in

Images of the Nvidia GeForce GTX 880 have been leaked online

It’s been over a year since the launch of Nvidia’s first GTX 700 series graphics cards, so in the usual yearly cycle of releases we should be expecting the launch of Nvidia’s next line – the GeForce GTX 800 series – sometime soon. What could be the first graphics card in that line, the GeForce GTX 880, has allegedly been pictured over on Chinese website GamerSky. At this stage it’s hard to verify what this card actually is, especially as a lot of the PCB has been pixelated, but it’s certainly not something we’ve seen before. In the center of the prototype board we see a new GPU that, according to some digging done by ComputerBase, is a Maxwell-based GM204. The estimated die area is larger than that of GK104, which was the first high-end Kepler GPU, but slightly smaller than GK110.

Some interesting news has come off the GPU rumor mill, including leaked photos of what’s purportedly the Nvidia GeForce GTX 880. The card is expected to offer up to 8GB of RAM with a 256-bit memory bus and possibly water cooling (though this may only apply to the engineering samples). Interestingly, there’s still a running debate over whether or not the next generation of Maxwell parts is built on 28nm or 20nm technology. Based on what we’ve heard, we’re guessing 28nm — 20nm at TSMC is not ramping well for graphics cards according to our sources, and neither AMD or Nvidia will have new big-core GPUs based on that technology this year. Videocardz published a side-by-side photo of the three chips, which would seem to confirm the fact that what they call a GM2xx is actually a 28nm chip. The GM2xx GPU shown is almost exactly the same size as the GK110 — far larger than the GK104. Given that Maxwell runs larger than Kepler, this would make sense — the GK107 is 118mm sq, while the GM107 (GTX 750 Ti) is 148mm sq. Based on what we know of Maxwell, this implies that the new core either packs far more cores than the old GTX 680 and GTX 770, or an absolutely enormous L2 cache. When Nvidia built GM107, it boosted the L2 from 256KB to 2MB — an equivalent step up for this core would work out to 4MB of L2. Take the GK104, add more cores, and octuple the L2 cache and you could easily hit GK110 die sizes on the same 28nm node. Consumer availability is being loosely forecast for Q4 2014, though we’re not standing behind that guesstimate.

What do you think?

Avatar of Alfie Joshua

Written by Alfie Joshua

Alfie Joshua is the editor at Auto in the News. Find him on Twitter, and Pinterest.

Leave a Reply

Your email address will not be published. Required fields are marked *

GIPHY App Key not set. Please check settings

This developer published an app on the Nokia App Store while paragliding

A privacy watchdog is complaining to the FTC about Facebook’s “experiment”