current position:Home>Greedy for Apple M1, the first large and small core processor of toothpaste factory will be on the market soon! Graphics card AI super sampling, TSMC N6 process

Greedy for Apple M1, the first large and small core processor of toothpaste factory will be on the market soon! Graphics card AI super sampling, TSMC N6 process

2021-08-27 09:27:59 Xinzhiyuan

New Zhiyuan Report

source :Intel

edit :Priscilla Good trapped

【 Introduction to new wisdom 】 In recent days, , Senior vice president of Intel joined hands with several Intel architects at the architecture day conference , A comprehensive introduction to CPU and GPU Significant progress in Architecture .

lately AMD Go all the way .

however , The toothpaste factory also brought a series of new breakthrough products on the construction day .

The first performance hybrid architecture processor ; New independent graphics card architecture ; Next generation Xeon processors for data centers ; new IPU;oneAPI tool kit .

Senior vice president and general manager of accelerated computing systems and graphics division Raja Koduri

As a consumer , Processors and stand-alone graphics cards with a completely innovative architecture are obviously more interesting .

New unique architecture , Based on TSMC N6

Intel has launched a new micro architecture of its independent graphics card ——Xe HPG.

The new unique display will provide fever level high performance for games and creative workloads .

Xe The kernel has 16 A vector engine and 16 A matrix engine 、 Cache and shared internal video memory

Support DirectX and Vulkan Ray tracing

Compared with the previous generation Xe LP Architecture implementation 1.5 Times the frequency increase and 1.5 Times performance improvement per watt

Optimize Architecture 、 logic design 、 Circuit design 、 Process technology and software

Using TSMC's N6 Process

Refactoring kernel graphics card driver components , Especially memory managers and compilers , The throughput of computing intensive games is improved 15% ( at most 80%), Game loading time is reduced 25%

Besides , Utilization based on AI Supersampling can change the resolution from 1080p Upgrade to 4k, And achieve almost the same image quality .

XeSS Technology utilization Alchemist The built-in XMX AI Speed up , Use deep learning to synthesize images very close to native high-resolution rendering quality .

From adjacent pixels , And motion compensation for the previous frame , To reconstruct sub-pixel details .

And with the original 4k Rendering comparison ,XeSS Technology can improve performance 2 times (FPS).

Games that could only be played at ultra-low resolution , Now you can even experience 4K Picture quality .

Graphics card roadmap

Intel Express , The first product is expected to be 2022 Listed in the first quarter of .

The first performance hybrid architecture : energy efficiency + performance

Energy efficiency nuclear (Efficient-core)

Facing today's multitasking scenario , Energy efficiency cores can improve throughput efficiency , Provides scalable multithreading performance .

Reduce the overall power consumption through low-voltage energy-efficient core , Provide power thermal space for higher frequency operation , This also allows the energy-efficient nuclear to improve performance , To meet more dynamic workloads .

At the same time, it can realize multi-core task load , And has a wide frequency range .

The energy efficiency core can prioritize workloads without consuming processor power , And through the number of instructions per cycle (IPC) Improved functionality directly improves performance , These functions include :

With Intel's most productive CPU kernel Skylake comparison , In single threaded performance , Energy efficient cores can be implemented at the same power consumption 40% Performance improvement of , Or when the power consumption is less than 40% Provide the same performance in the case of .

And two... Running four threads Skylake Compared to the kernel , With lower power consumption , The throughput performance provided by the four energy-efficient cores can bring 80% Performance improvement of , While providing the same throughput performance , Power consumption will be reduced 80%.

Performance core (Performance-core)

Intel's new performance core microarchitecture , It can break through the limitations of low latency and single threaded application performance .

Because the code volume of the workload is growing , Need stronger executive ability . Data sets also increase significantly with the increasing demand for data bandwidth , While the performance core increases significantly , It can better support applications with large code volume .

The performance core has a wider 、 deeper 、 The smarter architecture is enough to make it the highest performance Intel has ever had CPU kernel .

The performance core can break through the limits of low latency and single threaded application performance , It all depends on the following advantages :

Compared with the current second 11 Generation processor architecture , Achieves an average of about for a wide range of workloads 19% Improvement ;

Higher parallelism and increased execution parallelism ;

Equipped with Intel Advanced rectangle extension (AMX), Built in next generation AI Accelerated lifting technology , For learning reasoning and training ;

Reduce the time delay , Provide better support for applications with large data and large code volume .

Besides ,IPC There will also be 19% The promotion of .

Alder Lake

Intel's first performance hybrid architecture , Code name 「Alder Lake」.

Maximum supportable 16 Core ,24 Threads , And up to 30MB The cache of .

As Intel's next-generation Client Architecture ,Alder Lake be based on Intel 7 Made by process technology .

Whether it's an ultra portable laptop or a commercial desktop ,Alder Lake All support .

It uses a single 、 Highly scalable SoC framework , Provide three types of product design forms :

High performance 、 Dual chip 、 Socket type desktop processor , Leading performance and energy efficiency . Support high specification memory and I/O;

High performance notebook processor , use BGA encapsulation , And add the image unit , Bigger Xe Video card and Thunderbolt 4 Connect ;

Thin and light 、 Low power notebook processor , High density packaging , Configuration optimization I/O And power transmission .

To build such a highly scalable architecture , That means you need to... Without affecting power , Meet the calculation and I/O The agent's extraordinary demand for bandwidth .

So , Intel designed three independent internal buses , Each of them adopts a real-time heuristic post-processing method based on requirements .

The computing internal bus can support up to 1000GBps—— That is, every kernel or every cluster 100GBps, Connect the kernel and graphics card to memory through the last level of cache ;

I/O Internal bus support can be up to 64GBps, Connect different types of I/O And internal equipment , It can change the speed seamlessly without interfering with the normal operation of the equipment , Select the internal bus speed to match the required amount of data transfer ;

The memory structure can provide up to 204GBps The data of , And dynamically expand its bus width and speed , To support high bandwidth 、 Multiple operating points with low delay or low power consumption .

Alder Lake It will go on sale this fall .

Reference material :

copyright notice
author[Xinzhiyuan],Please bring the original link to reprint, thank you.

Random recommended