AMD Ryzen 9 4900HS Review: Mobile Ryzen goes full throttle
Mobile computing has become the next big target at AMD with its new series of Ryzen 4000 APUs. The new processors will arrive as the U-series for low-power and ultraportables and the H-series for high performance laptops. Today we have the first retail Ryzen 4000 laptop on the market to evaluate, so the focus will be on the benchmarks, so we can see for the first time how Zen 2 brought across to mobile stacks up against Intel’s tried and true line-up.
All Ryzen 4000 APUs are based on the Zen 2 architecture, built using 7nm technology with a monolithic die that also incorporates a Vega-based GPU. Earlier this month we published a full breakdown of Ryzen 4000 APUs, detailing some of its architectural features and specs. We’ll skip most of those details in this review, but you should definitely check that out for an overview of what else is coming in 2020.
Just to be clear, Ryzen 4000 parts do not use AMD’s Zen 3 architecture. The naming is a little confusing here – Ryzen 3000 for Zen 2 desktop chips, and Ryzen 4000 for Zen 2 APUs – but this is simply a continuation of what AMD started in 2017 when they launched the first Zen APUs as Ryzen 2000. Ryzen 4000 APUs are thus based on Zen 2 cores, same as the latest Ryzen desktop CPUs.
It’s no coincidence Ryzen Mobile U and H series mirror what Intel offers for different types of laptops. With ultraportables release still pending, today is all about the powerful H-series and how AMD will fare for content creation, productivity and gaming.
AMD’s Ryzen 4000 H-series consists of three distinct SKUs: the Ryzen 5 4600H delivers 6 cores and 12 threads for the mainstream, the Ryzen 7 4800H bumps that up to 8 cores and 16 threads, alongside a higher-clocked 8-core variant in the Ryzen 9 4900H. All of these APUs come with a 45W TDP, and are complemented by a low power 35W variant for laptops that also adhere to the rest of AMD’s HS-series design guidelines.
You’ll also find different GPU configurations for each SKU, using the refreshed Vega design with up to 8 compute units and higher clock speeds. Interesting to note, but we’d expect most H-series laptops to also include a discrete GPU, like our test system. Cache, while listed at 12 MB, is actually a combined L2 and L3 figure: all parts have 8 MB of L3 cache, Ryzen 7 and 9 get 4MB L2, and Ryzen 5 gets 3MB.
The chip we’re benchmarking today is the Ryzen 9 4900HS, a 35W SKU that provides slightly higher base and boost clock speeds than the Ryzen 7 4800H that sits below it at 45W, a bit of binning magic at play there. It’s not the outright flagship APU in AMD’s line-up, but it will give us a really solid look at Ryzen 4000 performance. We’ll also give you a sneak peak of Ryzen 7 4800H performance from an engineering sample, as we look to review the rest of processors in the line-up as new laptops make it to market.
Our benchmark testbed is the Asus Zephyrus G14, a neat portable 14-inch gaming laptop that packs in Ryzen 4000 HS APUs and up to GeForce RTX 2060 Max-Q graphics. We received the highest-spec model for testing, with the RTX 2060 Max-Q and the 4900HS, 16 GB of DDR4-3200 memory and a 1080p 120Hz display.
The focus on this review is purely CPU performance, so we won’t go into the design and other features of the Zephyrus G14. We will say this though, after a week of use we think it’s a very well built laptop, with a great keyboard and as we’re about to see, very compelling performance in such a small form factor.
You’ll also see across this review a number of different laptop CPUs in the charts and, at times, GPU configurations as well. The data in the charts is an average from the laptops we’ve tested with the given hardware. Testing laptop components is naturally a bit more difficult than desktops as each configuration can vary in cooling and other hardware, so these averages are meant to illustrate how a ‘typical’ system will perform. The averages do not include single channel memory systems or any other situations that heavily throttle the components at hand, we’ve tried our best to create apples-to-apples data where possible.
Let’s kick things off with a classic performance benchmark: Cinebench R20. In this test we see total domination by the red team. Not only is the 4900HS the fastest laptop processor we’ve tested in the multi-threaded workload, it’s also the fastest in single-thread performance. The 4900HS is 35% faster than the 8-core Core i9-9880H when using all cores, and 7% faster in single-core. There is the Core i9-9980HK that we haven’t tested that might put up a bit of a fight, but the 90W numbers from the 9880H suggest this might be a hard task. Also consider, these results are from the lower-TDP 4900HS, the 4900H at a full 45W should be another step ahead.
There are other brutal results in this Cinebench chart. The 4900HS crushes the Core i7-9750H, which is the most popular CPU used in slim and light gaming notebooks. Zen 2 offering is over 60% faster in the MT test. We also see the previous-gen AMD part get a bit humiliated, the 4900HS is so much faster than the Ryzen 7 3750H in the same power envelope it’s not funny.
We also tested with the legacy Cinebench R15 which has results from a wider range of CPUs going back more years. The 4900HS remains firmly placed at the top of the charts as expected. Anyone upgrading from a Core i7-7700HQ, for example, which was a popular H-series CPU in 2017 will be treated to more than double the performance with a Ryzen 9 4900HS.
One area where Zen 2 receives a particularly large upgrade is in how it handles AVX-256 instructions. It’s simply much faster at wide floating point operations than before. So the Ryzen 9 4900HS receives an enormous performance upgrade in our Handbrake x265 test, which uses AVX instructions. The 4900HS is a monstrous 179% faster than the 3750H in this workload, which is just unfathomable for a single generation performance leap.
On an AMD vs Intel front, it’s also highly favorable for AMD. Not quite the same margins as we saw in Cinebench, but we still see a 23% performance advantage of the 4900HS over the 9880H, and a 45% advantage over the 9750H. Frankly, these are massive deltas for a laptop form factor that often receives single digit gen-on-gen improvements.
For Intel to match the 4900HS with its 8-core offering, it needs to blow its power target out the window and instead use a 90W TDP, which is possible on some gaming laptops with a ‘Turbo’ mode or similar. The difference in power draw at the wall for these two systems is incredible: the G14 with the 4900HS ran comfortably at around 66W long term, compared to 150W for the power boosted 9880H in our HP Omen 15 test system. That just goes to show how much more efficient AMD’s Zen 2 design is at these long term workloads.
Blender tells us a similar story, with huge performance gains in the ~35% range for the 4900HS over the 9880H, and ~65% range for the 4900HS over the 9750H. Most Blender users will probably render on the GPU instead since it tends to be much faster, but it’s another benchmark that illustrates long term multi-core performance on these laptops.
In terms of decompression, we see AMD pull away with a significant victory in 7-Zip. Ryzen processors are known to work very well in this workload. However it does fall behind the 9880H in terms of compression, to the tune of ~8%, although AMD beats lower core count parts like the 9750H.
Adobe Photoshop performance is interesting. In our Iris Blur test which is mostly CPU limited, the 4900HS manages to beat the 9750H but falls behind the 9880H, sitting comfortably between those two processors.
This continues to be the case in the more comprehensive Puget Photoshop benchmark, which runs through a range of tests. In both of these workloads, which apply various effects to very high resolution images, the 4900HS is around 10 percent slower than the 9880H in a core-for-core battle. Granted, we are seeing 40% higher performance than AMD’s last-gen APUs.
Now let’s check out PCMark 10 numbers. We’re looking at the Essentials and Productivity workloads as they’re CPU limited, whereas the rest of the tests rely more on the GPU.
The Essentials workload covers things like app loading, web browsing and video conferencing. AMD manages to match the performance of the 9880H, which may not sound all that impressive, until you realize that the last-gen 3750H gets slaughtered in this test. Now AMD is at performance parity for these everyday workloads.
On the Productivity workload we see something similar as before, with the Ryzen 9 4900HS outperforming something like the Core i7-9750H. These sorts of tasks were less than decent on previous Ryzen, so even though Ryzen 4000 isn’t crushing Intel here, performance parity is a good result.
However there are some situations with productivity workloads where Ryzen 4000 isn’t as impressive. Our custom Excel benchmark features a lot of number crunching on a large dataset, and here that the 9880H outperforms the 4900HS again. The 4900HS is faster than the 9750H, so it’s not a terrible result by any means, but it does seem that large dataset workloads are a weakness for Ryzen.
MATLAB lets us reconfirm this with our ODE and FFT benchmarks. Again, lots of data crunching on large sets of data and Intel’s 9880H walks away with a victory.
Another workload where Ryzen performs well, but not enough to beat Intel’s 8-core competitor is our Acrobat PDF export test, which is fully single-threaded. The 9880H is marginally faster in Acrobat, although we still see a significant performance increase for the 4900HS over the 3750H with a healthy 20% bump to single-threaded performance.
One final workload before we look at GPU acceleration is AES-256 performance as provided by SiSoft’s Sandra benchmark. In the multi-threaded test, we see 15% higher AES performance from the 4900HS versus 9880H to make it the fastest CPU we’ve looked at for cryptography workloads. This gives Ryzen a neat advantage in two heavily utilized low-level tasks in decompression and cryptography.
Now let’s work through our Premiere tests which are mostly GPU accelerated, starting with our 1 pass encode which takes advantage of Intel’s QuickSync technology. Premiere does not support hardware accelerated encoding on AMD processors at this stage, so for those that like a quick export with slightly reduced image quality at the end, Intel is still the way to go. In particular, the 4900HS paired with Nvidia’s GeForce RTX 2060 Max-Q is ~20% slower than a Core i7-9750H with an RTX 2060 simply because hardware acceleration is not available.
On the other hand, our Core i9-9880H system doesn’t support QuickSync acceleration as the iGPU is fully disabled to support G-Sync through the Nvidia GPU. So we are left with a nice software encoding comparison between the 9880H and 4900HS, where the 4900HS pulls away strongly to the tune of 26%.
Then we get to our 2 pass encode which produces superior image quality and doesn’t support hardware acceleration. In this scenario, performance falls back to what we saw from most long-term workloads: the 4900HS is 30% faster than the 9880H, and 38% faster than the 9750H, despite our 9880H laptop packing a much faster GPU. This test isn’t GPU limited but even with a moderately capable discrete GPU it appears as though AMD’s Zen 2 APU is significantly faster for Premiere encoding.
But wait, there’s more Premiere tests. Here we have a single instance of Warp Stabilizer, which is a hugely demanding effect that runs on a single thread per instance. The 4900HS manages to stabilize the footage 14% faster than the 9880H, and 22% faster than the 9750H, which shows the power of the high single-thread performance we first saw with Cinebench.
And finally we have the Puget benchmarks. The Ryzen 9 4900HS is the best laptop CPU for live playback in Premiere, performing 12% better than the 9880H, meaning we’re seeing better editing performance. Then for their export benchmark, we see similar numbers to what we’ve just been talking about, with QuickSync acceleration assisting with some of this workload.
While we don’t think many people will actually be using the integrated GPU in these H-series processors given the vast majority of H-series laptops also include discrete graphics, it’s worth a brief look at how the iGPU fares in a small selection of low intensity games, just to see what improvements AMD has made to the Vega GPU.
The Ryzen 9 4900HS sports the fastest iGPU configuration available in the series with 8 compute units clocked up to 1,750 MHz. And it does deliver impressive results. In Grand Theft Auto V, the 4900HS delivered 36% more performance than the Ryzen 7 3750H, which featured 10 Vega compute units at 1,400 MHz. On paper, both CPUs have similar raw GPU performance, but with all the advantages AMD talked about like increased memory bandwidth, this new Zen 2 APU is able to pull ahead.
In Civilization VI, the 4900HS again provides 37% more performance than the 3750H within the same 35W power envelope. And moving to CS: Go, that margin remains identical at around 37%. The 4900HS running CS Go exclusively on the integrated graphics was able to achieve over 100 FPS on average using low settings, which is very nice.
And then finally we have our most performance-intensive iGPU test in Gears 5 running at Medium settings. The 4900HS is around 31% faster than the 3750H in this workload, which makes it faster than a low-end discrete GPU offering like Nvidia’s MX250.
It’s not common to see an H-series laptop paired with an MX-class GPU. Usually OEMs opt for more powerful configurations like a GTX 1650 or higher, but with Ryzen 4000 there’s really no need to bother with an MX250 or similar discrete GPU in an H-series type design.
Sustained Clocks and More Questions
At this point in the review we’d normally walk you through some comparison summaries between the Ryzen 9 4900HS and various other CPUs. But we still think there’s a few performance questions left on the table. One is what clock speeds the CPU actually runs at in practice, and how boost behaves. And the other is, why do we see lower performance in some data heavy workloads like Matlab, Excel and Photoshop. So let’s tackle this second part first.
We have two working theories as to why we see data heavy workloads perform this way. The first is a simple one based on the specifications of these processors, in particular cache sizes. The Core i9-9880H has a decent 16 MB of L3 cache, which matches what Intel offer on the desktop with parts like the i9-9900K. However the Ryzen 9 4900HS has just 8 MB of L3 cache, half of what Intel offers, and well below the 32 MB of L3 that AMD packs into their 8-core Zen 2 desktop processors like the Ryzen 7 3700X.
Having less cache means less data can be stored in super fast memory and accessed in an instant. When you have high core count, high performance CPU cores but not enough cache, this can become a bottleneck in some instances. And while this cache amount has doubled on Ryzen 4000 from previous mobile series, so has the core count. It’s probably not the whole story, but definitely part of it.
The other is the memory system. Yes, AMD offers higher memory bandwidth than Intel with the move to support DDR4-3200 speeds, Intel only offers DDR4-2666 with 9th-gen. In a benchmark like Sandra we see AMD providing around 35% more memory bandwidth. However, Ryzen 4000 appears to have inferior memory latency. This, like cache size can become a performance constraint. The Core i9-9880H has memory latency around 30ns for data sets above 32MB in size, while the Ryzen 9 4900HS has 46ns memory latency. That’s a substantial win for Intel.
It’s hard to say for sure whether these factors are part of the cause, or the entire cause, but looking through low level benchmarks these were the two things that stood out to us. When we get to benchmark more Ryzen 4000 APUs we’ll get a clearer picture of where these bottlenecks lie.
As for clock speeds and boost behavior, let’s take a look. From a cold start in our Handbrake AVX workload, the Ryzen 9 4900HS consistently boosted up to around 65W of power for a few seconds achieving 4 GHz boost clocks all core, before dropping down to 54W for a longer sustained period, with clocks around 3.7 GHz. Eventually the CPU settles down to 35W to provide 3.2 GHz all-core, just above this processor’s 3.0 GHz base clock. The boost period can vary depending on how warm the system is, but from a cold start we were generally seeing at least 2 minutes of ~53W boost which is generous. Temperatures were actually very well managed on the Zephyrus G14, with its air cooler providing a tick over 70C sustained, but of course, this will vary between laptops.
Ryzen 9 4900HS vs. Core i9-9880H
The big one here is the Ryzen 9 4900HS versus the Core i9-9880H. These are similar class processors, 8-core versus 8-core and should give us a pretty solid indication of what to expect from 10th-gen as well, if we get minor clock speed improvements from Intel as is rumored.
In most long term, multi-threaded workloads, we’re seeing 35% better performance from the Zen 2 chip. We also get impressive numbers for video encoding. Single-threaded workloads are also typically faster, or in the worst case scenarios like light productivity, equivalent to Intel’s. However, performance does fall behind in data heavy benchmarks like Excel, Photoshop and Matlab.
The amazing thing about most of those results is AMD is able to achieve, in many workloads, at least 25% better multi-thread performance while also coming in with lower sustained power draw, at 35W versus 45W. So what happens when you limit Intel’s 9880H to a PL1 of just 35W using Intel’s XTU software?
Well, the margins grow even more. While we do let these processors do what they normally do in the boost stage, some benchmarks don’t change too much, long term workloads now are strongly in favor of AMD. Over 50% better performance in some situations is possible, and a 40% improvement to Handbrake is impressive. This means that an OEM designing a slim and light system with limited cooling will generally get the best experience by far going with Ryzen.
Amazingly, even when we throw Intel’s 45W power limit out the window and boost up to 90W forever, Intel still can’t beat the 35W AMD processor in long term multi-thread workloads. In an astounding win for efficiency, the 4900HS is either equal to this 90W processor configuration, or up to 15% faster. With these two systems going at it, we observed an 80W power draw difference from the wall, so if Intel wants to match AMD with 14nm CPUs by just raising power limits, laptops will need much larger coolers to cope.
Ryzen 9 4900HS vs. Core i7-9750H
Comparing the 4900HS to the 9750H, there’s only one instance where the 9750H system can be faster, and that’s with QuickSync accelerated Premiere encoding. In every other benchmark, including those where previously the 9880H was faster, now the 4900HS is faster. This bodes well for any future battles in the Ryzen 7/Core i7 range where AMD has a core advantage in offering exclusively 8-core parts.
Ryzen 9 4900HS vs. Ryzen 7 3750H
Now let’s look at how far AMD has come with Zen 2 vs. Zen+ in a 35W mobile design. The difference in certain workloads is staggering. Not only is AMD producing a 30% performance improvement in single-thread workloads, they’re able to deliver 2.5x the performance in long term multi-thread tasks like Blender, Handbrake and Cinebench — with the same power draw.
Ryzen 9 4900HS vs. Ryzen 7 4800H
We did promise a sneak peek at Ryzen 7 4800H performance on an engineering sample laptop we had brief access to. Going on a small subset of benchmarks, performance is looking equal to, if not slightly better, than the 4900HS, which bodes very well for more mid-range systems that will use the Ryzen 7 part. Yes, the 4900HS is more efficient, but the 4800H should still deliver pure 8-core performance.
What We Learned
Ryzen 4000 is delivering the best performance for heavy productivity workloads you can get in a mobile form factor. We can say that without even testing the higher power Ryzen 9 4900H, because the 35W variant in the Ryzen 9 4900HS is already leaving Intel’s 8-core competitors in the dust.
The Ryzen 9 4900HS is a great choice for core-heavy work: we’re talking video encoding or transcoding, 3D rendering, file compression, that sort of thing. It’s much faster and does so at a lower power draw, allowing for better performance in a smaller form factor. Even in single-thread workloads, AMD generally comes out on top, which means that the Premiere editing experience is better on the Ryzen 9 4900HS, for example.
From a pure gen-to-gen upgrade perspective, AMD has done an amazing job. The Ryzen 7 3750H was difficult to recommend and it was handily beaten by Intel’s mid-range parts. That has all changed with the newer APUs.
Technically, Intel parts can still compete on the performance front if you remove the 45W long term power limit. We saw that with the Core i9-9880H and we expect it’d be similar had we tested the flagship Core i9-9980HK. But to get that sort of performance, you need a larger laptop with a beefier cooler, and it’ll likely run hotter, too. The Ryzen 9 4900HS can provide all that performance in smaller laptops — like this Zephyrus G14 — and run cooler at the same time.
However, AMD’s Ryzen 4000 CPUs aren’t the complete package, there are a few limitations. One is with data heavy workloads, like number crunching massive spreadsheets in Excel, working with huge photos in Photoshop or running Matlab scripts. Intel came out ahead in these tests. Lighter loads will do just fine on AMD, but if you’re a big number cruncher, Intel’s core-equivalent processors might be the way to go.
The other is with workloads that use Intel exclusive technologies like QuickSync, Premiere encoding being one. If that’s something you use all the time, Intel is the way to go. It’s worth mentioning that every other task in Premiere outside of QuickSync encoding is faster on Ryzen 4000, including editing and applying intensive effects like the Warp Stabilizer.
Outside of those two instances, you’ll have a better, faster productivity experience with these Ryzen 4000 8-core processors. That’s uncharted territory for AMD on a mobile PC form factor.
Also it seems that AMD will be offering this sort of performance at very competitive prices. The Zephyrus G14 with the Ryzen 9 4900HS and RTX 2060 Max-Q is set to retail for about $1,500, which is firmly the territory of six-core Core i7-9750H laptops. The Core i9-9880H laptop we bought for testing in this review cost $2,400 and was one of the least expensive we could find. Granted, it does have a much faster GPU for gaming, but right now you just won’t find 8-core laptops in the $1,500 price range. While most of our data has been focusing on Ryzen 9 vs Core i9, they don’t really compete head to head on pricing, at least for now as this could change with Intel’s 10th-gen.
A few additional notes to wrap up this review…
With previous Ryzen laptops we didn’t have the most stable experience, running into plenty of software bugs and crashes, which we’ve discussed a few times before. Completely different experience with the Ryzen 9 4900HS in the Asus Zephyrus G14. Even with a beta version of the GPU driver on this laptop, we ran into no stability issues, crashes or applications refusing to work correctly.
You also might be wondering about gaming performance with a discrete GPU. The Zephyrus G14 does include the RTX 2060 Max-Q after all. That’s something we’ll explore in a future review as we’re still working to get the best apples to apples comparison. Hoping to have that soon.
Battery life will also be on people’s minds, but it’s not something we set out to test as we didn’t have enough data for various laptops to make it a fair comparison.
Finally, we’ll close this one out with some comments on Intel’s 10th-gen. The new laptop CPUs are coming just around the corner, likely within a month. Rumors suggest Intel is sticking with 8-core parts for the H-series, and are still using 14nm technology. We just can’t see how they can compete on a performance or efficiency level. If all we’re getting is a small clock speed bump with the same 14nm efficiency as the last few generations, Ryzen is going to easily win this one. We’ll see that battle unfold soon, but based on what we’ve seen so far, we don’t hold high hopes for 10th-gen.