Benchmarks : Desktop GPAPU Performance: AMD "Llano" APU (CPU+GPU)

What is it?

In the previous article we analysed the performance of newly developed APUs from AMD in the Tablet/MID segment (see GPAPU Performance: AMD "Brazos" APU (CPU+GPU)), mobile (see GPAPU Performance: AMD "Llano" APU (CPU+GPU)) and this time we test the desktop market variant, called "Llano".

While it is a new design, the parts are from the past and present current lines from AMD. The APU consists of two or four CPU cores by a refreshed “Stars” K8F10 architecture, a "Sumo" GPU based on the “Redwood” core (Radeon HD5xxx family) and a memory controller with a dual dual-channel 128-bit DDR3 memory interface for both CPU and GPU.

"GPAPU" is General Processing (GP) using both the GPU (GPGPU) and the CPU (GPCPU) concurrently. We have worked to add support for APUs to all our GP benchmarks (Processing, Cryptography, Bandwidth) in Sandra 2011.

What are we measuring?

We are measuring the improvements in performance, power efficiency and cost efficiency ('bang-per-buck') of APUs compared to traditional discrete CPU + discrete GPU (if any) to end users. From a user’s point of view, it’s not important which chip in his system does the job, as long as they have to wait less, uses less power and costs less.

Thus we are measure GP performance (GPGPU, GPAPU) against native performance as well as software VM (.NET/Java):

• GPAPU/GPGPU Processing vs. Native Multi-Media Processing vs. .NET Multimedia Processing performance.
• GPAPU/GPGPU Cryptography vs. Native Cryptography performance.
• GPAPU/GPGPU Bandwidth vs. Native Memory Bandwidth.

Hardware Specifications

We are comparing an AMD A-Series "Llano" desktop APU against Intel Core i5 2400 series.

Standard Processing Performance

We are testing native and software VM processing performance of the processors themselves, i.e. not using (GP)GPU - aka "traditional" benchmarking.

Results Interpretation: Higher values (GOPS) mean better performance.

Base 10 Multipliers: 1GOPS = 1000MOPS, 1MOPS = 1000kOPS, etc.

Environment: Windows 7 x64 SP1, latest AMD Catalyst and Intel drivers.

CPU Processing Benchmarks		AMD A8-3850 (incl. Radeon HD 6550D GPU)	Intel Core i5 2400 (incl. HD 2000 GFX)	Comments
	Native Dhrystone/Whetstone (GOPS)	38.68 (41% lower)	65	This test shows a more than expected difference in performance between "Sandy Bridge" and the upgraded “Stars” architecture, suggesting that a major redesign would have been a better approach.
	.Net Dhrystone/Whetstone (GOPS)	13.07 (38% lower)	20.99	The VM environment is a little more favourable to the AMD chip, though the gap cannot be ignored.

Multi-Media Performance

We test raw native multi-media (SIMD) performance using any of the supported instruction sets (AVX, SSE, etc.) against GPAPU/GPGPU using any of the supported interfaces (OpenCL, DirectX ComputeShader/DirectCompute, CUDA).

Results Interpretation: Higher values (MPix/s) mean better performance.

Base 10 Multipliers: 1MPix/s = 1000kPix/s, 1kPix/s = 1000Pix/s, etc.

Environment: Windows 7 x64 SP1, latest AMD Catalyst and Intel drivers.

Multi-Media Benchmarks		AMD A8-3850 (incl. Radeon HD 6550D)	Intel Core i5 2400 (incl. HD 2000)	Comments
	GPAPU (CPU+GPU) Processing (MPix/s)	133.61 (25% higher)	-	Taken separately, neither the CPU nor the GPU from "Llano" cannot compete with "Sandy Bridge" and AVX. The GPU does close the gap somewhat but only when tested as an APU the power of AMD’s design ingenuity is revealed: the result is well above Intel’s new CPU even with AVX.
	GPGPU Processing (MPix/s)	95.21 (11% lower)	-
	Native Multi-Media (MPix/s)	63.68 (41% lower)	107 (baseline)
	.Net Multi-Media (Mpix/s)	15.25 (21% lower)	19.49 (baseline)	Again, in a virtual environment the AMD chip performs better than in the native one, still, to overtake Intel it needs help from the GPU.

Cryptographic Performance

We test cryptographic performance of the strongest common algorithms (AES256, SHA256) performance using any of the supported instruction sets (AVX, SSE, etc.) against GPAPU/GPGPU using any of the supported interfaces (OpenCL, DirectX ComputeShader/DirectCompute, CUDA).

Results Interpretation: Higher values (MB/s) mean better performance.

Base 2 Multipliers: 1MB/s = 1024kB/s, 1kB/s = 1024bytes/s, 1byte = 8bits, etc.

Environment: Windows 7 x64 SP1, latest AMD Catalyst and Intel drivers.

Cryptography Benchmarks		AMD A8-3850 (incl. Radeon HD 6550D GPU)	Intel Core i5 2400 (incl. HD 2000 GFX)	Comments
	GPAPU (CPU+GPU) (MB/s)	1535 (28% lower)	-	Having implemented hardware acceleration for AES and with help from AVX, "Sandy Bridge" is hard to beat. However, in the APU configuration, "Llano" manages an impressive leap in performance (almost 3x CPU only!). Very impressive.
	GPGPU (MB/s)	1231 (42% lower)	-
	CPU (MB/s)	585 (73% lower)	2127 (baseline)

Efficiencies

Because not all things in life are evaluated to their true value, the next measurements will take into consideration various efficiency aspects:

Live Results @ SiSoftware Live Ranker

• AMD A8-3850 APU Radeon HD Graphics
• AMD A6-3650 APU Radeon HD Graphics
• AMD A4-3400 APU Radeon HD Graphics

Related Articles

• AMD "Bulldozer" CPU/APU
• AMD Mobile "Llano" APU (CPU+GPU)
• AMD "Brazos" APU (CPU+GPU)

Final Thoughts / Conclusions

The benchmarks show the AMD approach to the once highly profitable desktop market is refreshing. However, is not enough to compete fully against "Sandy Bridge" itself a significant improvement over older "Bloomfield"/"Westmere" designs (see Intel Sandy Bridge CPU/GPU: AVX Performance): against these older chips it would have won if it came earlier.

AMD placed bet on the multimedia and won, but still lacks brand-new features (AVX, AES*) that bring large gains in performance: these will be added in their next iteration named "Bulldozer".

The APU is already much cheaper than the competition and mainboard prices are also generally cheaper even when providing the same features - making it a great buy overall.

* Note: Some Sandy Bridge CPUs (e.g. i3) have features disabled (e.g. AES, Turbo, Hyper-Threading, even AVX!) making low-end AMD chips an even better buy.

Please let us know what you thought of this article by voting using the icons/links below. Thank you for reading.