Nvidia's Titan cards have always walked a fine line, as manifested by their branding: Neither gamer-oriented GeForce nor professionally targeted Quadro, they're basically Quadro-power cards with GeForce-capability drivers. That historically plops them into the really, really expensive gaming GPU category or on the lists of video professionals who demand speed and value more than certification.
The new Turing-architecture-based Titan RTX still offers that come-hither glance at gamers, especially if you really want to play Metro Exodus in 8K when it arrives in 2019, but the architecture's optimized ray-tracing and AI-acceleration cores increase its attraction for more dataset-focused research, AI and machine-learning development and real-time 3D professional work that doesn't require workstation-class drivers.
The distinction between GeForce and Quadro is waning a little over time as applications drift away from OpenGL — Adobe's video applications like Premiere and After Effects, for example, use the CUDA cores directly for acceleration — but the elephant in that room remains Photoshop. You can still only get 30-bit color support (10 bits per channel) with the workstation drivers, which are restricted to Quadro cards.
It's hard to make direct comparisons solely based on specs, in part because Nvidia is inconsistent about the specs they provide at launch; you usually have to wait a little bit until people dig in and ferret them out.
Since most of the Nvidia-provided specs for the $6,300 Quadro RTX 6000 and the $2,500 Titan RTX are almost identical — the Quadro does have a faster base GPU clock speed and four DisplayPort connectors vs. the Titan's three — we can't wait to find out what magic the Quadro performs that merits an almost $4,000 premium. Given that neither are shipping yet (the Quadro's in preorder and the Titan is slated for the end of November), we'll have to wait a bit.
On the flipside, the much, much less well-endowed Quadro RTX 5000 only costs $200 less than the Titan RTX, so you give up quite a bit of power in exchange for those workstation certifications.
As a gaming card, it fits right in to its traditional slot as a power bump up from the highest-end GeForce, but unless it delivers a much bigger performance gap than last-generations GTX 1080 Ti/Titan Xp, it will be doubly not worth it at twice the price of the RTX 2080 Ti, at least until more games ship which take advantage of the ray-tracing processors.
Comparative specifications
| GeForce RTX 2080 Ti (Founders Edition) | Quadro RTX 5000 | Quadro RTX 6000 | Titan RTX | Titan Xp | |
| GPU | TU102 | TU104 | TU102 | TU102 | GP102 |
|---|---|---|---|---|---|
| Memory | 11GB GDDR6 | 16GB GDDR6 | 24GB GDDR6 | 24GB GDDR6 | 12GB GDDR5X |
| Memory bandwidth | 616GB/sec | 448GB/sec | 672GB/sec | 672GB/sec | 547.7GB/sec |
| GPU clock Speed (MHz, base/boost) | 1,350/1,635 | 1,620/1,815 | 1,440/1,770 | 1,350M/1,770 | 1,405/1,582 |
| Memory data rate/Interface | n/a/352 bit | n/a/256 bit | n/a/384 bit | 14Gbps/384 bit | 11.4Gbps/384 bit |
| Texture fill rate (gigatexels/sec) | 420.2 | 348.5 | 509.8 | 510 | 379.7 |
| Ray Tracing (Gigarays/sec) | 10 | 8 | 10 | 11 | n/a |
| RT cores | 68 | 48 | 72 | 72 | n/a |
| RTX-OPS (trillions) | 78 | 62 | 84 | n/a | n/a |
| CUDA Cores | 4,352 | 3,072 | 4,608 | 4,608 | 3,840 |
| Tensor Cores | 544 | 384 | 576 | 576 | n/a |
| FP32 (TFLOPS, max) | 14 | 11.2 | 16.3 | n/a | 12.1 |
| Price | $1,200 | $2,300 | $6,300 | $2,500 | $1,200 |
Now playing: Watch this: Our most cherished video game memories
8:00
Nvidia's Titan cards have always walked a fine line, as manifested by their branding: Neither gamer-oriented GeForce nor professionally targeted Quadro, they're basically Quadro-power cards with GeForce-capability drivers. That historically plops them into the really, really expensive gaming GPU category or on the lists of video professionals who demand speed and value more than certification.
The new Turing-architecture-based Titan RTX still offers that come-hither glance at gamers, especially if you really want to play Metro Exodus in 8K when it arrives in 2019, but the architecture's optimized ray-tracing and AI-acceleration cores increase its attraction for more dataset-focused research, AI and machine-learning development and real-time 3D professional work that doesn't require workstation-class drivers.
The distinction between GeForce and Quadro is waning a little over time as applications drift away from OpenGL — Adobe's video applications like Premiere and After Effects, for example, use the CUDA cores directly for acceleration — but the elephant in that room remains Photoshop. You can still only get 30-bit color support (10 bits per channel) with the workstation drivers, which are restricted to Quadro cards.
It's hard to make direct comparisons solely based on specs, in part because Nvidia is inconsistent about the specs they provide at launch; you usually have to wait a little bit until people dig in and ferret them out.
Since most of the Nvidia-provided specs for the $6,300 Quadro RTX 6000 and the $2,500 Titan RTX are almost identical — the Quadro does have a faster base GPU clock speed and four DisplayPort connectors vs. the Titan's three — we can't wait to find out what magic the Quadro performs that merits an almost $4,000 premium. Given that neither are shipping yet (the Quadro's in preorder and the Titan is slated for the end of November), we'll have to wait a bit.
On the flipside, the much, much less well-endowed Quadro RTX 5000 only costs $200 less than the Titan RTX, so you give up quite a bit of power in exchange for those workstation certifications.
As a gaming card, it fits right in to its traditional slot as a power bump up from the highest-end GeForce, but unless it delivers a much bigger performance gap than last-generations GTX 1080 Ti/Titan Xp, it will be doubly not worth it at twice the price of the RTX 2080 Ti, at least until more games ship which take advantage of the ray-tracing processors.
Comparative specifications
| GeForce RTX 2080 Ti (Founders Edition) | Quadro RTX 5000 | Quadro RTX 6000 | Titan RTX | Titan Xp | |
| GPU | TU102 | TU104 | TU102 | TU102 | GP102 |
|---|---|---|---|---|---|
| Memory | 11GB GDDR6 | 16GB GDDR6 | 24GB GDDR6 | 24GB GDDR6 | 12GB GDDR5X |
| Memory bandwidth | 616GB/sec | 448GB/sec | 672GB/sec | 672GB/sec | 547.7GB/sec |
| GPU clock Speed (MHz, base/boost) | 1,350/1,635 | 1,620/1,815 | 1,440/1,770 | 1,350M/1,770 | 1,405/1,582 |
| Memory data rate/Interface | n/a/352 bit | n/a/256 bit | n/a/384 bit | 14Gbps/384 bit | 11.4Gbps/384 bit |
| Texture fill rate (gigatexels/sec) | 420.2 | 348.5 | 509.8 | 510 | 379.7 |
| Ray Tracing (Gigarays/sec) | 10 | 8 | 10 | 11 | n/a |
| RT cores | 68 | 48 | 72 | 72 | n/a |
| RTX-OPS (trillions) | 78 | 62 | 84 | n/a | n/a |
| CUDA Cores | 4,352 | 3,072 | 4,608 | 4,608 | 3,840 |
| Tensor Cores | 544 | 384 | 576 | 576 | n/a |
| FP32 (TFLOPS, max) | 14 | 11.2 | 16.3 | n/a | 12.1 |
| Price | $1,200 | $2,300 | $6,300 | $2,500 | $1,200 |
Now playing: Watch this: Our most cherished video game memories
8:00
