The Hot Take: We'll have to see, they just keep milking Ai at the cost of liquidating a whole other industry.
The GeForce RTX 50-series desktop graphics lineup has remained unchanged for just over a year now (since the introduction of the RTX 5050), and none of the conversations we've had at major trade shows suggests that a mid-cycle Super refresh will occur any time soon. But there are still signs that such an update could still happen at some point, and the latest sign comes from Seasonic, which has listed an RTX 5080 Super, an RTX 5070 Ti Super, and RTX 5070 Ti Super in its PSU wattage calculator. Go deeper with TH Premium: GPUs(Image credit: Noctua)Desktop RoadmapEnterprise RoadmapRubin in-depthThe Stout Owl: The ultimate Noctua G2 PCYou can click through the calculator and assemble a hypothetical system with these unannounced products inside. To be useful, this calculator also provides board power numbers for these as-yet-unreleased cards, and that gives us another bit of juicy info. Given that no official specifications for these GPUs exist, it's impossible to say whether these figures are accurate. But it does allow us to speculate a bit on how they might stack up to existing products. Unannounced RTX 50 Super-series TGPsGraphics CardTotal Graphics Power (W)% ChangeRTX 5070250--RTX 5070 Super275*10% RTX 5070 Ti300--RTX 5070 Ti Super350*17%RTX 5080380--RTX 5080 Super415*15% *As listed in Seasonic PSU calculator. Unconfirmed by Nvidia. Seasonic gives this purported RTX 5080 Super a board power of 415W in its calculator, or 15% higher than the existing RTX 5080's 360W envelope. That makes sense, because the RTX 5080's GB203 GPU is already fully enabled, so any Super version of that card would have to lean on higher power limits and more aggressive clock speeds to see any baseline performance benefit. That figure could also partially account for slightly higher power usage from 8 GB more GDDR7 memory on such a card. Past RTX 50 Super-series rumors have suggested that Nvidia will boost VRAM capacity on those products by moving to higher-density GDDR7 modules with 3GB of capacity each. GDDR7's power consumption as part of the overall board picture is relatively small, but more of it will still matter. If Seasonic's figures are accurate, we should also expect a similarly sized TGP increase out of the RTX 5070 Ti Super, whose 350W rating is 17% higher than that of the RTX 5070 Ti. The RTX 5070 Super, meanwhile, gets only a 10% TGP bump over the RTX 5070, from 250W to 275W. Both of these cards rely on GPUs that are slightly cut down from their full available resources, so it's possible that Nvidia could boost their performance through a balance of enabling more Streaming Multiprocessors (SMs) in addition to boosting clocks through higher power envelopes. Beware of extrapolating performance improvements directly from these percentages, though. Our own testing has shown that any real-world performance benefits from these power limit increases are likely to be smaller than those figures would suggest. As our review of the MSI RTX 5090 Lightning Z showed, the largest performance boosts from higher power limits are likely to be concentrated in ray-traced and path-traced games, whose computational intensity is significantly higher than pure raster titles and is more likely to run the GPU into its power limits. In any event, we shouldn't expect to see these products any time soon. Nvidia has instead been focused on getting more out of existing Blackwell silicon with software improvements such as DLSS 4.5 upscaling and Multi-Frame Generation multipliers up to 6X. These technologies enable higher output frame rates and image quality with lower input resolution than past DLSS technologies, and the boost to both performance and image quality from those technologies in tandem has certainly made existing Blackwell products more appealing than they were at launch.But as monitor refresh rates continue to climb thanks to ongoing improvements to OLED and LCD panels, and next-generation HDMI 2.2 connectors looming over 2027, a hardware update of some kind that boosts baseline performance and potentially implements support for those standards seems practical at some point. Given these primarily consumer-focused improvements, an announcement at CES 2027 or Computex 2027 might make sense. As with any future product rumors, however, only time will truly tell.
The Hot Take: We'll see how long before they go to Intel because TSMC is filled up in Arizona FABs.
Nvidia will stay TSMC’s biggest customer in 2027, but AMD’s EPYC Venice could pinch the CPU bragging rights.
TSMC is seeing rising demand for 2.5D advanced packaging as CPUs become more important in the agentic AI bunfight. Apparently, GPUs alone are no longer enough to feed the machine.
Morgan Stanley reckons Nvidia will remain TSMC’s largest CoWoS customer in 2027. TSMC is expected to reach wafer capacity of 200,000 wafers a month that year.
Nvidia is using TSMC’s CoWoS packaging for two main product families. CoWoS-L is for AI GPUs such as Blackwell and Rubin, while CoWoS-R is for Vera CPUs.
CoWoS-L capacity is expected to hit about 910,000 units, up 40 per cent year on year. Vera shipments are expected to double, which would help Nvidia lift data centre revenue by 52 per cent.
Morgan Stanley said:“Nvidia uses TSMC’s CoWoS-L as the single source for all its AI GPU products (e.g. Blackwell and Rubin). Its 2027 CoWoS-L consumption could reach ~910k, up ~40 per cent year on year. Strong CoWoS-R bookings by Nvidia suggest room for AI GPM products (such as doubling). Taken together, we estimate Nvidia’s 2027 forecast for Nvidia’s data centre revenue to rise 52 per cent year on year.”
Nvidia has been pivoting harder into CPUs to claw back China revenue after GPU restrictions. Several customers have shown interest in Vera CPUs.
The company has hand-delivered the first Vera CPUs to Anthropic, OpenAI, SpaceX and Oracle. Nothing says “agentic AI era” like an expensive chip being passed around the usual suspects.
The problem for Nvidia is that AMD is not politely standing at the back. Its next-generation EPYC Venice platform is already in volume production at TSMC.
Venice is based on AMD’s upcoming Zen 6 architecture and is expected to deliver better performance and efficiency. It targets both AI and HPC, while Vera is being pitched squarely at agentic AI.
Morgan Stanley projects Nvidia’s Vera CPUs could reach 5.75 million units by 2027. AMD’s EPYC Venice, though, could reach 6.75 million units in 2027. That is 17 per cent more than Vera and 5.4 times its expected 2026 volume.
“Based on our CoWoS consumption forecasts, Nvidia’s 5nm Vera CPU could grow to 5.75mn units in 2027, while AMD’s 2nm Venice CPU may reach 6.75mn units in 2027 vs. ~1.25mn in 2026,” the beancounters said.
AMD has another advantage on paper, with Venice using TSMC’s advanced 2-nanometre process. Vera is based on a 3-nanometre process. The real headache for both companies may not be each other. It is custom silicon, where the cloud crowd is deciding that buying chips off the shelf is for the riffraff.
OpenAI, Google, Amazon and others are either talking up custom chips or already building them. That turns the AI supply chain into a fight between outside suppliers and in-house silicon vanity projects.
The Hot Take: It seems Nvidia is hedging two architectures against each other. x86 vs ARM. They've been working with MediaTek to create the Spark SoC.
According to an exclusive report by VideoCardz, Intel's first x86 system-on-chip (SoC) integrating an Nvidia RTX GPU has been added to its internal product roadmap and is expected to launch in the first quarter of 2028, potentially making its public debut at CES 2028.
The Hot Take: This is an interesting collaboration between the two seeing intel keeps saying they're not going to stop GPU development.
Intel's Serpent Lake SoCs featuring NVIDIA's RTX GPU tiles as integrated graphics are expected to roll out by Q1 2028. Intel & NVIDIA's Co-Developed Serpent Lake SoCs Featuring Next-Gen CPU & GPU Architectures Rumored For Q1 2028 Last year, Intel announced that it was working with NVIDIA on a custom SoC that would incorporate NVIDIA's RTX GPU tiles. Intel stated that these SoCs will power a wide range of PCs that require the integration of these levels of CPUs & GPUs together into a single package. It looks like we have our first timeline of when these SoCs will be […]Read full article at https://wccftech.com/intel-serpent-lake-socs-with-nvidia-rtx-gpu-tiles-reportedly-arrive-in-q1-2028/
The Hot Take: We'll see, they keep teasing it. But i feel they don't feel they have milked that Ai money cow enough to drop new hardware yet.
For almost a year, the RTX 50 Super series has been part of the rumor mill, but with the AI boom snatching production lines, causing memory prices to skyrocket, hype for the lineup had died down. Now, a potential RTX 5060 Super with 12GB of VRAM is apparently in the works, with the 50 Super series as a whole allegedly getting "back on track."
The Hot Take: Tell me something I didn't know already. Why else would the GPU market go crazy prices wise?
A business-intelligence researcher said that the Chinese military has been actively acquiring Nvidia AI chips, even after the U.S. put export controls on them. Public documents show that some institutions ask for these chips either through the specifications they demand or by directly asking for Nvidia chips by name.
The Hot Take: I'm interested in the Surface with this chip to get a decent GPU on an ARM setup and play with ARM Windows more personally. Only professionally worked with it and that was only an inch deep.
Computex 2026 and GTC Taipei will go down in history as the moment NVIDIA used to officially announce its entrance into the PC market. During his keynote at the Taipei Music Center, CEO Jensen Huang announced the RTX Spark – formerly codenamed N1 and N1X – which will power an array of premium laptops and small form factor systems coming this
The Hot Take: Will have to see what the final products show us.
The N1X reportedly comes in two SKUs: a top-end 20-core option with 6,144 CUDA cores matching the desktop RTX 5070, and a cut-down 18-core option with 5,120 CUDA cores. The standard N1 also has two configs, one with a 12-core CPU and 2,560 CUDA cores and a 10-core model with 2,048 CUDA cores.
The Hot Take: Was wondering why it was saying I didn't have Nvidia Control Panel! LOL
The interface lived for nearly two decades and served us well, but now it's time to switch to the latest NVIDIA App. NVIDIA Announces the Official Retirement of Control Panel, One of the Most Popular Tools for NVIDIA GPU-Based Systems The popular NVIDIA GPU tool, NVIDIA Control Panel, launched in February 2026. It has been just over two decades since the tool went live and has remained the most used utility on NVIDIA GPU-based systems. From offering simple settings to change refresh rate, resolution, and multiple-display setups to manage 3D settings, the Control Panel remained one of the easiest to […]Read full article at https://wccftech.com/nvidia-control-panel-is-officially-dead-after-two-decades/
Nvidia finance chief Collette Kress says rivals moaning about memory prices should have ordered the stuff ages ago.
Kress has blamed other firms for being caught short in the ongoing memory squeeze. She said Nvidia saw the price surge coming and acted before everyone started clutching their pearls.
In an interview with Tae Kim, Kress said her firm ordered memory in advance because it understood prices were about to go north. That is not the sort of line which will calm customers now paying more for DRAM.
The surge in memory prices has been driven by huge demand from AI chip companies. It has reshaped the industry, handing workers at SK hynix chunky bonuses while Samsung staff have been protesting.
AI GPUs have created chaos in high-bandwidth memory and DDR memory markets. These chips are greedy little beasts, and the supply chain is not enjoying the feeding frenzy.
Estimates suggest demand for Nvidia’s Rubin AI platform alone could exceed the combined memory needs of the Fruity Cargo Cult Apple and Samsung. Job’s Mob is expected to need 2.9 billion gigabytes of LPDDR memory in 2027, while Samsung may need 2.7 billion gigabytes.
Rubin AI chips could need as much as six billion gigabytes of LPDDR memory in 2027. That is the kind of number which makes procurement departments stare sadly at spreadsheets.
AI GPUs need DDR memory as well as HBM. Since both are made using the same machines, increased HBM demand pushes manufacturers to shift capacity away from DDR.
That capacity shift creates shortages in DDR too. Chinese firms are now believed to be eyeing the gap as a chance to become more important players in the memory market.
Nvidia has managed to stay ahead of the supply mess. Kress said the company had foreseen the disruption and lined up supply before the scramble began.
Discussing other firms complaining about high prices, Kress said many companies “are sitting here going, oh my gosh, the memory price went up… Nvidia knew that was going to happen. It was something everybody should have, at least we did, ordered a long time ago.”
That is a fairly blunt way of saying Nvidia thinks the rest of the industry failed its homework. It is probably true, though it is easier to say when your company is minting cash from the AI gold rush.
Kress said Nvidia was not simply buying chips off the shelf. It has been working directly with memory suppliers to design and build what it needs.
“They’re designing it with us. And then they go, now how much supply do we need? And we’re not just doing it with one. We’re doing it with all three memory suppliers. We say, here’s what we’re building. And then we’ve got to get them all in line and working with us. I don’t see another company doing that.”
The memory shortage is now another AI tax on the rest of the tech world. Nvidia gets first-class treatment because its GPUs are driving the boom which caused the mess in the first place.