Today NVIDIA officially launched their latest mobile GPUs, the GTX 960M and 950M; on a quieter note, the 940M, 930M, and 920M also showed up. NVIDIA launched the high-end GM204 back in September/October last year and brought out the mobile GTX 970M and 980M using the chips, and they more recently launched the desktop GTX 960 using a new GM206 part with up to 1024 CUDA cores. Considering the GTX 965M has 1024 CUDA cores many felt it would use GM206 as well, but that didn’t happen; in a similar fashion, those expecting (hoping) GTX 960M and 950M would use a derivative of GM206 are also going to be disappointed and it continues to use the GM107 chip.

The new GeForce GTX 960M and 950M end up being clock speed bumps from the existing GTX 860M/850M, though this time the 960M is purely Maxwell based and there’s no Kepler alternative. The GTX 950M has 640 CUDA cores running at 914MHz (+ Boost), and it can be paired with either DDR3-2000 (yuck!) or GDDR5-5000, both using a 128-bit interface. While the additional bandwidth of GDDR5 may not matter too much at lower resolutions, we’ve seen in the past that it can have a significant impact on performance at 1080p and medium or higher settings.

Interestingly, the clocks and core configurations mean that potentially the GTX 950M is actually a bit slower than GTX 850M (640 cores at 936MHz + Boost compared to 914MHz + Boost), but NVIDIA doesn't disclose the full range of base clocks and Boost clocks so the 950M will hit higher clocks on average compared to 850M. 960M meanwhile represents a minor bump in clock speeds over GTX 860M, going from 1029MHz + Boost to 1096MHz + Boost; the RAM stays the same with GDDR5-5000 providing 80GB/s of bandwidth. Keep in mind that both the 850M and 860M are marketed with "up to xxx MHz" language where NVIDIA has now switched to advertising the base clock (no "up to"). Even so, the newer parts are likely only 10% (give or take) faster than the previous generation parts.

NVIDIA’s product pages for the three lower tier GPUs reveal precious little in the way of specifications; all we know is that the parts are DirectX 12 capable and that the 920M is rated as being ~3X the performance of Intel’s HD 4400 while the 930M and 940M are 3.5X and 4X the performance rating of the HD 4400. The 940M and 930M are derived from GM108 and will replace the existing 840M/830M, while the 920M is a Kepler design that replaces the existing 820M. While NVIDIA’s specifications rate the 840M at 5.0X the performance of the HD 4400 where the 940M is "only" 4.0X the performance of HD 4400, NVIDIA changes the games used for the performance score calculations each generation so they're not directly comparable; the newer parts however are all designed to be faster than the existing parts.

Here are the full specifications for the old and new GM107 based parts, the GeForce GTX 960M/950M and the 860M/850M:

NVIDIA GeForce GTX 960M/950M Specifications
  GTX 960M GTX 950M
GTX 950M
GTX 860M
GTX 860M
GTX 850M
Processor GM107 GM107 GM107 GM107 GK104 GM107
CUDA Cores 640 640 640 640 1152 640
GPU Clock (MHz) 1096 + Boost 914 + Boost 914 + Boost Up to 1029 + Boost Up to 797 + Boost Up to 936 + Boost
GDDR5 Clock 5GHz 5GHz 2GHz 5GHz 5GHz 5GHz
Memory Interface 128-bit 128-bit 128-bit 128-bit 128-bit 128-bit
Maximum Bandwidth 80GB/s 80GB/s 32GB/s 80GB/s 80GB/s 80GB/s
Maximum Memory 4GB 4GB 4GB 4GB 4GB 4GB
eDP 1.2 Up to 3840x2160
LVDS Up to 1920x1200
VGA Up to 2048x1536
DisplayPort Multimode Up to 3840x2160

Along with the new parts, NVIDIA and their partners have announced several new laptops using the “new” GPUs. Below are images of the new Acer Nitro V, Alienware 13, ASUS G501, HP Omen, and Lenovo Y50. Obviously there won’t be a major difference in performance between the new GTX 900M parts and the existing GTX 800M parts, so this is most likely just NVIDIA aligning the product names for use with the upcoming Broadwell systems.

Source: NVIDIA

Comments Locked


View All Comments

  • The_Assimilator - Thursday, March 12, 2015 - link


  • hammer256 - Thursday, March 12, 2015 - link

    Ha! in the short run this can be nice for the authors, all the blame goes to Purch ;)
  • JarredWalton - Thursday, March 12, 2015 - link

    Sadly, Word's spell check has let me down again. Yes, "launce" (plural "launces") is actually a word, though it's not one I would likely ever use. "Hey, can you get me some launce sushi? I love the taste of sand eel!" :-p
  • Amoro - Thursday, March 12, 2015 - link

    This is really rebadging at it's worst. The performance difference between the GTX 860M Maxwell and Kepler variants is massive. 1389 GFLOPS Maxwell versus 2108 GFLOPS on Kepler
  • Meaker10 - Thursday, March 12, 2015 - link

    Yet the maxwell performs on par at stock in games and overclocks much better. Theoretical max FLOPS is not a good measure and this is proof.
  • yannigr2 - Thursday, March 12, 2015 - link

    If the overclock is not locked in BIOS based on latest rumors.
  • Amoro - Thursday, March 12, 2015 - link

    I agree, that GFLOPS are not the best indicator of performance, especially between different architectures. But it's just so hard to believe that it is 10% faster on average with nearly half the CUDA cores, 232+Mhz clock, and architectural improvements. The only review I could find was from notebook check:

    Something still doesn't seem right.
  • JarredWalton - Thursday, March 12, 2015 - link

    Each Maxwell core has been optimized to perform certain calculations better than a Kepler core, so even with many extra CUDA cores Maxwell can often keep up and surpass Kepler -- depending on the game, of course. The real problem here is that we now have a second generation of mobile "Maxwell 1.0" parts where the desktop is moving to "Maxwell 2.0" for most of their current generation GPUs.

    Bottom line: get a GTX 965M or above if you care about gaming performance.
  • Penti - Thursday, March 12, 2015 - link

    So no HDMI 2.0 600MHz yet?
  • Tegeril - Saturday, March 14, 2015 - link

    Which then begs the question: what is Apple going to put in the next 15" rMBP? Might we see another flip back to AMD? The 965m is probably too power hungry.

Log in

Don't have an account? Sign up now