A few years ago, Nvidia was expected to release the RTX 3070 Ti with as much as 16GB of VRAM instead of the 8GB we actually ended up getting. Only database listings ultimately indicated its existence until a physical prototype showed up, but we never really got to see it working. Fast forward to today, however, and Brazilian YouTuber fmklab, aka Fabian, has recreated the card on his own. Using an RTX 3070 PCB with a transplanted 3070 Ti core and GDDR6 chips from Samsung, fmklab broughall those parts together to achieve a 16GB memory pool.

The card was stripped down to its PCB, and the RTX 3070 GPU was removed and replaced with a 3070 Ti chip. The reason for the swap is that the RTX 3070 uses GDDR6 instead of the GDDR6X the retail version of the RTX 3070 Ti is equipped with, which presented an interesting challenge.

Fabian could've put all 16GB of GDDR6X memory on a 3070 Ti card, but it would've been far too expensive, so he swapped the Samsung-made HC14 (8Gb) GDDR6 chips that were already on the board with HC16 (16Gb) chips. That changed the VRAM capacity from 8GB to 16GB, but when the card booted, GPU-Z only recognized 8GB. The PCB's memory straps needed to be reconfigured in order for it to use the entire memory pool.

Using the Nvidia BIOS Reader tool, Fabian figured out exactly what straps to adjust and went ahead with the soldering process. Afterward, MATS properly recognized the full 16GB VRAM capacity with no errors. GPU-Z also showed the RTX 3070 Ti equipped with 16GB of GDDR6 memory, similar to the prototype that popped up years ago. But verifying that the card recognized the additional VRAM was just the first step.

To actually test his creation's performance, Fabian ran a local LLM on the GPU to see if it would allocate more than 8GB and found that it used nearly the entire 16GB pool. Call of Duty: Warzone ate more than 13GB of VRAM while paired with a Ryzen 7 9800X3D, exhibiting no signs of instability.

Fabian even linked a Unigine Superposition benchmark in the video's description as proof of the card's existence, where he scored 5,367 points, which is actually lower than most real 3070 Ti runs. That's probably because of the decreased memory bandwidth of GDDR6 versus that chip's stock GDDR6X, which is the biggest tradeoff for the higher capacity.

Overall, this was a successful experiment and possibly the only working 3070 Ti 16GB in the world, but there comes a point of diminishing returns. A 3070 Ti is still a powerful card, but it gets smoked by current-gen Blackwell options even with less memory. It would be much more reasonable to buy, say, an RTX 5060 Ti 16GB if you actually need better performance with some future proofing today, not to mention the superior efficiency, feature set, and after-sales support you'd get from a new Blackwell card.

There are now several mods enabling DLSS 5 on older Nvidia GPUs, along with even AMD's RDNA 4 cards. Similarly, Nvidia's multi-frame gen has also been ported to prior generations. It can get a bit confusing to keep track of all these developments, and that's where "DLSS Unlocked" comes in. It's a new mod from ShyVox that combines OptiScaler_DLSSNR with DLSS Enabler to bring DLSS 5 and DLSS MFG to all RTX GPUs: Turning, Ampere, and Ada.

DLSS Unlocked serves as a sort of all-in-one package for owners of RTX 20-, RTX 30-, and RTX 40-series looking to enjoy the latest Nvidia goodies on both Windows and Linux. It takes two separate mods developed by other community members and organizes them into one seamless, unified release. You download the executable from GitHub or Nexus Mods and just select the root directory of the game you want to try it on.

As for the requirements, the game must use the DirectX 12 API and already support both DLSS upscaling and frame generation natively. You need to provide your own patched DLSS 5 DLL as well, for legal reasons. The two mods comprising DLSS Unlocked also have their own peculiarities. OptiScaler_DLSSNR is simple; it just uses OptiScaler to add neural rendering on top, but DLSS Enabler works in a more complicated manner.

On RTX 20- and RTX 30-series GPUs, you'll be using AMD's FSR 3.1 to interpolate all of the frames instead of using Nvidia's pipeline. On RTX 40-series, the second frame is generated through DLSS-G (regular frame-gen), but the rest of the frames are handled by FSR 3.1 to get you up to 6X MFG. That handoff is itself possible thanks to another mod called DLSSG to FSR3. See how this web of tweaks can get a bit confusing?

You also need to keep in mind that this is a different approach compared to DLSS 5 Swapper, which enables DLSS 5 inside any game through ReShade (using RenoDX). For DLSS Unlocked, the game must already support DLSS 2/3 upscaling for it to be patched into DLSS 5 and unlock MFG. That's how OptiScaler_DLSSNR works, which DLSS Unlocked relies on.

None of these mods should be used with multiplayer games because the anti-cheat software can detect them as threats and potentially ban you. Ultimately, when you have DLSS Unlocked installed, a DLSS-compatible game should now have DLSS 5 neural rendering while having its frames interpolated up to 4x. You definitely need that, considering DLSS 5 comes with a massive compute cost that can purportedly even melt RTX 5090s.

Nvidia's GeForce RTX 50-series (codenamed Blackwell) graphics cards continue to rank among the best graphics cards available today. As a result, these highly sought-after Blackwell graphics cards have become prime targets for criminals seeking to scam unsuspecting buyers. Although scams involving hollow graphics cards are not new, their growing frequency and spread are cause for concern. The latest reports from Germany have revealed at least two buyers who fell prey to the same scam, reportedly from the same seller.

Hollow or phantom graphics cards may not be colloquial terms you hear every day. They basically refer to graphics cards that lack the silicon and memory chips. Criminals remove these components and sell non-functional graphics cards to their prey. Logically, you have no way of knowing these components are not present until you disassemble the graphics card, something that the average consumer does not do. Swindlers have run similar scams with fake memory modules without integrated circuits, or SSDs with microSD cards inside.

The first report, as recounted by respected hardware enthusiast Roman “der8auer” Hartung, uncovers a poor victim who bought a second-hand Palit GeForce RTX 5090 GameRock OC for around $2,090 (1,800 euros). The seller’s asking price was suspicious since a brand-new one starts at around $5,805 (4,999 euros) in Germany. Even for a used one, the dramatic price difference indicates a potential scam.

Many of these scams happen through online transactions. So the buyer went so far as to meet the seller in person, hoping a face-to-face interaction would ensure the graphics card was legit. It did not make a difference, though, because the buyer did not have a system to test the graphics card on the spot. In the end, it was a matter of taking the seller's word for it, and that turned out horribly.

A second report has emerged from a member of the Hardwareluxx forums, who reportedly fell victim to the same deceptive seller. Tragically, this individual ended up paying close to $2,613 (2,250 euros) for the phantom graphics card.

In both cases, the scammer put on a friendly facade and provided what seemed like legitimate paperwork to bolster the buyers' trust. For the first victim, the seller shared an invoice showing the purchase of two graphics cards, claiming to sell one because they no longer needed it. Similarly, in the second case, the scammer furnished an invoice to convince the buyer of the transaction’s legitimacy.

The use of realistic invoices is a growing trend among fraudsters. The first buyer uncovered that the serial number printed on the graphics card itself did not match the serial number listed on the packaging. The second buyer discovered that the serial numbers on the invoice and the packaging matched, but the seller deliberately scratched off the serial number on the graphics card.

The AI gold rush will not be over anytime soon, so we do not expect graphics card pricing to improve. This gives scammers more motivation to prey on buyers' desire for a graphics card and lure them in with prices that seem too good to be true. The two incidents serve as a cautionary tale for anyone trying to procure a graphics card, especially from second-hand sources.

Not that long ago, we reported that Nvidia was dusting off the blueprints for the RTX 3060, in its 12 GB form. At the time, we'd spotted it in stores for about $339.99, but just like with ever-climbing memory, hard drive, and SSD prices, two months passed is an eternity. The same cards are now selling for $489 new, and one particular specimen is the SRhonyra RTX 3060 12 GB Low Profile card, for $495.59.

• SRhonyra RTX 3060 12 GB Low Profile card - $495.59

This card and its price may raise more than a few eyebrows, but there are reasons why it exists. First off, it's a one-slot model, making it easy to put many of them to work in the same machine with relatively little concern for airflow. They have no power inputs and rely on 70 W delivered by the PCIe slot alone, eschewing the need for a high-end PSU and lots of cables. Third, the low-profile form factor makes it possible to place them in potent puny personal computers.

Attentive readers might surmise that one (or more) of these would be good candidates for an entry-level local LLM rig. That's precisely how SRhonyra is pitching the card, calling it "capable of local AI" and "running 7B [to] 13B LLMs." A standard-dimensioned, fully powered RTX 3060 is capable of drawing a maximum of 175 W, so it's fair to assume the performance of the diminutive variant will sit below that of its full-sized brethren, given it ought to only draw 70 W of juice from the PCIe slot.

Even then, it's likely that people interested in these cards are looking to use them either as secondary GPUs, or use more than one in the same box to be able to virtually pool their VRAM and use larger models than you'd otherwise be able to. The low power draw also means they're a simple, thoughtless drop-in to an existing system, whereas larger, more power-hungry cards require careful consideration with physical spacing (or lack thereof), power supply sizing, and ever-annoying cables.

Nvidia's DLSS 5 Neural Rendering technology has taken the PC gaming world by storm since its recent leak and then official rollout in recent days. In what might be one of the most impressive technical applications and mods of the feature yet, one developer has showcased DLSS 5 running on a second GPU to share the processing load, rendering the game on the first GPU before applying Neural rendering at the end of the frame, thus boosting the performance of neural-rendered frames.

Marcelo Guibout shared the demonstration online, with videos showing the process running on a cinematic video from The Blood of Dawnwalker, as well as Cyberpunk 2077. Guibout was quick to clarify that the videos are technical showcases, not benchmarks. However, they did share some performance figures. More exciting still, you can download the project from GitHub and try it for yourself. However, the technique does require a second display and doubles the display latency.

The above demonstration features a Ryzen 7 7800X3D, 32GB DDR5 setup with two Nvidia RTX 5060 Ti 16GB GPUs, with both cards using PCIe 5.0 x8, and a display attached to each card. The ReShade add-on, dubbed MGPU Bridge, reads each finished frame before applying Nvidia's Neural rendering. "Neural rendering happens at the end of the frame: it takes a finished frame and hands a finished frame back," they explain. "That is what makes it possible to pick it up and run it somewhere else. The add-on creates its own D3D12 device on your second GPU, sends each finished frame across to it, runs DLSS-NR there, and displays the result on that card's own monitor — so nothing has to come back. The render GPU does no neural work at all and runs cooler for the same reason."

In the TBOD demo at 1080p, they shared the following performance numbers:

Guibout clarified that DLSS super-resolution is running in every column of the test, with the rows showing each mode tested. The game rendered at the same internal resolution in all three columns, with the first column giving figures with DLSS 5's neural rendering switched off, representing the ceiling for performance. As you can see, running DLSS 5 on the second card in this dual-GPU setup drastically increases performance in every mode.

"The frame rates are not the finding. The slope is: what you gain by going from DLAA down to Ultra Performance, and how much of that available gain each arm keeps," Guibout explains. "Neural post-processing saturates whatever device it runs on, and it always runs at output resolution. Its cost barely falls as you drop the DLSS mode, while the render work collapses. On the render card it therefore eats a larger and larger share of every frame, and upscaling stops paying for itself: you keep about a third of what the machine actually had to give. Move it to the second card and you keep 86% of it."

Other benefits include temperature reductions on the rendering card, with the first GPU running 21 degrees cooler without the neural load.

Guibout is cautious to note this isn't a return to Nvidia's SLI technology, which would split the workload of frame rendering between two GPUs. They also noted plainly this is not Nvidia's vision for DLSS 5 and lacks the native integration of official DLSS 5 support.

You can see the original demo for TBOD below:

As commenters online have noted, the method seems more akin to using Nvidia's dedicated PhysX cards, with plenty of potential for gamers who have the capacity to run a dual GPU setup. The second demo features a Ryzen 5 5600 on a DDR4 system, so top-of-the-line hardware isn't required to make it happen.

The RTX 5090 is no stranger to controversy thanks to the incendiary nature of its 16-pin connector. When you add something as taxing as DLSS 5, which comes with a heavy increase in power draw, to an already dangerous mix, you're just cooking up a recipe for disaster. And it seems like this recipe has likely claimed its first victim. HardOCP forum member Erek claimed their RTX 5090 fell prey to the woes of the 16-pin connector yesterday, as it melted while testing DLSS 5 in NBA 2K27.

The irony is not lost on us that NBA 2K27 is the only game that officially supports DLSS 5 right now — everything else is modded — yet it still reportedly killed the user's MSI RTX 5090 Gaming Trio OC, which carries a 575W TDP. However, during testing, erek reported his GPU going past 600W, with GPU-Z showing it hovering above 610W consistently, not just as a spike. Before this, the card only consumed around 450W while gaming, implying that DLSS 5 added more than 150W to the power draw.

While the 16-pin connector is capable of carrying up to 600W of power, the PCIe x16 slot the card is plugged into also provides 75W on its own. So, the connector wasn't necessarily drawing over 600W alone. Tragedy struck right after as Erek's RTX 5090 started exhibiting signs of instability and eventually stopped working entirely. The owner smelled something burning and, sure enough, found the 16-pin connector pulling an age-old classic: melting.

The plastic on the connector had melted enough to bond with the cable, making it very difficult to separate the two. Once they did come undone, the damage was clearly visible, with at least five pins on the 16-pin cable scorched; one pin's housing had completely melted away. This was a yellow-tipped cable, too, meant to ensure it was plugged all the way in. Despite the fact that we only got a blurry picture of the connector on the GPU, the little yellow dot of melted plastic almost comically stands out amidst a sea of black and grey.

We can't confirm whether DLSS 5 was actually responsible for burning Erek's card. After all, an RTX 5090 meeting its maker is a monthly tradition at this point, and the cause is usually tied to a loose fit between the connector and the cable. However, our own testing does align with how the enthusiast came up with their DLSS 5 hypothesis. We saw the MSI Lightning Z variant of the 5090, which has 2x 16-pin connectors and a 1,000W XOC BIOS, pull 800W in Control with DLSS 5 enabled. Without DLSS 5, the same GPU pulled just 691W.

That wasn't even the most egregious difference; in Hogwarts Legacy, we went from 480W without DLSS 5 to a whopping 720W with it enabled. That's a 50% increase in power draw for just one feature. The same game made the Founder's Edition of the 5090 pull 417W without the neural rendering tech involved, and 547W with it toggled on. So even in the standard version of the GPU with a single 16-pin connector, there's a 31% jump in power consumption because of DLSS 5.

Considering Erek's 5090 model lands somewhere between those two, and given his own testing showing a severe power budget imbalance, it's not unreasonable to assume DLSS 5 was the culprit. Had the user's GPU been acting weird before they tried the new feature, we might be having a different conversation. But it's clear that Nvidia has a lot of work to do before DLSS 5 makes its way onto the RTX 40-series down the line. Performance improvements aside, the safety concerns stemming from poor efficiency also need to be addressed.

Nvidia launched the GeForce RTX 3060 (12GB) in early 2021 as an affordable mainstream gaming GPU with a 170W TDP, before re-releasing two months ago to offer relief during the ongoing component crisis. Throughout this time, not one person thought that the 3060 consumed too much power... except someone in China who decided to make a 75W version of the card. This blower-style variant has no 6- or 8-pin power connectors and runs entirely off the PCIe slot it'll be connected to, resulting in some expectedly underwhelming performance.

The RTX 3060 takes very well to undervolting and, therefore, can already be taken all the way down to just 100W at the cost of modestly reduced performance. A single PCIe x16 slot can provide up to 75W of power, so forcing a 3060 down to that number requires a shunt mod. This practice is usually associated with unlocking power limits on a GPU to chase overclocking feats where you decrease resistance, but in this case, you'd be increasing it.

This is a single-slot, low-profile card with a blower-style cooler about the size of a modern smartphone. You get just 1x HDMI and 1x DisplayPort in terms of connectivity. There is no fin stack present either, the PCB lacks a backplate, and the shroud is a thin metal sheet responsible for all the heat dissipation with a rudimentary heatsink in the middle.

A teardown of the GPU on BiliBili shows it's under-equipped from the inside, too. The memory chips have no thermal pads or paste on them. Instead, the metal shroud just touches the core directly to keep the entire thing cool. Remember that there's just one fan at the far end to blow hot air; there is no intake or proper airflow with positive pressure. So far, everything about this card, except its compact size, is looking subpar.

Once we get to testing, any skepticism is validated as the 75W RTX 3060 GPU is barely able to edge past 900 MHz, despite being rated for 1,770 MHz boost clocks. This results in a Time Spy score of just 4,821 points whereas a regular RTX 3060 easily scores upwards of 9,000 points in the same benchmark. Even older budget GPUs like AMD's iconic RX 580 and Nvidia's equally-popular GTX 1060 score more.

During the Time Spy run, the GPU hotspot also went past 90 degrees Celsius, proving that the cooler is barely performing if it's struggling to tame even a 75W card. Clearly, this GPU was suffering from severe thermal throttling on top of already having its power budget more than halved. One could make an argument that it's still an Ampere GPU, so it could make sense in low-power, compact systems. But there are now mini PCs with similarly performing or far more potent integrated graphics in 2026.