RAM Speeds and SecondGeneration Ryzen CPUs
WE TALKED ABOUT memory performance and timings a few months back, but it’s a topic worth revisiting thanks to the launch of the secondgeneration Ryzen processors. For this discussion, I used the top-of-theline Ryzen 7 2700X, with additional testing using a Ryzen 5 2400G APU.
First-gen Ryzen CPUs were more finicky with RAM sticks, especially at launch, and performance could also vary. I spoke with motherboard manufacturers, asking about memory performance and Ryzen CPUs in particular. One of the interesting tidbits is that Intel platforms apparently offer far more options for tweaking memory performance than AMD Ryzen platforms—about twice as many timings, with AMD microcode handling many elements of DDR4 training. That’s apparently part of why memory compatibility on socket AM4 was initially so dodgy.
With further updates to the AGESA (AMD Generic Encapsulated Software Architecture) software, plus the refinement in the second-gen Ryzen parts, we get better performance and compatibility. One area where AMD cites specific improvements is cache and memory latencies—up to 16 percent lower L3 cache latency, 34 percent lower L2 latency, 13 percent better L1 cache latency, and 11 percent better memory latency. Combined, these improve per-clock performance by around 3 percent.
The reason for the modest overall improvement is that modern processors have a cache hierarchy designed to mitigate the pains of dealing with slow memory. Ryzen was an exception, with the Infinity Fabric running at the same speed as the DDR4 memory. I wondered if the improved cache and memory latencies would make DDR4 speed less of a factor, so I tested six 16GB kits ranging from DDR4-2133 CL15 to DDR43400 CL16, with several in between.
The results are as I expected. Applications that predominantly run from the CPU cache show very little change in performance, regardless of the RAM used. 3D rendering, data encryption/decryption, and even video encoding only saw 1-2 percent gains. Calculating pi digits with y-cruncher and data compression/ decompression show larger gains of 5-10 percent. What about games?
I tested 20 games, and the gains can be quite large. Dishonored 2,Fa rC ry5 ,K ingdomC ome: Deliverance, and Vermintide2 frame rates improved by up to 15 percent. But if we eliminate the lowbar DDR4-2133 CL15 (which no one should buy at this point, as DDR42666 CL15 sells for roughly the same price), the benefit is halved.
Overall, fast memory with tight timings improved gaming performance by 7.5 percent on average, compared to the baseline DDR4-2133 CL15. Move to DDR4-2400 CL15, and the difference is only 3.8 percent. DDR4-3400 CL16 ended up a bit slower than DDR4-3200 CL14, as expected—that’s a real-world latency of 8.75ns versus 9.41ns.
With the Ryzen 5 2400G APU, I tested gaming performance with the integrated Vega 11 graphics. The reduction in memory bandwidth has a far greater impact, with faster memory improving performance by 20 percent on average, and up to 33 percent in games such as Shadow
ofWar. If you’re thinking about building a budget gaming rig with a Ryzen APU, don’t skimp on RAM.
It’s a good thing that top-shelf memory is less of a requirement with the latest Ryzen CPUs, as RAM prices continue to be a sticking point. 16GB DDR4 kits could be had for $60 in 2016, with exotic kits at $100. The cheapest 16GB kits now cost $150—though it was $185 for a similar kit two months ago!
There are accusations of price fixing, with a class-action lawsuit in the works. We can only hope prices continue to drop, because the amount of DRAM going into smartphones, cars, PCs, graphics cards, and AI is only set to increase.