AMD has really been bringing the heat to Intel this year, with incontestable wins for its 7nm CPUs in the desktop space, high-end desktop space, and server space. The one thing everybody has been waiting with bated breath for is mobile—while Intel brought limited supplies of high-performance 10nm Ice Lake parts to market, AMD has remained pretty silent about mobile.
The most I could ever get out of my AMD folks was a sort of “we can’t talk about that yet,” with suspicious little yellow feathers floating out of their mouths, but no real detail.
Yesterday at CES, that final shoe dropped—Ryzen 4000 mobile is here, and it brings AMD’s recent trademark of high core and thread counts and jaw-dropping low TDPs to the mobile arena. The flagship U-series part, Ryzen 4800u, offers eight cores/16 threads on only 15W TDP, and although we’ve got nobody’s word for it yet but AMD Performance Labs’, it appears to whip the high-end Ice Lake i7-1065G7 solidly across the board in tests ranging from Cinebench R20 to 3DMark, Adobe Premiere, and more.
Of course, performance is only half the battle in ultralight form factors—power consumption is the other. It shouldn’t be any surprise that AMD is showing massive performance-per-watt increases over the first two generations of mobile Ryzen, given those performance numbers with a 15W TDP. The bigger question—and one that can’t be so quickly answered—is how well Ryzen 4000 series systems will idle. And unfortunately, that’s not a question AMD can entirely control themselves.
In the mobile arena, integration is crucial to system performance—everything from motherboards to firmware to cooling is incredibly one-off and proprietary to each final system build. When designing a new laptop, just slapping a processor and some RAM on a reference board design and calling it a day won’t cut it.
This “insufficient integration” problem has plagued AMD laptops for years, with OEMs not doing the same level of integration work on AMD builds as they have on Intel. The common “wisdom” among buyers has been that AMD laptop CPUs just sucked—but Microsoft proved that line of thinking wrong with 2019’s Ryzen-powered Surface 15, which have the years of integration work and attention to detail necessary to make a great mobile system.
We won’t really know how well the OEMs have—and will—do with Ryzen 4000 series CPUs until we get some systems on hand to test. But we have high hopes that the sheer, unprecedented power the new 7nm mobile designs offer will leave OEMs more excited and willing to build premium, well-designed products around AMD than they have been in the past.
Ryzen 7 4800H challenges Intel’s i7-9700K desktop gaming CPU
You might be thinking the Ryzen 7 4800u, with its eight cores, 16 threads, and Ice Lake i7-1065G7 whippings would be the primo Ryzen 4000 CPU. If you are, that’s a mistake—an understandable mistake, but a mistake nonetheless. The H-series equivalent to the 4800u offers the same thread count and boost clock but increases the idle clock from 1.2GHz to 2.9GHz, and the TDP from 15W to 45W.
It also handily nosed past Intel’s most recent full-on gaming CPU, the i7-9700K, on both content creation and physics engine benchmarks, despite being a mobile form factor with under half the TDP.
For all the differences between AMD’s and Intel’s current marketing strategies—with Intel relying heavily on their investments in AI software and hardware, and AMD focusing on pure, unadulterated power-efficient grunt performance—they do have one striking similarity. Intel and AMD both seem intensely focused on pushing the idea of serious content creation happening in ultralight laptop form factors, positioning them as real alternatives to traditional desktop designs.
We’re still not 100% sure how that’s going to work out—no matter how many times Jason Levine does a neat Photoshop transform on stage, or how many bar charts we get out of AMD, it seems unlikely to us that the interface and peripheral challenges presented by ultralight laptops will really lend themselves to content creation work. But we’ll concede that advances in CPU design are certainly making it look more possible than it used to.
The Radeon 5000 series adds the 1080p-focused 5600 XT to its lineup
AMD also announced the Radeon 5600XT. This is fairly weak tea compared to the Ryzen 4000 series launch—the 5600XT really just offers a new price point in the existing Radeon 5000 lineup at $280 with a focus on being powerful enough for uncompromising AAA gaming at 1080P.
We’ve tested a few price points of Radeon 5000 series cards and Geforce GTX cards, and in our opinion, it’s hard not to love the Radeons. The image quality is fantastic, and when running the incredibly demanding Unigine Transposition benchmark, we see fewer immersion-destroying artifacts on Radeon systems than we do on GTX systems. The Radeon line also offers far, better Linux support than Nvidia cards do; screen-tearing during Linux video playback, among many other irritations, becomes a thing of the past when you yank out your Nvidia GPU and replace it with a Radeon.
With all that said—and we can already hear the groaning from the peanut gallery—nobody seems to be building machine learning inference or training platforms that support Radeon. The uber-popular Tensorflow platform targets CUDA architecture specifically, so no matter how many FPS your Radeon card gets compared to its nearest Intel counterpart when gaming, it’s a poor choice if you’ve got an itch to do some hard-core deep-learning work on your system.
AMD also had a slide declaring itself the only manufacturer of both premium performance CPUs and GPUs—which doesn’t seem like a line that will work for long, since Intel also announced its first discrete Xe series GPU, the DG1. It’s still very early days to estimate just how powerful Intel’s new line of GPUs will be—for either traditional GPU work like gaming and content creation, or for AI acceleration—but we saw a laptop with Intel’s DG1 discrete graphics doing a good job playing , so it’s equally early to rule Intel of the high-performance GPU game.
On the GPU front, it may very well be Nvidia who’s feeling the squeeze at this point, with AMD eating into its gaming mindshare and Intel apparently determined to close the gap on AI acceleration.
Threadripper 3990x—when too much is just enough
AMD also announced the Threadripper 3990x. We got a chance to build and hands-on review a test system using the 3970x, and it was pretty nuts—although it drastically outperforms both Intel’s HEDT parts and its Xeon Scalable parts for most workloads, the sheer thread count of the thing meant the 140mm fans on our NZXT Kraken liquid cooler were in leaf-blower mode every moment the system was on, and it raised the temperature in the office its workbench was in when running under load.
We haven’t seen TDP ratings on the Threadripper 3990x yet—but while we’re sure they do a good job measured in perfomance-per-watt, on a relative scale to normal systems, they’ve got to be pretty entertaining.
Meanwhile, if you actually that kind of firepower, the 3990x kicks the snot out of a dual Xeon Platinum 8280 rig running an easy five times the CPU cost—AMD Performance Labs shows a 3990x system completing a V-ray scene render 30% faster than a dual 8280 could. It’s also, unsurprisingly, much faster than the previous performance king, the 32 core/64 thread Threadripper 3990x. We may be seeing some limitations in the overall architecture beginning to show up, however. The 64 core/128 thread 3990x is about half again as fast as its 32 core/64 thread little sibling—which was nearly twice as fast as the 16 core/32 thread 3950x.
There’s no doubt about it, AMD is running rampant through the market in everything it’s focusing on right now—whether you’re looking for a gaming or general-purpose desktop CPU, a content creation powerhouse or server CPU, a laptop or even a graphics card, AMD is leading in performance.
However, it’s not entirely out of the woods on the laptop front yet. We only have AMD Performance Labs’ word on how well the Ryzen 4000 series performs, but we’re inclined to take its claims at face value—it didn’t steer us wrong on the Ryzen 3000 series or Epyc series benchmarks, after all, and we feel it clearly knew how damaging it would be to its own brand to get too sketchy with preliminary numbers. What we’re more worried about is the OEMs’ commitment to building premium systems around them—if OEMs don’t put in the hard work to integrate mobile Ryzen well with cooling, motherboard, peripheral, and firmware designed and tested to work well together, the result will be shoddy laptops.
We do like the fact that we’re seeing already-built Ryzen 4000 laptops at the show from Asus, Lenovo, Acer, and Dell—along with AMD’s brag that we’ll see 100+ systems by the end of the year. Hopefully, this marks the end of a long, dark period for AMD’s mobile prospects.
Finally, there’s the question of whether Intel will eventually succeed in what appears to be more of a “flanking” strategy on its part. While AMD is focusing—very successfully—on building high-performance systems targeting current markets at great prices, Intel is investing in both hardware software ecosystems that may eventually change what the market is in the first place. We hope to see an implementation of the DLB machine-learning acceleration instructions in AMD’s CPUs before too much longer, and perhaps some work porting AI frameworks like Tensorflow to the Radeon GPUs as well.