Not logged in » Login
Aug 11 2014

Heralding a Higher HPC Era: AMD Launches Server GPUs FirePro S9150 and S9050

AMD is expanding its portfolio of high-end GPGPU cards with the new flagship FirePro S9150 and its smaller cousin S9050. Designed to power simulations, signal processing, video transcoding and any kind of Big Data application, both cards can easily handle even the most compute-intensive tasks and massively speed up calculations while improving the utilization of predefined power budgets.

In its introductory press release, AMD bills the FirePro S9150 as "the most powerful server GPU ever built for High Performance Computing." Judging from the specs, it easily lives up to that promise by delivering a massive 5.07 TFLOPS of peak single-precision performance and – perhaps even more important – a whopping 2.53 TFLOPS of peak double-precision performance, thus becoming the first card to ever break the 2 TFLOPS double precision barrier. Moreover, at 10.8 GFLOPS per watt, it also offers a 77% better performance/watt ratio than competing products. To this end, the FirePro S9150 boasts 2,816 stream processors in 44 GCN compute units, 16GB of GDDR5 memory, and a 512-bit memory interface with up to 320 GB/s memory bandwidth. Equipped with AMD PowerTune technology, the S9150 may also perform real-time analysis of GPU utilization and automatically boost GPU clock speed by up to 30% if it finds that "an application is not making the most of the power available to the GPU." In terms of SPIs and operating systems, the card supports DirectX® 11.1, Open GL 4.3, and OpenCL™ 1.2 as well as the latest enterprise-ready releases of Windows, Windows Server, and Linux (32- and 64-bit editions). However, the record-breaking performance comes at a cost – despite its excellent GFLOPS per watt ratio, the FirePro S9150 isn't exactly a power saver and may eat up as much as 235 watts, which puts it on par with NVIDIA's Tesla K20X modules.

AMD's second new release, the FirePro S9050, features 1,792 stream processors, 12GB of GDDR5 memory, a 384-bit interface and up to 264 GB/s memory bandwidth. Given these specifications, it's not really a surprise the card turns in a considerably weaker performance than the flagship product, although 3.23 TFLOPS of peak single-precision and 806 GFLOPS of double-precision performance can't be considered slow by any standards and make it a solid contender for NVIDIA's Tesla K20 cards. API and OS support is the same as above, with support for WinXP and Vista added to the mix – which doesn't seem to make much sense in light of the fact that the former reached its EOL in April and that the latter has a mere 3 percent market share. The S9050's power consumption is only slightly lower and clocks in at 225 watts.

Both cards are available immediately, initially from AMD distribution partners and OEMs. Prices were not disclosed.


Comments on this article

No comments yet.

Please Login to leave a comment.


Please login

Please log in with your Fujitsu Partner Account.


» Forgot password

Register now

If you do not have a Fujitsu Partner Account, please register for a new account.

» Register now