The first model will go on sale after Nvidia has posted Nvidia RTX 6000 Ada specs on its website. As is so often the case with graphics cards formerly known by the Quadro name, the GPU is the same chip used in GeForce derivatives. As for the RTX 6000 Ada, in the Geforce RTX 4090 we have the Ad 102 chip.
More efficient than the GeForce RTX 4090
Of the maximum possible 144 SMs (streaming multiprocessors), 142 are active on the card, which is significantly more than the Geforce card, which has to be operated with 128 SMs, but the clocks are probably higher. Nvidia's workstation GPUs are said to draw only 300 watts and are cooled by DHE (Direct Heat Exhaust) coolers with vapour chambers and radial fans. This is already known from the Ampere generation.
The graphics memory is connected to a 384-bit interface intended for the Ad-102 GPU, but doubles up at 48GB compared to the 24GB of the regular RTX 4090. It also supports ECC error correction to detect and correct bit errors. With speeds of 20 Gbps, the memory is comparable to the upcoming Radeon RX 7900 GPU.
| Model | RTX 6000 | GeForce RTX 4090 |
|---|---|---|
| GPU | AD102 | AD102 |
| Shader multiprocessors / RT cores | 142 / 142 | 128 / 128 |
| Compute cores FP32 / Tensor | 18.176//568 | 16.384//512 |
| Base/Boost Clock | 2175 MHz / 2535 MHz | 2235 MHz / 2520 MHz |
| Computing power (FP32) | 91.1 TFLOPS | 82.6 TFLOPS |
| Raster power amplifiers | 192 ROPS | 176 ROPS |
| Memory amount/type | 48 GB GDDR6X | 24 GByte GDDR6X |
| memory interface/transfer rate | 384 bits/960 GB/s | 384 bits/1008 GB/s |
| Power consumption | 300 watts | 450 watts |
With a 700 MHz higher clock and 25% higher memory bandwidth compared to its predecessor, the Nvidia RTX A6000, it delivers a massive boost in computing power. There are many other familiar features like Framelock support with Quadro Sync II, NVENC with dual AV1, and PCIe Gen 4 x16 designs like the Geforce model. However, NVLink connectivity is omitted for the first time. It is not yet known whether so-called video memory pooling for large-scale rendering or AI models is still possible.
Like all Ada GPUs, the card is powered via a 16-pin ATX 3.0 connector. Improper plugging in, which Nvidia claims is responsible for many of its Geforce graphics card defects, should be rare in professional environments. The US store lists GPUs in the $7,400 to $10,000 range.
The RTX 6000 will also be available in our store at the end of January 2023.