"Ever since AMD launched its first APU family, it’s been clear that faster RAM had a significant impact on performance. With the Ryzen 5 2400G, the potential impact of faster DDR4 could be even more significant. The CPU-GPU interconnect system uses AMD’s Infinity Fabric as well, and Infinity Fabric’s clock is directly tied to DRAM clock.
We decided to put the Ryzen 5 2400G to work at both DDR4-2133 and DDR4-3200 to see how much performance improved from using faster RAM. Normally, such a shift would also incur a hefty financial penalty, since RAM is typically priced by performance binning. In any given era, you’ll find a fairly flat area under the price curve — in this case, we would expect the slowest DDR4 (2133) to be priced quite similarly to moderately quicker RAM, like DDR4-2667 or DDR4-2933. Then the price curve would bend upwards at an increasing rate. If DDR4-3200 was 15 percent more expensive than DDR4-2667, we’d expect DDR4-3866 to be 30-40 percent more expensive than DDR4-3200 (all numbers are approximate).
But with the DRAM market overheated and expected to stay that way, we’ve seen some price compression around DRAM clocks. That’s not to say DDR4-3200 isn’t more expensive than DDR4-2133, but a pair of 8GB Crucial sticks will set you back $165, while a pair of DDR4-3200 G.Skill DIMMs will run you $185. That’s only about a 12 percent price difference, which isn’t much — if the Ryzen 5 2400G can use that additional performance.
We’ve run through the same suite of game tests that we used in our Ryzen 5 2400G review, only this time we’ve rerun them at DDR4-2133 clocks as well as DDR4-3200.
"In aggregate, we see very significant improvements between the two RAM clocks. In some cases, our minimum frame rates literally double. Even in the situations where this does not occur, the improvement is substantial. Average frame rates improved by 1.2x by shifting from DDR4-2133 to DDR4-3200, while minimum frame rates increased by 1.37x.
This is far better DRAM scaling than you’d expect to see in all but the most bandwidth-constrained scenarios if we were examining CPU performance. The gains are more than enough to visibly impact game performance. The typical rule of thumb is that people start to notice a performance gap of 10 percent, and DDR4-3200 improves the Ryzen 5 2400G’s performance on average by 2x that goal. Keeping high minimum frame rates is also important to maintaining strong performance, and those rates also shift upwards by an even larger amount.
Given the compressed nature of the DRAM stack and the sky-high prices on dGPUs, if you’re looking to the Ryzen 5 2400G’s GPU as a pit stop on your way to something better, I’d go ahead and buy the faster RAM. The overall performance improvement from doing so is much better than the additional cost, given current market conditions."
DDR4 Scaling and the Ryzen 5 2400G: Do Higher Clocks Boost Performance? - ExtremeTech
Know what I think? The 7nm Zen 2 generation will finally have enough space for HBM2/3 modules on the APU itself and finally get rid of the RAM factor. It's already being done with Vega M GL, but those are too big to fit into Socket AM4. I have a feeling AMD could have done it with this generation, but between the already high TDP and the cost and shortage of HBM2 requiring their allocation to crypocurrency cards and Intel Vega M GL chips, something had to give.
Actually, I wonder if any of the guys at AMD who came up with Threadripper have thought up a prototype Threadripper APU with the two stabilizers replaced by a graphics core and HBM2 stack, all tied together by Infinity Fabric..