AMD’s CPU division has had a heck of a few months. Its Ryzen 5 and Ryzen 7 CPUs established new levels of bang for the multithreaded buck. The Ryzen Threadripper 1920X and Ryzen Threadripper 1950X put the company back in the high-end desktop game for the first time in many years. Benchmarks suggest that Epyc server CPUs present compelling values versus the Xeon Scalable Processor family, too.
Those processors are all impressive enough, but AMD’s most important Ryzen CPUs will likely be those with an integrated graphics processor on board. A chip with integrated graphics is critical to addressing the needs of the vast majority of PC users at home, in the office, and on the go. For most users, the integrated graphics processors in Intel’s CPUs offer more than enough pixel-pushing power, and those IGPs drive the vast majority of displays out there. An integrated graphics processor is also essential in the thin-and-light notebooks that make up the bulk of PC sales these days.
Try as it might have, AMD’s past APUs with Radeon graphics on board didn’t set buyers’ hearts on fire. Consequently, and much as it has with Xeons in the data center, Intel has long enjoyed a vast market-share advantage over AMD in the attractive mobile market thanks to its finely tuned process technology, architectural advantages, and highly scalable CPU designs.
The Raven Ridge APU die. Source: AMD
Today, though, AMD is putting the final puzzle piece in place for a fully competitive Ryzen CPU lineup with its Ryzen 5 2500U and Ryzen 7 2700U mobile APUs, both of which use the system-on-chip formerly code-named Raven Ridge. Although the official series name for these parts—Ryzen CPU with Radeon Vega Graphics—is a tad unwieldy, AMD’s latest attempt to fuse CPU and graphics processing power seems more elegant and compelling than it’s ever been.
See, APUs have always boasted far better integrated graphics performance than Intel’s CPUs, but that graphics power was paired with CPU cores that trailed Intel’s in the kinds of single-threaded workloads that make up the bulk of most regular folks’ computing tasks. Outside of a couple niche cases like HTPCs or truly bargain-basement gaming builds, past APUs never really found a natural home in our recommendations for system builders thanks to that unbalanced performance.
A block diagram of the Zen core
The Zen CPU architecture seems poised to change all of that. As the past few months have shown, AMD finally has a CPU core that’s competitive with Intel’s latest and greatest, even if it can’t quite match Skylake and its derivatives clock-for-clock and in every workload. The Haswell or Broadwell-class performance that Zen provides has proven sufficient for many PC users, and folks with even older chips have long been digging in their heels. We also know that Zen is quite power-efficient, since Ryzen CPUs have had no trouble keeping up with Intel’s latest on that front in our estimations.
The Vega 10 GPU
AMD’s Vega graphics architecture may also mark an important step forward for the Radeon graphics on APUs. Although the vaunted High-Bandwidth Cache Controller has been replaced with a custom memory controller for Raven Ridge APUs, other Vega architectural features could still make Radeon graphics a better fit for APUs than ever. For example, the Draw Stream Binning Rasterizer, a form of hybrid tiled rendering, helps to reduce pressure on off-chip memory bandwidth by keeping the data associated with each tile (or bin) in on-chip cache and minimizing data movement between on-chip and remote memory. In a relatively bandwidth-rich implementation like that of the Radeon RX Vega discrete graphics card, the DSBR’s economizations might not be keenly felt, but I’d be willing to bet that they’ll be useful in the bandwidth-restricted confines typical of Ryzen APUs.
Outside of its core CPU and GPU IP, AMD’s design teams are working with a more consistent set of building blocks than they have been over the past few years, too. Although the company didn’t talk much about this fact in its presentation, it’s worth noting that the Zen and Vega architectures have been tuned for and produced on the same 14-nm LPP process from GlobalFoundries from the get-go. That common production process would seem to make integrating both parts on a single die much easier than it may have been in the past, where AMD apparently had to prioritize performance on some parts of the chip (e.g. Kaveri’s integrated GPU) at the expense of peak CPU frequencies.
That 14-nm process could also provide better performance scaling as TDPs for Ryzen APUs climb well past the 15W envelope that AMD is initially targeting, too, an important consideration for chips that will presumably drop into desktops built on the Socket AM4 platform. Carrizo APUs, for example, were designed around a 28-nm high-density logic library that was typically used for GPUs. That high-density library reduced die area and power consumption to fit those chips into their optimal 15W TDPs, but that choice resulted in less-than-ideal frequency scaling in more relaxed thermal envelopes. Recall that the only Carrizo chip offered for the desktop was the graphics-free Athlon X4 845 at 65W. That sub-optimal scaling doesn’t seem like it should be a problem with Raven Ridge.
AMD has also started using its Infinity Fabric interconnect in shipping products. The use of Infinity Fabric means the implementation inelegancies from past APUs should be a thing of the past. Recall that the Llano APU used several purpose-built interconnects to join its Radeon graphics with the Stars CPU cores of the time. Trinity and Kaveri APUs refined this approach with wider, more numerous pipes and coherent methods of accessing main memory, but those interconnects were still specifically built for use in APUs and couldn’t be reused in other AMD products. The Infinity Fabric, on the other hand, has already played a key role in the eight-core Zeppelin die of Ryzen 7, Ryzen 5, and Ryzen 3 fame, as well as the multi-chip modules for Threadripper and Epyc CPUs. it also features in the Vega 10 GPU that powers the Radeon RX Vega 56 and RX Vega 64 graphics cards.
All told, the use of the Infinity Fabric to tie together Zen CPU cores and a Vega graphics processor in an SoC fabricated on a common 14-nm process would seem to be the most complete realization so far of the “SoC-style approach” that CTO Mark Papermaster sketched out all the way back in 2012, and it may well be the best example of the revitalized AMD firing on all cylinders. Depending on how one looks at it, Ryzen APUs are the culmination of a years-long journey or the first step into a new era. Probably both. Let’s see how AMD plans to tackle Intel’s mobile-market dominance.
Getting Ryzen ready for Raven Ridge
The Ryzen family of APUs are smarter about revving up CPU clocks compared to the Precision Boost behavior in their desktop brethren, thanks to a new per-core boost mechanism called Precision Boost 2.
Much like Intel’s Turbo Boost 2.0, AMD describes this new approach as an opportunistic scheme that will extract the maximum possible boost speed from every core given thermal and workload characteristics, using the same 25-MHz granularity available from existing Ryzen chips.
As a result, the “two-core boost” and “all-core boost” figures that AMD has provided for past Ryzen CPUs won’t apply to Raven Ridge SoCs or future Ryzen CPUs. Instead, active CPU cores will gracefully throttle up or down as workloads and thermals allow.
Also thanks to the SenseMI suite of technologies in the Zen architecture, AMD’s Extended Frequency Range (XFR) technology will be coming to Ryzen APUs as Mobile XFR. Like their desktop counterparts, Raven Ridge parts will be able to boost above their specified maximum boost clocks if a given system’s cooling hardware, thermal conditions and workloads allow.
Don’t expect to see the same XFR headroom from every system, though—assuming you see it at all. AMD suggests the feature is always active, but it also noted that XFR will only be available in “premium notebooks with great cooling solutions,” and it further notes that a product will have to meet the company’s own performance criteria in order to ship with mobile XFR enabled. The end notes of AMD’s presentation suggest that one of those criteria will be the system with mobile XFR enabled could apparently handle 25W of waste heat from the CPU, so we wouldn’t expect to see it in the thinnest and lightest systems with Ryzen APUs inside.
To manage power from the host system, Ryzen APUs will take advantage of integrated low-dropout regulators (LDOs) on board the SoC itself. Voltage regulation on Ryzen desktop CPUs is performed by external circuitry on the motherboard (even though LDOs are present in desktop Zen dies), but the integrated LDOs take center stage in the move to mobile.
AMD claims the combination of a shared voltage rail and regulator from the motherboard, in tandem with the on-die LDOs, allow it to reduce current requirements from the external voltage regulator and to build smaller and lighter laptops. Because each CPU core has its own LDO, as does the IGP, those regulators can also act as fine-grained power gates to let unused elements of the chip power down when they’re not needed.
The use of the LDOs on board Raven Ridge also gives AMD full per-core frequency and voltage control for all CPU cores and the integrated GPU based on workload. If only some CPU cores are under heavy load, for example, the chip can direct power and raise clocks where needed while throttling back less-stressed functional units.
These tradeoffs can also happen across the SoC if the graphics processor is under heavy load and the CPU isn’t, according to the company. Since all this monitoring and adjustment is occurring on-die, AMD claims that adjustments happen more quickly and accurately than they otherwise might.
When the SoC is largely idle, AMD can employ a range of power-saving techniques. Each core can enter the C6 deep-sleep state when it’s idle, and if all cores enter the C6 state for a long enough period, a CPUOFF state reduces the power flowing to the L3 cache, as well. Similarly, the LDOs allow the graphics portion of the die to be power-gated when it’s not being used, and a deeper GFXOFF state reduces power to the GPU’s uncore. Finally, if both functional units enter deep-sleep states, a VDDOFF state will halt the external voltage regulator. Power-saving schemes like these have been present in past APUs, but this implementation seems to be the most integrated and refined that AMD has shipped so far.
The Infinity Fabric on Raven Ridge is subdivided into two regions whose members share distinct characteristics: those that can remain gated off during display refreshes, and those that need to become active for short periods during those events. AMD’s designers worked to separate and minimize components of the fabric that need to be able to become active during display refreshes from those that can remain gated off in order to save more power.
If deep-sleep states are important for power saving, being able to quickly respond to a user’s demand for processing power is just as critical. AMD claims that the Raven Ridge SoC is generally able to wake up gated-off components much faster than the Bristol Ridge FX-9800P could, thanks to state saving and restoring optimizations in the Zen pipeline, retention registers in the Vega GPU that reduce the need for saving operations, and a bypass for the phase locked loop clock that prevents that component from slowing the elements of the SoC during wakeup.
A comparatively easier-to-explain change is that the Raven Ridge SoC package itself is only 1.38 mm thick, down from 1.82 mm for similar chips from the Bristol Ridge generation of APUs. Every millimeter counts these days, and paring off even fractions of one will likely help AMD’s design partners put Raven Ridge APUs in chassis that one could shave with.
The Ryzen APU duo
Now that we’re sufficiently up to date with AMD’s foundations for Raven Ridge, let’s talk about the two implementations of the SoC that AMD is announcing today.
|Ryzen 7 2700U||4/8||2.2||3.8||10||1300 MHz||4MB||15W||Two channels
|Ryzen 5 2500U||2.0||3.6||8||1100 MHz|
The Ryzen 7 2700U will tap four Zen cores and eight threads running at 3.8 GHz boost and 2.2 GHz base speeds. The integrated GPU boasts 10 Radeon Vega compute units for a total of 640 stream processors. AMD calls this integrated GPU “Vega 10 Processor Graphics” in its footnotes, so expect that branding to show up in marketing materials and on spec stickers. AMD suggests a 1300 MHz boost clock for the integrated GPU, but it isn’t going into more detail about base or typical speeds today.
The Ryzen 5 2500U will offer the same four cores and eight threads as its bigger brother, but they’ll run at lower clock speeds: 3.6 GHz boost and 2 GHz base, to be exact. The Ryzen 5 chip also loses two Vega compute units in the bargain, so it could carry a Vega 8 Processor Graphics sticker. The Vega 8 integrated GPU has 512 stream processors, and AMD claims peak clock speeds of up to 1100 MHz for this IGP.
To get their quad-core layouts, both CPUs implement one Zen core complex (or CCX) provisioned with 4MB of shared L3 cache. That’s the smallest L3 cache we’ve seen on a Zen CPU so far, down even from the 8MB in the Ryzen 5 1400. Each Zen CCX can have as much as 8MB of L3, so it’ll be interesting to learn whether this choice was made for power savings, market segmentation, or some other reason. Each chip will retain 512KB of L2 cache per core, though, and they’ll be able to run with DDR4 memory at speeds up to 2400 MT/s.
Both of these chips will nominally run within 15W power envelopes, although system integrators will be able to squeeze them into TDPs as low as 12W or open up headroom to as much as 25W. Intel’s eighth-generation Core i7s can squeeze into machines dissipating as little as 10W, or as much as 25W, so AMD’s chips should offer designers about about as much flexibility.
A sneak peek at performance
No CPU announcement is complete without performance numbers, and AMD offered us a few tidbits from its internal labs to whet our appetites. As usual, I’d caution reading too much into these results. In part, that’s because the company chose an Acer Spin 5 notebook as its host for the Core i7-8550U that’s going up against the Ryzen 7 2700U in these tests. It’s a 13.3″ notebook that’s just 0.63″, or 16 mm, thick. My guess is that this laptop is likely quite thermally constrained given its compact body and small footprint.
I only bring this up because AMD included a set of test numbers for an Acer Swift 3 notebook with a Core i5-8250U in in its notes that it didn’t graph. I’m never one to let good test data go to waste, so I went to the trouble. Swift 3 notebooks come in 14″ and 15.6″ bodies, so they could offer better cooling potential than the 13″ Spin 5. With the same four cores and eight threads as the i7-8550U, the potentially better-cooled i5-8250U doesn’t prove as easy a match for the AMD side of the court. Keep these numbers in mind as we examine AMD’s internal performance figures, and note that AMD didn’t offer any details of its reference notebook for either of its SoCs. We have no idea how slim or well-cooled that machine might be.
Cinebench is first up. The company’s internal testers pitted the Ryzen 7 2700U against Intel’s second-best Core i7s from the Kaby Lake and Kaby Lake Refresh generations: the Core i7-7500U and Core i7-8550U. The Ryzen 7 2700U basically matches the Core i7-7500U in the single-threaded portion of Cinebench, and it’s about 10% short of Intel’s latest-and-greatest. That performance bodes well for the kinds of lightly-threaded workloads most people tend to spend most of their time doing on the average PC.
Zen CPUs tend to do well in Cinebench’s multi-threaded test, as well, and Ryzen APUs are no exception. For heavier-duty use, the Ryzen 7 2700U puts up an impressive Cinebench all-thread score of 719, or a whopping 44% faster core-for-core and thread-per-thread compared to the Kaby Lake Refresh part. This best-case showing for Raven Ridge suggests prodigious multithreaded performance potential for the 15W power envelope.
In the range of applications that AMD provided (but didn’t graph) data for the Core i5-8250U, however, it becomes clear why the Spin 5 and its Core i7-8550U might not be the most representative implementation of that chip. Whether in POV-Ray, PCMark 10, TrueCrypt, or PassMark 9, the Core i5-8250U in the Swift 3 beats out the Core i7-8550U in the Spin 5. In whatever reference design AMD is using for the Ryzen 7 2700U and Ryzen 5 2500U, both SoCs clearly have enough room to stretch their legs, and that might explain the rather wide gaps between the Ryzen 5 2500U, the Ryzen 7 2700U, and the Core i7-8550U in some tests. We’d be curious to see what would happen if one were to put the i7-8550U in the Swift 3. Still, the Ryzen APUs remain quite competitive with the Intel parts given these caveats.
Of course, many prospective Ryzen APU buyers will be more interested in these chips’ gaming chops than anything, and if 3DMark Time Spy is any indication, the Vega 10 processor graphics in the Ryzen 7 2700U could prove quite compelling if the price is right for notebooks built with this chip. The 2700U more than doubles the performance of Intel’s UHD Graphics 620 IGP in the Core i7-8550U, and it matches a result taken from the public 3DMark database for a machine with a Core i7-7500U and a GeForce GTX 950M inside. Of course, a scroll through 3DMark’s databases also shows a variety of similar Core i5-7500U-and-GTX-950M systems with scores well over 1000. Don’t throw out your gaming notebook just yet.
AMD didn’t provide competitive numbers for Intel’s UHD Graphics 620 IGP in real-world games, but the Ryzen 7 2700U still puts up a good showing. Acceptable performance with reasonable graphics quality in immensely popular games like League of Legends and Dota 2 at 1920×1080 would probably seal the deal for many mainstream gamers, but potentially playable performance at 1920×1080 with Counter-Strike: Global Offensive might not hurt, either.
The seemingly magical results for Overwatch are tempered by the fact that the Ryzen 7 2700U achieved them with low settings at 1280×720 and with a 79% render scale. Some might be able to tolerate those settings, but full-HD gaming this is not. Comparison data with Intel’s UHD Graphics 620 IGP would be nice to see, but if frame-time performance jives with these FPS numbers, gaming on a high-end Raven Ridge machine seems likely to prove a better experience overall than with the common UHD Graphics 620.
Although it’s not pitting Ryzen Mobile CPUs against Intel parts in the battery life department, either, AMD does claim large increases in longevity from its Ryzen APUs compared to Bristol Ridge parts. That most drastic improvement comes in VP9 video playback, where the Vega GPU is likely able to accelerate decoding that the Radeon graphics on Bristol Ridge can’t. Likely thanks to that help, the Ryzen 7 2700U delivers twice the battery life as the FX-9800P does. In H.264 playback and MobileMark 14, the gains are less drastic, but 15% to 26% better battery life is still nothing to sneeze at.
HP, Acer, and Lenovo hop on board
SoCs are no good without OEMs to build systems around them, and AMD is announcing three major design wins today from Acer, HP, and Lenovo. These machines will be Raven Ridge’s pioneers during the holiday season of this year. AMD expects a full ramp of partner designs built around Ryzen APUs in 2018.
The HP Envy X360 is a 15.6″ convertible with a 1920×1080 touch screen. HP will offer it with a Ryzen 5 2500U SoC, as much as 8GB of dual-channel DDR4-2400 RAM, and solid-state storage ranging up to 512GB or a 1TB mechanical hard drive. The Envy X360 is a hair over three-quarters of an inch thick (or 19.5 mm), and it weighs about 4.7 lb (2.15 kg).
Lenovo’s Ideapad 720S is a 13.3″ traditional notebook. It’ll come with a 1920×1080 IPS display, up to 512GB of NVMe storage, and options for Ryzen 5 2500U and Ryzen 7 2700U SoCs. In a choice that’s notable for all the wrong reasons, Lenovo will offer the machine with some amount of single-channel DDR4-2133 RAM. Given what’s likely to be a bandwidth-hungry SoC, the choice to stick with one channel of DDR4 seems unfortunate. At just half an inch thick (13.6 mm) and 2.5 lb (1.14 kg), though, the Ideapad 720S definitely delivers on AMD’s thin-and-light promise.
Acer’s Swift 3 rounds out the wins AMD has secured for the moment. This is a 15.6″ traditional notebook with a 1920×1080 IPS screen. Acer will offer it with as much as 8GB of dual-channel DDR4-2400 RAM and SSDs as large as 256GB. At 0.7″ thick (18 mm) and four pounds (1.8 kg), the Swift 3 seems like the kind of straightfoward machine that’s just right for the middle of the bell curve of PC users.
Conclusions – for now
If its internal performance numbers are to be believed, Ryzen APUs seem to have what AMD needs to take the fight to Intel in the largest and most attractive segment of consumer PCs today. The combination of Zen CPU cores and Vega graphics power in Raven Ridge seems to make for the kind of truly balanced APU we’d want to recommend as an alternative for Intel CPUs in thin-and-light notebooks. With competitive single-threaded performance, high multithreaded potential, and promising gaming performance, the Ryzen 7 2700U and Ryzen 5 2500U seem like a fine opening salvo as AMD re-enters the high-performance mobile market.
On top of these apparently solid SoCs, AMD can already count design wins from major manufacturers in its corner, and those systems mostly seem as though they’ll put the best foot forward with Ryzen APUs. We’ll need to see where pricing for these systems lands, of course, and buyers will still have to bite after years of Intel dominance in notebook PCs. Still, the final puzzle pieces seem to be falling into place for AMD to complete its x86 CPU renaissance.