AMD announced that they will hold a new CPU-related announcement at 1 AM KST on October 9, 2020.
I personally like these surprise announcements without any prior notice. Unlike Cyberpunk 2077, which teased its release for years, making us impatient and eventually indifferent, a sudden reveal like this is much more exciting.
According to the preliminary information, there have been some improvements.
However, those meaningless statistics charts with no concrete vertical axis criteria are useless, and until the product is released, the only thing we can really rely on are the hardware specs.
Roughly speaking, Ryzen’s unique design has historically caused latency issues. Previously, a single die contained two 4-core CCXs, making a total of 8 cores in one die. This time, they’ve put an 8-core CCX into one die, significantly reducing latency.
Moreover, while the L3 cache size remains the same, the shared cache area has doubled due to changes in the CCX structure. The IPC (Instructions Per Clock, a measure of CPU performance, though highly variable) has also improved.
The images show the previous generation’s CCD with two 4-core CCXs and the new generation’s single 8-core CCX. Previously, they had to synchronize the speeds of the CCXs and the CCDs, but now, there’s no need to distinguish between CCX and CCD. Just synchronizing the speed between dies connected via Infinity Fabric remains necessary, which is an unavoidable design aspect. Although the synchronization process has one less step now, there will still be some latency.
AMD has a history of misleading consumers with inflated specs (though many loyal users convince themselves the products are good), so actual performance specs will only be known once the product is released. However, it’s true that, according to the structural explanation, latency has been reduced compared to the previous generation (though the extent is unknown).
Seeing them shift towards bundling cores to enhance performance, despite their heavy marketing on the benefits of multiple dies for multitasking, suggests to me that it might be due to lacking process technology to achieve better power efficiency in a single die design.
Technically, AMD’s multi-die should theoretically be superior to Intel’s monolithic die in terms of heat dissipation. However, Intel CPUs can often achieve stable overclocks of 4.8 to 5.1 GHz with water cooling, whereas Ryzen struggles to hit its factory boost clocks and overheats even with water cooling when slightly overclocked, indicating a technological gap.
This generation, though, is said to have significantly improved power efficiency. Compared to the first-generation 1800X, the 3900XT has twice the power efficiency, and Zen 3 is reported to be 2.4 times more efficient. Compared to Intel’s i9-10900K, it’s claimed to be up to 2.8 times more efficient.
This implies they’re claiming to have surpassed Intel’s 10th generation in power efficiency since the 3000 series. But is this not false information? Or did they measure power consumption with the system idle?
Such claims foster distrust and criticism.
The 5900X is said to feature 12 cores and 24 threads, with a boost clock of 4.8 GHz. Given their history of exaggerated claims, actual performance will be confirmed only after release. The TDP specs are similarly not very meaningful as they vary widely depending on the task, a vagueness that applies to Intel as well.
The FHD gaming performance comparison between the 3900X and 5900X seems to reflect the structural latency improvements. However, these graphs are misleading as they normalize the heights of the 3900X bars, similar to political statistics in Korea.
In gaming, consistent frame rates without stuttering are more important than just high frame rates. Therefore, real-world use will be the best test.
Next is the Cinebench score. It claims to be the first CPU to surpass 600 points in single-thread performance. However, a 10900K can easily exceed this when overclocked. The meaningful overclocking capabilities of Intel CPUs contrast with Ryzen’s minimal overclocking margin, highlighting stability and performance concerns.
When comparing the 3950X with the 10900K in multi-thread performance, the price disparity undermines the significance of such comparisons.
Finally, the gaming performance comparison between the 10900K and 5900X also shows a meaningless, aesthetically normalized graph.
Price is another factor. As mentioned before regarding Nvidia’s 3000 series, AMD initially positioned itself as a consumer-friendly brand with lower prices, only to raise prices once they could compete on performance. This price increase suggests confidence in competing with Intel at the factory settings.
Although I will still prefer Intel for fast FPS games due to its low latency and smooth performance, those who do not mind these factors may now choose based on personal preference.
The price is a more honest indicator of a product’s performance than those arbitrary comparison charts.
To conclude, AMD’s repeated empty promises and misleading marketing, akin to political campaign pledges, lead to this critical view. However, if any company, including AMD, offers genuinely good products at reasonable prices, it will be positively received by consumers.
AMD’s political marketing aside, I am grateful for the increased cores, higher clocks, and lower prices in Intel’s products, driven by this competition.
It’s not the products but AMD’s misleading marketing that I find distasteful. As long as competition results in better products, it’s a welcome development.