AMD Debuts New Chips and Boards
August 7, 2012, Siggraph, Los Angeles—We talked with Adam Glick from AMD about their accelerated processor products. The new integrated processors can accelerate the workflows for graphics workers by offering high performance at low power. AMD is positioning the latest chip as the first part in this next generation of computing.
The entry-level workstation platform, the A300 FirePro APU uses a 4 core x86 plus a GPU and can support up to 2 GB in the frame buffer. It supports the latest open APIs and out performs the discrete graphics competition performance in most benchmarks. Due to the performance and reliability levels of the APU, it has been certified for use with the main creation software tools like Autodesk and many other software packages.
The integration of the base x86 cores with graphics acceleration enhances performance and capabilities without increasing power. The low-end A300 unit draws 65W, and the high-end A320 only 100W. The chip supports the latest interface standards like USB 3.0 and 6 Gb SATA.
In comparison with the Intel Xeon chips, they deliver better results with Specviewperf 11 and other tests benchmark numbers in similar configurations using 16 GB DDR-3 1600 memory and a 1GB graphics frame buffer. In engineering design and analysis jobs, the AMD chip outperformed the competition again. Using Sony Vegas as a rendering test, the CPU/GPU combination outperformed a discrete V3900 GPU-A320 CPU combination.
In addition to the APU, they also released new FirePro graphics boards. Simon Vivera described the “Southern Island” class of GPUs as new architectures that move the graphics cores into the next compute focus areas, called Graphics Core Next.
Although the professional graphics market is dwarfed by the consumer products in unit volumes, the revenues are close to a third of the $4B graphics TAM. The workstation business unit is working to develop a good product mix for key vertical markets. They are getting support from the software vendors and have certifications from many software vendors for graphics and GPU accelerated computation on their tools.
The GPUs have evolved from fixed functions machines to simple shaders, and now to graphics parallel cores. The latest chips and boards work to balance workloads and combine compute and render functions for greater speed and efficiency. The heterogeneous computing systems help to address the performance limitations due to power constraints without needing to resort to higher clock frequencies.
The resulting products offer scalable performance at reasonable power levels. Compared to the previous generation V7900 boards, they show between 1.5 and 4.5 times better performance. The results for very large, highly textured images demonstrate that the new engines can handle very large data sets like textures up to 32 TB ( 16k x 16k x 8k x 128-bits). The GPUs all operate with both OpenGL and OpenCL.
The hardware tessellation units enable increased vertex re-use, off-chip buffering improvements, and larger parameter caches for on-chip storage between tessellation output and consuming pixel shaders. The higher performance provides pixel-accurate rendering without visible artifacts. Another feature is the thermal control that adjusts the clock for maximum performance at the lowest active power. Active power management can also power down the GPU when it is not in use for long periods of time, like during blank screens.
The latest products start with the FirePro W5000, a $600, 65W board for CAD and DCC work. Next is the W7000 for $900, the W8000 for $1600 and the W9000 for $4000. The top end board runs at 300 W, so this is not a board for your average PC. The all of the boards can drive up to 6 displays and can connect to additional cards via framelock/genlock. The first set of OEMs planning to incorporate the FirePro boards includes HP and Dell.
