CPU Benchmarks: Synthetic
With the graphics-heavy metrics behind us, let’s break new ground with CPU benchmarks. We’ll start with the synthetics, and PCMark Vantage is first in line:
PCMark’s graphics component is pretty much the only taxing 3D workload in this benchmark, so it’s interesting to see that the A8-3500M appears to perform so closely to the Phenom II X4 at 1.5 GHz. With the GPU unencumbered, we would have hoped there would be room in the chip’s thermal budget for Turbo Core to accelerate the processing cores as close to 2.4 GHz as possible.
Unfortunately, it’s obvious that the 2.4 GHz Phenom II X4 is running significantly faster than the A8-3500M. Even when AMD’s APU is paired with the discrete Radeon HD 6630M to remove the graphics load entirely, Turbo Core doesn’t do much of anything.
To be clear, the A8-3500M has the lowest thermal ceiling in the quad-core Llano notebook-oriented lineup. Still, with the graphics subsystem unloaded, we’re disappointed that Turbo Core appears to be more of a marketing gimmick than a feature with teeth. In comparison, it’s clear when Intel’s Turbo Boost technology is affecting performance—verifiable by the tool Intel provides for monitoring the feature in action.
Intel’s Core i5-2520M is making up for all of its last-place finishes in the game testing, with a sizable lead over the A8-3500M. We do question the applicability of PCMark’s gaming test score after seeing how badly the Sandy Bridge chip was beaten in actual games.
In Sandra’s Arithmetic bench, the A8-3500M performs on par with the 1.5 GHz Phenom II X4. The Core i5-2520M lands closer to AMD’s 2.4 GHz Phenom II, but with iSSE4.2 extensions enabled, Intel’s CPU takes a huge lead.
The situation is the same in the Multimedia benchmark, except that the Core i5 is unleashed with AVX extensions.
The Cryptography result is similar, except that the Core i5-2520M literally jumps off the charts thanks to its AES-NI support, yielding a 10x performance lead in AES256-ECB hashing bandwidth!
GPGPU/GPCPU Processing
Let’s take a look at OpenCL and Direct3D compute capabilities that can leverage Llano’s CPU and GPU resources at the same time.
Intel only recently released an SDK to support OpenCL, but it’s limited to the CPU and cannot be run on the HD Graphics execution units. Intel’s GPU can run DirectCompute shaders though, so we can test this aspect of its feature set.
The A8-3500M APU demonstrates a colossal advantage over both Intel’s Core i5-2520M OpenCL result and the Intel HD Graphics 3000GPU DirectCompute score. But hold on; let’s look at the GPGPU/CPGPU Cryptography benchmark:
The Core i5-2520M CPU achieves parity with the A8-3500M APU in this discipline, and Intel’s GPU isn’t even contributing here! The bottom two results show Radeon HD 5570 DirectCompute versus Intel HD Graphics 3000 DirectCompute scores.
Finally, let’s consider memory bandwidth. As you can see, the A8-3500M suffers due to its low clock, while a 2400 MHz Phenom II X4 does much better, approaching the Core i5-2520M.