by Anton Shilov
04/01/2013 | 05:18 PM
UPDATE: Adding more explanations from AMD.
Contemporary microprocessors from AMD and Intel can dynamically change their clock-speeds in order to reduce power consumption or improve performance. Still, there are certain clock-speeds that are declared to be default and at which the chips should operate most of the time in all applications. However, it looks like under certain conditions AMD’s latest microprocessors for desktops run at lower clock-speeds, which negatively affects performance.
AMD’s latest A-series accelerated processing units code-named Trinity have default clock-speeds, idle clock-speeds as well as maximum clock-speeds that are activated only in certain cases and when the Turbo Core technology in enabled. However, it was discovered in the “Contemporary APUs: AMD Trinity vs. Intel Ivy Bridge” article, under high multi-threaded loads, Trinity drops clock-speed below base levels. For example, AMD A10-5800K APU has a base clock rate of 3.8GHz and capable of turbo-boosting to 4.2GHz, however, at high multithreaded loads the clock-speed is dropped to 3.40GHz, not to 3.80GHz as the processor’s TDP exceeds the default level. Other Trinity microprocessors behave in similar way.
During the test of AMD Trinity APUs, X-bit labs observed stable drop of clock-speed in Linpack benchmark and irregular drops in various video encoding apps, 3D renderers, etc. It is interesting to noted that the frequency never drops when Turbo Core is disabled and the processor operates at 3.80GHz all the time. As it turns out, in certain cases AMD’s chips deliver lower performance than they are supposed to.
AMD appears to know about the decrease of clock-speed below the base level and claims that it occurs in very rare cases under loads that are not typical for client microprocessors.
“Linpack is one of the few synthetic applications that may exceed the de-rating defined for our TDP. In the synthetic instances where base frequency causes our APU to run above its TDP, the part will throttle down to a frequency below base. Similar to Intel, thermal design power (TDP) is a realistic power target for partners to use in order to provide the best balance between cost and performance. However, as stated above, some unusual and synthetic workloads can cause a thermal event where our APU briefly drops below the base frequency,” said Peter Amos, a spokesman for AMD.
The world’s second largest maker of PC microprocessor also does not consider the drop of clock-speed below the officially stated level a problem, but believes it is a good design decision.
“Trinity-based APUs use workload calculations to set frequency and protect the processor. There are some benchmarking apps that trigger the extreme of those calculations. In this case the APU hits the max for a split second, reduces frequency before jumping up, and repeats this pattern until temps improve. When TDP is exceeded, graceful clock throttling via the Turbo Core technology is a designed defense to protect the processor. By design, the clocks may be throttled even below the base frequency when worst-case high-power applications (non-typical to market segment e.g. HPC app in this case) are run. In other words, base frequency is designed such that it is sustainable (i.e. APU operating at or above base) for most, if not all, typical applications relevant for the market segment. […] It would be a problem and unfair if these “tests” ended up ruining the life span of our customers’ products,” concluded Mr. Amos.