TECH TALK The Pros and Cons of Integrated Graphics

A YEAR AFTER THE LAUNCH of AMD’s Zen architecture as a CPU-only solution, AMD returns to the APU market with its first Ryzen APUs. Packing more graphics cores than ever, they promise to eliminate the need for dedicated graphics. And they can do that. Sort

The big problem with most integrated graphics solutions is that the CPU and GPU portions of the processor must share limited system memory bandwidth. Even with a dual-channel 64-bit DDR43200 interface, that’s only 64GB/s. If that sounds like it should be more than sufficient, consider that even Nvidia’s budget GTX 1050 delivers almost twice that much, thanks to a 128-bit GDDR5 interface—and all 112GB/s on a GTX 1050 gets devoted to GPU duties. In the computing playground, GPUs are the big, greedy bullies that don’t like to share with other children.

That doesn’t make AMD’s Ryzen APUs bad—on the contrary, I like what they bring to the table. But as a gaming solution, they’re scraping the bottom of the barrel. In benchmarks, the faster Ryzen 5 2400G with Vega 11 graphics sometimes wins, sometimes loses against the ultra-budget GT 1030 and RX 550. Those are $80 parts, so you’re paying $170 for a CPU and integrated GPU that end up slightly slower than a CPU and dedicated graphics card. But that’s only if you opt for high-performance DDR4-3200 memory; go with bog-standard DDR4-2400 or DDR4-2133, and you lose 15-20 percent of the graphics potential, thanks to the bandwidth bottleneck.

AMD’s 2200G loses some of the graphics and CPU cores, but gets the price even lower. At $100 for a CPU and graphics solution that’s twice as fast as any of Intel’s current crop of integrated graphics, that’s hard to beat. But 1080p gaming will struggle at times—you’ll often end up at minimum quality, and sometimes be forced to run at lower resolutions. Mostly thanks to that memory bandwidth bottleneck.

Ironically, Intel is pursuing the higher performance integrated graphics solution, with its upcoming Eighth-Gen Kaby Lake-G parts. They’ll use AMD’s Radeon Vega graphics, the major difference being the inclusion of twice as many GPU cores and a dedicated stack of HBM2. They will truly eliminate the need for a dedicated graphics card for moderate gaming needs, but the Kaby Lake-G processors will be for laptops and NUCs, not larger desktops. And that’s fine, because larger desktops will almost always benefit from the lower pricing that comes from not having to cram as much performance as possible into the smallest space possible.

If there’s one area where AMD’s APUs really shine, it’s HTPC use. Thanks to AMD’s graphics experience, both video decode and encode acceleration are present and work well. I tested the chips in MSI’s B350I Pro AC motherboard, a compact Mini-ITX form factor, and it worked great for HTPC duties. Even better is that it idles at around 20W of power, and only consumes about 30W when watching 1080p video streams. Combine this sort of system with a good home NAS that holds all your media, and you’re set.

Long-term, it will be interesting to see where AMD and Intel go with integrated graphics. Intel’s Iris Pro Graphics tried to boost GPU performance, but drivers and cost were never strong points. Intel’s drivers have improved, however, and with additional work, we could see Intel go back to its own GPU solution for high-end integrated graphics. AMD, on the other hand, just needs to reach a point where it’s economically feasible to stuff more GPU cores and some dedicated memory (HBM2 or even HBM3) into a package. Eventually, that’s bound to be more cost effective than using a CPU with a dedicated GPU— though the performance conscious will always opt for a discrete graphics card. Jarred Walton has been a PC and gaming enthusiast for over 30 years.