Break the core number of false impressions

Intel has always been obsessed with revealing details of the company's embedded graphics processor (GPU) core, but it actively introduced its own GPU at the recent annual developer forum. The HD Graphics 5300 is the first product derived from the Intel 8th Gen 8 processor graphics processing architecture and appears to exist in Intel's new Core M Series processors that lock tablets and other portable devices.

Core M uses a ring bus between the CPU core, cache, and GPU. Each linked processor or cache has a dedicated local interface. The ring interconnect is a 32-byte wide, two-way data bus with individual lines responsible for requests, snoops, and acknowledgements, allowing the GPU to act as a "first-class citizen."

GTI (graphics technology interface) is the gateway between the GPU and the other functional blocks of the processor SoC, facilitating communication between CPU cores and possibly other fixed-function components such as camera imaging pipelines; In some versions of the GPU, Intel doubles the write bandwidth via the GTI and achieves shared virtual memory sharing between the CPU core and the GPU.

This architecture is based on so-called execution units (EUs). It has 7 threads and each thread contains 128 general-purpose buffers. The main operating unit in the EU is a pair of SIMD floating-point units (FPU) and supports floating. Point and integer operations, each SIMD FPU can complete the addition / multiplication floating-point instructions in each cycle. Those EUs present a clustered distribution known as a subslice, which is further clustered into slices. These elements are modular functional blocks that Intel uses to create various derivatives.

"International Electronic Business"

Intel Gen 8 Execution Unit contains execution units with multiple common caches, some of which support specific architecture caches

The Core M Series with the Iris Pro 5200 GPU will include 128MB of embedded DRAM, but not within the processor chip, but above the chip; the Gen 8 graphics processor uses 576KB L3 cache per slice, which is higher than the capacity 384KB of the previous generation.

A new feature of the Gen 8 GPU is the global memory consistency between the GPU and the CPU core; the processor product with this new GPU integrates support for Intel's Intel VT-d virtualization technology (Virtualization Technology for Directed I/O) The hardware components. This specification represents an extension of Intel's existing approach to imaging virtual machines to physical resources.

Intel is very cautious about the 180-degree turnaround of GPU products; the company has many engineering talents who design excellent GPUs, but because they may never see investment compensation in marketing, manufacturing, and finance, they believe that No one will pay extra for this product.

Intel’s strategy has changed since the Gen 6 GPU took up more chip space; now Gen 8 GPUs occupy more than 60% of the area on the processor chip, not including optional graphics processing external DRAM . Now you can almost say that Intel has built a GPU and forced certain CPUs to go with it. As for the number of Intel GPU cores, because each EU has to perform 16 32-bit floating-point operations in each cycle, some people think it is 16 cores, but with this algorithm, the GPU may also have 384 cores.

AMD once boasted of having 512-core GPUs. Nvidia said that their GPUs have 192 cores. Although Qualcomm will not say, we think that their GPUs have more than 24 FPU cores; however, when you see the GPU The performance benchmark test results, not measured by the number of core, this comparison is really boring.

Unfortunately, most processor SoCs do not develop an open system such as a PC. Such a device can be evaluated in a more equitable environment; however, in the case of a portable device, the performance of the graphics processor is not. The test results may not matter, but more important is the performance per watt.