Reviews

Ryzen, not Zen: Is this AMD's return of performance CPUs?

The salami slices continue to pile up happily and we can finally write a little more about the upcoming CPU generation Zen today. But stop! A little bit AMD has filled the pun box office and Zen is only part of the we...

The salami slices continue to pile up happily and we can finally write a little more about the upcoming CPU generation Zen today. But stop! AmD has filled the pun box office a little bit and Zen is only part of the real name in the end. Zen, Ryzen, new Horizon? Or even Ryzen = Risen?

Lying quite identical in the ears, it would almost announce the rebirth of powerful and competitive AMD CPUs, which the attendees acknowledged with a knowing smile at the Technology Summit in Sonoma. Of course, only AMD's marketing department alone knows which molecular compounds had to be consumed in order to meander so gallantly around the corner. We can look forward to the result, because only success counts.

But let us remember a few times! It was 2011, when AMD wanted to verbally flatten everyone else with bulldozers and then it was only enough for itself. AmD's CEO, Dr. Lisa Su, then recalled in Sonoma on Thursday last week, quite aptly and, above all, self-critical[…]ly: "We believe that there has been no innovation in desktops over least the last five years." Of course, she is not wrong, because apart from a few cosmetic changes, nothing has changed in the basic problem of architecture.

So now it should and must be judged by Ryzen, a generation of US people who see AMD in the high-end desktop and server market and which will ultimately reach the mobile sector and be used, for example, in notebooks. The company has already told us that its top model would be an eight-core design with true multi-threading support (analogous to Intel's SMT), allowing 16 threads to make better use of the processor's resources.

Early leaks initially showed engineering samples (ES) at 2.8 GHz. Later, AMD showed a blender benchmark with a Summit-Ridge CPU running on three GHz, which reached a tie against an Intel Core i7-6900K, which was also limited to 3 GHz. And now, at the Technology Summit in Sonoma, we were shown a Ryzen with 3.4 GHz base clock, whose boost was not yet activated according to AMD.

This gives us hope, because in a new benchmark, this time it was the relatively good HandBrake, which scales over all threads, the same Intel CPU, which also ran out-of-the-box (3.2 GHz base, Turbo Boost up to 3.7 GHz), could even be beaten wafer-thin. A classic head-to-head race. In addition, AMD's fade-in showed a power consumption that was more than 20 watts lower, although we can't estimate how much of the difference was due to the different motherboards and their voltage transformer losses.

 

Apart from that, we would of course have liked AMD to have tested Ryzen with Skylake and/or Steamroller in other situations and in purely IPC-dominated benchmarks. Not only to clarify the IPC improvements, but also to see if the company has managed to overcome the overall weaknesses of bulldozers.

 

First technical details of Ryzen

 

At Hot Chips 28 in August this year, AMD's architecture briefing showed that each core has its private L2 cache, each with a size of 512 KB. Thus, a so-called CPU complex (CCX), which is composed of four cores each, would have a total of two MByte L2 and eight MByte L3 caches, which consist of four parts. An eight-core CPU like the one presented here consists of two CCXen, which in the end would result in a total of four MByte L2 and a whopping 16 MByte L3 cache, i.e. 20 MByte for the entire CPU.

 

This looks pleasing in direct comparison to the current competitor portfolio, as the tested Intel Core i7-6900K (Broadwell-E) has a total of only 18 MB, while the smaller consumer CPUs have a maximum of only up to 9 Mb of cache. That makes Naples, the professional version of Summit Ridge, even properly server-ready, especially since the L2 cache is twice as large as that of the current Intel CPUs.

 

The responsible employee in the name-finding department at AMD must have had a very creative run, because it was just a lot of new buzzwords and names when presenting. Under the umbrella term SenseMI, AMD is bringing together various new technologies into one package. Three of the new features are designed to optimize efficiency or maximize performance in the processor's (thermal and electrical) boundary ranges. Two other highlights are artificial intelligence, which should inherently make this architecture faster.

 

 

The first feature, Pure Power, is a modern control system capable of monitoring temperatures, clock rates and voltages in a new quality. AMD's CTO Mark Papermaster told us there were hundreds of sensors in various parts of the chip. These sensors lead the determined telemetry data to a management unit (arbitrator) that can fine-tune all necessary adjustments and make them in real time. This very comprehensive control system from Pure Power is designed to help to adjust the power consumption even faster and more optimally at every required power level. In the end, therefore, only what is really needed is to be delivered.

 

 

Conversely, Precision Boost aims to maximize the respective clock rate within this control circuit in such a way that the total power consumption is not exceeded. The algorithm for this is also new, which now enables virtually "on-the-fly" finest frequency adjustments in 25 MHz increments. In other words, you use the remaining room for manoeuvre within a current load state to keep the clock and thus the performance as high as possible in compliance with all specifications.

 

 

Since the popular sport of overclocking has not crept unnoticed by AMD, the appropriate possibilities are made available to enthusiasts on the appropriate platform to change the framework conditions of Pure Power in such a way that Precision Boost is then can be brought to higher clock rates.

 

Cooling plays a very important role in ensuring the lowest possible operating temperature of the processor. AMD also told us that the more aggressive and better cooling rates are allowed beyond the normal limits of Precision Boost. The company calls this option the Extended Frequency Range (XFR), which is also automatically activated by default.

 

Details about this technology are unfortunately rare, but we wonder if there is any defined limit for XFR at all and, if so, where it lies. It also seems to us that the demanding workloads, which would primarily benefit from higher frequencies, will also be the ones that could push temperatures up. The radiator industry will certainly thank AMD and never before has water been as valuable as it is today.

 

 

Regardless of the features mentioned so far, the fourth feature of SenseMI is also intended to increase performance in general. At the moment, however, it is still unclear whether what AMD calls Neural Net Prediction is really a real innovation or is just a modified marketing term for already known improvements to (Ry)Zen's front end.

 

 

The same applies to Smart Prefetch. Mark Papermaster admitted that while he discussed (Ry)Zen's back-end hardware, he could not yet name the exact algorithms that are ultimately responsible for ensuring that data is always (only) cached when it is effectively stored are also needed. For example, Smart Prefetch is probably AMD's effort to sell as a rhetorical buzzword exactly what the new architecture is supposed to help minimize latency and pipeline utilization.

 

 

AMD also reaffirmed the interoperability between Ryzen and its new AM4 platform, which will come with DDR4 memory, PCIe 3.0, USB 3.1 Gen2, NVMe and SATA Express. However, it was still not clear how many memory channels, lanes or ports the first motherboards could have.

 

 

That would be all there would be to say about Ryzen from the Technology Summit in Sonoma, where a salami could give you a lot more slices. The inclined reader can therefore continue to be curious.

 

 

For all those who want to consume all this again as a video handy packaged and visually puffed up, we would also have a stream, which AMD has now determined to consume as soon as possible. Who wants – just click on the picture or the link below:

 

 

Link: Follow live stream

Danke für die Spende



Du fandest, der Beitrag war interessant und möchtest uns unterstützen? Klasse!

Hier erfährst Du, wie: Hier spenden.

Hier kannst Du per PayPal spenden.

About the author

Igor Wallossek

Editor-in-chief and name-giver of igor'sLAB as the content successor of Tom's Hardware Germany, whose license was returned in June 2019 in order to better meet the qualitative demands of web content and challenges of new media such as YouTube with its own channel.

Computer nerd since 1983, audio freak since 1979 and pretty much open to anything with a plug or battery for over 50 years.

Follow Igor:
YouTube Facebook Instagram Twitter

Werbung

Werbung