AMD launches mighty server/supercomputer GPU with 16 GB GDDR5 VRAM

AMD logoSo far, the High Performance Computing stage has mostly been ruled by NVIDIA when it came to GPU compute acceleration, but Advanced Micro Devices has been doing its best to catch up, and it seems it just managed it.

And it isn't even holding onto the relatively unknown FireStream brand anymore. Even though that's the company's brand for stream processing and/or GPGPU in supercomputers, the company chose to sell its new server GPU under the FirePro brand instead.

In fact, AMD even says that the new product, the FirePro S9150 server GPU, is the “first AMD architecture designed specifically with compute workloads in mind.”

AMD FirePro S9150

Looking at its hardware specs, we have to say that the newcomer definitely has what it takes to compete in the same league as NVIDIA's Tesla series.

And that means a lot, seeing as how Tesla GPU compute accelerators power some of the most powerful supercomputers on the Top500 list.

Anyway, the new AMD FirePro S9150 server GPU has 2,813 stream processors in 44 GCN compute units (graphics core next), as well as 16 GB of GDDR5 backing it up through an interface of 512 bits.

The memory bandwidth is, thus, of 320 GB/s, while the power consumption can go as high as 235W (which isn't as high as it could have been). In fact, it is the same as the 235W of NVIDIA's best Tesla card, the Tesla K40.

The peak single-precision floating point performance is of 5.07 TFLOPS, 18% more than the Tesla K40, which has 4.29 TFLOPs. Meanwhile, double-precision floating point performance is of 2.53 TFLOPs vs. 1.43 TFLOPs. Clearly, AMD is playing for keeps here.

There's a second HPC FirePro card from AMD, called AMD FirePro S9050. This one has 1,792 stream processors divided into 28 GCN compute units, plus 12 GB VRAM, a 384-bit memory interface, and a bandwidth of up to 264 GB/s. The power draw is of 225W.

Both newcomers utilize the PCI Express x16 card form factor and have passive cooling solutions. They also support AMD STREAM technology, which actually lets a server/supercomputer divert parallel processing tasks from the CPU (or CPUs) to the GPUs. Basically, it's the answer to NVIDIA's CUDA.

With the launch of the FirePro S9150 and S9050 server GPUs, AMD has reaffirmed the long-lasting rivalry between it (or rather what used to be ATI Technologies) and NVIDIA. Sapphire will be the main distribution partner, with ASUS, GIGABYTE, Supermicro, and TYAN making compatible motherboards starting this quarter (Q3, 2014).

Source: Softpedia

Tags: AMD, servers, videocards

Add comment

Your name:
Sign in with:
Your comment:

Enter code:

E-mail (not required)
E-mail will not be disclosed to the third party

Last news

Apple has made to the iPhone 7/7 Plus is by making the home button a solid state button
Links to quoted tweets will also not be counted against the 140 characters
You’re not supposed to expose the iPhone to water anyway
Microsoft only planning for a launch by Holiday 2017
The announced the Hero 5 Black and Hero 5 Session 4K cameras
It has been estimated that the cost to Samsung of recalling and replacing the defective Samsung Galaxy Note 7 units
The Pixel models may represent an end to Google's cut-rate Android phone line
New patent hints at upcoming tech for Microsoft phones
Samsung Galaxy TabPro S - a tablet with the Windows-keyboard
The first Windows-tablet with the 12-inch display Super AMOLED
June 7, 2016 /
Keyboards for iOS
Ten iOS keyboards review
July 18, 2015 /
Samsung E1200 Mobile Phone Review
A cheap phone with a good screen
March 8, 2015 / 4
Creative Sound Blaster Z sound card review
Good sound for those who are not satisfied with the onboard solution
September 25, 2014 / 2
Samsung Galaxy Gear: Smartwatch at High Price
The first smartwatch from Samsung - almost a smartphone with a small body
December 19, 2013 /
HP Slate 7 is a 7-inch Android 4 Tablet PC with good sound
A cost-effective, 7-inch tablet PC from a renowned manufacturer
October 25, 2013 / 4

News Archive



Do you like Windows 10?
or leave your own version in comments (32)