Today there aremedia broke the Intel Xe graphics parameters, its terror specifications are shocking, then 16384 core in the end performance will reach what extent! Before we look at this issue, let’s look at the performance of Intel’s nuclear display.
Single watt: a total of 512 execution units, 4096 cores, TDP 150W, frequency 1.5GHz, single-precision floating point 12.2TFlops
Double tile: total 1024 execution units, 8192 cores, TDP 300W, frequency 1.25GHz, single-precision floating point 20.48 TFlops
Four-watt: total 2048 execution units, 16384 cores, TDP 400/500W, frequency 1.1GHz, single-precision float inglot 36TFlops
This is the test score of the 3DMark FSE for dell XPS 13 2in1 notebook, which features what appears to be an Intel i7-1065G7 processor with 64 built-in EU execution units with a total of 512 cores and a frequency of 1.1GHz. Its 3DMark FSE graphic score is 1404 points, which is about 10% worse than the 1500 score of the full-blooded MX150 (blood-filled MX150 Boost frequency 1532MHz).
Of course, there will be a lot of students to question, Intel nuclear performance is very strong, but the game frame rate performance is not satisfactory.
In fact, this is a general disease of nuclear display, nuclear display not only with the CPU to share 15W TDP, but also with the system to share memory bandwidth, and the bloody MX150 can not only enjoy 25W TDP, but also exclusive 2GB gDDR5 memory. This is the most important reason why nuclear display in the game performance disorder.
Not only Intel’s nuclear display, AMD’s nuclear display also faces the same problem! The Vega8 GPU with 512 stream processors built in the Ryzen 5 3500U is parameterly comparable to the mobile version of the RX 550, and the 3DMark’s running score is not much different, but in the actual game, the two will show a very large gap.
So we can imagine that the iris Pluss of the i7-1065G7 will be taken out to make a stand-alone graphics card, providing 25W TDP and 2GB GDDR5 memory, and that the actual performance will not be much worse than the MX150 for the 384 stream processors.
Now back to Intel Xe graphics, up to 2048 EU execution units, 16,384 cores, the overall specification is 32 times i7-1065G7, converted to at least 12288 Turing stream processors, that is, 2.8 RTX 2080 Ti. The premise, of course, is that Intel’s drivers can fully perform the GPU.
The 2048 EU Intel Xe graphics single-precision floating-point performance is 36 TFlops, and the RTX 2080 Ti’s 13.4 TFlops single-precision floating-point performance is exactly 2.7 times the difference.
After 12 years, Intel’s determination to return to the field of dominance is extraordinary, and we believe that Intel has the strength to challenge NVIDIA’s monopoly in terms of solo! After all, it’s already hard for consumers to buy cheap graphics cards in a single big situation. According to NVIDIA’s latest financial results, Q4 gross profit reached $2.03 billion in fY20, with a gross margin of 64.5%.