While the response to the RTX consumer cards from Nvidia has been underwhelming, professional GPU customers should sit up and listen. Nvidia is taking no prisoners with the new generation of Quadro products.
Professional GPUs are, obviously, not for everyone. While they may share much of the same hardware as their consumer counterparts, they command a much higher asking price for a reason. These cards are built to exacting standards. Their driver software is certified to work with specific software packages. In order to minimize downtime, which is worth far more than the asking price of the card. You also get access to direct support from the GPU company, which isnít something sold with gaming cards!
With this generation of Quadro cards something curious has happened however. The technology mix across professional and consumer cards in the RTX family is about the same, but when you look at what these RTX cards are designed to do it makes much more sense from a professional perspective than a gamer perspective. Itís such a paradigm shift that plenty of outlets are telling gamers to hold off on RTX cards until there are actual games that support the new features that Nvidia have brought to the table.
On the professional side however, the new tech can benefit pro users NOW and there is no such rationale for holding off on them.
The Turing Test
The RTX cards are based on a new GPU architecture known as Turing. This an evolution of the Volta chips that didnít go beyond the Titan V.
The RTX 8000,6000 and 5000 cards are real monsters from a performance perspective. They sport massive core counts, but also introduce a wider mix of core types as we see with Volta. While Volts introduced AI-specific tensor cores, Turing adds a world first - ray tracing cores.
Ray tracing is widely considered the gold standard for photorealistic rendering, but has never been feasible in real-time. These new cards from Nvidia allow for a type of hybrid rendering, mixing ray tracing and rasterization, optimized by the machine learning cores.
The Specs Say it All
If we look at the flagship RTX 8000 card, thereís a lot to drool over here. It has 4608 CUDA cores and 576 tensor cores. This is backed up by another first - GDDR6 VRAM. 48 GB of the stuff!
Itís hard to gauge the raw performance compared to previous generation Quadro cards, simply because RTX is so different. However, a single-precision rating of 16 TFLOPS is 4 TFLOPS faster than the previous flagship, the P6000.
In terms of of how many rays the GPU can cast, the figure is 10 giga rays per second. Which sounds impressive, but as the first of its kind we have nothing to compare this to.
Professional Use Cases
Professional Use Just as with Volta, these cards offer tantalising acceleration for projects that involve AI or deep learning. Those AI cores are also being used to optimise workflows and workloads to do more per TFLOP than before.
We still need to see how it plays out in terms of actual performance gains, but when it comes to GPGPU workloads we expect an improvement. Especially since the RTX 8000 has fewer CUDA cores than the Titan V. Despite this, the Titan V is 2.2 TFLOPS behind. Thatís impressive.
The same of true of tensor cores. The 8000 fewer tensor cores than Volta, but 15 FP16 TFLOPS more under the hood.
Another important fact to note is that NVlink provides almost perfect scaling for non-graphical workloads. IN other words, if you put two RTX quadros in the same machine for GPGPU jobs, youíll get a pretty good return on your investment.
To configure a Workstation using any Quadro RTX please visit our Workstation Computer line: