On Computex 2022, according to AMD's public roadmap, AMD will be very busy in the second half of 2022. It will not only launch the new Zen 4 Ryzen 7000 desktop processor, but also launch the fourth generation EPYC Xiaolong Enterprise Processor and RDNA3 Radeon 7000 series graphics cards. However, some parts are not very clear, and AMD's technical marketing director Robert Halllock later introduced more new details about the Ryzen 7000 series in an interview. Let's take a look together.
First, let’s quickly review the Ryzen 7000 series, which is AMD’s new generation desktop processor, with a new Socket AM5 LGA package built-in, requires a new motherboard, supports DDR5 memory and PCI-Express Gen 5. At the heart of the Ryzen 7000 is the new Zen 4 microarchitecture, which has higher IPC and new features than the Zen 3. The Zen 4 CPU core is made using TSMC's 5nm EUV process, while the I/O chip is made using 6nm process.
In the Computex 2022 demonstration, AMD showed an unnamed Ryzen 7000 processor engineering sample that can play AAA-level games. The chip is shown as 16 cores and 32 threads. Is this the largest number of cores when the Ryzen 7000 series was released? Robert Halllock gave a yes answer.
When discussing CPU overclocking, Robert Hallock said “5.5GHz is very easy for us” and used early Ryzen 7000 series CPUs in their 5.5GHz clock speed demonstration, without overclocking or using over-exceptional cooling settings. If AMD's 5.5GHz clock speed is "easy" to achieve, AMD's flagship Ryzen 7000 series CPUs may be able to provide higher acceleration clock speeds. Perhaps AMD's Ryzen 7000 series CPUs will be able to reach 5.7GHz,
On Computex, AMD showed a 15% improvement in single-threaded performance. In response, Robert Hallock said that this is just a conservative number and will release detailed scores of IPC and frequency later. AMD 15+% single-threaded performance data comes from a single benchmark Cinebench R23, and in short, one benchmark data is not enough to fully describe the performance improvements provided by the new CPU architecture.
Robert Hallock also clarified the meaning of “AI acceleration” mentioned in the speech, which will rely on AVX 512 VNNI and BLOAT16 instructions, formats that are widely used by TensorFlow, AMD ROCm and even NVIDIA CUDA libraries.
Regarding CPU integrated graphics, Robert Hallock said that RDNA2 graphics that are not in the computing chip but in the I/O chip will become standard configuration. These will provide basic display output capabilities as well as video encoding and decoding, including AV1. There have been rumors that Raphael's RDNA2 graph may have 4 computing units. Hallock said the specifications of the RDNA2 iGPU should be consistent across all CPUs. Whether the
AM5 platform will live as long as the AM4 platform, Robert Hallock said it is not sure at the moment. It is still in the early stages of building AM5 and there is still a long way to go. As for the photos that look like the computing chip is gold-plated? Robert Halllock said they are not gold-plated, calling it a "back metallization" process that uses it to solder the die to the radiator. Depending on the manufacturing method, different colors of light can be refracted.
Computex 2022 is over, and the next time AMD may announce more information on the financial analyst day in early June, and super hardware will continue to pay attention.