Admit it. You love underdog tales. The Cleveland Cavaliers coming back from a 3-1 deficit against the Golden State Warriors. The New York Giants defeating the 18-0 New England Patriots, and the Average Joes beating the heavily favored Purple Cobras in the dodgeball finals.
Well, you can now add AMD’s highly anticipated Ryzen CPU to that list of epic comebacks in history. Yes, disbeliever, AMD’s Ryzen almost—almost—lives up to the hype. What’s more, it delivers the goods at an unbeatable price: $499 for the highest-end Ryzen 7 1800X. That’s half the cost of its closest Intel competitor.
But before AMD fanboys run off to rub it into Intel fanboys’ faces, there’s a very important thing you need to know about this CPU and its puzzling Jekyll-and-Hyde performance. For some, we dare say, it might even be a dealbreaker. Read on.
What Ryzen is
We can’t get into a review of Ryzen without first recalling the tragic circumstances that came before it: AMD’s Bulldozer and Vishera CPUs sold under the FX brand. Intended as a competitive comeback to Intel’s own epic comeback chips—the Core 2 and Core i7—AMD’s FX series instead went down in flames performance-wise.
The failure of Bulldozer and Vishera left AMD languishing for years, all but abandoning the high end and nursing itself on an ARM CPU for servers. In fact, the last time AMD had a truly competitive CPU, people still listened to INXS, the Weakest Link was a thing, and George W. Bush was president. The technical term for that is: a hell of a long time ago.
Ryzen is nothing like its star-crossed predecessors, though. The FX CPUs used a technique called clustered multithreading (CMT) that shared key components of the chip; they were built on an uncompetitive 32nm, and later 28nm, process; and the 8-core versions more often than not lost to Intel’s 4-core chips.
As anyone who has wallowed in failure only to return to greatness knows, tragedy and loss only make an underdog story sweeter. With Ryzen, AMD rebooted its CPU design, tossing aside CMT. It even adopted a technique from Intel’s playbook called simultaneous multithreading (SMT), which virtualizes CPU resources.
Whereas AMD’s design once had every two cores share resources, each Ryzen core is now a distinct entity built into a four-core complex. In the shot below, two core complexes make up an 8-core Ryzen chip.
Also gone are the 32nm process of Bulldozer and the 28nm process of Vishera. Ryzen CPUs are built on a state-of-the-art 14nm process by AMD’s spun-off fab, Global Foundries. In short, the core design seems to have set the stage for an AMD return to glory.
Ryzen: Heavy-duty CPU, light-duty chipset
It should be pointed out that Ryzen’s chipset isn’t exactly heavy-duty. Because Ryzen is more of a system-on-chip (SoC) than a CPU, it contains many interface features on-chip, which are augmented by a particular AM4 socket chipset (to learn more about the various chipsets, see our guide to choosing an AM4 motherboard.) The upshot is that while Ryzen provides up to eight physical cores, the surrounding infrastructure is more consumer-oriented. Ryzen has 24 PCIe lanes total, 16 of which are dedicated to the GPU. If you go with the highest-end motherboard chipset, the X370, in order to run two video cards, that single x16 is split into two x8 connections.
The remaining PCIe lanes can be used by the motherboard maker for NVMe or other I/O options. That’s not much different from what Intel does with the current Core i7-7700K, which also has 16 lanes of PCIe for the GPU. While Intel’s highest-end consumer Z270 chipset appears to have a lot more I/O with up to 24 lanes of PCIe, a bottleneck between the consumer-grade chipset and CPU keep it from being utilized.
For most people, there is still plenty of speed on tap with Ryzen and Core i7-7700K. But for those who need insane numbers of PCIe lanes, for, say, multiple NVMe drives that will be used simultaneously, Intel’s Broadwell-E and its X99 chipset have the advantage.
Broadwell-E and Haswell-E feature up to 40 lanes of PCIe directly wired to the CPU, plus an additional eight lanes of PCIe in the X99 chipset. But, again, for the vast majority of consumers, even prosumers, that’s overkill.
How fast is it? There’s only one way to find out
AMD has already said Ryzen exceeds its goal of a 40-percent increase over previous designs—in fact, it has hit a 52-percent increase in clock-for-clock performance over the Piledriver cores. But no one cares about that. The only thing you want to know is how it does against Intel.
How we tested
For the benchmark-o-rama, I set up four separate PCs. All featured clean installs of the latest version of 64-bit Windows 10. Each of the PCs was also built using the same SSD and GPU, and the latest BIOS was used on each board.
I turned to what we believe are Ryzen’s natural competitors: Intel’s $1,089 8-core Broadwell-E Core i7-6900K; the $441 6-core Broadwell-E Core i7-6800K; and the $349 4-core Kaby Lake Core i7-7700K. And, although its well beyond its prime, I also included an 8-core AMD FX-8370, which is currently the top Vishera-based CPU you can get without wading into the crazy range (meaning AMD’s insane FX-9590 chip, which works with only a handful of motherboards due to its excessive power consumption).
For the pair of Broadwell-E processors, I tested on an Asus X99 Deluxe II board. I used an Asus Z270 Maximus IX Code for the Kaby Lake chip. I paired the Vishera with an ASRock 990FX Killer. The Ryzen CPUs were tested with an Asus Crosshair VI Hero board.
I used a Founders Edition GeForce GTX 1080 on all of the builds, and the clock speeds were checked for consistency.
RAM configuration
I opted to test each with 32GB of RAM, with the memory controllers fully loaded using standard JEDEC-speed RAM. On the Ryzen and Kaby Lake systems, that meant four DIMMs of DDR4/2133 for a total of 32GB of RAM. The Broadwell-E systems were stuffed with eight DIMMS of DDR4/2133 for a total of 32GB RAM. The FX CPU had four DDR3/1600 DIMMS in it for a total of 32GB of RAM. The Ryzen, Kaby Lake, and FX machines were in dual-channel mode, whereas the Broadwell-E box was in quad-channel mode.
Cooling configuration
One final disclosure: I expected to test all the builds using closed-loop coolers. But because AMD didn’t send me a CLC for Ryzen until late into testing, I had to test all the PCs using air cooling. AMD fans might suspect this hobbled the Ryzen parts, which have a mode called eXtended Frequency Range (XFR) that allows the chip to clock up to the capabilities of the cooler.
I should add here that XFR only really adds up to 100MHz to the chip’s speeds today. The Ryzen 7 1800X on XFR, for example, would hit 4.1GHz over its Precision Boost speed of 4GHz. The Noctua air cooler I used is itself a fairly burly cooler, and I did see XFR speeds kicking in on occasion.
Cinebench R15.037 performance
First up is Maxon’s Cinebench R15 benchmark. This test is based on Maxon’s Cinema4D rendering engine. It’s heavily multithreaded, and the more cores you have, the more performance you get. AMD has been showing off this benchmark for a couple of weeks now, with AMD’s 8-core Ryzen 7 1800X exceeding Intel’s 8-core Broadwell-E chip.
My own tests don’t quite match AMD’s results. First, my Core i7-6900K scores are slightly faster than AMD’s. AMD’s own tests, in fact, showed the midrange Ryzen 7 1700X matching Intel’s mighty 8-core. While I didn’t have time to test the Ryzen 7 1700X, I did test the lower-wattage and lower-priced Ryzen 7 1700.
The insanity? Not only do we have a $500 Ryzen besting an Intel chip that costs twice as much, the $329 Ryzen comes pretty damn close. This test is all kinds of win for AMD—and Intel fans, it’s now independently confirmed. Damn.
Cinbench R15 also features a single-threaded test that only loads up a single CPU core. Although the results here are quite good for AMD, the winner is the Core i7-7700K chip. We can attribute that to the higher clock speed of the Intel CPU (4.2GHz to 4.5GHz) and the greater efficiency of Intel’s newest core at work. Still, Ryzen can stand tall against Intel’s best and brightest.
My final Cinebench test involved an update that came out in December, at the request of AMD, Maxon officials told PCWorld. When asked what changed between Cinebench R15.037 and Cinebench R15.038, Maxon declined details, saying they were “proprietary” to AMD. Strangely, AMD officials told me they didn’t even know there was a change. In fact, the company’s recent demonstrations were done on the older version.
Still, when a benchmark is changed to correct something for a CPU, alarms get set off and conspiracy theories get spawned, so I also ran Cinebench 15.038 on all the CPUs. The result? Basically nothing changed for all CPUs involved. Cinebench, of course, was one of the benchmarking applications named in the Federal Trade Commission's suit against Intel for allegedly cooking tests to hurt AMD CPUs. Intel ultimately settled, but Cinebench has long been blamed by AMD fans for being crooked. After seeing these results, it appears most of the blame is misplaced.
Blender 2.78a performance
The Ryzen hype train kicked off last year when AMD showed its chip equaled Intel’s 8-core in the open-source Blender render engine. It’s a popular application used in many indie films for effects. You’d expect Ryzen to perform well here, and it does. Although I'm seeing the Ryzen 7 1800X run just a tad behind the Core i7-6900K, it’s not enough to justify spending twice as much, is it? And that’s basically all win for AMD.
The thing is, at CES the media heard that maybe, just maybe, the issue was a bug in Blender that had been documented for some time. Birds whispered that Ryzen might not perform as well once the bug was ironed out. Fortunately, the bug was ironed out and a new version of Blender was compiled and released. To see if the birds were right, I also ran the CPUs on Blender 2.78b and saw a significant increase in performance for the Intel CPUs—but the AMD FX and the Ryzen also benefited. I left the axis scale the same so you could see the improvement.
In the end, though, it didn’t matter, and when you factor in price, Ryzen is all over this one too.
POV-Ray performance
Our last rendering test uses the POV-Ray Raytracer. This open-source benchmark dates back to the days of the Amiga and, like the two previous tests, loves cores. First up is the multithreading performance test using the built-in benchmark scene. The result is presented as pixels per second and the higher the number, the faster the render.
It is, again, all good news for AMD, as the Ryzen 7 1800X is slightly faster than the Intel CPU that costs twice as much. Even better, peep that Ryzen 7 1700, which is almost a third the cost of the Core i7-6900K chip. That’s an ouch in every single way you can measure it.
POV-Ray also includes a single-threaded test that shows the Broadwell-E chips pulling slightly ahead of the Ryzen 7 1800X. The fastest, no surprise, is the newest Core i7-7700K CPU, which runs at a higher clock speed and features Intel’s newest core. The truth is, though, you won’t be running it in single-threaded mode when doing a render, you’ll be using it in multithreaded mode.
But wait, there’s more. Like Cinebench and Blender, POV-Ray has also been recently updated. The version is in beta but available for testing. I downloaded and ran the POV-Ray 3.7.1 beta 3 and gave it a go on all of the machines. Unlike Cinebench and Blender, whose updates don’t move the needle at all, I saw a sudden swing here for the Broadwell-E chips. While version 3.7 had the Ryzen chips ahead, with 3.7.1 beta 3 they were suddenly behind: The Ryzen 7 1800X was slower than the Core i7-6900K, while the Ryzen 7 1700 came in awfully close to the 6-core Core i7-6800K.
What gives? I spoke with POV-Ray coordinator Chris Cason, who told me the current beta adds AVX2 support, and while 3.7 was compiled with Microsoft’s Visual Studio 2010, the beta version is compiled with Visual Studio 2015.
POV-Ray was fingered by Extremetech.com’s Joel Hruska a few years ago for appearing to favor Intel. POV-Ray officials, though, have long denied it. Cason told PCWorld: “We don’t care what hardware people use, we just want our code to run fast. We don’t have a stake in either camp—in fact, for the past few years, I’ve been running exclusively AMD in my office. I’m answering this email from my dev system, which has an AMD FX-8320.”
So conspiracy theorists, go at it.
Handbrake performance
Moving on from 3D rendering tests, my next test is the popular Handbrake test. This is a wonderful free encoder that’s handy for converting video and is heavily multithreaded. It also has support for Intel’s QuickSync, but my tests today will stick with the standard CPU test. For that I take a 30GB 1080p MKV file and convert it using the Android preset. This version is older than what’s available now, but still a relevant test for encoding performance. And yes, Intel fans, enabling QuickSync on the Core i7-7700K (by turning on the integrated graphics in your board’s BIOS) would let the Kaby Lake chip wail on all others here.
But, if you use your CPU for encoding with Handbrake, and not the graphics chip, here’s what you’d get: a Ryzen-colored win with unicorns and rainbows. Sure, the Ryzen 7 1800X is only a tad faster, but half the price amirite?
Adobe Premiere CC 2017 performance
Here’s why I think the Handbrake encode is pretty valid: It’s pretty close to the results using Adobe Premiere Creative Cloud 2017. For this project, I take a real-world 4K video shot by our video team on a Sony A7S and encode it using the 1080p Blu-ray preset in the Premiere. For the Premiere test, I actually put the project on a Plextor M8Pe drive to ensure the drive speed of the SATA SSD I used wasn’t a bottleneck.
Because I’m changing the resolution of the video, I use the “Maximum Render Quality” option. For this encode, I use the Mercury software engine in Premiere, which hits the CPU cores rather than the GPU. This may seem unrealistic, but driver drift for the GPU can impact render quality, and some still feel the CPU encodes look better. Either way, it’s better to have more cores. The Ryzen 7 1800X has a bit more speed than the Intel chip that costs twice as much. The Ryzen 7 1700 that costs about a third of the Intel chip is looking pretty, too. I mean damn.
But wait, video editors are tsk-tsk-ing the results because everyone uses the GPU for the render today. That’s a good point, so I rendered the project using the Mercury Playback engine using Nvidia’s CUDA. GPU rendering does indeed take off a nice chunk of time from the render. Keep in mind, the video (which you can watch here) for this test is 4K, but one minute, 43 seconds long. If this were a one-hour video, you’d throw money at the CPU to cut down your render times. The question is, which CPU would you throw the money at? The one that basically gets you a “free” $500 GPU, or the one that doesn’t?
0 comments:
Post a Comment