Nvidia's RTX 6000 Ada Tested in 3DMark: AD102 with 18,176 Cores

On paper, Nvidia's RTX 6000 Ada professional graphics card has more compute horsepower than the GeForce RTX 4090 flagship board for gamers. Yet, when tested in 3DMark TimeSpy Graphics benchmark, it could not actually beat it, according to a Reddit post noticed by @harukaze5719. 

Nvidia's RTX 6000 Ada 48GB graphics board is based around AD102 GPU with 18,176 CUDA cores enabled. While Nvidia and its partners do not specify clocks of the GPU, they disclose that the card has a peak compute performance of 91.1 FP32 TFLOPS (which suggests clocks of around 2,505 MHz). By contrast, Nvidia's GeForce RTX 4090 24GB uses AD102 GPU with 16,3840 CUDA cores clocked at up to 2,520 MHz and rated for up to 82.575 FP32 TFLOPS. 

There is one major difference between the two cards (although they seem to use the same PCB): the professional one is rated for up to 300W total board power, whereas the gaming one can consume up to 450W of power, which means that the latter can run at high clocks considerably longer than the former. On the one hand, this limits performance, but on the other hand this increases longevity of the RTX 6000 Ada versus GeForce RTX 4090. 

This is exactly the case with performance of the two cards in 3DMark Time Spy Graphics benchmark. The RTX 6000 Ada's Graphics score was 30,158 points with the latest Windows update (as seen in the screenshot), but it could score 36,844 points before the update, according to the individual who posted the score, Reddit user Healthy-Blood-54. Meanwhile, GeForce RTX 4090 can easily score 40,964 points. 

The reason for such relatively low performance are of course comparatively low GPU clocks that drop from a whopping 3+ GHz in the first couple of minutes of the test to around 1.5 GHz later on. Interestingly, the owner says the board does not overheat, so perhaps it starts to drop clocks at relatively low temperatures. 

"When running full usage, it is about 85°C on the core with the fans at around 70%, explained Healthy-Blood-54. "No louder than the [RTX] A6000 [based on GA102]. But again, remember these are battle cards. These can probably run at 95°C core 24/7 for 10 years before any issues. The cost is not just in the VRAM." 

While the RTX 6000 Ada 48GB graphics card cannot beat the GeForce RTX 4090 despite the higher number of enabled CUDA cores and higher peak FP32 compute throughput, the rumored GeForce RTX 4090 Ti with 18,176 stream processors enabled will likely beat the current flagship by a wide margin.