Featured Articles

AMD SVP John Byrne named turnaround exec of the year

AMD SVP John Byrne named turnaround exec of the year

Director of AMD’s PR Chris Hook has tweeted and confirmed later in a conversation with Fudzilla that John Byrne, Senior Vice…

More...
Shield Tablet 8 launching on Tuesday July 22nd

Shield Tablet 8 launching on Tuesday July 22nd

We knew the date for a while but as of right now we can confirm that Nvidia’s new Shield Tablet 8,…

More...
AMD confirms 20nm in 2015

AMD confirms 20nm in 2015

Lisa Su, Senior Vice President and Chief Operating Officer, AMD, has confirmed what we told you back in May 2014 – …

More...
AMD reports loss, shares tumble

AMD reports loss, shares tumble

AMD’s debt load is causing huge problems for the chipmaker -- this quarter it had another substantial loss. The tame Apple Press…

More...
AMD A8-7600 Kaveri APU reviewed

AMD A8-7600 Kaveri APU reviewed

Today we'll take a closer look at AMD's A8-7600 APU Kaveri APU, more specifically we'll examine the GPU performance you can…

More...
Frontpage Slideshow | Copyright © 2006-2010 orks, a business unit of Nuevvo Webware Ltd.
Thursday, 13 February 2014 18:09

Nvidia Maxwell brining new Streaming Multiprocessor

Written by Slobodan Simic

Packs 128 CUDA cores per SM

According to a report from Videocardz.com, it appears that Maxwell is a bit different than the Kepler GPU architecture when it comes to the number of CUDA cores per SMX, or in Maxwell's case, SMM. Apparently, all those GPU-Z readings were off, as a Maxwell SMM packs 128 CUDA cores unlike a Kepler SMX which packed 192 CUDA cores.

With this in mind, the upcoming GM107 will pack 5 SMMs with 128 CUDA cores per SMM, which adds up to a total of 640, which we will see on the upcoming GTX 750 Ti graphics card. The GTX 750 will have one SMM less for a total of 512 CUDA cores. As we wrote earlier, Maxwell will be all about power efficiency, and according to latest reports, should deliver twice the performance per Watt compared to Kepler. Apparently the GPU is quite small and die should be somewhere around 148mm2 and provides up to 30 percent higher density of CUDA cores per mm2 of die.

The GM107 GPU will have a TDP of 60W which means that it will not even touch the 75W provided by the PCI-Express power connector but some graphics cards will feature it due to stability and additional overclocking headroom.

According to Videocardz.com, 28nm Maxwell will be available as GM107 and GM108 GPUs while 20nm will be focused on higher end parts and will arrive at a later date. The Maxwell GPU will also bring larger L2 cache, will have improved workload balancing and compiler-based scheduling, has increased number of instructions per clock cycle, faster H.264 encoding with improved NVENC and new GC5 low sleep power state. 

As reported earlier, first GM107 Maxwell-based graphics cards are expected on 18th of February.

You can check out more info over at Videocardz.com.

Nvidia-GM107Maxwelldiagram-1

 

Last modified on Friday, 14 February 2014 09:19
blog comments powered by Disqus

 

Facebook activity

Latest Commented Articles

Recent Comments