 |
Q & A : Benchmarks : CPU vs. GPGPU: Arithmetic Performance |
What are they?
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.
Why do we measure it?
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.
What do the results mean?
- The arithmetic results are in pixels/s, i.e. how many pixels can be computed in 1 second.
- In all cases, as higher indexes mean better performance (pixels/s) the higher the result the better the performance.
Typical Single Floating-Point Results
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 2010, for US market, in USD, via JustRelevant and is provided as an example only. Please check prices in your own region.
| Rank | CPU Name | Cores / Speed / Power (TDP) / Price | Performance | Power Efficiency | Cost Efficiency | Comments | |
| #1 |  |
Intel Core i7 975 | 4C / 8T / 3.2GHz / 130W / 1040$ | 193 Mpixels/s | 1486 kpixels/W | 186 kpixels/$ | The best CPU if you want performace, but is it worth the cost? |
| #2 | |
Intel Core i5 760 | 4C / 2.8GHz / 125W / 205$ | 99 Mpixels/s | 1037 kpixels/W | 481 kpixels/$ | A very good performer, reasonably priced if you don’t want GPU solution. |
| #3 | |
Intel Core i3 550 | 2C / 4T / 3.2GHz / 73W / 130$ | 80 Mpixels/s | 1092 kpixels/W | 613 kpixels/$ | Efficiency at its best with a very good price among all CPUs. |
| #3 |  |
AMD Phenom X6 1055 | 6C / 2.8GHz / 125W / 180$ | 73 Mpixels/s | 582 kpixels/W | 404 kpixels/$ | A powerful AMD solution with a very attractive price and acceptable power efficiency. |
| #4 | |
AMD Phenom X4 970 | 4C / 3.5GHz / 125W / 185$ | 60 Mpixels/s | 478 kpixels/W | 323 kpixels/$ | It does not have extraordinary performance nor efficiency but it’s an alternative if you want a quad core. |
| #5 | |
AMD Phenom X3 720 | 3C / 2.8GHz / 95W / 148$ | 36.4 Mpixels/s | 383 kpixels/W | 246 kpixels/$ | Being the last in this test does not mean it’s a bad choice, especially if you consider the black edition. |
| Rank | GPU Name | Shaders / Speed / Power (TDP) / Price | Performance | Power Efficiency | Cost Efficiency | Comments | |
| #4 |  |
nVIDIA GeForce GTX 480 | 480 / 760MHz / 250W / 420$ | 1,798 Mpixels/s | 7192 kpixels/W | 4280 kpixels/$ | The best Fermi result in the test, however you may want to wait a price drop to consider it. |
| #7 | |
nVIDIA GeForce GTX 470 | 448 / 625MHz / 215W / 240$ | 1,675 Mpixels/s | 7790 kpixels/W | 6979 kpixels/$ | This middle range solution should be considered if you don’t want to spend too much. |
| #10 | |
nVIDIA GeForce GTX 460 | 336 / 675MHz / 150W / 140$ | 1,066 Mpixels/s | 7106 kpixels/W | 7614 kpixels/$ | The last among GPUs but not the least in this test, having a good market value and popularity. |
| #8 | |
nVIDIA GeForce GTX 570 | 480 / 1.46GHz / 219W / 350$ | 1,532 Mpixels/s | 6995 kpixels/W | 4377 kpixels/$ | Below average result from this chip with a refreshed Fermi arhitecture and the last at power efficiency I’m afraid. |
| #5 | |
nVIDIA GeForce GTX 580 | 512 / 1.54GHz / 244W / 510$ | 1,714 Mpixels/s | 7024 kpixels/W | 3360 kpixels/$ | The high end chip from the latest GeForce generation of boards does not offer astounding performance or efficiency. |
| #1 | |
AMD Radeon HD 5870 | 1600 / 850MHz / 188W / 250$ | 2,517 Mpixels/s | 13388 kpixels/W | 10068 kpixels/$ | If it’s about performance and power efficiency on single floating point, look no further. |
| #3 | |
AMD Radeon HD 6870 | 1120 / 900MHz / 150W / 240$ | 1,845 Mpixels/s | 12300 kpixels/W | 7687 kpixels/$ | Middle range AMD card which has all three characteristics evaluated here at top level. |
| #9 | |
AMD Radeon HD 5770 | 800 / 850MHz / 108W / 120$ | 1,325 Mpixels/s | 12268 kpixels/W | 11041 kpixels/$ | Best buy GPU solution if the cost is the most important to you, paired with good performance and power efficiency. |
| #2 | |
AMD Radeon HD 6970 | 1536 / 880MHz / 250W / 370$ | 1,880 Mpixels/s | 7520 kpixels/W | 5081 kpixels/$ | Interesting that the latest model from AMD it’s not even as efficient as previous generation, not to mention the performance. |
| #6 | |
AMD Radeon HD 6950 | 1408 / 800MHz / 200W / 300$ | 1,711 Mpixels/s | 8555 kpixels/W | 5703 kpixels/$ | Little difference between this and the XT version in the absolute way. But that puts it though a lot behind it in the chart, having in exchange a better energy and cost efficiency. |
You can check your own processor or video adapter or view how other processors/video
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 2010, for US market, in USD, via JustRelevant and is provided as an example only. Please check prices in your own region.
| Rank | CPU Name | Cores / Speed / Power (TDP) / Price | Performance | Power Efficiency | Cost Efficiency | Comments | |
| #2 | |
Intel Core i7 975 | 4C / 8T / 3.2GHz / 130W / 1040$ | 78 Mpixels/s | 598 kpixels/W | 75 kpixels/$ | A CPU on a par with GPUs also having the best power usage; unfortunately way too expensive by far. |
| #2 | |
Intel Core i5 760 (CPU) | 4C / 2.8GHz / 125W / 205$ | 50 Mpixels/s | 527 kpixels/W | 244 kpixels/$ | Acceptable output considering power drawn, but quite costly. |
| #3 | |
Intel Core i3 550 | 2C / 4T / 3.2GHz / 73W / 130$ | 30 Mpixels/s | 413 kpixels/W | 232 kpixels/$ | Even it’s the last as a performer, it has good power and cost efficency. |
| #1 | |
AMD Phenom X6 1055 | 6C / 2.8GHz / 125W / 180$ | 65 Mpixels/s | 522 kpixels/W | 363 kpixels/$ | The best performance among CPUs considering the cost and power consumption. |
| #2 | |
AMD Phenom X4 970 | 4C / 3.5GHz / 125W / 185$ | 54 Mpixels/s | 433 kpixels/W | 293 kpixels/$ | A good quad core performer reasonably priced. |
| #3 | |
AMD Phenom X3 720 | 3C / 2.8GHz / 95W / 148$ | 33 Mpixels/s | 344 kpixels/W | 221 kpixels/$ | Almost the last solution if you consider all the aspects of the evaluation. |
| Rank | GPU Name | Shaders / Speed / Power (TDP) / Price | Performance | Power Efficiency | Cost Efficiency | Comments | |
| #4 | |
nVIDIA GeForce GTX 480 | 480 / 1.4GHz / 250W / 420$ | 268 Mpixels/s | 1072 kpixels/W | 638 kpixels/$ | A fast solution, but this is an expensive graphics card and also it’s not so power efficient. |
| #6 | |
nVIDIA GeForce GTX 470 | 448 / 1.2GHz / 215W / 240$ | 250 Mpixels/s | 1162 kpixels/W | 1041 kpixels/$ | First generation Fermi chip with an average performance and efficiency combined with a good price. |
| #8 | |
nVIDIA GeForce GTX 460 | 336 / 1.35GHz / 150W / 140$ | 112 Mpixels/s | 746 kpixels/W | 800 kpixels/$ | A Compute 2.1 capable card that has decent performance giving its price on the market. |
| #7 | |
nVIDIA GeForce GTX 570 | 480 / 1.46GHz / 219W / 350$ | 229 Mpixels/s | 1045 kpixels/W | 654 kpixels/$ | Second generation Fermi design having less than expected results (only a quater of full Fermi double precision capability is enabled). |
| #5 | |
nVIDIA GeForce GTX 580 | 512 / 1.54GHz / 244W / 510$ | 257 Mpixels/s | 1053 kpixels/W | 503 kpixels/$ | High-end card from nVIDIA that costs accordingly given the fact that is the latest, but it does not offer here a result that is to impress. |
| #3 | |
AMD Radeon HD 5870 | 1600 / 850MHz / 188W / 250$ | 366 Mpixels/s | 1946 kpixels/W | 1464 kpixels/$ | Very good performance paired with the best possible efficiency on price and power consumption. |
| #9 | |
AMD Radeon HD 6870 | 1120 / 900MHz / 151W / 240$ | 60 Mpixels/s | 397 kpixels/W | 250 kpixels/$ | A new entry that lacks hardware double precision so it’s not to compare it directly with the other cards. |
| #10 | |
AMD Radeon HD 5770 | 800 / 850MHz / 108W / 120$ | 43 Mpixels/s | 398 kpixels/W | 358 kpixels/$ | Better efficiency in software emulation of double precision than the newer entry, unfortunately it’s the last here. |
| #1 | |
AMD Radeon HD 6970 | 1536 / 880MHz / 250W / 370$ | 414 Mpixels/s | 1656 kpixels/W | 1118 kpixels/$ | Astounding double precision performance from the AMD sustained with good use of cost and power resources. |
| #2 | |
AMD Radeon HD 6950 | 1408 / 800MHz / 200W / 300$ | 370 Mpixels/s | 1850 kpixels/W | 1233 kpixels/$ | Top efficiency and performance from the newly launched AMD code Cayman Pro chip. |
You can check your own processor or video adapter or view how other processors/video adapters measure up using the Sandra benchmarks.