Monday, April 24, 2017

XDA Takes on Ryzen: In-Depth Look of AMD’s New Processors On the Linux Side

Index

In 2016 readers wanted to know how AMD's FX processors performed when it came to Linux and Android building – and thanks to the folks at AMD we were happy to find out and share the results with you. The lessons we learned from that article helped us learn how to better configure our systems and realize build times far better – our FX-8350 build times dropped by over 50% thanks to your tips! But even then we reached a wall. AMD's performance, while better than expected, felt like it had room to improve in comparison to its competition in Intel.

For some time we had been told by AMD that something new was on the horizon: a processor that offered performance improvements of 40% or more versus the current lineup. Now with the Ryzen lineup making its way out to the public we're also getting our hands on these processors. Do they perform as well as AMD has been claiming they would? It's time to find out.

Click Here to Return to the Index (Top)

Background

Advanced Micro Devices has been one of the few competitors to Intel in the x86 realm in no small part due to patent licensing. I still recall times in the late 90s of having AMD, Intel and even a few classmates with Cyrix CPUs. Unfortunately by the early 2000s Cyrix as a brand had disappeared from the scene. The rivalry between AMD and Intel lent itself to healthy competition and innovation throughout the first years of the new millennium. AMD disrupted the market by introducing the x86-64 architecture to mainstream CPUs in 2004. It was something that would take several more years to mature and kept the competition strong. Both also innovated beyond the frequency barriers by multi-core architectures in the same time frame.

As Intel dialed in their performance with multi-core simultaneous multi-threading (branded Hyper-Threading) AMD began to struggle to keep up with its rival. In 2011 the Bulldozer architecture was supposed to help close this gap and bring AMD to a more competitive stance against Intel. And while Bulldozer and its relative Piledriver did offer improvements, it failed to get AMD to where they wanted to go. This began the long process of a "bottom up" redesign of their processors, their implementation of simultaneous multi-threading which became the Zen architecture and Ryzen CPU lineup. Given that they didn't offer many updates to their AM3+ lineup for several years, Ryzen offered great risk with the chance of great reward.

Over the past year they've been slowly detailing the technical details, which Anandtech has been very helpful in explaining this beyond AMD's own technical slides and presentations. Just as the Polaris architecture proved competitive abilities in the graphics technology scene against NVIDIA the goal was simple with Ryzen: Make sure they accomplish what they sent out to do with Bulldozer and become a real contender against Intel once again. Many analysts and partners had even gone as far as taking a bearish stance in advance with Ryzen. After the experiences with Excavator and Piledriver they expected it to do about as well as its predecessor. Now it's time to let the guesses fall to the wayside and the benchmarks continue the rest of the tale.

Click Here to Return to the Index (Top)

Everything AMD sent to put Ryzen to the Test

Unboxing

If it looks like this is a lot of stuff, it is. AMD actually sent this all in multiple batches. The reasoning behind that was to offer the samples for testing in time for the availability of Ryzen 7 and Ryzen 5 into the marketplace. Most of those reviews though are performed in a Windows environment. But if we want to test building Android from source we're still looking at Linux, and so this is where the focus of our review will be. It has also been an interesting journey that taught a very valuable lesson in the end.

Gigabyte AM4 Motherboards and Linux

This unfortunately caused issues with both of the pictured Gigabyte motherboards. Both refused to boot into any version of Ubuntu without using the noacpi flag. Since that takes away many of the enhancements of Ryzen – including simultaneous multi-threading – we started digging for answers. Through a collaboration of other reviewers, Phoronix founder Michael Larabel and a few XDA readers who had run out and purchased Ryzen on day one, we were able to confirm that this is an issue specific to Gigabyte. MSI boards and ASUS boards that have been released do not appear to have these issues. So after reaching back out to AMD we received the MSI X370 XPower Gaming Titanium and ended up using this to test everything as it arrived.

Since we didn't get to test the Gigabyte boards out in Linux we did put them to brief use in Windows. If you're not worried about Linux, seemed to offer the highest stable overclocks of the three boards, achieving 3.9 GHz with the Geil RAM reaching its full clock speed of 3200 MHz – and staying cool thanks to the Wraith Spire cooler, one of the two new LED fan cooled solutions released by AMD with Ryzen. Gigabyte's X370 board also offered Sound Blaster MB5 support for a better surround experience. Both also featured some great LED lighting features with additional headers to connect peripherals to match the lighting pattern of the motherboard.

Our thanks go to the collaborative efforts of many who helped troubleshoot the issue in a quick manner. AMD also deserves some thanks for its quick response once the issue was confirmed. If anything this reinforces the fact that more reviews of PC hardware are sorely needed on day one for Linux users. We also plan to review and inform users should the situation improve with Gigabyte and Linux.

Click Here to Return to the Index (Top)

Photos

Click Here to Return to the Index (Top)

Testing Specifications

While the test bench survived a treacherous journey halfway across the world, the AMD Radeon HD6450 was not the graphics card we wanted to test with. So out came the Radeon and in went the best graphics card in the available arsenal – a NVIDIA GeForce GTX 1080. The reason for this was simple: While AMD knows that it has quite a fanbase for its own graphics cards, it also knows that many others will go with NVIDIA.

Since the lowest Ryzen processor is a 4 core, 8 thread CPU I also performed a benchmark using my personal PC. I have listed the specifications and testing methodologies below.

FX-8350 Testing Platform

AMD Ryzen (AM4) Testing Platform

Intel Core i7-6700K Testing Platform

Note: Items not mentioned as provided by a third party were purchased on own. In addition, additional drives were physically installed in each platform but were not used for testing purposes.

Software

  • Ubuntu 17.04 with all current updates.
    • Ryzen Note: Ryzen was tested with an updated 4.10 kernel and 4.11 kernel as it has code implementation for Ryzen's simultaneous multi-threading. Users of AMD and NVIDIA GPUs will want to steer clear of the 4.11 kernel for now as both proprietary packages currently do not have code merged to easily install and run the proprietary packages.
  • Phoronix Test Suite v7.0
  • Unigine Heaven v4.0
  • Unigine Valley v1.0
Click Here to Return to the Index (Top)

Testing Methodology

In order to try and minimize the number of test variables, all RAM was kept at 2133 MHz for the testing. The exception to this was the FX-8350 and AM3+ platform as its maximum was 1600 MHz. CPU speeds were overclocked for the 1800X, 1700X to 3.9 GHz and 1600X to 4.025 GHz to come as close as possible (or, in the case of 1600X, an extra 25 MHz to test viability) to both the FX-8350 and i7-6700K stock speeds of 4.0 GHz. The 1700 and 1500X were attempted to overclock to these speeds but would not remain stable. As a result they were tested at their highest stable speeds: 3.7 GHz for the Ryzen 7 1700 and 3.8 GHz for the Ryzen 5 1500X. For Intel a brief test of the i-6700K at 4.0 GHz and an easily achieved overclock of 4.2 GHz based on build times yielded slightly better numbers for Intel but not enough to change the findings of this review (one build was faster, another was not.) I decided to run the Phoronix Test Suite at 4.2 based on these findings. All overclock adjustments mentioned are by clock multiplier only.

Android build times are an average of 3 builds of Lineage OS branch cm-14.1 for the Nexus 6P (angler). All builds were done using ccache and make 3.82 based on previous findings. It should be noted that we also did some tests with make 4.1 and did not notice the performance issues with the current build process that we did during our FX processor testing. As the times of our sample tests matched those of 3.82 they were not included in the test.

Click Here to Return to the Index (Top)

Benchmarks

Benchmark Notes: Phoronix Test Suite's CPU suite offers a plethora of tests and not all will be included in this review. The full test results from the Phoronix Test Suite are available on Openbenchmarking.org site. I am adding a link here for the comparison of all results.

FFTW v3.3.4

FFTW is a single-threaded benchmark of fast Fourier transform. Much like we have seen in the Windows benchmarks, Ryzen's single-threaded performance seems to be similar across all variants with frequency being the main difference. It's a significant improvement over the previous generation FX-8350 – over double the performance – but still struggles even against the Skylake i7.

During the press briefing AMD admits that Ryzen "wins some and loses some" and here we can see why. While in Windows many of the tests stayed within 15% of Intel, AMD's Ryzen processor seems to be over 20% less performing than the Skylake i7-6700K.

GMP Benchmark v0.2

GMP's benchmark is a composite of arithmetic operations and while multi-threaded the CPU frequency makes a large difference here. Again, we see that even the Ryzen 5 1500X is a stark difference versus the FX-8350. Surprisingly our two highest were the highest clocked CPUs during our benchmark tests, showing that core and thread counts do not seem to make much of a difference over raw speed in this test.

SciMark 2 (Java) v1.1.1 Composite

The SciMark 2 benchmark utilizes Java for arithmetic operations and then provides scoring based on a composition of the various operations.  AMD's Ryzen CPUs shine brightly here. Even the 1500X, at a slower speed than its Intel equivalent, outperforms it by roughly 12 percent.

Bork File Encrypter v1.4

Bork tests encrypting files in Java, measuring how long it takes to encrypt a sample file. Here we see that AMD's Ryzen does improve against its own previous architecture, but that Intel clearly wins this round.

CacheBench

Cachebench is a tool to help measure performance of the underlying cache for a given architecture. It's noteworthy because at face value it seems the additional L1 cache helps give AMD an edge here, even with the older FX-8350. But this alone doesn't tell the entire story; we see in other benchmarks that doesn't always translate into real-world wins across the board.

John The Ripper

For readers who wanted to know about cryptography performance, John The Ripper has some good news for you. The multi-threaded nature of this benchmark allows it to tap well into the additional cores and threads for maximum performance. Ryzen's additional cores certainly make it look attractive here – and even more so when considering cost as an additional factor versus Intel.

GraphicsMagick v1.3.19 Sharpen

GraphicsMagick is one of the more robust open source image tools in the market; it's one I have actually even found use for as an IBM i RPG programmer. The sharpen results here do seem to be skewed slightly against the Intel CPU and I have to believe that this is in part due to the GPU difference. I'm including it here because it's noteworthy to compare the results of the 1600X, 1700, 1700X and 1800X together. The pricing for these 4 processors is different and shows that in some cases the additional horsepower may not offer too much more in the way of benefit.

C-Ray v1.1

C-Ray is another example of this same theme: The jump from 8 threads to 12 makes a dramatic improvement and well justifies the 1600X in its position. But those returns diminish between the 6-core Ryzen 5 1600X and its Ryzen 7 counterparts.

7-Zip v9.20.1

The 7-Zip test is for compressing a file but more clearly defines the difference between the four, six and eight core variants of Ryzen. What is a bit disappointing is to see the 1500X performance less than its predecessor the FX-8350. Thankfully this can be explained by the 200 MHz clock difference between them. The PBZip benchmark follows a similar trend.

GZip 1.1.0

Not all compression routines do better with Ryzen. Take for example GZip, which saw the Intel i7-6700K outperform them all. A similar result was found with the LZMA benchmark. In both cases the differences between all Ryzen models are marginal. When we realize that these are both typically single-threaded operations we can then make sense of this as it matches the trend of single-threaded performance in other benchmarks.

Click Here to Return to the Index (Top)

Benchmarks: Build Performance

This last page of benchmarks focuses on builds – including Android builds on some (but not all) cases. We also provide 2 graphics benchmarks to help get an idea of how Ryzen will affect overall graphics performance.

Build Test: ImageMagick

While the i7-6700K does better than all of Ryzen the margin easily falls within the 200 to 500 MHz clock speed differences. Here the 1500X compile time decreases by 25% from the FX-8350, but we see diminishing returns as we cross into the 6 and 8 core thresholds.

Build Test: Linux Kernel 4.9

When we get to the Linux Kernel the above case no longer applies and the additional cores here get a chance to flex their muscles. We also see a more clear definition in regards to performance between the four, six and eight core models. CPU clock speed also makes a difference in this build test.

Build Time: LineageOS cm-14.1 ccache

As we have noticed in the benchmarks the Ryzen 7 processors all seem to behave similarly when working on the same clock speed. So with that we put the best of each to the test for our LineageOS build. The FX-8350 build time has increased with the cm-14.1 branch and is typical of each newer version of Android. We see a tremendous improvement in this with the 1500X, but as we cross into the 6 core threshold the CPU bottleneck is removed. Times improve with the 8 core but the returns quickly diminish and could probably be better invested in I/O performance, just like we observed with the RAID5 array and Xeon counterpart to our i7-6700K.

Graphical Benchmark: Unigine Heaven 4.0

Unigine's graphics benchmarks are a well-known benchmark often used in reviews. We want to limit the variables in the build configuration in order to get a better idea of how much Ryzen improves over the FX-8350. Then by changing only the CPU we can get an idea if the cores offer an additional boost. And in the case of Heaven we see a significant improvement from FX-8350 to the 1500X, then a slight bump as we add additional cores and threads to the mix.

Graphical Benchmark: Unigine Valley 1.0

Valley is a newer benchmark and again we see an improvement from the FX-8350 to the Ryzen 5 1500X. But once we cross this threshold it seems to be no longer about additional cores and threads but again about raw speed. Our 1800X sees an improvement over the 1500X with an additional 100 MHz but is beaten by the 1600X running at 4.025 GHz.

Click Here to Return to the Index (Top)

Ryzen and Coreboot

Shortly after the release of Ryzen AMD took to Reddit in an AMA to answer questions about its new product. During that discussion came up the request from several Linux and open source users requesting that AMD release its Platform Security Processor (PSP) code to the open source community. The reason for this is to enable support of options such as Coreboot and Libreboot, among others. Even Edward Snowden chimed in on the matter via Twitter:

AMD's Product Manager James Prior replied during the AMA that they are investigating this and provided a further update earlier this month regarding the status of the investigation. We have previously offered support for AMD as they released GPUOpen to the community and hope they will take another big step forward by supporting this request. We will follow this and provide additional updates as they are available.

Click Here to Return to the Index (Top)

Final Thoughts

There's a few ways to think about Ryzen and how it fits into the new ecosystem. The first, which many have discussed, is how it fares against Intel in a core/thread matchup. There is a definite improvement from the FX-8350 to the Ryzen 5 1500X. If looking at performance alone we don't see it always keeping up with the Intel Core i7-6700K. But AMD's strategy has never been about raw performance alone. Instead, AMD's business strategy has largely been to offer the best value at the price point at which it offers its products. They have continually followed this overall strategy in both its processor and graphics technology. This becomes even more important at the moment when considering the overall costs of the PC.

When wanting to overclock an Intel motherboard currently the choice is generally limited to the Z series of their lineup, though some have found ways around this barrier. AMD has taken an alternate approach, offering 3 motherboard chip sets that enable overclocking and then enabling multiplier overclocking on every AM4 processor. This offers tremendous selection at lower price points. The challenge that I found with offering all of the processors is that once XFR is no longer a factor, the processor models all behave similarly.

From there it gets down to how well you can overclock that particular processor. You may hear of this from time to time as the "silicon lottery" because not every processor will overclock as well as others. For example, I was able to either increase the multiplier or run the RAM at its XMP profile speed on all 3 motherboards. Recent BIOS updates have helped improve the matter, but it doesn't resolve it. Take for example the Ryzen 7 1700. Higher clock speeds would boot, such as 3.9 GHz, but would end up failing for reasons other than overheating.

The choices are very clear when we look at stock speeds. AMD now offers an alternative to Intel's extreme consumer lineup at half the cost of its competitor. As we move into overclocking the same goals can be achieved with even a Ryzen 7 1700, so long as the processor will allow. The cost savings of 66% less than Intel's i7-5960X or i7-6900X are enough to even buy a very high end graphics card or additional SSD storage. The same case can be made for the Ryzen 1600. It is priced very competitively against the Core i5 and it offer an experience beyond many consumer uses of the Core i7. Consumers who are looking at getting the most value out of Ryzen should not immediately dismiss these "non-X" options. That cannot be emphasized enough if they intend to overclock their PCs.

There are cases though, at the moment, where Intel can still justify its price point. One of those cases comes in PCI Express lanes, which Broadwell-E and the upcoming Skylake-E offer additional PCI-E lanes beyond what AMD does. This means in some high-end usage there isn't an alternative – yet. AMD is expected to announce an X390 motherboard and Naples, the enterprise lineup for Ryzen. I fully expect here to see a challenge against X99 and not only the additional PCI Express lanes but also memory capacity.

AMD's own challenges in this market meant taking a very big gamble on its return to form by sacrificing many short-term sales. Sometimes this pays off even bigger – the Chicago Cubs and Cleveland Cavaliers are testaments to that in the sports world. If AMD's goal was to achieve this competitive stance, I believe they have achieved their goals. But this isn't the end of the story; change and innovation need to continue. Intel has hinted at such a change by retiring the Core branding in 2019. Could this hurt AMD's promise to stay on the AM4 platform until 2020? Only time will tell. In the meantime pass the popcorn; a competitive AMD is back to their old form and that means an exciting competition for your hard earned dollars. Who wins in this case? We, the consumers, do. And it's about darned time.

Click Here to Return to the Index (Top)

IMPORTANT: We want to know: did you enjoy this review? Would you like to see more of this coverage on XDA? Or maybe you don't think this is relevant on our site? Sound off on this and other parts of reviews in the comments below, Facebook, Twitter or Google+! We look forward to your feedback!



from xda-developers http://ift.tt/2pY0wAT
via IFTTT

Posted by at

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)