Ryzen 4000U / H mobile processors: Eight stripped Zen-2 cores for notebooks

Desktop processors from the Ryzen (Threadripper) 3000 series and the Epyc 7002 server models are modularly constructed as groups of one I / O die and up to eight CPU chips. The combined processors Ryzen 4000U and Ryzen 4000H (code name Renoir), on the other hand, each consist of a single 7-nanometer silicon chip that combines CPU cores, GPU and chipset functions. This increases production costs, but otherwise offers many advantages.

The small cache is striking compared to previous Zen 2 processors: The Eight-core mobile processors like the Ryzen 7 4800U AMD calls 12 MB, divided into 512 KB level 2 cache per processor core and 4 MB L3 buffer memory per CPU complex (CCX). With the eight-core Ryzen 7 3700X desktop, the L3 cache is twice as large at 16 MB twice.

If you have problems playing the video, please activate JavaScript

(Source: AMD)

On the presentation of the Zen 2 architecture, AMD explained that the doubled L3 cache from Ryzen 3000 (desktop) compared to Ryzen 2000 compensates for the higher latencies by the communication between several silicon chips. The more data the computing cores can hold locally in the cache, the less often they have to reload data via the I / O die. In the Ryzen 4000U / H combination processors, the cores and memory controllers are located in the same silicon chip, which reduces the latency for memory queries – AMD achieves similar performance with less buffer memory.

Another cache halving saves AMD chip area. The manufacturer took this step with the previous generations Ryzen 3000U / H (Picasso) and Ryzen 2000U / H (Raven Ridge). For space reasons, AMD in Ryzen 4000U / H has probably also reduced the integrated graphics unit from a maximum of eleven compute units (704 shaders) to eight compute units (512 shaders). Thanks to significantly higher clock frequencies and faster memory, the GPU still manages higher frame rates than the predecessor Ryzen 3000U / H, according to AMD. The 4000 series supports either DDR4-3200 or LPDDR4X-4266 instead of DDR4-2400, which increases the transfer rate from around 36 GB / s to up to 64 GB / s.

According to AMD, the Ryzen 4000U / H series should be significantly above the performance curve of recent years.

(Picture: AMD)

With the design decisions, AMD is reducing the chip area of ​​the Renoir-Die to an estimated 150 mm² – 60 mm² less than the 12 nm predecessor Picasso (Ryzen 3000U / H). Earlier we assumed four computing cores for the new combination processorsto squeeze the chip area and thus the cost. Has eight slightly stripped down cores AMD, on the other hand, is a trump card against competitor Intel up its sleeve, which offers a single six-core in the same 15-watt class, the Core i7-10710U. At the same time, the production costs should remain moderate at 150 mm². AMD’s CEO Lisa Su emphasized lower selling prices compared to Intel at the CES 2020 IT trade fair, which means that market share can be gained.

The data exchange between several silicon chips costs more energy than the communication within a single die. In notebooks, low power consumption is particularly important for long battery life. AMD has to save every possible watt, which currently speaks against a chiplet design.

The Ryzen 4000U / H mobile processors are said to be twice as efficient as the predecessor – mainly through TSMC’s 7-nm process, but also by optimizing the CPU and GPU architectures.

(Picture: AMD)

Opposite the website AnandTech AMD has confirmed that Ryzen 4000U / H works with PCI Express 3.0 at most and not with PCIe 4.0, This should also be about saving electricity, since the transfer of twice the amount of data costs energy. In everyday life, an NVMe SSD with PCIe 3.0 is probably fast enough even in premium notebooks. Only the connection of an independent graphics chip like the Radeon RX 5700M or Radeon RX 5600M could be a compromise with PCIe 3.0.




Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.