Back in July NVIDIA launched Fermi in to the professional space, introducing the first of their Fermi-based Quadro cards. The Quadro 4000, 5000, and 6000 were all based on GF100, and like the GeForce and Tesla lines used cut down GPUs in order to meet NVIDIA’s TDP and yield needs.
Notably, all the Quadros launched with their FP64 capabilities uncapped, something we weren’t sure would be made available outside of the Tesla line. Along those same lines, the 5000 and 6000 models also had ECC support enabled, again another feature initially promoted for Tesla. The result of this was that the first Fermi Quadro cards were capable of behaving a great deal like Tesla cards on top of their traditional professional graphics duties.
Now less than 3 months down the line NVIDIA is launching the rest of the Quadro series. Today marks the launch of the 2000 and the 600, which extend the Fermi Quadro lineup to the smaller Fermi GPUs. In the process, these cards also move away from Tesla-like compute capabilities and focuses more on Quadro’s traditional graphics roles such as modeling, CAD, digital video production, and the more recently emerging field of GPU-accelerated professional applications.
|NVIDIA Quadro Lineup|
|Memory Type||6GB GDDR5||2.5GB GDDR5||2GB GDDR5||1GB GDDR5||1GB DDR3|
|Memory Bus Width||384-bit||320-bit||256-bit||128-bit||128-bit|
|FP64||1/2 FP32||1/2 FP32||1/2 FP32||1/12 FP32||1/12 FP32|
Unlike the first Quadro cards, the last two members of the Fermi Quadro lineup are fairly straightforward Quadro versions of NVIDIA's other GPUs. Quadro 2000 is GF106 based while Quadro 600 is GF108 based, with the latter GPU launching here in the professional space before it launches in the consumer space. GF104 is absent from NVIDIA’s Fermi Quadro lineup as NVIDIA has decided to go with a heavily cut down GF100 part for the Quadro 4000 to fill that role rather than to use GF104, due to GF100’s vastly superior FP64 speeds. Meanwhile Quadro 2000 and Quadro 600 will support FP64 (as do all other NVIDIA GPUs) but as GF104-derrived GPUs their FP64 speed is limited to 1/12th FP32 on account of the limited ratio of FP64 capable CUDA cores.
For the entire lineup of Fermi based Quadros, NVIDIA has been heavily pushing the enhanced geometry capabilities of Fermi, and the 2000 and 600 are no exception. Fermi’s Polymorph engines give the derrived GPUs a big step up in geometry performance, and while the advantage isn’t as great for these lower-end parts, it’s still something NVIDIA believes is of particular importance for the CAD market. CUDA is of course also a big player here, particularly since Adobe’s Premiere Pro CS5 offers GPU acceleration through CUDA. And of course we can't ignore NVIDIA's Quadro drivers which are specifically tuned on a per-application basis and where much of the Quadro "magic' occurs for professional applications.
As is traditional for the Quadro lineup, both of these cards are on the lite side when it comes to power consumption compared to their desktop counterparts. The 2000 is rated for 62W, and the 600 a mere 40W. NVIDIA doesn't list the clockspeeds for these cards, but clearly they're lower than their desktop parts in order to make up the power consumption difference. DisplayPort outputs will also be making an appearance here after being absent on NVIDIA's desktop cards, where they went with HDMI ports instead of DP due to the vast proliferation of HDMI-enabled budget monitors.
Pperformance aside, the 2000 and the 600 will have some feature differentiation. The 2000 will support NVIDIA’s SLI Multi-OS system, while the 600 – lacking any SLI capabilities in the first place – will not. Along the same lines, the 600 will be a half-height card (as is common for NVIDIA’s entry-level Quadros) while the 200 will be full-height – the 600 drops a DisplayPort output to make this happen.
Finally, NVIDIA has put the MSRP on these cards at $599 for the 2000, and $199 for the 600. However expect street prices to vary some, as Quadro pricing isn't nearly as cut-throat and can afford a larger margin to set prices against.