AMD FirePro Graphics Series Are Between 50% and 500% Faster Than NVIDIA Tesla ~ The World is Mine

Aug 7, 2012

AMD FirePro Graphics Series Are Between 50% and 500% Faster Than NVIDIA Tesla

8/07/2012 07:12:00 PM

dannzfay

Texas-based fabless CPU and GPU designer, American company AMD has just launched the new workstation graphic cards using the famous Graphics Core Next (GCN) architecture. The company calls these the “world’s most powerful workstation graphics line” and, seeing the numbers, we tend to agree.

AMD has simply stumbled onto a treasure with its GCN architecture. While the company was never so much involved with professional graphics and GPU compute like Nvidia, AMD’s first attempt to build a GPU compute-optimized design has been more successful than anybody would have expected. Looking at the numbers, the W9000 is able to push out 4TFLOPS single-precision IEEE 754-2011 and 1TFLOPS double-precision IEEE 754-2011. The card uses a Tahiti GPU made at TSMC in the foundry’s well-performing 28nm technology and has 4.3 billion transistors inside working with 6GB of GDDR5 memory on the usual 384-bit BUS. AMD’s W9000 is the company’s fastest and most powerful FirePro professional adapter and, if we compare it with whatever Nvidia has best in its lineup, we can’t help but be amazed.

Nvidia now has two flagships in its Tesla product line. One is the Fermi-based Tesla 2090 and the other is the new K10 dual-GPU professional Tesla card. The first one excels at double-precision floating point operations and is able to push out an impressive 665 Gigaflops in DP and 1.331 Teraflops in single-precision floating point calculations. These numbers put AMD’s W9000 50.3% ahead in DP FP64 and show it to be 300% as fast in single-precision floating point operations. The thing is that Nvidia’s newest Tesla adapter is the Kepler-based K10 model that features two full-0blowing Kepler GPUs inside and it is excelling at single-precision floating point operations.

The K10 is, indeed, 14% faster in single-precision floating point operations, but the new AMD FirePro W9000 shows an impressive 526% DP FP64 performance when compared with K10’s meager 190 Gigaflops performance. Many said that this might be a dual-GPU card from AMD and, although the company has not made any such information available, we tend to disagree and say that we believe AMD’s FirePro to be a single-GPU card. The $4000 (€3230) W9000 is not the only card launched today, as AMD is also offering new adapters for various price points.

The AMD FirePro W8000 comes with 4 GB of GDDR5 memory and is able to display 3.2 TFLOPS of single-precision floating point performance and 806 GFLOPS of double-precision floating point performance and that still makes it considerably faster than the Tesla 2090. Priced at $1600 (€1300), the W8000 is actually the competitor suited to fight the Fermi-based Tesla 2090 and the W9000 is clearly in a class of its own. The AMD FirePro W7000 comes with 4 GB of GDDR5 memory and is able to display 2.4 TFLOPS of single-precision floating point performance and 152 GFLOPS of double-precision floating point performance. This is the first card in AMD’s new lineup to display lower DP FP64 performance than Nvidia’s Tesla K10.

Considering single-point performance, the W7000 is still twice as powerful as the Tesla 2090. The smallest brother in the new FirePro generation is the AMD FirePro W5000, which comes with 2 GB of GDDR5 memory and is able to display 1.27 TFLOPS of single-precision floating point performance and 80 GFLOPS of double-precision floating point performance. The FirePro W7000 and FirePro W500 are priced at $900 (€730) and $600 (€490), respectively. AMD also has a premium server partner to integrate its new professional card, and that is the famous SuperMicro.