Nice
So, anyone remember the good old times when the now-lamented Mr. Saddam was rumored to buy up PS/2 en masse to create reality-bending weapons of mass destruction?
I want to go back.
AMD has upped the ante in its battle with Nvidia for the data center and high-performance computing (HPC) markets, with the launch of two new GPU cards for servers. The company's new big gun is the FirePro S9150 card, which maxes out at a blistering 5.07 TFLOPS peak single-precision floating-point performance and 2.53 TFLOPS …
From the specs tab on the website:
Cooling/Power/Form Factor-Max Power: 235W
-Bus Interface: PCIe® x16
-Slots: Two
-Form Factor: Full height/ Full length
-Cooling: Passive heat sink
Then below that:
System Requirements-20 CFM airflow cooling at 45° C maximum inlet temperature
Just curious. I notice some of the GPUs experience a 1:4 performance penalty when going from single- to double-precision, but many (bot notably not all) of the AMD GPUs manage to reduce the penalty to 1:2. Can anyone explain how they do this and why it isn't consistent across the board (unlike nVidia, whose GPUs seem to be consistently 1:4)?
GPUs are single precision beasts at their core. DP is achieved by 'double pumping' the SP paths. 1:2 suggests AMD has all the necessary plumbing in HW to do so efficiently (VERY efficiently.) Less than that suggests the shader compiler is doing all the heavy lifting, making the HW purely SP.
Bravo AMD, nice to see DP taken seriously now.