AMD Ryzen 4000 Renoir in detail

– At CES 2020 in January, AMD gave a first look at the new mobile Renoir architecture, now the manufacturer is presenting the platform with Zen 2 cores and 7 nm Vega GPU in detail. BitcoinMinersHashrate participated in Ryzen Mobile Tech Day 2020 in Austin, USA in February and gives a comprehensive overview.

Today AMD allows a look at the technical details of the new notebook CPU product series “Renoir”, which uses Zen-2 cores and Vega graphics unit – each in 7 nm production. The company had already announced the new platform for Ryzen 4000U, 4000H and 4000HS at CES, but at that time without any technical details.

Benchmarks were planned, then the virus came

Originally there should have been independent tests today, in February BitcoinMinersHashrate was promised a sample from Lenovo. However, the corona virus caused delays in production in Asia, which is why, contrary to planning, there are no goods today – and therefore, according to AMD, there is still no sample. The pre-order phase for new notebooks starts today.

The information published today was for selected journalists in February. Four weeks ago, AMD invited them to the company’s headquarters in Austin, Texas, and the Renoir Mobile Tech Day held. This alone shows the importance of the product launch in the mobile segment this year, because such events are usually only available for the most important new products.

Renoir’s development began before Zen launched

Right at the start, AMD surprised with a statement: The developments at Renoir, the code name of the APU, began before the market launch of the underlying processor architecture Zen (Ryzen 1000) and of course also before Zen 2 (Ryzen 3000). The original plans therefore did not include any experience that AMD had been able to collect with Zen in the mass market since February 2017. But in this case, too, the manufacturer relied on simulations that Renoir found possible with Zen 2.

This makes it clear once again that a new design of processors and graphics cards can still take around four years. And that, although AMD relies on a well-known approach at Renoir: The processor and graphics solution are packed together with the memory controller and I / O elements on one die. This has existed since Llano in 2011. An overview from AMD shows the long way to date.

Renoir has the complete notebook package in view

A notebook is more than just a CPU and a graphics unit, all parts such as display, battery, housing, cooling and more must work together to enable a coherent device. With Renoir, AMD wants to go this route even more clearly than last. In the past, AMD notebooks were often good in terms of CPU and GPU performance, but the surrounding area was anything but suitable. In the end, the devices were often sold out, and the reputation of an AMD notebook suffered greatly.

Recently, AMD has been more concerned about the ecosystem. Flagship notebooks like the Surface from Microsoft were a start from the beginning, now it should also go into the mass market and use Renoir, the best mobile platform from AMD so far. The development of the entire IT industry helps. SSDs have long been standard, the prices of LPDDR4 have fallen sharply, 1-watt displays are widely available. The importance of these technologies in the notebook is great, coupled with the optimizations and adjustments to the processor and graphics results in a significantly rounder overall package. AMD calls the status quo up to 3.4 times more efficient than Kaveri in 2015. A catch: Picasso was already 2.9 times more efficient, so the increase from Renoir to Picasso is only as big in terms of metrics as the years before.

In any case, Ryzen 4000 will not be able to be used immediately and everywhere. AMD is cautiously expecting a little more than a dozen solutions with the new chips. A total of 170 products are to rely on AMD in the OEM environment, more than 100 of them on Ryzen. Older AMD solutions will continue to be used.

Identical Zen-2 cores, however, with only a quarter L3 cache

The underlying processor architecture for Renoir is Zen 2, which made its debut with Matisse in July 2019 in the form of the Ryzen 3000 and is also used for Rzyen Threadripper 3000 and Epyc 2. Zen 2 represents a massive revision compared to Zen, which starts with elementary things like jump prediction, continues with larger and faster buffers and caches and continues in the “UnCore area” like the memory controller and the PCI Express interface. BitcoinMinersHashrate had already summarized all details about the Zen-2 architecture in June 2019 in the report “Zen-2 architecture: Ryzen 3000 are AMD’s most powerful desktop processors”.

Renoir uses the Zen-2 architecture, but there is a big difference: per CCX with four cores, there are no longer 16 MB L3 cache in the form of four 4 MB slices, for the mobile market it is only 4 MB , divided into four 1 Mbyte pieces. This saves space and energy. The impact on performance is extremely small and negligible for the market environment, but AMD did not give an exact number.

When asked why AMD chose two CCXs and therefore eight cores for Renoir, the manufacturer explained the overall achievable performance in the unchangeable power budget set from the start. The motto is: No IPC is bought at the expense of increased power consumption. Four cores with a very high clock rate were left behind in comparison to eight cores with less clock in the same TDP budget, which is why AMD went for eight cores.

The Vega graphics are significantly more powerful and efficient

When asked AMD CEO Dr. Lisa Su at CES 2020, why the manufacturer still relied on Vega and not the successor Navi, she explained in January that it was the logical choice at the time – and RDNA was not yet finished. Vega, on the other hand, was already largely known and established, scaling the next step there. Exactly this procedure was also suitable for an APU and was implemented in the same way by AMD.

The new Vega graphics solution at Renoir was scaled in two areas. The 7 nm production allowed the individual parts that make up a GPU to move much closer together, which automatically reduced the leakage currents. This in turn allowed for a significantly increased clock margin, which is why AMD even opted for 20 percent fewer CUs on a maximum of eight units and still offer more performance than the previous generation. The manufacturer promises up to 59 percent more power per CU compared to previous 12 nm Vega solutions.

But in addition to the pure GPU, AMD has also worked on optimizations that affect support for multimedia content. The encoder should work up to 31 percent faster, VP9 support is given.

All of this results in a GPU that offers almost 1.8 TFLOPS graphics performance in maximum configuration and can also use a memory interface that achieves 77 percent more bandwidth. Graduations in the GPU – both in terms of the number of CUs and the clock – clearly limit the performance. The table on the second page shows which solution is used for which model.

More and faster storage

The already briefly mentioned memory is also undergoing a further innovation. AMD’s APUs have always relied on well-functioning and above all high-clocking memory, otherwise bottle necks would quickly occur, especially when using the graphics. In recent generations, this was no longer as serious as with Kaveri and similar subjects, but AMD wanted to give the new solutions the best option this year.

This means not only the support of DDR4-3200, as can also be found in the desktop with Zen 2, but above all of LPDDR4-4266 as memory, which is predestined for use in the mobile market environment. This benefits the U series in particular, which previously only used DDR4-2400, which ensured a bandwidth of 38.4 GB per second on a dual-channel memory interface. Renoir can now provide a maximum of 68.3 GB of bandwidth per second with LPDDR4-4266 – that’s 78 percent more. The other option with classic SDRAM offers 51.2 GByte / s when using DDR4-3200.

The latest I / O (without PCIe 4 and Thunderbolt 3)

Renoir offers slightly more equipment than the previous generation. The bottom line is that there are four additional PCI Express lanes that can be used freely by the customer, for example, for additional mass storage space. Renoir has a total of 20 PCIe lanes according to the Gen 3 standard.

The PCIe lanes are not required for additional USB connections, because Renoir natively has two additional ports here, for example, two 4K displays can also be operated via native USB-C.

As confirmed by AMD at CES 2020, the platform will not yet have a fourth-generation PCI Express. This is reserved for the “real” CPUs; the possibilities would hardly be exploited in the notebook anyway. The company should save this step for the next generation.

AMD also commented on the topic of Thunderbolt. The manufacturer would like to use this, but the license-free solution from Intel is not so free in the end. This simply means that Intel prohibits AMD from installing Thunderbolt chips in the devices. Partners like Lenovo will still do it.

The complete package holds almost 10 billion transistors

If all the individual parts are packed together, Renoir presents itself as follows: Manufactured in 7 nm at TSMC, 13 layers are stacked on top of each other and a total of 9.8 billion transistors are installed. The die uses an area of ​​156 mm² and thus offers almost twice the amount of transistors compared to its predecessor Picasso on an even 25 percent smaller die. In order not to overstress the ecosystem, AMD continues to use the same BGA socket with the dimensions 25 × 35 × 1.38 mm.

A word about the security of the APU

With the introduction of Zen 2, AMD had already made adjustments to the hardware for protection against Specter, which require less power than pure software solutions or subsequently patched microcodes.

On the next page: battery life, SmartShift, models and first impressions