After rolling out its new GPU family on consumer products, AMD has turned to the workstation market and has taken the opportunity of including the Trinity APU in the FirePro range.
The professional graphics card market is difficult and dominated by Nvidia but AMD remains combative. In the absence of an Nvidia GPU fully adapted for the pro world (the GK110 is scheduled for 2013), AMD once again has an opportunity to break into this market.
The FirePro V9800, V8800, V7900 and V5900 have thus been replaced by Wx100 models from the GCN (Graphics Core Next) generation, which support all the latest technologies such as OpenCL 1.2 or PCI Express 3.0.
The FirePro W9000 is the new top end model in the pro range and corresponds to the Radeon HD 7970 with the Tahiti GPU. Its 2048 processing units are slightly overclocked at 975 MHz up from 925 MHz. This gives processing power of 4 Tflops in single precision and 1 Tflops in double precision, which is almost double what you get with the FirePro V9800. The memory interface is equivalent to the corresponding Radeon with a 384-bit bus and GDDR5 clocked at 1375 MHz. The video memory is however up from 3 to 6 GB, which makes a difference in certain domains, and there's ECC support. The TDP has been announced at 274W and the video connectivity is identical to that on the previous generation: 6 mini-DisplayPorts.
A step down in the range, we have the FirePro W8000, which is the pro equivalent of the Radeon HD 7950: its 1792 processing units are clocked at 900 MHz with raw processing power of 3.2 Tflops in single precision and 0.8 Tflops in double precision. The memory bus has however been cut to a 256-bit model, while the GDDR5 is still clocked at 1375 MHz, for a similar memory bandwidth to that offered by the previous generation. This 256-bit bus does however allow AMD to equip the FirePro W8000 with 4 GB of video memory. It also has ECC support but just four DisplayPort outs and a TDP of 189W.
The FirePro W7000 is based on the Pitcairn GPU and corresponds to a Radeon HD 7870, which is slightly underclocked. Its 1280 processing units are thus clocked at just 950 MHz giving 2.4 Tflops of single precision processing power. This GPU doesn’t offer double precision support to speak of and gives just 0.15 Tflops of processing power in this mode. Its 4 GB of GDDR5 are clocked at 1200 MHz with a 256-bit bus. ECC is not supported but the card is just single slot, with a TDP of under 150W and four DisplayPort outs.
The smallest model in the new family, the FirePro W5000 is based on a heavily cut down version of the Pitcairn GPU which doesn’t yet have a consumer equivalent. It has just 768 processing units clocked at 825 MHz and giving 1.3 Tflops of processing power in single precision. The point of this configuration, in comparison to the smallest GCN GPU, which without being cut down offers similar processing power at a lower cost, is that it allows AMD to include a 256-bit memory bus and twice the geometric processing power. The FirePro W5000 is relatively compact, short and on a single slot and has a TDP of under 75W. Connectivity wise it has two DisplayPort outs and a DVI Dual-Link out.
In comparison to the Radeons, these new FirePros come with drivers optimised for pro applications, notably with a set of specific profiles. They are moreover guaranteed for three years and will be available for a long period. Current pricing stands at $4000 for the FirePro W9000, $1600 for the FirePro W8000, $900 for the FirePro W7000 and $600 for the FirePro W5000.
At the same time as the launch of these new pro graphics cards, AMD has rolled out an APU in the FirePro family for the first time, which probably represents the most serious threat to Nvidia. As the ridiculously weak specs of some of the Quadros suggest, a good number of pro applications don’t require enormous processing power. Such entry-level products, often billed at high-end consumer prices, function simply as a license to take advantage of drivers that have been let off the leash in terms of pro application support.
Intel was the first to cotton on to this by rolling out its integrated graphics cores as pro (P) versions for certain Xeons. The highest performance of these is currently the Xeon E3-1275V2, which is based on Ivy Bridge architecture, and the HD Graphics P4000 graphics core which has drivers certified for some professional applications.
With its APUs, AMD has the opportunity of offering a similar solution and standing out with its rich and long experience in the domain and wider support of the different available graphics applications. This is now happening with the FirePro A320 and A300, which are derivatives of the Trinity with identical specifications to those for the A1-5800K (including the unlocked multiplier) and A10-5700. According to the figures published by AMD, the Intel solution suffers greatly by comparison and an entry level Quadro would also be beaten. It remains to be seen if the market will be receptive and put confidence in the AMD drivers and support, two areas where Nvidia is very strong…