LightBlog

mercredi 27 décembre 2017

XDA PC Hardware Analysis : Intel Core X and AMD Threadripper Face Off in Our Linux and Android Test Suite

Back during Computex we first got word from AMD and Intel that they were both making dramatic changes to their high-end desktop lineups. AMD finally brought Ryzen into a whole new core and thread count with Threadripper. Intel responded with the introduction of the Core X lineup and also increasing its core and thread counts at the previous generation's pricing.

These new lineups looked like it would offer great processors for enthusiasts, team build machines and those who work on custom Android builds for daily work. You have been asking us to try and find out how well this will work for your own use cases — and we've been working since then to make that happen.

While we were unable to source these directly from Intel and AMD we then went out searching to see would be willing and able to help us complete the testing. The challenges with this though are having all of those processors available as well as a stable testing environment to easily limit the variables during testing. I reached out to Shannon Robb at AVADirect as part of this. After discussing what we were looking for they were willing to help out. So we want to give a hearty thanks to both Mr. Robb and AVADirect for allowing this testing to be conducted — and hope that we might be able to collaborate again in the future!

But you're not here for the long story — we promised tests and benchmarks! So let's get to it!


Test Configurations

The components listed below are the parts that Mr. Robb used to conduct the testing and were either sourced from his personal hardware and/or AVADirect. All processors were tested at stock speeds due to the large amount of benchmarks and tests required.

Shared Platform Configuration

AMD Ryzen 7 1700X (AM4) Testing Platform

AMD Ryzen Threadripper 1950X (TR4) Testing Platform

Intel Core i7-7700K (LGA 1151) Skylake/Kaby Lake Testing Platform

Intel Core i7-8700K (LGA 1151) Coffee Lake Testing Platform

Intel Core i7-6950X (LGA 2011-3) Broadwell-E Testing Platform

Intel Core X-Series (LGA 2066) Testing Platform

Software/Operating System

  • Ubuntu 17.10
  • Phoronix Test Suite – current using apt

Testing Methodology

While our testing has gotten a lot more consistent with the Intel Kaby Lake and Coffee Lake reviews this was a unique situation where we needed to be able to coordinate our testing specifications. The steps used to conduct the testing are currently available here.  We are not quite confident in these yet to fully automate it, but we do want readers to have this information in case they would like to conduct their own testing.

We were also given the opportunity to review all test results to confirm if there were any abnormalities that arose during testing. This allowed us to help identify possible issues without having to go back to the test cases at the end.


Non-Build 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 for the tests from the Phoronix Test Suite are available on OpenBenchmarking.org.  Benchmarks are color grouped by Ryzen/Threadripper, Intel 7700K/8700K and then Intel high end desktop CPUs separately in a color scheme that more closely follows other graphs on XDA.

FFTW

FFTW is a single-threaded benchmark of fast Fourier transform. Intel's mainstream Kaby Lake and Coffee Lake do quite well here, along with Ryzen and Threadripper. The clock speeds seem to be the largest factor in performance here as many of Intel's HEDT processors are clocked lower than their mainstream counterparts.

GZip Compression

GZip is a common compression method and so it makes sense to check out the performance here. Since both the Ryzen 7 1700X and Intel Core i7-6950X fared the worst, it seems that clock speed is a factor but not the only one. Core counts also do not seem to be a sole determining factor as the i7-7700k outperforms all of the other processors and, out of this group, has the lowest core/thread count of them all.

SciMark 2 (Java) v1.3.0

The SciMark 2 benchmark utilizes Java for arithmetic operations and then provides scoring based on those results. Intel continues seeing its wins in this new version. The trend for single threaded performance continues though: More cores doesn't equate to better performance and clock speed matters once again.

John The Ripper

On the cryptography front, John The Ripper is mostly what we expected to see. We see core counts seem to make quite the difference here except with the Threadripper 1950X.  On the Intel side we can see that is the driving factor and, with exception of the i7-7800K, see a good stepping on each addition of cores and threads. Hopefully if the situation changes and we get a 1950X in-house we'll be able to test that further and determine why it did not perform as well as we expected.

C-Ray v1.1

The C-Ray results from this group shows a different story than our previous results with the mainstream processors. Clock speed becomes less of a determining factor and instead core/thread counts do become the main one instead. Looking at the data from this perspective it's very interesting how the 8 cores of the Ryzen 7 1700X outperform the 10 cores in the i9-6950X — and that may be a rare case where clock speed is what help set one over the other. But the biggest surprise is the Threadripper 1950X outperforming everything else Intel could offer in its consumer lineup!


Benchmarks: Build Performance

Build Test: ImageMagick

Build tests were working in AMD's favor in our mainstream processor reviews until the Coffee Lake i7-8700K took the lead. But now that we step into the high-end desktop processor range we see something that was reflected in most of the build tests for Phoronix Test Suite. AMD's Threadripper 1950X sees performance levels similar to the 7900X but gets edged out as we step into the higher core and thread counts for Intel. We'll come back to why that's noteworthy when we review current pricing for each of these.

Build Test: GCC

We highlight the GCC build test here because it's an exception to the rule. In this case the Threadripper outperforms all. The grouping suggests that this is not a multi-threaded build test and some further exploration into the matter confirms it. Clock speeds seem to make a difference here as well, but are not the only determining factor. 

Build Test: LineageOS cm-14.1 Pixel XL

We stay with the Pixel XL for timed build benchmarks. As requested by readers the graph will display both build times with and without caches.

There's a lot to take away from this particular test. It also shows why we need to review this at the various processor tiers. First, it confirms that bottlenecks that we see when building Android from source have changed thanks to the new build process. In every level we see a decrease in cached build times, showing that the higher core and thread counts make a difference now at all levels. We also see how clock speed makes a difference with the Core i9-7980XE being edged out by the 7960X due to those clock speed differences. That doesn't remove the fact that an I/O bottleneck remains, but it at least gives us clues into how the overall build process now behaves.

The non-cached builds though are simply impressive. It's amazing to think that in raw power alone, a developer could push out a full build in under 30 minutes per device. Is Threadripper the fastest here? No — it gets edged out by 3 processors on the Intel side. But as we talked about in our original Ryzen review, AMD's strategy has never been one to claim a win over every performance benchmark.


Pricing and Final Thoughts

Since I had all of the other data in a spreadsheet to generate the graphs I figured I'd use that same spreadsheet to also offer one more data point. The graph below reflects pricing on Newegg in US Dollars.

There's typically two types of buyers in the consumer market. The first is the consumer that finds price is not an issue, and will throw down for the absolute best performance. In the case of building Android from source the i9-7960X seems to be this logical choice for this type of buyer. The 7980XE does have additional cores and threads but suffers from a lower overall clock speed that offers less in our benchmarks. This allowed Intel to keep its claim to the highest core and thread count for consumer processors.

The second type of buyer out there is the one on a budget. And when you look at the pricing above it's clear that AMD is aggressively now pricing the Threadripper 1950X to not only be the choice against the 7900X. We highlight the two because at one point they were both listed at the MSRP of $999.99 — and that seems to have changed at this point. The 1950X does not perform as well as Intel's best performer, the i9-7960X. But one on a budget has to ask: Is the 20% difference in performance worth almost double the price? For this buyer the AMD Threadripper 1950X is a clear winner due to performance per dollar. This remains both Ryzen and Threadripper's biggest strength.

Regardless of which type of buyer you are you have something to celebrate here. And thanks to the help of Shannon Robb and AVADirect we're glad we were able to finally prove that. We have a few more ideas in the incubator on how you can use those additional cores and threads if your workload isn't as demanding, and we'll keep working with others to try and get those tested and share the results with everyone. But as we close out 2017 let this be yet one more reminder: what a great year it's been for consumers in the CPU marketplace!

A friendly plug here for AVADirect since they were kind enough to both offer time and manpower to help test these cases out for us. Please consider showing your support for them by either contacting Mr. Robb and/or AVADirect on Twitter or even possibly a purchase from their website. (Editor note: This is being added out of gratuity for their support, AVADirect did not request this nor did they know I was going to post this in the article.)

Readers, we'd like to know: Was this article helpful in making a recent decision on choosing a processor? What do you think of us working and collaborating with others to help accomplish some of the testing? Please sound off in the comments below or on one of our social media posts for this!


 



from xda-developers http://ift.tt/2l7YFsi
via IFTTT

Aucun commentaire:

Enregistrer un commentaire