Double-punch hard-handed four-hand – this truth we all understand, if a graphics card performance is not enough, then why not add a graphics card? This is AMD’s CF firefight, NVIDIA’s SLI speed technology concept, is multi-way graphics card parallel. CF/SLI Doka was once a symbol of AMD/NVIDIA graphics players, such as the four-way Titan of the year, is definitely a symbol of the trench.
However, CF/SLI doka technology this two years is not to be seen, in addition to showing off, in practice there are many problems, many games do not support multi-card, drivers, game development are trouble, so that AMD/NVIDIA two also snubbed this technology, such as SLI technology is only limited to a few high-end graphics cards.
Intel will also launch high-performance stand-alone GPUs this year, their attitude on the doka issue is different, Intel is still very enthusiastic about the multi-card interconnection, especially the nuclear-explicit and solo parallel work, which will become a selling point for them.
At the GDC 2020 online event, Intel recently released Multi-Adapter: Integrated and Integrated and Live GPUs Together, which describes their technical work in parallel with nuclear and solo.
Simply put, Intel is supporting the parallel operation of the solo-based sione-by-one, which can be divided into three processes: simulation, replication, and rendering, supporting the sharing of data in different GPUs and co-accelerating work together.
For Intel, the work of solo-and-nuclear parallels is not just to render 3D, but also to do physical operations, AI, shadows, mesh deformations, and so on, and a lot of asynchronous work can be done.
Now Intel just demonstrated this technology, the specific effect is not yet certain, for nuclear display, solo, how much of this parallel ascension is the key, before AMD / NVIDIA in CF / SLI is facing the trouble is efficiency, especially AMD has also engaged in the technology of solo display and nuclear display, the results are not satisfactory.
Intel’s first stand-alone will be the DG1, which will use the Xe architecture, which won’t change, but the performance won’t be very high,probably at a moderate level.
According to the previously leaked information, DG1 has 96 sets of EU execution units, a total of 768 core, the base frequency of 1GHz, acceleration frequency 1.5GHz, 1MB secondary cache and 3GB of memory, TDP is 25W.
If compared to the Gen11 core of the i7-1065G7, the number of stream processors is 50% higher, and the acceleration frequency is 1.5GHZ to 1.1GHz, an increase of 36%.
Presumably, the DG1’s performance is comparable to that of the GTX950, which is about 15% worse than the GTX 1050.