AMD has just officially announced the brand new Zen 3 CPU architecture and the brand new Ryzen 5000 series desktop processors. The gorgeous specifications and performance did not disappoint us.
This is the third-generation new architecture after Zen, Zen+, and Zen 2, and it is also the third-generation new processor after the Ryzen 1000, Ryzen 2000, and Ryzen 3000 series. In order to be consistent with the Ryzen APU naming sequence, at the same time To facilitate consumer identification, AMD skipped the Ryzen 4000 series on the CPU this time and entered the Ryzen 5000 series directly.
Not surprisingly, at the CES 2021 exhibition early next year, we will see the Ryzen 5000U/H APU, which is also based on the Zen 3 architecture, for thin and light notebooks and gaming notebooks.
Looking back at history, the Zen architecture born in 2017 has achieved the biggest leap in AMD processor history. Compared with the previous bulldozer architecture, IPC (instructions per clock cycle) performance has increased by an incredible 52% (which can be roughly understood as same-frequency performance) Changes), coupled with the subsequent optimized version of Zen+, basically laid the foundation for AMD processors for many years.
Zen/Zen+ is manufactured using GlobalFoundries 14/12nm process, the basic architecture module is CCX (CPU Core Complex), each module has 4 physical cores, and supports SMT synchronous multi-threading, that is, 8 logical cores, and supports new dynamic acceleration Algorithm, and each CCX integrates 8MB L3 cache.
On the one hand, Zen 2 upgraded the manufacturing process to TSMC 7nm, and on the other hand continued to reform the underlying design, bringing a 15% increase in IPC, and the highest acceleration frequency has also moved from 4.35GHz to 4.7GHz.
It continues to use the 4-core CCX modular design, and the three-level cache capacity of each CCX is doubled to 16MB. At the same time, the new chiplet design is introduced, which can easily expand more cores. The mainstream desktop comes to up to 16 cores, and the enthusiast Desktops and data centers are up to 64 cores.
So what will Zen 3 do next?
First of all, Zen 3 architecture continues to be matched with TSMC’s 7nm process. There is no change in this aspect. Therefore, any specifications and performance improvements are purely the contribution of the architecture itself, without the bonus of the process . This is undoubtedly a great challenge to the architecture design. .
To this end, Zen 3 re-adjusted the CCX, core layout, and cache system, once again significantly improved IPC, and the acceleration frequency continued to increase.
Of course, high frequency is not the top priority of Zen 3, and more energy is put on improving energy efficiency.
Specifically, the Zen 3 architecture doubles the number of cores of each CCX module to 8 (16 threads), and the capacity of the third-level cache is also doubled to 32MB, and all 8 cores are shared and can be directly accessed, which is equivalent to each The core's three-level cache capacity has also doubled, which greatly accelerates the communication connection between the core and the cache in practical applications and significantly reduces memory latency.
At the same time, each CCD chip is still 8 cores, and the CCX modules contained inside have changed from two to one . CCX and CCD basically have the same meaning.
To some extent, you can simply understand the Zen 3 CCX module as the product of the two CCX modules in Zen 2. Of course, the actual situation is much more complicated than this, involving connection speed, bandwidth, delay and other aspects Adjustment and optimization. The details of this will be announced later…
AMD claims that under the same 4GHz fixed frequency and 8-core configuration, combined with 25 application load test results, the IPC of the Zen 3 architecture has increased by at least 19% compared to the Zen 2!
With this accumulation, the same frequency performance of the Zen 3 architecture has more than doubled compared to the bulldozers of the year! Compared with the original Zen architecture, it has also increased by about 37%.
And we just said that the 19% improvement this time has nothing to do with the manufacturing process. It is entirely from the optimization of the architecture itself. The front-end unit and load/storage unit contribute the most to the improvement, followed by the execution engine and cache prefetching. Unit, micro-manipulation cache unit, branch prediction unit.
It can be said that every module of the Zen 3 architecture is redesigned from the inside to the outside, which is also the biggest change since the birth of the Zen architecture.
The Ryzen 5000 series still continues the small chip design, including one or two CCX (CCD) Die, one IO Die, and the latter has not changed at all. It continues with the 12nm process, 16 PCIe 4.0, dual-channel DDR4-3200 memory, which is completely the same as the previous generation. the same.
In terms of energy efficiency (performance per watt), AMD claims that the Zen 3 architecture has reached 2.4 times that of the original Zen architecture. Compared with Zen 2, it has also increased by 20% . Compared with the mainstream flagship i9-10900K, AMD has reached an astonishing 2.8 Times!
The Ryzen 5000 series first released four models. The flagship is the " Ryzen 9 5950X ". Like the current Ryzen 9 3950X, it has dual CCD modules, 16 cores and 32 threads, 8MB L2 cache, and 64MB L3 cache. Of course, these three The level cache has changed from four blocks of 16MB to two blocks of 32MB, which are shared by eight cores.
The highest acceleration frequency has gone from 4.7GHz to 4.9GHz , almost touching the 5GHz threshold, and the base frequency is 3.4GHz.
While achieving all this, the thermal design power consumption remains at 105W , which is completely unchanged!
According to official indicators, compared with Ryzen 9 3950X, the content creation performance of the Ryzen 9 5950X can be increased by 27%, and the game performance can be increased by 29%. Compared with the i9-10900K, the creative performance is up to 59% and the game performance is higher. Up to 11%.
AMD also admits that it can lead its opponents without pursuing the performance of every game, but it will narrow the gap as much as possible, and at the same time, those who can lead will continue to lead.
Next, look at " Ryzen 9 5900X ". Same as the current Ryzen 9 3900X, it has dual CCDs, 12 cores, 24 threads, 6MB L2 cache, 64MB L3 cache (naturally changed), and the base frequency is lowered from 3.8GHz To 3.7GHz, but the maximum acceleration frequency is increased from 4.6GHz to 4.8GHz , and the thermal design power consumption remains 105W. If you compare Ryzen 9 3900XT, the base frequency is lowered by 100MHz and the acceleration frequency is increased by 100MHz.
According to official data, the Ryzen 9 5900X CineBench R20 single-threaded score is as high as 631, which is the first time in the history of desktop processors to break the 600-point mark. Whether it is Ryzen 9 3900XT or i9-10900K, the score is around 544, a comparative improvement. Is 16%.
The frequency is fine-tuned to 100MHz and the single-core performance is improved so much, I have to admire the Zen 3 architecture is really good.
In terms of game performance (1080p resolution), Ryzen 9 5900X is up to 50% higher than Ryzen 9 3900X, and up to 21% higher than i9-10900K. Of course, there are a few slightly inferior ones.
Accordingly, AMD claims that the Ryzen 9 5900X is the best gaming processor in the world -which makes Intel, which has always given this title to its flagship, to be so embarrassed…
Then there is " Ryzen 7 5800X ", single CCD, 8-core 16 threads, 4MB L2 cache, 32MB L3 cache, base frequency 3.8GHz, maximum acceleration 4.7GHz, which is 100MHz lower than Ryzen 7 3800XT, the latter No change, the thermal design power consumption is 105W.
The last is " Ryzen 5 5600X ", 6 cores and 12 threads, 3MB L2 cache, 32MB L3 cache, base frequency 3.7GHz, maximum acceleration 4.6GHz , compared with Ryzen 5 3600XT, the former is reduced by 100MHz and the latter is increased by 100MHz. The thermal design power consumption is 65W.
AMD did not give specific performance data for the Ryzen 7 5800X and Ryzen 5 5600X, but it advertised a wave of cost performance:
Compared with i9-10900K, Ryzen 9 5900X is 13% higher in single thread, 23% higher in multi-thread, and 3% higher in 1080p games.
Compared with i7-10700K, Ryzen 7 5800X is 9% higher in single-thread, 11% higher in multi-thread, and the 1080p game is the same.
Compared with i5-10600K, Ryzen 5 5600X is 19% higher in single thread, 20% higher in multi-thread, and 13% higher in 1080p games.
Of course, Intel also has relatively cheaper KF series. Currently, the Bank of China offers i9-10900KF, i7-10700KF, and i5-10600KF which are 200 yuan (USD $29) , 150 yuan (USD $21) , and 600 yuan (USD $86) lower than the standard version, but they are selected comprehensively The price-performance ratio of the new Ruilong is still good, and this is only the initial stage. The follow-up will definitely become cheaper and more prominent.
The Ryzen 5000 series will be launched on November 5th. In terms of prices, Ryzen 9 5950X is 799 US dollars, Ryzen 9 5900X 549 US dollars, Ryzen 7 5800X 449 US dollars, Ryzen 5 5600X 299 US dollars, compared to Ryzen 3000 Corresponding models in the series have increased by $50.
It should be noted that models with thermal design power consumption equal to or higher than 95W will continue to not come with the original radiator and need to be purchased separately by the player; those with a thermal design power consumption of less than 65W will have their own ghost radiator.