"Just yesterday AMD second generation thread torn flagship threadripper 2990WX appearance officially lifted, AMD official website also exposed part of its performance, benefited from 64 threads with up to 32 cores, Threadripper 2990WX easily won the consumer market The title of the king of the CPU, I9-7980XE sitting on the throne is relaxed, which can be described as an unlimited scenery between the time.
But the saying goes well, now there are many scenery, before, there is a lot of discount (the meaning of the lover), AMD is no exception. Just 2 years ago, AMD is also an opponent that is ignored by Intel. The product performance and Intel have a huge gap. The stock price is only 2 US dollars, which is on the verge of bankruptcy. AMD will be so down, one of the main reasons is because the BullDozer, bulldozer processor micro architecture launched in the year.
In the first two days, foreign media EXTREMETEC has summarized the worst 10 CPUs so far, with Bulldozer entered the position of the top three, which can be seen that the failure of the bulldozer is a relatively recognized fact. Today, in Threadripper 2990WX, PConline Xiaoyou led everyone to review the failed architecture of AMD, seeing it from birth to the story, as a story of a loser.
Birth of the bulldozer: the mission of carrying revenge
AMD is an adventurous company. Although his revenue is only 1/10 of Intel, it dares to innovate in HT bus, DDR memory, multi-core processor and other technologies. With the K8 architecture in this century, The processor leads a big lead in the IPC and performance, in which INTEL is "teaching people" during this time. But Intel is still the big brother of the CPU, rich in financial resources, massive talent reserves, huge industry influence and strong relationship with OEM core partners have become Intel turn capital.
Finally, in 2006, the Core series processor was shocking, and the first batch of Core 2 EXTREME 6800 and Core 2 Duo E6300 showed unparalleled performance. Not only don't transcend the upper generation of CHANPIN40%, but the performance of the power consumption is also made of tongue, Intel's products complete a gorgeous counterattack. If the first processor model, the first launched processor model is the horn of the countert, then the Core I Series processor for Nehalerem is the foundation of Intel, which is thoroughly laid the foundation of the CPU market for 13 years.
Powerful intel
In November 2008, Intel released the Core i7 965E / 920 processor, native quarter-core, internal integrated memory controller, using QPI bus architecture and HT super-thread technology to break the memory bandwidth transmission bottleneck, with a smart Turbo Turbo acceleration Techniques increase the energy consumption ratio of the processor to the ultimate, the release of the Core processor is a syndic change, it is convincing KO's K10 architecture processor at the time of AMD, forcing the other party can only rely on cost performance or "open nuclear", etc. Sao campaign can have a lot of share in the market.
Taste the sweetness of success, AMD naturally will not be so easy to accept failure, in order to change the market situation, AMD is fully investigated into the development of the next generation revolutionary architecture, and launched the first generation of bulldozer processors in 2011, but At that time, no one did not expect it, the launch of the bulldozer architecture processor is the beginning of AMD completely defeated in the CPU market.
Bulldozer architecture: embarrassing "innovators"
The micro-frame structure and production process of the CPU directly determines the efficiency of the CPU, optimizing the micro-architecture and update production process to become the most important way for CPU vendors to improve CPU performance, and the bulldozer architecture is the CPU micro-architecture made by AMD.
Early AMD proposed a lot of innovation points when promoting the bulldozer architecture, summarized 1, new modular design, more efficient, core extension easier. 2,32nm SOI production process, power consumption control is more excellent. 3, a new multi-threaded architecture, multi-threaded computing performance is stronger. 4, the command 4 is transmitted (K10 only 3 emissions) and the AVX instruction, the integer / floating point operation is stronger, and the single core performance is improved. 5. Second generation Turbo Core technology, better adapt to a variety of applications.
Among them, the core foundation and soul of the bulldozer architecture are modular design. Everyone knows that the CPU has more physical cores in the traditional sense, and the performance will be stronger, but the cost will be higher, and the Intel is applied to the Core processor applied SMT technology, which is the hypercale. SMT allows a plurality of threads of a CPU core to share resources and perform synchronously, hardware does not need to increase costs, but efficiency is definitely no more physical cores.
AMD "Bulldozer" micro-architecture
AMD is unique to launch CMT technology in order to balance cost and multi-line effectiveness. AMD encapsulates two cores and related units into a module on the "bulldozer", and two cores share a floating point arithmetic unit, but each core has a complete integer arithmetic unit, and the FX-8150 consists of eight cores. The floating point unit is actually only four, and the CPU in the past is a floating point unit for each core. Thus, the four cores are composed of dual modules, and the six cores are composed of three modules, so push. The benefits of using modular design can reduce redundant circuits, and the CPU core is easier, which is really a creative of a sky. AMD called this physical multi-core, and therefore, the 8-core CPU called the bulldozer called 8-core processor. The author is more willing to call it 4 module 8 thread processors because each "core" is actually Incomplete disabled core, there is no separate floating point arithmetic module, but the benefit of doing this is that the CPU obtains 8 complete integer arithmetic units with relatively small costs, and 8 threads with high-end processors.
AMD is designed in the bulldozer. In fact, there are three reasons: one is that he believes that more than 80% of the calculations in the CPU are integer operations. The advantage of adding an integer unit is obvious, using a 5% core area of the core area. In exchange for 80% of the integer performance, the floating point operation can be handed over to the GPU, which is more efficient. Second, in the future, the general operation will continue to develop multithreaded, and the requirements for threads are unlimited. The third is that the future of the CPU frequency is achievable, which can make up for the problem of single-threaded processor.
Amd's gambler corporate style reflected in the development and design of the bulldozer, and he hit three gambling in the future processor development, but unfortunately did not have a bet.
8 nuclear bulldozer architecture
The advanced desktop processor equipped with four modules has a large number of integer threads, but most users' workload is still unable to be allocated to eight threads (simple to say that most programs multi-thread optimization is not good. ). Single threaded operation still occupies the most use scenarios of users. On the other hand, the sharing of floating-point units means that there is not enough operating resources with floating point arithmetic applications. Although GPU-based calculations are very important in some specific work - such as scientific supercarcies - but mainstream applications are more dependent on CPUs to do floating point operations.
The bulldozer's architecture brings the retrogression of the processor monoconuclear performance, even if you don't go to your own on behalf of the six-core flagship Phenom II X6 1090T, do not say to the SNB processor compared with Intel. The reverse retreat of monoconuclear performance in AMD can improve the frequency of the processor, but the bulldozer processor using a more backward 32nm process is a big fire furnace, resulting in a large fire in power consumption (this situation The ultimate the ultimate when the pile driver).
Bulldozer flagship FX-8150
Another result caused by such a design is that most game performance of the most floating point calculation performance and monoconuclear performance, the bulldozer is even less than "fat dragon 2", can only eat dust behind the SNB processor, this is in the DIY market It is fatal, after all, most DIY players have to play DIY hardware for the game. The consumer market does not buy, reputation is gradually collapsed, and the ending of the bulldozer architecture seems to have been foreseen.
3 Development of bulldozer: Agricultural machine series processor
Pile driver: unyielding challenger but finally a laughter
AMD's development road map of CPU at the time
After the introduction of the bulldozer architecture processor, the market feedback is not very good, but this is just the first generation, or the market and software environment is still not adapted to the new species in this processor. So AMD launched a version of the bulldozer small repair, the PileDriver pile driver architecture processor. Compared with the bulldozer, the same "pile driver" core changes are mainly reflected in: 1. New FMA3, AVX1.1 and F16C and other new instructions; 2. Strengthen power management, reduce product power consumption; 3 , First-level, secondary caching optimization; 4, the core frequency is improved, the TDP remains unchanged.
The most mainly this power consumption is optimized, which allows the processor to compare the last generation of overclocking spaces at the same voltage. Although the pile driver represented by FX8350 is only created by Groford's 32nm process, but it benefits from the long-standing pipeline, the pile driver can be easily overclocking to 4.5GHz, so that the time square The a meal is called "The performance is not enough, overclocking is coming", "super, not". " In order to play the performance of the pile driver, AMD even launched a nuclear bomb FX9590 with TDP up to 220W, dynamic frequency up to 5.0GHz, but its terrible power consumption and heat dissipation power supply requirements, and 5.0GHz single nuclear performance only The i7-4770K single nuclear performance at 3.8GHz is a mainstream market, and the consumer is selected.
The FX9590 is just AMD to emit an unyielding roar. The high-frequency pile driver can only bring the title of the big stove to AMD, but still can't change the fate of the bulldozer family failure, and finally settle the shame title of "I3 Meriditian".
The failure of the pile driver made AMD aware of himself. To this AMD gave up the CPU high-end market, transfer to another large pit heterogeneous computing research (Details Click here to jump related articles). Since then, AMD is launched, the Excavator excavator is only applied to the APU and the low-end processor model, and the focus is to reduce the power consumption of the CPU and improve the energy consumption of the processor. Although these low-end APU products are favored by many consumers in the market, consumers choose its reason most because of their high performance nuclear, and even many netizens laugh, they buy APUs to buy GPUs to send CPUs, buy Intel's CPU It is to buy a CPU to send GPU.
AMD's CPU has also been completely lost in the market. No matter from various perspectives such as market or performance performance, the bulldozer architecture is very failed, and it can even be said to be nailed to a shame column of CPU history, so until today AMD I don't dare to enable the FX suffix that once symbolizes the highest performance processor. After all, FX is easy to let people think of the bulldozer series processor.
4 Influence of the bulldozer: Failure is the mother of success
Modular design: in front of the past
The popular discourse is always more reasonable. For example, failure is the mother of success, and, for example, digital hardware information comes to PConline. Since the failure of the push machine, AMD has a sustablishment, and finally in 2017, it has made a surprising power in 2017, and it has achieved counterattack in the case of everyone who is not optimistic, and the Ruilong series processor comparable to the Core i series processor is released. At the time, AMD has been in the top of the top for 13 years.
ZEN architecture
The Ruilong series processor can be able to use the architecture of the CPU Complex, which is well adopted by the Architecture of the Ruilong Series. The architecture is also cleverly adopted a modular design called CCX (CPU Complex). Modular this word is not very familiar? You have not mistisred, although the modular design concept is the root cause of the failed bulldozer, but the modularity is not a flood beast, and the Zen architecture is a more mature product under the modular design concept.
Although the modular design concept, the Zen architecture sucks the lessons of the bulldozer, each of the core is a complete core, and greatly strengthens the floating point calculation performance, on the integer line, ZEN has 4 arithmetic logic units (ALU ) And two address generating units (AGUs). On the floating point line, the concept of shared floating point units is discarded: Now each core has a pair of independent 128-bit multiply superimposed unit (FMA). Separated addition and multiplication lines in the floating point unit are used to deal with more mixed instructions when the multiplication is not performed. However, the 256-bit AVX instruction set is also separated on two FMA units, and all floating point units are used. The final result is that the Ryzen series processor contrast the bulldozer series of processors has more than 40% of the huge increase in IPC.
ZEN architecture
There are 4 cores in a CCX module in the zen architecture, each with its own independent L1 and L2 cache, and a CCX module shares 8MB L3 cache. Each core can selectively turn the SMT function, or selectively close the part core. The benefits of bringing the zen architecture compared to Intel's annular wiring, it is obvious. On the latest processor layout of AMD, each generation from the highest server epyc to the lowest end-level Ryzen 3 series processor, they only need to design a Die (kernel), then constantly stack, this comparison Intel will save There are a lot of design and flow costs, so we can also see that the market's core Ryzen processor price will be low compared to Core. In addition, modular design also allows the CPU core to increase as simple as stack building blocks, and the birth of Threadripper 2990WX is also a matter of water to the stream.
Improved the shortcomings of the bulldozer single thread, relying on high frequency, and adhering to the modular design concept, I believe that the future general operation is still very demanding for multi-thread, after 6 years, AMD once again relying on the Zen architecture Enter the horn of the high-end CPU market.
The AMD relies on Threadripper 2990WX to win the consumer-level CPU market throne, it is the result of the most failed processor gamma footstone in history.
Summarize
The hero will always be late, not to mention that the bulldozer is more like a traditional "dog bears", but no matter what, when the bulldozer 29990wx is so powerful, I believe that there will be no regret to leave, disappear There is a huge flow in history. And I can only use this.Text, commemorating this I powder mock, is a story of a hate, bringing it to you.
Original address: https://www.eeboard.com/news/amd-100/
Search "" Love Board "", pay attention, daily update development board, intelligent hardware, open source hardware, activity and other information can make you fully master. Recommended attention!
[WeChat scanning picture can be paid directly] "
Our other product:
