NVIDIA Vs AMD : Which GPU Is Better For Server-side Parallel Processing

Server-side parallel processing using GPUs is more commonly known as High Performance Computing or HPC. By utilizing millions of GPU cores linked together HPC actually computes huge amounts of raw data better and faster than the CPU. This is particularly useful in calculating scientific research and analysis along with the growing amounts of Big Data.

NVIDIA and AMD have been the traditional providers (and rivals) in this field but Intel has also recently announced their HPC solution to the world as well. While Intel’s HPC offering does offer some interesting features (such as easier coding and higher memory), in terms of raw power it still has a ways to go before it can match either NVIDIA or AMD. As such our comparison only covers the latest HPC developments from NVIDIA and AMD with a blow by blow account of each solution.

Latest HPC Offerings
NVIDIA’s Kepler based HPC offering is named the K20X and utilizes NVIDIA’s Tesla GPU technology to accelerate the computational processing of HPC data. It runs in a 235 watt package at 1.3 teraflops for 5.6 gigaflops per watt. AMD’s recently released FirePro S10000 card on the other hand uses two Tahiti GPUs for 1.48 teraflops in a 375 watt package for 3.9 gigaflops per watt.

For single precision FP, the AMD FirePro S10000 delivers 15.8 gigaflops per watt while the NVIDIA Tesla K20X delivers 16.8 gigaflops per watt. In terms of memory, the K20X provides 6GB and 250GB per second while the FirePro S10000 delivers the same 6GB at 450GB per second. Both platforms were tested without any error correction so these results may vary after the following processing. Again, both platforms require the computing to be processed in such a way as to make the GPU cores able to process them, and both are equally difficult and complex code for.

Crunching the numbers
A look at the raw numbers shows that NVIDIA’s Kepler based K20X provides much better energy efficiency than the AMD FirePro S10000 card. However, in terms of sheer FLOPs processing power the AMD FirePro S10000 card is slightly higher. In terms of bandwidth though, the AMD FirePro S10000 has nearly double that of the NVIDIA K20X offering

Conclusion
Overall, both NVIDIA and AMD have some impressive hardware to show off in the HPC field. On one hand you have the new Kepler designed NVIDIA Tesla K20X which is more energy efficient and has plenty of potential to grow, while on the other hand you have the AMD FirePro S10000 which delivers in the sheer brute force aspects of HPC processing but draws a significant amount of power to do so.

Since the HPC field isn’t that concerned with energy efficiency AMD’s better overall computational processing power is a better HPC solution than Kepler’s more efficient gigaflop per watt performance, for now. However, this conclusion is only for the short term as each company may very well come up with a solution to each of their issues very soon. Or Intel’s HPC solution could make a sudden uprising and beat both AMD and NVIDIA to the punch. Only time will tell for sure.

Jim Bibby is a consultant and a PC Technician for PC and MACs Computers, tablets and Servers. As an author his training  articles have been published regarding  hardware and software products and A+ training news. This article is courtesy of netwind learning center.