NVIDIA GeForce GTX 870M vs NVIDIA GeForce GTX 850M vs NVIDIA GeForce GTX 860M SLI

NVIDIA GeForce GTX 870M

The NVIDIA GeForce GTX 870M is a high-end DirectX 12-compatible graphics card for laptops unveiled in March 2014. Similar to the GTX 880M, it is based on the 28 nm GK104 chip (Kepler architecture), but features less shader cores (1344 instead of 1536) as well as smaller and slower memory (6 GB GDDR5).

Architecture

The Kepler architecture is the successor to the Fermi architecture that first appeared in laptops with the GeForce 400M series. The GK104 Kepler core offers eight shader blocks, called SMX, that are clocked at the same speed as the central core. In the GTX 870M, only seven blocks are active for a total of 1344 CUDA cores and 112 TMUs. The chip can access up to 6 GB of GDDR5 memory (1250 MHz, 5000 MHz effective) connected via a 192-bit interface.

Performance

Compared to the old GeForce GTX 770M, the GTX 870M offers significantly more performance thanks to its higher clock rate and the additional shaders. Therefore, the 870M is just slightly slower than the GTX 680MX and the Radeon R9 M290X. The GPU has enough power to run demanding games of 2013/2014 (e.g. Battlefield 4 or Thief) fluently in FullHD resolution and very high detail settings.

Features

The feature set includes support for up to 4 active displays. Furthermore, high resolution monitors of up to 3840x2160 pixels can be connected using DisplayPort 1.2 or HDMI 1.4a (if available). HD-Audio codecs, such as Dolby TrueHD and DTS-HD, can be transmitted via bitstream mode through the HDMI port. However, as most laptops will feature Optimus, the integrated GPU will likely have direct control over the display ports and may limit the feature set available by the Nvidia Kepler cards.

The 5th generation PureVideo HD video processor (VP5) is also integrated in the GK104 core and offers hardware decoding of HD videos. Common codecs such as MPEG-1/2, MPEG-4 ASP, H.264 and VC1/WMV9 are fully supported up to 4K resolutions while VC1 and MPEG-4 are supported up to 1080p. Two streams can be decoded in parallel for features such as Picture-in-Picture. Another novelty is the inclusion of a dedicated video encoding engine similar to Intel QuickSync that can be accessed by the NVENC API.

New features of the GTX 800M series include Battery Boost (longer battery life while playing games), Shadowplay (recording of in-game videos) and Game Stream (stream PC games on the Nvidia Shield console).

Power Consumption

The power consumption of the GeForce GTX 870M should be similar to the the GTX 780M, leading to a TDP of at least 100 W. Therefore, only large desktop replacements can handle the power demands of a GTX 870M.

NVIDIA GeForce GTX 850M

The NVIDIA GeForce GTX 850M is an upper mid-range, DirectX 12-compatible graphics card for laptops unveiled in March 2014. It is one of the first cards based on Nvidia's new Maxwell architecture (GM107 chip), but is still manufactured in 28 nm. The GTX 850M offers 640 shader units clocked at 876 MHz (+ Boost) as well as 4 GB of GDDR5 memory (128 bit, 5000 MHz effective). In addition, there is a slightly higher clocked version (936 MHz + Boost) using slower DDR3 memory (2000 MHz effective).

Architecture

Compared to Kepler, Maxwell has been optimized in several details to increase power efficiency. Smaller Streaming Multiprocessors (SMM) with only 128 ALUs (Kepler: 192) and an optimized scheduler should lead to better utilization of the shaders. Nvidia promises that a Maxwell SMM with 128 ALUs can offer 90 percent of the performance of a Kepler SMX with 192 ALUs. GM107 features 5 SMMs and thus 640 shader cores, 40 TMUs and 16 ROPs (128-bit interface).

Another optimization is the massively enlarged L2 cache (GM107: 2 MB). The larger size can process some of the memory traffic to allow for a relatively narrow memory interface without significantly reducing performance.

Similar to Fermi and Kepler, Maxwell / GM107 supports DirectX 12 with feature level 11_0 only.

Performance

Although the GTX 850M has less shaders, it often outperforms the old Kepler-based GeForce GTX 765M while drawing less power. Despite the lower core clock, the GDDR5 version usually takes the lead over the DDR3 model and sometimes even beats the Kepler variant of the GeForce GTX 860M. Many games of 2013/2014 (Tomb Raider or BioShock Infinite, etc.) can be played fluently in FullHD resolution and high detail settings. However, very demanding games such as Crysis 3 or Battlefield 4 will require lower resolutions and/or details.

Features

GM107 integrates the sixth generation of the PureVideo HD video engine (VP6), offering a better decoding performance for H.264 and MPEG-2 videos. Of course, VP6 supports all features of previous generations (4K support, PIP, video encoding via NVENC API). Unfortunately, HDMI 2.0 is still not supported.

Power Consumption

The power consumption of the GeForce GTX 850M should be somewhat below the old GTX 760M at about 40 - 45 Watts. Therefore, the 850M is best suited for 15-inch laptops or larger. The 800M series also supports Optimus to automatically switch between an integrated graphics card and the Nvidia GPU.

NVIDIA GeForce GTX 860M SLI

The NVIDIA GeForce GTX 860M SLI is a high-end DirectX 12-compatible graphics solution for laptops, consisting of two GeForce GTX 860M. Each card has a 28 nm GM107 core based on the Maxwell architecture. The clocks for the SLI-setup are identical to a single GTX 860M. With SLI, each card usually renders a single frame (AFR mode). Therefore, it may suffer from micro stuttering in low fps ranges of 30fps. This happens because of different timespans between two frames (e.g., irregular delays between sequential frames).

Alternatively, there is also a version based on the Kepler chip GK104, which has 1152 shader units clocked at 797 MHz.

Architecture

Compared to Kepler, Maxwell has been optimized in several details to increase power efficiency. Smaller Streaming Multiprocessors (SMM) with only 128 ALUs (Kepler: 192) and an optimized scheduler should lead to better utilization of the shaders. Nvidia promises that a Maxwell SMM with 128 ALUs can offer 90 percent of the performance of a Kepler SMX with 192 ALUs. Each of the two GM107 features 5 SMMs and thus 640 shader cores, 40 TMUs and 16 ROPs (128-bit interface).

Another optimization is the massively enlarged L2 cache (GM107: 2 MB). The increased size can handle some of the memory traffic to allow for a relatively narrow memory interface without significantly reducing the performance.

Similar to Fermi and Kepler, Maxwell / GM107 supports DirectX 12 with feature level 11.0 only.

Performance

The GeForce GTX 860M SLI is about 70 - 80 percent faster than a single GTX 860M, leading to a performance somewhere between a single GeForce GTX 680 (desktop) and Radeon R9 280X (desktop). Accordingly, the 860M SLI competes with the high-end single-GPU GeForce GTX 880M. Most games (as of 2014) are playable at the highest detail setting and in FullHD resolution. In many cases it is even possible to activate additional anti-aliasing.

Features

GM107 integrates the sixth generation of the PureVideo HD video engine (VP6), offering a better decoding performance for H.264 and MPEG-2 videos. Of course, VP6 supports all features of previous generations (4K support, PIP, video encoding via NVENC API). Unfortunately, HDMI 2.0 is still not supported.

Power consumption

The power consumption of the GeForce GTX 860M SLI should double from 60 to about 120 watts (including board and memory) compared to a single GTX 860M. Therefore, only large desktop replacements can handle the heat.