Intel HD Graphics (Cherry Trail) vs NVIDIA GeForce GTX 1070 Mobile vs Intel HD Graphics 4000

Intel HD Graphics (Cherry Trail)

The Intel HD Graphics (Cherry Trail) is an integrated graphics card in the Cherry-Trail SoCs (e.g. Atom x7-Z8700). Based on a Broadwell GPU (Intel Gen8) and supports DirectX 11.2. Depending on the SoC offers 12 or 16 EUs.

Atom x7-Z8700: 16 EUs @ 200 - 600 MHz, dual-channel memory
Atom x5-Z8500: 12 EUs @ 200 - 600 MHz, dual-channel memory
Atom x5-Z8300: 12 EUs @ 200 - 500 MHz, single-channel memory

In Windows, gaming is limited to less demanding or quite old games. Using Android, even high end 3D games should run fluently. For example Team Fortress 2 was running on the Surface 3 (x7-Z8700) only in lowest settings and 1280x720 fluently (11 - 69 fps range).

The integrated video decoder is able to display even 4K videos (H.264 tested) fluently.

The Intel HD Graphics GPU in our review sample of the MS Surface 3 (Atom x7-Z8700) only supported DisplayPort 1.1 as it seems, as 4K (3840x2160) with 30 Hz was the maximum Refresh Rate we could achieve (no 60 Hz selectable with the pre-installed drivers).

NVIDIA GeForce GTX 1070 Mobile

The Nvidia GeForce GTX 1070 is the second fastest Pascal based graphics card in mid May 2016. It comes in two variants, the desktop version (see here for the detail page) with 1920 shaders and the mobile variant with 2048 shaders and lower clock rates. Both cards are based on a cut down GP104 chip (like the GTX 1080) and are manufactured in 16 nm FinFET at TSMC.

Compared to the GTX1080, the GTX1070 features less CUDA cores (2048 versus 2560) that are clocked a bit lower (maximum Boost 1645 versus 1733). Instead of the new GDDR5X, the GTX 1070 uses slower GDDR5 graphics memory - but still 8 GB. The TDP is also reduced and rumored to be 10 Watts higher than the predecessor, the GTX 980M.

For thin and light laptops Nvidia offer a Max-Q called version of the GTX 1070 with lower TDP and also significantly lower performance. Sadly this variant im sold with the same GTX 1070 name, making it hard to judge the performance.

The performance depends on the cooling solution of the laptop. With a good fan system, the performance of the laptop version is on par with the desktop GTX 1070 (Founders Edition). Nvidia claims that partner solutions can be up to 10 percent slower however. In our benchmarks with various laptops the performance ranged from -5 to -14% slower performance compared to partner models of the GTX 1080 desktop models with slightly higher clocks.

On average the GTX 1070 is around 20 percent faster than the old mobile GeForce GTX 980 and therefore the second fastest mobile graphics card (non SLI). Therefore it is capable of WQHD and 4K gaming in high settings (see game benchmarks below).

The GP104 chip is manufactured in 16nm FinFET process at TSMC and offers a range of new features. DisplayPort 1.4 (ready), HDMI 2.0b, HDR, Simultaneous Multi-Projection (SMP), improved H.265 video en- and decoders (for PlayReady 3.0) are only some of the improvements. See our detailed Pascal architecture article for more details.

The power consumption of the GTX 1070 is specified with 115 Watt TGP according to Nvidia and therefore slightly above the old GeForce GTX 980M. This leads to rather big and clunky gaming laptops that use the GTX 1070. The later released Max-Q version of the GTX 1070 uses only 80 - 90 Watt TGP and is therefore better suited for thin and light laptops.

Intel HD Graphics 4000

The Intel HD Graphics 4000 (GT2) is a processor graphics card that is included in the Ivy Bridge processors of 2012 (3rd generation of core, e.g. Core i7-3770). The base clock can be automatically overclocked using Turbo Boost technology. Depending on the processor model, the base and turbo clock rate may differ greatly resulting in different graphics performance of ULV parts compared to high-end desktop and laptop quad-core parts.

Compared to the Intel HD Graphics 3000 in Sandy Bridge CPUs, the HD 4000 card was completely redesigned and offers improved DirectX 11 capable shaders, Hardware Tessellation, a dedicated level 3 cache (before the Last Level Cache LLC of the CPU) and DirectCompute support. The IPC (instructions per clock) can therefore be even 2x as fast as with Sandy Bridge and overall up to 60% more performance (3DMark Vantage) should be possible.

First benchmarks position the HD Graphics 4000 (in a fast quad core desktop CPU) on a level with a dedicated Nvidia GeForce GT 330M and therefore above the AMD processor graphics Radeon HD 6620G. In our extensive tests with games the HD Graphics 4000 was able to beat the HD 6620G in a fast Core i7-3820QM by about 15%. In the slower i7-3610QM and a dual core i5 it was on a similar level as the 6620G. Therefore, casual gamers that wont mind reducing the quality settings in high end games, may be happy with the performance of the HD Graphics 4000. Beware, that the HD Graphics 4000 is used with different clock speeds depending on the CPU model. The ULV CPUs (Core ix-3xx7U) for example feature lower clock speeds and cant maintain the Turbo frequency as good as the 35 - 55 Watt models. Therefore, the ULV version is about 30% slower on average.

A speciality of the Ivy Bridge GPUs is that 4x MSAA is supported in hardware now. However, 2x is only supported through software.  The algorithm to support 2x is going through the 4x pipeline with a software algorithm, so performance is similar to 4x MSAA.

The integrated video decoder called Multi Format Codec Engine (MFX) was also improved and should allow even simultaneus 4K video decoding. DXVAChecker lists MPEG2, VC1, WMV9, and H264 as supported codecs. QuickSync for fast transcoding of videos was also optimized for higher performance and better image quality.

Another new feature is the support for up to 3 independent displays (depends on how the HD 4000 is used in the laptop - maybe only with a DisplayPort / eDP) as AMD offers with theirs Eyefinity support (up to 6 displays). DisplayPort 1.1 (max 2560x1600) and HDMI 1.4 (max 1920x1080 without hacks) are supported by the chip according to Intel.

Due to the 22nm 3D Tri-Gate production process, the power consumption should be relatively low (the development was focused on performance per Watt). The TDP of the whole package (including processor and memory controller) varies between 18 Watt (ULV) up to 45 Watt (mobile quad core) for the consumer laptop CPUs.