System Performance: Testing the AMD Ryzen 7 2700U

It’s fair to say that AMD’s notebook processor architecture has been somewhat lackluster in recent years, and as we saw back when Ian visited AMD, as the company tried to get by with Kaveri and Carrizo, it was somewhat hamstrung by the OEMs as well, with low-cost, low-quality devices, and performance constraints like single-channel RAM. That’s not the case with Ryzen Mobile though. Acer is offering it in a proper laptop, with an IPS display, SSD storage, and dual-channel memory.

As it did on the desktop, AMD is launching Ryzen Mobile as a turning point for their products. They are offering full quad-core processors in their 15-Watt range, with eight threads. That’s the same as Intel who only recently switched from dual to quad-core in the U series. But while Intel still has a CPU advantage with Kaby Lake Refresh, they’ve been let down with their UHD 620 graphics. AMD on the other hand, was able to leverage their Vega graphics processor and package it with the Ryzen Mobile CPU.

AMD has also been able to move to the 14 nm FinFET process by GlobalFoundries, so they are no longer as far behind on the foundry side either.

With the Zen core, AMD has been able to achieve the scalability of low-power processors such as the Ryzen Mobile with one CPU Complex (CCX) of four cores, all the way up to their larger desktop processors. They leverage their Infinity Fabric for memory, I/O, and PCIe, but also for their SenseMI interconnected sensor system which provide the telemetry to adjust the processor power and frequency on the fly. As we recently saw in the desktop Ryzen 2nd generation, AMD introduced Precision Boost 2 to the desktop, but that technology first launched on Ryzen Mobile. Precision Boost 2 is a telemetry-aware DVFS system which provides finer control over the CPU core frequency, with 25 MHz increments. The new system allows for higher boost levels under multi-core loads when the overall workload is not as demanding, whereas in the original technology they only offered two boost levels depending on if two or more than two cores were loaded. The advantage of the new system is that it listens to the telemetry to ensure that the CPU is boosted to the highest it can be without hitting a thermal, electrical, or utilization boundry, which should improve performance quite a bit over the original version.

AMD is offering two versions of Ryzen Mobile, in the Ryzen 7 2700U, and the Ryzen 5 2500U.

AMD Ryzen Mobile
  Ryzen 5 2500U Ryzen 7 2700U
CPU Cores 4 Cores, 8 Threads (1 CCX)
CPU Base Clock 2.0 GHz 2.2 GHz
CPU Boost Clock 3.6 GHz 3.8 GHz
GPU Cores 8 Radeon Vega CUs 10 Radeon Vega CUs
GPU Clock Up to 1100 MHz Up to 1300 MHz
TDP 15W Nominal, 12-25W Configurable
DRAM Up to DDR4-2400

Although the CPU frequencies are similar, the GPU is a larger unit in the 2700U, which is the model we have for review. It’s also important to note that unlike Intel, AMD only supports DDR4 on Ryzen, which is going to have an impact on laptops power usage compared to LPDDR3 that we still see in most Ultrabooks.

We’ve run the Acer Swift 3 through our standard laptop suite. We’re going to be updating our test suite soon with a few new tests, such as PCMark 10, and new web tests, but for this review it’s important to have more data to fall back on for comparisons. If you’d like to check out the Acer against any laptop we’ve tested, please use our online Bench.

PCMark

PCMark 8 - Home

PCMark 8 - Creative

PCMark 8 - Work

As an all-around test, PCMark tests every aspect of the system, so the CPU, GPU, memory, storage, and even display can impact the resuts. Here we see some interesting results. The Ryzen can’t quite hang with the latest Intel quad-core laptop processors, although it’s very competitive with the Skylake based notebooks, at least in terms of this test.

Cinebench

Cinebench R15 - Single-Threaded Benchmark

Cinebench R15 - Multi-Threaded Benchmark

Cinebench is a pure CPU test, where higher IPC, higher frequencies, and more cores are the main ingredients to a strong result. Thanks to the four cores and eight threads, the multithreaded results are very good on the Acer, and in fact despite it not having as strong single-threaded performance as the ASUS Zenbook 3, it surpasses it in multithreaded. AMD says they have a “highly efficient SMT implementation vs. other 4C8T SoCs” and it does appear to be the case. Even compared to the Surface Book 2 which has the fastest KBL-R quad, the ratio of single-threaded to multi-threaded is higher on the AMD part. No, it can’t quite hit the same levels of performance, but it’s significantly faster than any of the 7th generation Core results.

x264

x264 HD 5.x

x264 HD 5.x

This test converts a video using software, and like Cinebench, is a CPU bound test which relies on more cores and higher frequency to go with strong IPC. Here we see a similar story to the Cinebench results, where the easier pass one result is not quite as strong as KBL-R, but the multithreaded result is much stronger. AMD has done some nice work to keep the Zen cores fed under multi-threaded loads even if they don’t quite have the IPC of KBL-R.

Web Tests

Web performance is important due to the volume of web use on a typical PC, so good performance here can translate to a good user experience, with less scrolling hiccups, and more responsive pages. We use Microsoft Edge on the current build of Windows 10 as our test vehicle. Web performance is one part CPU, and one part browser engine though, so over time, the same system should trend upwards in performance on the same tests, as improvements are made to the underlying browser.

Mozilla Kraken 1.1

Google Octane 2.0

WebXPRT 2015

Here the results aren’t quite as rosy for AMD. The performance on the web benchmarks is a significant departure on what was seen on the previous benchmarks, with results firmly lower. Some of that will be down to Intel being able to boost to higher frequencies than the AMD chip, with Intel having about a 10% advantage in maximum boost on the Core i7-8650U, but Intel has also spent quite a bit of effort on their SpeedShift to be able to ramp up the frequency very quickly. This is especially important for the web tests which do a short amount of work. Precision Boost 2 just can’t seem to keep up.

CPU Conclusion

Even though the AMD chip can’t quite offer the same levels of performance when compared to the latest Intel U series, it still offers competitive results, and is a massive step forward from Carrizo. We never got to test a Carrizo laptop, other than the one article Ian wrote, so we don’t have comprehensive results from our laptop suite, but even just looking at the Cinebench results, the HP Elitebook 745 G3 only got a ST score of 71, and a MT score of 205, compared to 141 ST and 540 MT for the Ryzen 2700U. That’s a 98.6% increase in single-threaded performance and a 163.4% increase in multi-threaded.

But Ryzen isn’t just about the CPU of course. AMD’s Vega GPU is packed in too.

Design GPU Performance: Vega at 15-Watts
POST A COMMENT

78 Comments

View All Comments

  • jaydee - Thursday, May 3, 2018 - link

    Looking forward to Ryzen hitting the business line notebooks (Elitebook/Latitudes). Lot of employees at my company are still being given the woeful Elitebook 745 G3 (Carrizo) as of today, as band new laptops for a 4 year cycle... Reply
  • Jimster480 - Monday, May 28, 2018 - link

    They did, and the offerings are atrocious and insanely overpriced. Reply
  • Valantar - Thursday, May 3, 2018 - link

    This is very promising in terms of performance, even if AMD still needs to convince OEMs to build more premium SKUs with Ryzen mobile. Also, that base platform power draw needs fixing, for sure. Wonder how much of that is RAM, but I suppose that would be pretty much impossible to measure.

    Related to that: why doesn't RAM have some sort of turbo/power saving implementation like CPUs? Given that this likely runs dual channel 1.2V DDR4 at ~2400MT/s, couldn't the base power draw benefit massively from downclocking the RAM to, say, 1600MT/s at idle and simultaneously lowering voltage? I get that this isn't part of the DDR4 spec, and that the ICs and DIMMs as such aren't tested or certified for lower voltages and speeds, but the lack of dynamism in RAM is starting to feel old. The same goes for high-speed desktop RAM, really - why run 3600MT/s RAM idling at the desktop? It can't possibly be /that/ hard to implement a two-state (e.g. "low power" and "normal") dynamic system that's directly tied to other system loads (not just CPU, but GPU, RAM, network and disk activity too). Tuning the boost/de-clock triggers and control algorithms would likely require a bit of work, but is this really that hard to do?

    I suppose this could trip up a few applications that are highly reliant on timely RAM access, but those can't be very common in normal consumer usage - and I don't envision this catching on in datcenters and the like. Might that be why we still haven't seen something like this, as businesses and server OEMs don't care?
    Reply
  • DanNeely - Thursday, May 3, 2018 - link

    If idle link throttling of DDR ever happens, I'd expect it to happen first on mobile, and then work its way up though LPDDR, to regular DDR specs. OTOH GDDR has been able to do this for a while, so it's definitely doable without seriously impacting peak performance. GPUs are power pigs, and lower idle power has been a major improvement over the last decade; but it's a segment not willing to sacrifice performance at the top to get it.

    You're probably right about it never making it to the data center. I've read that a number of the major cloud platforms have disabled CPU clock throttling and run at full speed 24/7 regardless of the actual load because serving responses at low speed and/or boosting back up from idle to full speed have measurable impacts on latency and the major web companies care about every millisecond there.
    Reply
  • neblogai - Friday, May 4, 2018 - link

    Gaming benchmarks show mobile Raven Ridge processors doing exactly that: RAM clock changes in the game, 933-1066-1200. You can see such behaviour in TechEpiphany's videos on youtube. However- because this is not something seen before- these readings are semi-discarded. In would be great if someone tested it properly, or asked AMD if this is how it really works, and is not an error by monitoring software. Reply
  • HStewart - Thursday, May 3, 2018 - link

    I think for comparison, it would have been nice to test the Intel version of same model - I found it interesting that 2700 version was more expensive than the Intel with MX150. Also the Intel model was less weight which tells me Intel components are more compact. The 2700 model had more ssd but that should not be that way. Reply
  • Cooe - Thursday, May 3, 2018 - link

    It's because of the doubled SSD size that the 2700 model is more expensive obviously. Reply
  • Jimster480 - Monday, May 28, 2018 - link

    Its not that because the AMD machines are relegated to cheap SATA bottom of the barrel drives and ultra cheap screens. The price is just Intel making sure that nobody is going to make a competitive notebook.
    This laptop doesn't even land in "mainstream" it lands in "ultra budget with such a big price nobody will buy it". The screen, SSD, battery, wifi, chassis, keyboard are all unacceptable for this price range and Acer knows it aswell. But the check they receive from intel outweighs the obvious lack of sales they will have for this model.
    Reply
  • Stuka87 - Thursday, May 3, 2018 - link

    I would be surprised if you could tell the weight difference between the two. Its 4 ounces. And the Intel is cheaper because its SSD is half the size. Reply
  • Krysto - Thursday, May 3, 2018 - link

    Still only 8GB of RAM?! Come on. Reply

Log in

Don't have an account? Sign up now