NVIDIA GeForce GTX 980M SLI vs Apple M2 8-Core GPU

NVIDIA GeForce GTX 980M SLI

The NVIDIA GeForce GTX 980M SLI is a high-end DirectX 11-compatible graphics solution for laptops, consisting of two GeForce GTX 980M. Each card has a 28 nm GM204 GPU based on the Maxwell architecture, which is the same chip as found on the GTX 970M as well as several desktop cards. The clocks for the SLI-setup are usually identical to a single GTX 980M (1038 MHz GPU, 1250 MHz memory).

With SLI, each card renders a single frame (AFR mode). Therefore, it may suffer from micro stuttering in low fps ranges of 30 fps. This happens because of different timespans between two frames (e.g., irregular delays between sequential frames).

Architecture

GM204 is based on Nvidia's Maxwell architecture, which replaces the previous Kepler generation. Among other changes, Nvidia have restructured the streaming multi processors (SMM) in order to increase workload and efficiency. Now each SMM contains only 128 shader units and 8 TMUs, which is noticeably less than an equivalent Kepler GPU. The GM204 features four GPCs, and each consists of four SMMs. But, in the GeForce GTX 980M, Nvidia have deactivated 4 of 16 SMMs. As a result, only 1536 shader ALUs and 96 TMUs are active. In addition, there are 64 ROPs in four clusters with 16 units each.

The memory interface has a bandwidth of 256-bits (4x 64 bit) for each of the two GPUs. Thanks to an improved compression algorithm, the connection to the GDDR5 memory should be significantly more efficient per MHz. Furthermore, the L1 (96 KB per SMM) and L2 caches (2 MB) have been significantly increased to lower the needed bandwidth. Compared to the mid-range GM107, the newer and larger GM204 has several additional features and changes. Aside from the improved polymorphic engine in the SSMs, it is also the first Nvidia chip to support DirectX 12 Feature Level 12_1.

Performance

With good driver support, SLI almost doubles the performance of a single card. The GTX 980M SLI outperforms the GeForce GTX 970M SLI by about 30 percent and is the fastest mobile graphics solution in late 2014. Compared to the former GTX 880M SLI, the advantage is even bigger (40 - 50 percent). Thus, the GTX 980M SLI is powerful enough to run even the most demanding games with highest details, AA/AF and resolutions exceeding FullHD.

Features

Most of the features are identical to the GeForce GTX 800M series. With the GTX 980M, there can be up to 4 active displays possible, but this will likely decrease due to Optimus. Displays can be connected with a maximum resolution of 3840 x 2160 pixels via DisplayPort 1.2 or HDMI 2.0 (no HDCP @ 4K). HD audio formats like Dolby True HD and DTS HD can be sent to a compatible receiver as bitstream. As with previous models, 3D Vision is not supported with Optimus.

The updated video processor VP6 supports the usual formats like H.264 and can also decode and encode H.265 with a resolution of up to 4K via the NVENC API. Several parallel streams, such as picture-in-picture in a Blu-Ray movie, are also possible.

Further features of the GTX 900M series, which are also partly supported by older cards, include Battery Boost (longer battery life during gaming), Shadowplay (recording of gaming videos up to 4K60p and 130 Mbit/s) and Gamestream (game streaming on Shield console). More information on these features are available here.

Power Consumption

The power consumption of each GeForce GTX 980M including its MXM board and memory should be about the same as a GTX 880M with a specified TDP of 122 W, leading to a total TDP of about 240 W. Given the significantly higher performance, this can be considered a massive increase in performance-per-watt. Nevertheless, the SLI setup can only be used in very large and heavy gaming notebooks with powerful cooling systems. Under low load, dynamic clock rates help the GPUs to save energy.

Apple M2 8-Core GPU

The Apple M2 8-core GPU is an integrated graphics card offering 8 of the 10 cores designed by Apple and integrated in the Apple M2 SoC. It uses the unified memory architecture of the M2 SoC (up to 24 GB LPDDR5-6400 with 100 GB/s bandwidth) and should offer 128 execution units.

The performance is positioned right between the old 8-core model in the M1 GPU and the higher end 10-core model in the M2. The theoretical performance should be around 2.9 Teraflops, as the 8-core version offers the same 1,398 MHz maximum clock rate as the 10-core version. Compared to other iGPUs from AMD and Intel, the M2 benefits greatly from the unified memory architecture and the high bandwidth.

The M2 chip is manufactured in the second generation 5nm process at TSMC (most likely N5P). The power consumption is around 10 Watt (in our tests under load of the MBP13) compared to the 13.5W of the 10-core version.