in board market". Meanwhile, NVIDIA holds a dominant 88% of the market.

Not many gamers of today remember how much the industry was changed by 3dfx in the 90s, but now we've got a modder who has developed a PCB that features the VSA-100 chip, the chip that powered the Voodoo4 graphics card back in the day.

The 3dfx Voodoo4 graphics card was released 24 years ago, designed for power efficiency in a market that was quickly adapting to the mega-success of 3dfx at the time (NVIDIA and ATi, which was later acquired by AMD). Modder "sdz" talked about this modding experience on the Vogons forums, going through the steps and issues that he came across during the custom modification of the Voodoo4 GPU.

The modder created a working Voodoo4 GPU for laptops through a nifty MXM connector, designing the custom MXM carrier and using an FPGA to setup the required hardware and software connections, which is paramount in using older hardware like this. The new Voodoo4 GPU inside of a laptop was made to work with Windows XP, which was an operating system released closer to when 3dfx was still around.

This year's Computex might have been light on next-gen GPU announcements from NVIDIA and AMD, but that doesn't mean GPU makers aren't prepping for the GeForce RTX 50 Series. Odds are we're going to see the RTX 5080 and RTX 5090 later this year, and at Computex 2024, Palit had a range of new 'prototype' GPU designs it was ready to showcase - albeit with current-gen hardware inside.

First up is the GeForce RTX 4080 SUPER MASTER, which features a full-metal die-cast shell made from aluminum where you can't see the screws. If you're familiar with Palit's RGB-bling GameRock series of GPUs, this is the complete opposite - a minimal, brushed metal look with a fully enclosed GPU and some minor angles on the end to give it character.

As a prototype, Palit notes that the design still has work left before it's finalized and ready for production - namely, fine-tuning the cooling to ensure it can handle a GeForce RTX GPU with up to 450W of power running at around 60 degrees. Palit wouldn't tell us if that means we'll see this design with the GeForce RTX 50 Series, but it's safe to assume that finalizing the design will happen once the team gets their hands on actual hardware and boards for the next generation of GeForce cards.

Intel has officially unveiled its next-gen Xe2 GPU which will drop into Lunar Lake CPUs later this year, while next-gen Arc "Battlemage" discrete GPUs will arrive after.

During the Intel Tech Tour 2024 event, Intel fellow and ex-NVIDIA staffer Tom Peterson talked all things next-gen Xe2 GPU architecture. Intel is making things easier to understand with its next-gen GPU architecture, where instead of using LP, LPG, HP, and HPG naming schemes (which can be confusing for people), Intel is calling its next-gen GPU lineup Xe2. Internally, the company will still use codenames, but they won't be used on the client side moving forward.

Intel has designed its next-gen Xe2 GPU from the ground up, with higher utilization, improved work distribution, and less software overhead. There are multiple (major) problems that were inside of Xe "Alchemist" GPUs, which are now fixed in Xe2. Intel has promised some major IP performance efficiency for Xe2, with performance gains of up to 12.5x in particular use cases.