Last week ASUS and Acer presented their new notebook models with the first dedicated GPU from Intel. Technical details about the Iris-Xe-Max-GPU (previously called DG1) were not provided by either the notebook manufacturer or Intel. Today they come from Intel and people are talking about the alternative to the entry-level GPUs, which are used when the integrated variant is no longer sufficient.

As part of a preview of the Tiger -Lake processors we could use the Core i7 – 0915 already look at G7 with Iris-Xe graphic. This offers 96 Execution Units (EUs) with a cycle of 1 , 35 GHz. In our benchmarks, the integrated GPU doubles the graphics performance compared to its predecessor.

The Iris-Xe-Max-GPU offers with 96 ESs have the same expansion and range of functions, but can set the clock to up to 1, 65 screw up GHz. This corresponds to an increase of 22, 2% and at about 20% should be the performance plus of the dedicated variant. As for the Tiger Lake processors themselves, the GPU is manufactured in nm SuperFin instead. This is the improved production in 10 nm, which Intel should lead to a breakthrough in this area.

The GPU is supported by 4 GB of dedicated memory . This is LPDDR4X memory, which has a memory bandwidth of 68 GB / s is connected. In the case of the integrated variant, theoretically 63 GB / s reached. However, this is a shared memory that is made available via the processor.

The connection between Iris-Xe-Max-GPU and the processor of the notebook takes place via four PCI-Express 4.0 lanes. The support for Variable Rate Shading, Adaptive Sync and Async Compute should not go unmentioned. There is also the option of AV1 decode, output via eDP 1.4b, DisplayPort 1.4 and HDMI 2.0b as well as support for a maximum resolution of 4. 096 x 2. 304 at 60 Hz via HDMI and eDP as well as from 7. 680 x 4. 320 Pixel at 60 Hz via DisplayPort. All this is also made possible by the integrated version of the Xe-GPU.

Intel Deep Link

Between a Tiger Lake processor and the external Iris-Xe-Max-GPU there should be a certain cooperation via deep link. This includes functions such as the accelerated calculation of AI applications and increases the encoding performance by outsourcing the encoding.

Apparently, the available performance is used in a shared manner. In comparison of the calculations in combination of a Sunny Cove and (simulated) Tiger Lake processor in interaction with a GeForce MX 350 the Tiger Lake processor with Iris-Xe-Max-GPU in CPU encoding should be 20% and in the AI ​​acceleration by 40% faster. It is precisely in these two areas that Intel sees the greatest advantages.

The relatively high encode performance is probably one of the reasons why Intel wants to use the DG1 or, better, the Xe architecture in the data center. As an SG1, it should offer the same performance as four DG1s and still be used as a small form factor card. Intel is targeting cloud gaming and video streaming applications that are designed for particularly low latencies. The SG1 should go into production shortly and, as I said, be available this year.

In terms of gaming performance, a notebook with Iris-Xe-Max-GPU should in most cases be on par with a notebook with MX 350 – be GPU. In a few cases the Intel solution is even significantly faster.

Information about the power consumption of the Iris-Xe-Max-GPU would be interesting. For a Core i7 – 1185 G7 with a GPU clock of 1, 35 GHz we have a maximum power consumption 14 W measured. The TDP of the Iris-Xe-Max-GPU should be 23 W lie.

Just 20% faster in games

Intel sets the For the most part, the integrated graphics unit does not focus on gaming performance, but advertises with fast encoding and hardware acceleration for AI applications. Even if, or precisely because, the gaming performance has been doubled compared to the previous generation, a significant increase in performance should be expected from using it as a dedicated GPU with a higher power limit. 96 ES for an order 22, 2% higher However, tact leave less leeway than would have been expected.

Intel confirms this in its own benchmarks. The Iris Xe Max in Metro Exodus is about 15 % more quickly (35 vs 42 FPS) as the integrated Iris-Xe solution. In DOTA 2 you are with the NVIDIA GeForce MX 349 on par, but here is the stronger Iris-Xe-Max variant slower than the in-house Iris-Xe competition.

The combination of Tiger Lake processor and dedicated Iris Xe Max GPU will initially be used in three notebook models. These would be the Acer Swift 3X, the ASUS VivoBook TP 470 and the Dell Inspiron 15 7000 2in1. When the opportunity arises, we will certainly get one of these models and carry out further benchmarks.

Intel is currently working with partners on bringing another GPU based on the Xe-LP architecture onto the market, then in the first half of the year 2020 will be available in desktops – “In addition to mobile, Intel is working with its partners to bring Xe-LP-based discrete graphics to value desktops in the first half of 2021. “

Products based on the Xe-HP and Xe-HPG architecture should also 2021 consequences.