Multi-core performance of the CPU, this is no longer repeated, but can the CPU's single nuclear performance can be increased unlimited? Before answering this question, let's talk about a little story. Many years ago, VIA VIA can also have Intel hard. At that time, the Pentium 4 era, Intel's CPU frequency continued to break through 1GHz, 2GHz, 3GHz to do a higher frequency CPU, said that Run 4 frequency 4GHz, later there At the front of Intel, CEO Barrot, because Intel launched a 4GHz high-frequency product as the promise of the 4th era.
But many people don't know that 4GHz is not the ultimate goal of Intel. In 2001, IdF meetings at Intel once said that the 4 processor can be 10GHz. Today, 18 years have passed, this goal has not been achieved, (Silicon-based era) may never be achieved.
This thing can show that the CPU frequency does not want to improve, and the 4 times have been in the past for so many years. In fact, the mainstream frequency of the CPU is still around 4GHz. Although Intel has achieved official 5GHz frequencies in Core i7-8086k, but is most Most processors are not so high in daily use, and the frequency under high load is not bad in 4GHz.
The first problem that constrained monoconuclear performance super-strong CPU is that the frequency cannot be greatly improved, and this factor is also related to the current process technology, essentially Moore's law has been invalid, which affects the 50 years of the semiconductor industry 50 years of Jinke Yuqi The arrival of the silicon-based chip physical limit has been invalid, and there is no significant performance improvement from the 28 nm node, and the power consumption problem is getting more serious.
Everyone knows the theoretical process process, the more advanced process, the CPU performance will be higher, power consumption, fever will be lower, but in fact this problem is complicated, the power consumption of the CPU can be divided into static power consumption (Static) Power) and dynamic power, the former is mainly caused by leakage current, the more advanced process, the leakage current has increased trend, and the dynamic power consumption can be calculated using 1/2 * CV2F, the F frequency is more High, the higher the dynamic power consumption.
For the higher frequency, the voltage increase is inevitable, but the voltage is high, the static power consumption, the existence of dynamic power consumption determines that the CPU frequency is higher, the power consumption will increase speed, will be serious The performance performance of the processor is influential because it is to be reduced.
Speaking of this, Intel's 14nm craftsmanship was excluded, but from the technical contest, from the first generation of Skylake architecture, 14nm ++ process, performance increased by 26%, or work The cost is reduced by 52%, which is rare in the case where the basic structure is not changed.
The slowdown in the process technology causes the CPU frequency impossible, and many people think that there is a very cow's CPU architecture that makes IPC greatly improved? Theoretically, this idea is ok, but the reality is cruel, the CPU architecture is still obeying the physics law of the semiconductor process, there is no advanced process, and the good CPU architecture is not possible.
In addition, even if the process on the CPU architecture is not considered, the CPU architecture is simply, whether the X86 is still the ARM architecture, in the 64-bit era CPU unit is nothing more than the ALU unit, cache, I / O, etc., but regardless of the improvement The part of the unit, is still in the end, it is necessary to calculate the number of transistors, and consider the cost of improvement - this cost is not just money, such as lifting L1 / L2 / L3 cache can improve performance, but the core area of the cache is very Big, and there is also a problem of hit rates and punishment, not just adding units.
In addition, the inside of the CPU can also be divided into an integer part, a floating point part, and the former is important for daily use. Floating point performance is more important to calculation, but the floating point performance of the CPU is not required for daily, so everyone is universally can't feel This part is improved.
Support AVX512 Core i9-7900X floating point performance improved
Fairly said that the advancement of the CPU floating point unit in recent years is a single-core superior requirement that meets the topic, because from SSE to AVX to AVX2 to the latest AVX-512, CPU floating point performance is substantial . Such as Intel said: "With up to two 512-bit fusion multiplication (FMA) units, the application can package 32 double precision and 64 single-precision floating point operations per second per second in the 512-bit vector. And eight 64-bit and sixteen 32-bit integers. Therefore, compared with Intel Advanced Vector Extension 2.0 (Intel AVX2), the width, quantity, and the width of the FMA unit are doubled. "
But the front also said that the floating point performance of the CPU is not required, and the integer performance is more important, but the performance improvement of the integer unit is not so obvious, causing many people to squeeze for the CPU architecture for many years.
Be
Hot recommendation:
Micro-core AVR-IOT WG AC164160 evaluation board, safety connection millions of smart devices
Yadno Semiconductor LT6274 / LT6275 90MHz 2200V / μs 30V low power operation
Texas Instrument TPS65218EVM Evaluation Module, Evaluation of TPS65218 Power Management Device Performance Full Assembly Platform
British Flying CoolGanTM Gallium Nitride HEMT, set new power conversion benchmark
Be
Xiaobian Recommended Activities:
[Raspberry Seaf 4 guess] predicting correctly available new version of the Raspberry Party
Double reward | watch videos, learn TOF, understand Yingfailing Real3
Mouser Send a gift - participate in users have a prize, there is a Raspberry School 3B + waiting for you!
Be
Invite you to join the panel engineer group (group number 135513647)
More activities, consultation, technical exchange
Our other product:
