2017 has been the best year for AMD since at least 2009-2010 and arguably the best for over a decade. The Ryzen CPU family has forced Intel to make significant adjustments to its own product stack, but so far, these changes have been confined to the desktop side of the market. Leaked benchmarks purportedly from an AMD Raven Ridge APU suggest the new CPU core is a vast improvement over Carrizo, though there are a few caveats to address before we tackle the leaks.
First, we don’t know if Geekbench is as optimized for Ryzen as it is for Intel; Ryzen having been on the market for much less time than Intel’s Skylake and Kaby Lake processors. Existing optimizations may need to be tweaked to address Raven Ridge’s smaller L3 cache (GB reports 4MB instead of 8MB).
Second, it’s possible that turbo modes aren’t implemented on pre-production hardware. It’s not uncommon for UEFI developers to push multiple UEFI updates out to reviewers as a product nears launch, and these often fix features like improper turbo clocks. Third, we don’t know the Ryzen 5 2500U’s TDP. Intel refers to CPUs with a “U” designation as being ultra-low power and typically keeps them to 15W or less, but while AMD is probably using similar nomenclature that’s obviously not confirmed.
With all of that in mind, we’ve put together a graph showing how AMD’s new Ryzen 5 2500U compares to the Carrizo-derived FX-9800P, the Intel Core i5-8250U, and the Core i7-8550U.
The Ryzen 5 2500U supposedly has a base clock of just 2.0GHz, which puts it on-par with the base clock on the 8550U. We don’t know how high these chips will turbo or how long they can hold their turbo clocks, but one thing is immediately clear. The 2500U beats the stuffing out of the Carrizo / Bristol Ridge-derived 9800P, despite the fact that AMD’s earlier chip runs a 2.7GHz base clock; 1.35x higher than AMD’s newer core. Ryzen 5 is 1.48x faster than Carrizo in single-threaded performance and 1.83x faster than Carrizo in multi-threaded performance.
The scaling factor between the three chips also shows significant improvement for AMD. The 9800P is just 2.14x faster in Geekbench 4’s multi-threading test compared to its single-threaded performance. This implies the CPU is severely power-limited when attempting to use more than one core and must scale back its own clock rate as a result. Part of the discrepancy is caused by the scaling performance penalty intrinsic to Bulldozer’s design, but not that much of it. Excavator’s multi-core scaling penalty was better than Steamroller, and Steamroller bested the old Piledriver architecture’s average of 1.8x scaling as compared to standard 1.9-2x with traditional dual-core architectures.
Ryzen 5 2500U’s scaling factor is 2.64x — not as strong as Intel’s 3.03, but significantly better than anything the company has fielded in recent years.
Intel still sweeps these tests overall, but we’ll wait on TDP values and production-ready UEFI to draw firm conclusions on the relative performance of the chip (I prefer real-world application tests to synthetic metrics in any case). Graphics performance will also play a significant role in how the two chips compare to each other; AMD’s Vega may draw significant amounts of power at higher clocks and core counts, but as we’ve seen with cards like the Radeon Nano, there’s not a linear relationship between performance and TDP. Knocking 10% off a GPU’s performance can cut its power consumption by as much as a third, which means Vega could be far more competitive in an APU in terms of both performance and performance per watt.
Even assuming these results are accurate, they still represent a distinct improvement for AMD. When Intel’s Core family offered very nearly 3x the performance of AMD’s best APU, it was quite difficult to recommend the latter. Ryzen 5 slashes that gap dramatically. Compared against a lower-end Core i5-8250U, the gap is even smaller. If AMD’s Vega opens a lead between itself and Intel’s integrated graphics, it could make the company’s hardware a viable choice for those who would rather trade some CPU performance for better graphics and gaming.
I’m optimistic about AMD’s long-term mobile chances for much the same reason I’m optimistic about its server prospects. AMD’s mobile market share may not have been as low as its server share, which had fallen to below 1%, but it’s more than low enough. AMD had to choose where to focus its efforts, and it decided to emphasize the consumer desktop and server markets first. With its limited RD budgets, the company has had to make tough choices about where to focus its limited funds.
AMD will announce its Q3 2017 results later today, and I expect them to be fairly strong. The purpose of these first-generation Ryzen parts is to put the company back on the board in markets where it hasn’t been remotely competitive in the recent past. If AMD continues to execute well it should be able to shrink the gap between itself and Intel in many areas, and offer compelling performance per dollar in many others.