A potential game changer?

AMD is announcing the Ryzen Mobile processors with integrated Vega graphics today. Should Intel and NVIDIA be worried?

I thought we were going to be able to make it through the rest of 2017 without seeing AMD launch another family of products. But I was wrong. And that’s a good thing. Today AMD is launching the not-so-cleverly-named Ryzen Processor with Radeon Vega Graphics product line that will bring the new Zen processor architecture and Vega graphics architecture onto a single die for the ultrathin mobile notebook platforms. This is no minor move for them – just as we discussed with the AMD EPYC processor launch, this is a segment that has been utterly dominated by Intel. After all, Intel created the term Ultrabook to target these designs, and though that brand is gone, the thin and light mindset continues to this day.

The claims AMD makes about its Ryzen mobile APU (combination CPU+GPU accelerated processing unit, to use an older AMD term) are not to be made lightly. Right up front in our discussion I was told this is going to be the “world’s fastest for ultrathin” machines. Considering that AMD had previously been unable to even enter those markets with previous products, both due to some technological and business roadblocks, AMD is taking a risk by painting this launch in such a light. Thanks to its ability combine CPU and GPU technology on a single die though, AMD has some flexibility today that simply did not have access to previously.

From the days that AMD first announced the acquisition of ATI graphics, the company has touted the long-term benefits of owning both a high-performance processor and graphics division. By combining the architectures on a single die, they could become greater than the sum of the parts, leveraging new software directions and the oft-discussed HSA (heterogenous systems architecture) that AMD helped create a foundation for. Though the first rounds of APUs were able to hit modest sales, the truth was that AMD’s advantage over Intel’s on the graphics technology front was often overshadowed by the performance and power efficiency advantages that Intel held on the CPU front.

But with the introduction of the first products based on Zen earlier this year, AMD has finally made good on the promises of catching up to Intel in many of the areas where it matters the most. The new from-the-ground-up design resulted in greater than 50% IPC gains, improved area efficiency compared to Intel’s latest Kaby Lake core design, and enormous gains in power efficiency compared to the previous CPU designs. When looking at the new Ryzen-based APU products with Vega built-in, AMD claims that they tower over the 7th generation APUs with up to 200% more CPU performance, 128% more GPU performance, and 58% lower power consumption. Again, these are bold claims, but it gives AMD confidence that it can now target premium designs and form factors with a solution that will meet consumer demands.

AMD is hoping that the release of the Ryzen 7 2700U and Ryzen 5 2500U can finally help turn the tides in the ultrathin notebook market.

  Core i7-8650U Core i7-8550U Core i5-8350U Core i5-8250U Ryzen 7 2700U Ryzen 5 2500U
Architecture Kaby Lake Refresh Kaby Lake Refresh Kaby Lake Refresh Kaby Lake Refresh Zen+Vega Zen+Vega
Process Tech 14nm+ 14nm+ 14nm+ 14nm+ 14nm 14nm
Socket BGA1356 BGA1356 BGA1356 BGA1356 ? ?
Cores/Threads 4/8 4/8 4/8 4/8 4/8 4/8
Base Clock 1.9 GHz 1.8 GHz 1.7 GHz 1.6 GHz 2.2 GHz 2.0 GHz
Max Turbo Clock 4.2 GHz 4.0 GHz 3.8 GHz 3.6 GHz 3.8 GHz 3.6 GHz
Memory Tech DDR4/LPDDR3 DDR4/LPDDR3 DDR4/LPDDR3 DDR4/LPDDR3 DDR4 DDR4
Memory Speeds 2400/2133 2400/2133 2400/2133 2400/2133 2400 2400
Cache 8MB 8MB 6MB 6MB 6MB 6MB
System Bus DMI3 – 8.0 GT/s DMI3 – 8.0 GT/s DMI2 – 6.4 GT/s DMI2 – 5.0 GT/s N/A N/A
Graphics UHD Graphics 620 UHD Graphics 620 UHD Graphics 620 UHD Graphics 620 Vega (10 CUs) Vega (8 CUs)
Max Graphics Clock 1.15 GHz 1.15 GHz 1.1 GHz 1.1 GHz 1.3 GHz 1.1 GHz
TDP 15W 15W 15W 15W 12-25W
15W Nominal
12-25W
15W Nominal
MSRP $409 $409 $297 $297 ? ?

The Ryzen 7 2700U will run 200 MHz higher on the base and boost clocks for the CPU and 200 MHz higher on the peak GPU core clock. Though both systems have 4-cores and 8-threads, the GPU on the 2700U will have two additional CUs / compute units.

The TDP numbers will warrant more digging as we get hardware in hands. Both processors are listed as 12-25 watts (configurable) with a nominal TDP of 15-watts. I’m not sure what the means exactly. I can say that based on looking at the scores provided by AMD for performance, I am guessing that Cinebench runs are going to be hitting that 25 watt level for short bursts of higher clock speeds. More on that later.

There is a lot of technology that goes into create a monolithic die with a quad-core CPU design along with a 10 Compute Unit GPU based on Vega. First, the SenseMI technology that tended to be overlooked in the desktop environment where power and thermal limits are less impactful makes a bigger impact on the Ryzen mobile release. Precision Boost gets upgraded to version 2 as the algorithm to dynamically adjust the clock speed of cores within thermal limits of a notebook design. It aids the system to achieve higher single threaded performance when possible but constrains to power limits in multi-threaded situations.

This graph provided by AMD shows an example workload that scales thread count and tells us corresponding clock speeds on the various cores. It shows the single threaded test running at just over 3.6 GHz with lower clocks as more threads are loaded on to the system. At four threads, the three operating cores are running around 3.0 GHz. Fully loaded with 8 threads, the 2700U processor appears to sit firmly at 2.9 GHz.

To be quite blunt – those are extremely high clock speeds for a notebook form factor that typically demands 15-watt TDP levels. For comparison, the latest 8th Generation Intel Core processors (Kaby Lake-R) are only able to run at 2.4 GHz when running in a sustained many-thread workload like Handbrake. Without hardware to test AMD claims, we have to take them at their word, but the complexity of system design in a notebook is very different from desktop system. The question will be how LONG can the Ryzen mobile processors support running at these clocks?

Mobile XFR is another SenseMI technology that will help Ryzen in the mobile segment. This feature boosts clock speeds of the processor when thermal headroom is available, and AMD tells us that when a notebook is designed with adequate cooling, it can see as much as 23% application performance increase because of it. Leveraging the temperature awareness of the Precision Boost 2 technology, Mobile XFR will require specific AMD performance and cooling criteria to be enabled on a per-system basis.

Performance claims from AMD for this launch are likely to set Intel’s product teams back a step. Though AMD admits to a 10% deficit in Cinebench R15 single-threaded results, it claims to have 44% better performance in the multi-threaded score, and this is with a true-quad core Core i7-8550U based on the latest Intel 8th generation Kaby Lake-R. That’s not a little gap, and since we are looking at a matched core/thread count comparison, the difference HAS to indicate that the Ryzen 7 2700U is running at higher clocks than the 8550U, much higher. (We already know that from an IPC perspective, the KBL design exceed Zen.)

AMD also claims performance victories in POV-Ray, PCMark10, and TrueCrypt 7.1a and a minor loss in PassMark9. These are impressive results from AMD, though honestly, the all-core loaded performance of ultrathin notebooks is less important for this market than nearly any other. Responsiveness and battery life are the kings for productivity machines and it appears we’ll have to get some hardware in our hands before we can judge AMD’s new baby in that regard.

From a GPU perspective, it should come as no surprise to anyone that follows the market even slightly that the Ryzen APU with Vega graphics has the edge here. The Ryzen 7 2700U with 10 CUs will offer similar performance to a notebook with a discrete NVIDIA GeForce 950M GPU and 2.6x that of the Intel Core i7-8550. AMD also showed data where the Ryzen 7 2700U was able to hit 60 FPS or near it for many popular games running at 1080p: League of Legends, DOTA 2, CS:GO. For hardcore PC gamers that isn’t going to impress but for thin and light designs with Intel integrated graphics that will be a significant increase. Also, the advantage of getting drivers from a company that also sells discrete GPUs means more performance upgrades and reliability improvements on a regular basis.

This advantage for AMD is going to be key to its sales pitch. What would previously have required both an 8th generation Core processor from Intel, along with a discrete GPU like the 950M from NVIDIA, can now be done with a single chip. This COULD save money, space on the board design, and power. I say COULD because we haven’t been able to get hands on it the hardware yet, and proving out this advantage is more crucial for consumers to have confidence in the story AMD is telling us.

How AMD makes it all happen

A lot of work was done by AMD engineers to improve the power efficiency of Ryzen Mobile, starting with a two-stage voltage rail for the entire SoC. There is an off-chip stage that is part of the motherboard voltage regulation coupled with a second stage that utilizes an on-chip voltage regulator and digital low-dropout regulators. These LDOs are divided into regions for CPU cores, graphics, and other system functions. These help reduce current requirements on the chip itself while also doubling as power gates when certain portions of the SoC are not in use.

If you remember back to the Haswell design, Intel had integrated a voltage regulator on the processor as well. However, with Skylake, the design has since reverted, with Intel going back to an on-motherboard solution. At the time of Skylake’s launch Intel claimed that removing the FIVR “enabled improved power efficiency across a much broader range of thermal design powers.” We will see if AMD’s implementation has longer legs than Intel’s. AMD believes that its integration only requires a 45A maximum draw while the Intel 7th generation Core parts can pull as much as 95A through both the CPU and GPU rails.

Ryzen Mobile introduces per-core frequency and voltage scaling that is based thread utilization. Fine grained to 25 MHz increments, this allows AMD to steer power to the most demanding threads and workloads without sacrificing power efficiency and needing to increase power to all or most of the cores on the die. It also allows for power sharing between the CPU cores and the GPU – not a new idea for thermally limited mobile processors. But AMD claims that on-die regulation and frequency control from Precision Boost 2 allows Ryzen Mobile to handle this faster and more accurately than any previous integration.

With the implementation of the digital LDOs in place, AMD has integrated enhanced gate states to allow CPU cores, GPU cores, and the entire chip to reach lower power states. Each CPU core can individually enter CC6 states through gating and when all cores are idle it can drop to a CPUOFF state that allows lower power to the L3 cache. For graphics, LDO-based gating can power down as much as 95% of the GPU functionality. When it enters both GFXOFF and CPUOFF states, the entire chip can halt the voltage regulator and provide 99% residency at a static Windows screen. This should translate into much improved battery life over previous AMD APU designs.

Infinity Fabric gets more credit with this processor implementation as it allows for some interesting region to region communications and power gating configurations. AMD has built the Ryzen Mobile processor with SoC elements divided into regions, each of which can be gated and controlled independently. While this might make the power and communication diagrams look more complex, this allows AMD the flexibility to shut down as much of the chip as possible. Type A regions on the slide above can be gated during screen refreshes while Type B must be active, if only briefly.

AMD touts gate exit speed as another improvement at the hand of Infinity Fabric. Gate exiting is what allows the chip to get out of a sleep state and into an active working state and is critical to improve the experience and responsiveness of a system. Handling the telemetry from the six unique clients on the SoC is a significant task and one that AMD claims has been in the making for more than four years.

From a battery life perspective, AMD is claiming to have made a lot of progress. For VP9 video, which has been hardware accelerated on Vega, we should see more than a 2x increase compared to the AMD FX-9800P design previously. For H.264 playback and MobileMark that difference is bit more modest at 2-3 hours, but if they can become competitive with Intel’s latest offerings that is a huge win for AMD.

The overall performance story from AMD is very aggressive with Ryzen Mobile. The interior line in the chart above indicates the performance of the Intel Core i7-8550U, a quad-core KBL-R part. Claiming that the HD 620 will have only 38% of the performance of the Vega integrated graphics is a great result! There are some caveats here to note. For example, the Content Creation result is based only on POV-Ray, Productivity is based only on PCMark10, and Data Security is only based on TrueCrypt. Maybe the most troublesome is the power efficiency metric in which AMD only takes the Cinebench R15 multi-threaded score and divides by 15-watts. Because we are looking at MEASURED power, and only nominal results from AMD’s specs, I question if that result will hold up.

Most importantly for AMD, there are design wins they can talk about. HP, Lenovo, and Acer will all have systems in the channel by the end of this year, in time for the holiday season. The HP Envy X360, Lenovo Ideapad 720S, and Acer Swift 3 will all be available with Ryzen 7 2700U or Ryzen 5 2500U hardware and various other configuration options. Though these aren’t quite as thin and light as some other Intel machines just yet, they are offering high end options like NVMe SSDs, IPS screens, USB Type-C, and more. No pricing is set yet, but I am assume they will be in line with current selling Intel options. (Note that only the Lenovo machine is listed as single channel memory – possible a big detriment to GPU and gaming performance.)

I am hoping to have a system or two in my hands in the very near future powered by the Ryzen Processor with Radeon Vega Graphics. There is a steep curve for AMD to get past for acceptance into the thin and light market of notebooks, but AMD feels confident they have made the processor that can finally do it.