Q & A : Benchmarks : CPU vs. GPGPU: Arithmetic Performance

A set of benchmarks designed to measure arithmetic performance, i.e. how fast it can crunch numbers. These are native ports of the traditional CPU benchmarks that have been available in Sandra since 1995. They allow us to directly compare CPU with GPGPU performance by using the same algorithms and the same data.

We believe the industry is seeing a shift from a model where the vast majority of workload is processed on the traditional CPU: in a wide range of applications developers are using the power of GPGPU to aid business analysis, games, graphics, and scientific applications.

Coupled with the charts added to the latest version of the software, we can work out whether a CPU or GPU would be faster, more power efficient or cost efficient.

1. The arithmetic results are in pixels/s, i.e. how many pixels can be computed in 1 second.
2. In all cases, as higher indexes mean better performance (pixels/s) the higher the result the better the performance.

Testing the single floating-point (32-bit float) performance of various current desktop processors and GPGPU-capable video adapters reveals quite interesting results.

Note: Prices fluctuate all the time; the below table was correct as of December 2008, for US market, in USD, via JustRelevant and is provided as an example only. Please check prices in your own region.

Device Name Cores / Speed / Power (TDP) Performance Power Efficiency Cost Efficiency Comments AMD Phenom 9850 (CPU) 4 / 2.5GHz / 125W 68 Mpixels/s 544 kpixels/W 404 kpixels/$ Excellent cost efficiency due to aggressive pricing, but last in raw power and power efficiency. AMD Phenom II 940 (CPU) 4 / 3GHz / 125W 85 Mpixels/s 680 kpixels/W 363 kpixels/$ Excellent cost and power efficiency, better than Core 2 but not quite reaching Core i7. Intel Core 2 QX9650 (CPU) 4 / 3GHz / 130W 92.8 Mpixels/s 713 kpixels/W 93 kpixels/$ Excellent raw performance and power efficiency, too expensive and now replaced by Core i7. Intel Core i7 965 (CPU) 4+4 / 3.2GHz / 130W 125 Mpixels/s 961 kpixels/W 125 kpixels/$ Best raw power for a CPU, good power efficiency but far too expensive compared with competitive solutions. Competitive with most GPGPUs but the very latest! ATI Radeon 3850 (GPU) 320 / 669MHz / 104W 152 Mpixels/s 1461 kpixels/W 1688 kpixels/$ Competitive in both power and cost efficiency, though not the fastest; superseded now. ATI Radeon 4870 (GPU) 800 / 750MHz / 160W 431 Mpixels/s 2693 kpixels/w 2873 kpixels/$ Just a little bit faster than its competition (GT200), with better power/cost efficiency. ATI Radeon 5870 (GPU) New! 1600 / 850MHz / 188W 912 Mpixels/s 4851 kpixels/w 2406 kpixels/$ Pummels everything into dust with fantastic performance, power and cost efficiency. The very best! nVidia GeForce 9600 GT (GPU) 64 / 1.6GHz / 95W 194 Mpixels/s 2042 kpixels/W 1940 kpixels/$ Faster than its rivals (2XXX and 3XXX series) with great power and cost efficiency, though bettered by newer cards. nVidia GeForce 260 GTX (GPU) 216 / 1.3GHz / 182W 426 Mpixels/s 2340 kpixels/W 2302 kpixels/$ Great performance though bettered on power and cost by the competition

You can check your own processor or video adapter or view how other processors/video adapters measure up using the Sandra benchmarks.

Most Popular Processors, Video Adapters

Most popular Processors as benchmarked by users (past 30 days):		Most popular Video Cards as benchmarked by users (past 30 days):
1. 20% Intel Core i7 920 269.99 USD 2. 17% Intel Pentium 4 88.00 USD 3. 5% Intel Core i5 750 194.99 USD 4. 5% Intel Core i5 670 294.78 USD 5. 3%   Intel Core 2 Quad Q9450		1. 12%   NVIDIA GeForce 8800 GT   2. 12%   NVIDIA GeForce GTX 295   3. 12% ATI Radeon HD 4870 X2 451.99 USD 4. 12% NVIDIA GeForce GTX 260 202.99 USD 5. 8% ATI Radeon HD 4800 174.99 USD
For a complete list of statistics, check out the Most Popular Hardware page. For a list of more products, see SiSoftware Shopping.

Typical Double Floating-Point Results

Not all GPGPUs support 64-bit floating-point numbers (doubles) which means higher precision results have to be emulated which is very computationally intensive that yields somewhat different results.

Note: Prices fluctuate all the time; the below table was correct as of December 2008, for US market, in USD, via JustRelevant and is provided as an example only. Please check prices in your own region.

Device Name Cores / Speed / Power (TDP) Performance Power Efficiency Cost Efficiency Comments AMD Phenom 9850 (CPU) 4 / 2.5GHz / 125W 37.7 Mpixels/s 301 kpixels/W 224 kpixels/$ Excellent cost efficiency due to aggressive pricing, similar in performance with GPGPUs supporting single-precision only. AMD Phenom II 940 (CPU) 4 / 3GHz / 125W 46.5 Mpixels/s 372 kpixels/W 198 kpixels/$ Better than the original Phenom and Core 2, not quite catching with Core i7 yet. Intel Core 2 QX9650 (CPU) 4 / 3GHz / 130W 47.5 Mpixels/s 365 kpixels/W 48 kpixels/$ Excellent raw power, good power efficiency but far too expensive compared with competitive solutions. Intel Core i7 965 (CPU) 4+4 / 3.2GHz / 130W 68.7 Mpixels/s 528 kpixels/W 69 kpixels/$ Best raw power for a CPU, good power efficiency but far too expensive compared with competitive solutions. Competitive with most GPGPUs but the very latest! ATI Radeon 3850 (GPU) 320 / 669MHz / 104W 72 Mpixels/s 692 kpixels/W 800 kpixels/$ Faster than all (but the very latest) processors and competitive in both power and cost efficiency. Superseded by the 48xx series. ATI Radeon 4870 (GPU) 800 / 750MHz / 160W 213 Mpixels/s 1331 kpixels/W 1420 kpixels/$ Excellent raw performance in double-prevision, until superseded recently; still more cost efficient as much cheaper. ATI Radeon 5870 (GPU) New! 1600 / 850MHz / 188W 459 Mpixels/s 2441 kpixels/W 1211 kpixels/$ Best raw performance, best power and cost efficiency, the fastest solution. If double-precision is required, this is the card to pick. nVidia GeForce 9600 GT (GPU) 64 / 1.6GHz / 95W 14.5 Mpixels/s 152 kpixels/W 145 kpixels/$ The lack of native double-precision support really hurts its performance, going from 1st to last. Only suitable for single-precision work. nVidia GeForce 260 GTX (GPU) 192 / 1.3GHz / 182W 52 Mpixels/s 285 kpixels/W 281 kpixels/$ Big performance hit on double-precision makes it lose its crown to the competition; recent price drop has helped.

You can check your own processor or video adapter or view how other processors/video adapters measure up using the Sandra benchmarks.

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

	Good article
	Bad article
	Back to Q & A