Graphics Cards - Page 66

China's Zephyr has released the world's first ITX-ready GeForce RTX 4070, a dual-slot version of the card that is only 172mm long and arrives with a single fan. Dual-slot GeForce RTX 4070 cards aren't uncommon thanks to the Ada generation's excellent power efficiency, but the Zephyr GeForce RTX 4070 ITX Sakura Blizzard graphics card is the smallest we've seen.

Granted, the card's look isn't exactly minimal - it features a white shroud with holographic patterns, a pink PC, and a pink fan. The dimensions are impressive: 172 x 123 x 42mm - the world's smallest GeForce RTX 4070 GPU - powered by the same single 8-pin power connector. Under the hood, you also have the same AD104-250 GPU and 12GB GDDR6X memory, but with stock settings and no out-of-the-box OC.

As spotted by VideoCardz, this tiny GeForce RTX 4070 sold out on its first day of sale. Zephyr noted that a restock will be available in mid-July.

Intel Arc graphics driver 31.0.101.5590 WHQL for discrete GPUs and Intel Core Ultra rigs with integrated Arc graphics is here - adding support for a pair of new titles, Destiny 2: The Final Shape and the highly anticipated Elden Ring Shadow of the Erdtree expansion.

As per the headline, there's some added goodness for legacy titles, including Crysis (2007) and Crysis 2 (2011). The latest Intel Game On driver release boosts the original Crysis performance at 1080p with Very High settings by up to 10% on a discrete Arc GPU. Crysis 2 also gets up to a 6% increase in 1080p performance - this time on a Core Ultra chip with Arc graphics.

With that, the answer to the question of whether Intel Arc GPUs can run Crysis is now "better than before."

Compact builds are becoming increasingly popular each year. Small form factor (SFF) cases can lead to a smaller desk footprint or a rig that can sit snugly underneath a TV. The big concern, at least when trying to assemble an SFF build that is powerful enough to play the latest PC games, is whether or not the hardware will fit.

Noticing the trend, NVIDIA announced, as part of its Computex 2024 announcements, a new 'SFF Ready' label that denotes which GeForce RTX 4070 through to GeForce RTX 4080 SUPER GPUs can fit in smaller SFF cases. It all boils down to GPUs and SFF cases meeting specific dimension criteria (NVIDIA told us that pretty much all GeForce RTX 4060s are SFF Ready), and you can see all of the GeForce RTX cards (and compatible cases) here.

With that, ASUS is introducing a new line of budget-friendly GeForce RTX GPUs that are SFF Ready and called ASUS Prime. So far, the company has announced three models: the Prime GeForce RTX 4070 SUPER, Prime GeForce RTX 4070, and Prime GeForce RTX 4060 Ti.

AMD has teased a future-gen GPU architecture that would use a "multi-chipset" GPU design in a new patent filing... this is NOT RDNA 4... rather, this would be RDNA 5 or further out in AMD's Radeon GPU roadmap.

The company is very experienced in multi-chiplet designs, with AMD's Instict MI200 AI accelerators being the first to feature an MCM (multi-chip module) design, using multiple chiplets stacked onto a single package. This includes the GPCs (Graphics Processing Cores), HBM memory stacks, and the I/O die.

The newly-spotted patent describes the use of three different "modes" of chiplet utilization, with the differences being in how the resources are allocated and then managed ahead. The patent shows three different modes: the first is the "single GPU" mode, similar to how graphics cards with GPUs function now, all of the on-board chiplets would act as a single, unified processing unit, with resources shared in a collaborate environment.

XFX has 'something cool' coming, or more specifically something coming for the cooling solutions on its higher-end AMD Radeon RX 7000 graphics cards - namely replaceable fans.

VideoCardz spotted XFX publicizing this move on X, noting that the graphics card manufacturer has already made swappable fans a reality over in China with its Phoenix Nirvana products (a few weeks back, so not long ago).

It looks like the same idea will be coming to the US (and we'd imagine other regions as well), and if XFX follows in the footsteps of what it did in the Chinese GPU market, we will see RX 7900 XTX, RX 7900 GRE and RX 7800 XT graphics cards that have these swappable fans.

AMD's Radeon RX 6500 XT is an entry-level budget option for those looking to get into PC gaming without breaking the bank. The card and model are a few years old now, and with reference specs of 4GB of GDDR6 memory at 18 Gb/s on a 64-bit interface, it's not exactly something we could recommend in 2024.

That is, until now. Sort of. ASRock has launched a brand-new Radeon RX 6500 XT variant with the ASRock Radeon RX 6500 XT Phantom Gaming 8GB OC (RX6500XT PG 8GO). As the name suggests, this variant increases the VRAM capacity to 8GB, making it far more useful for playing modern titles at 1080p with low to medium settings.

Outside of the memory, you've got the same Navi 24 GPU, albeit with a boost to the out-of-the-box Boost Clock to 2825 MHz. Double the memory and an overclocked GPU? It sounds great for a GPU that retails for $169.99 USD (which is substantially lower in price than the entry-level GeForce RTX 3050 from the same generation). However, there is a catch.

AMD is running a new promotion for those who pick up a new MD Radeon RX 7800 XT or Radeon RX 7700 XT graphics card before August 17, 2024. The 'Why We Game' promo offers new Radeon GPU owners their pick of two games from a solid list of blockbuster titles: Avatar: Frontiers of Pandora, Starfield, Lies of P, and Company of Heroes 3.

AMD notable a promotion last year for the launch of Starfield, so being able to pick up two complimentary games with an eligible GPU purchase does add value to both the Radeon RX 7800 XT and Radeon RX 7700 XT - to the tune of around $100 USD when you factor in the cost of two games.

The promotion kicked off this week, so if you've been on the fence about what mid-range GPU to buy for 1440p gaming, this could make that decision much easier. Several models for both Radeon GPUs are sold at or slightly below MSRP. The Radeon RX 7700 XT probably presents the best value, as it can be picked up for under $400 USD.

NVIDIA's next-generation flagship GeForce RTX 5090 has some new leaked specs, with the GB202 GPU configuration being leaked onto X by kopite7kimi:

GB202 is going to be an absolute freaking beast, offering 33% more CUDA cores than the AD102 GPU inside of the GeForce RTX 4090, with GB202 rumored to pack in an incredible 24,576 CUDA cores. AD102 has just 18,423 CUDA cores in comparison. GB202 will power the new ultra-enthusiast flagship GeForce RTX 5090 graphics card, which is rumored to have a monster 512-bit memory bus with ultra-fast, next-generation GDDR7 memory.

RTX 5090 performance expectations: I'm expecting to see at least a 50% improvement in 4K gaming performance, with that coming from a mix of Blackwell GPU architecture improvements over Ada Lovelace, and the wicked new, ultra-fast GDDR7 memory on that deliciously huge 512-bit memory bus. We should expect at least 50-70% more 4K gaming performance with Blackwell + GDDR7 + 512-bit memory bus, and that's without DLSS 4 in the mix.

SK hynix is far ahead of its HBM competitors, but they're simultaneously behind their GDDR7 memory competitors, where the South Korean memory giant won't have GDDR7 memory ready until Q1 2024.

The last time we heard about SK hynix's GDDR7 memory was back at NVIDIA's GTC 2024 event (GPU Technology Conference) where we saw GDDR7 modules at 16-24Gb, up to 40Gbps speeds, and up to 160GB/sec memory bandwidth per module. The baseline of GDDR7 memory is 32Gbps and 128GB/sec bandwidth per module, so SK hynix has a rather nice 25% bandwidth improvement for future variations of GDDR7 memory.

Last week at Computex 2024, SK hynix said that its GDDR7 memory will enter mass production in Q1 2025, which should be after NVIDIA launches its GeForce RTX 5090 and GeForce RTX 5080 graphics cards. If mass production is coming in Q1 2025, then we should expect that SK hynix is already sampling customers (namely NVIDIA) right now.

NVIDIA has a fleet of next-gen Blackwell-based GeForce RTX 50 Laptop GPUs on the way, with new leaks suggesting we'll see 6 different variants, codenamed GN22.

The new leaks are coming from a hacking group that is interested in Clevo laptops, leaking out 6 x RTX 50 Laptop GPU variants codenamed X11, X9, X7, X6, X4, and X2: from fastest, to slowest. The two boards are being designed by Clevo as "GN22 Board 1" and "GN22 Board 2" which will house 3 individual SKUs of RTX 50 Laptop GPU.

According to the leaks, we can expect "GN22-X11" with 16GB of GDDR7 memory and "GN22-X9" also featuring 16GB of GDDR7 memory, while "GN22-X7" will sport 12GB of GDDR7 memory. Each of these RTX 50 series Laptop GPUs will be made onto Clevo's upcoming "GN22 Board 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.

GB202 will reportedly have a total of 12 SMs, each with 8 TPCs for a total of 96 TPCs and each TPC features 2 SMs for a total of 192 SMs. GB202 will support a massive 512-bit memory bus, but it can be sliced up into different configurations depending on the SKU: like the recent rumors of a 448-bit memory bus, but also down into 384-bit, 320-bit, and other designs.

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.

In the new shipping manifest, AMD's next-gen RDNA 4-based Navi 48 XTX GPU was spotted, along with some new board numbers. We've got:

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.