We go over every GPU in the RTX 30-series lineup to determine whether or not it's the right time to leave the Ampere platform for newer Blackwell and RDNA4-based pastures.
Northwest Repair was sent an Asus ROG Strix 4090 that had a dead core and memory chips with fake markings to make it look like the real thing. Unfortunately, there's no way of telling what these counterfeit parts actually were, but the main takeaway is just how good these scammers are getting as of late.
New leaks claim that Intel's upcoming Xe3P graphics architecture won't feature any discrete gaming GPUs, and even the next-gen Xe4 lineup isn't confirmed to. Intel is instead prioritizing the datacenter and workstation segments for new graphics IP, and is featuring them on mobile parts.
An RTX 5090 plugged into this eGPU using the CopprLink standard was able to get essentially the same level of performance as the GPU directly plugged into the motherboard. The setup requires a $1,300 enclosure and a $1,000 adapter card, and it's only meant for enterprise markets, but it shows what the future of eGPUs could be for consumers.
A research team at Specops benchmarked an RTX 5090, against two AI datacenter GPUs with Hashcat to see which GPU is the better password cracker. Spoiler alert, the RTX 5090 was the winner.
Intel's latest GPU driver is capable of booting Crimson Desert and can reportedly run at nearly 60 FPS at ultra settings on an Arc B580. But the game suffers from occasional crashing and visual artifacts.
Intel is working on its own neural texture compression with similar compression performance to Nvidia's counterpart. Best of all, Intel has a fallback version of its compression tech that will work with GPUs that don't come with Intel's XMX engine.
The company behind the tiny box AI accelerator says that its macOS driver for Nvidia eGPUs has just been signed by Apple, making it a legitimate software for Macs and no longer needs workarounds to work with the device.
Nvidia has just demoed its Neural Texture Compression technique again at a GTC talk, where it showed VRAM usage dropping from 6.5 GB to just 970 MB in a scene. NTC uses a neural network to decompress textures instead of standard block-based compression, reducing texture size and VRAM usage while also improving final image quality.
Two new Rowhammer attacks for GPUs have been discovered that can cause bit flips in VRAM to gain arbitrary read/write access over it. These attacks target page files and the page directory that are otherwise protected from electrical disturbance by the driver. By "massaging" these data structures into vulnerable regions where a bit flip can occur, the attacker can access even the CPU memory.
A YouTuber resurrects a damaged RTX 4090 with a bent PCB, using a custom BIOS to disable one of its memory channels, reducing its VRAM to 20GB but stabilizing it in the process.
The Nvidia App can now automatically recompile shaders for you in the background after every GPU driver update. This should save gamers several minutes across different titles, especially blockbuster ones, where shader compilation can often delay your session. You still need to compile shaders for the first time after a new install, however.
The arrival of Nvidia's DLSS 4.5 Dynamic Multi Frame Generation mode and its extended 5X and 6X multipliers promises more control and higher generated frame rates for GeForce RTX 50-series graphics cards. We went hands-on to see just how far AI can stretch one input frame.
登录
