The idea of placing a small SRAM memory chip on top of the chiplet that contains the processor cores and caches, all connected via vertical silicon pathways, is undoubtedly one of the best ideas AMD has ever had. This allows them to increase the L3 cache, which means an increase in the number of successes when looking for data and instructions in it. The consequences? A performance boost by reducing latency.
How are the Ryzen 7000X3D?
What interests us is the main course and it is that contrary to what happened in the last generation where only the 5800X3D was the only CPU of its type that was launched on the market. This time we have three different models, which They will be available from February of this year.. That is, in a few weeks. A launch date that has surprised us due to the fact that we expected these brown beasts in a few months and a demonstration that AMD, if it has acquired excellence in anything, is in the times when it comes to developing new processors.
| Model | Cores/Threads | total cache | Basic GHz | GHz Boost | TDP |
|---|---|---|---|---|---|
| Ryzen 9 7950X3D | 16/32 | 144MB | 4.2GHz | 5.7GHz | 120W |
| Ryzen 9 7900X3D | 12/24 | 140MB | 4.4GHz | 5.6GHz | 120W |
| Ryzen 8 7800X3D | 8/16 | 104MB | 4.6? GHz | 5GHz | 120W |
Despite being variations of the Ryzen 7000, the development of the Ryzen 7000X3D has been carried out in parallel, since it has certain improvements, such as being more efficient per watt and having lower TDP, 120 W, than the models that came out a few months ago. This is crucial to prevent V-Cache placement from causing a thermal throttling problem on the processor. Regarding the amount of cache, there is a detail that suggests that these processors could be the first asymmetric in the history of the company led by Lisa Su.
Regarding its performance, for the moment being the designers of this chip we have to take it with tweezers, but compared to the Ryzen 7 5800X3D, AMD has said that the 7800X3D in particular exceeds its predecessor, the 5800X3, by 20% in performanceD. And with it it is placed as the most powerful 8-core processor in the history of the PC. As for the top of the range, they have stated that the 7950X3D outperforms Intel’s i9-13900K by 24%.
Is it a processor with heterogeneous cores?
Well, we have to clarify that each of the cores in the Ryzen 7000X3D are symmetrical with the others and, therefore, they are of Zen 4 architecture, however, AMD has done something that cannot be seen with the naked eye, but it can be seen. you can see in the promotional images of the chip. Of the two CCDs that the Ryzen 9 have, only one of them has the V-Cache installed and the other does not. As a result, processes running on one of the chiplets will run faster than on the other. Hence the heterogeneous, since not all perform in the same way.
Why has AMD done this? Well, the normal thing would be that they would have created a single CCD for the 16 cores. This would have forced them to change the internal topology and redo the ring that communicates the cores and that is designed to support up to 16 clients, threads of execution, to 32 clients. However, this would have slowed down the processor. Wouldn’t it be better if both chiplets had V-Cache? No, from the moment in which being physically distanced completely destroys her intentions. Our conclusion is that they already tested it and realized that the distance between the CCDs eliminated the advantage of the V-Cache in the interoperability scenarios between one chiplet and another inside the processor.
Can Intel respond with a similar solution?
Power could, using its Foveros technology to place the two chips on top of each other to communicate with each other. We already saw how they did it with Lakefield back in the day, of course, we haven’t seen it applied yet in high consumption processors, which is typical in desktop systems. The interesting thing about the V-Cache is not only the performance it provides, but the ability to have a processor with a large amount of last-level cache in a world where the latency between processor and RAM is increasingly high.
It would not be the first time that Intel copies an idea that works to AMD and vice versa. The two most famous cases are the integrated memory controller and the 64-bit extension of the x86 architecture. So I would not rule out a future Intel processor with said architecture. Of course, it would be a daunting job for them to have to change the entire cache structure of the processor, as well as the intercommunication between the cores.
