Nvidia's partners have quietly started sales of Nvidia's RTX 5000 Ada Generation graphics card designed for professional visualization (ProViz) applications. The board carries the company's flagship AD102 GPU for client PCs, but in a severely cut down configuration to reduce its power consumption to 250W. Surprisingly, retailers are selling PNY's Nvidia RTX 5000 Ada boards at inflated prices.
The RTX 5000 Ada graphics card is now available from PC Connection, Ebay, ProVantage, ShopBLT, and ThinkMate, among other workstation-oriented retailers. Meanwhile, none of them currently sell this add-in-board at its MSRP of $4,000, with prices hovering 10–20 percent above this figure. ProVantage even goes so far as to suggest the list price for the card is $6,999 and that you're saving 37%. Nice try.
Nvidia's RTX 5000 Ada Generation graphics card is based on the AD102 graphics processor with 12,800 CUDA cores spread across 100 SMs. It offers peak compute performance of 65.3 FP32 TFLOPS and is mated to 32 GB of GDDR6 memory using a 256-bit interface. By contrast, Nvidia's range-topping RTX 6000 Ada powered by the same AD102 GPU boasts with 18,176 CUDA cores and delivers a whopping 91.6 FP32 TFLOPS. Furthermore, the RTX 6000 also has 48 GB of GDDR6 memory connected to the processor using a 384-bit bus.
Cutting down 30% of the GPU may seem a bit odd, but perhaps this was the only way to build a product that would sit below the RTX 6000 Ada and still deliver a noticeable, yet not dramatic performance gap with the top-of-the-line ProViz graphics card. Nvidia's full AD103 has a maximum of 10,240 CUDA cores spread over 80 SMs, so its potential performance would be far lower than a cut-down AD102.
Interestingly, despite severely cutting down the GPU configuration, Nvidia retained a rather high power rating of 250W for the part. As for the cooling system, the board is equipped with a dual-slot cooler with one blower. The board also uses a 12VHPWR PCIe auxiliary power connector.
Meanwhile, it remains to be seen how Nvidia might choose to leverage the AD103 chip for it's professional offerings. RTX 6000 Ada uses nearly the full AD102, with 142 (out of 144) SMs and the full 384-bit memory interface. RTX 5000 Ada uses AD102 with 100 SMs and only a 256-bit interface. Then there's a big jump down to RTX 4500 Ada with a full AD104 implementation of 60 SMs and a 192-bit interface. RTX 4000 Ada and RTX 4000 SFF Ada wrap things up with a partial AD104 chip using 48 SMs and a 160-bit interface.
There's room for an RTX Ada card between the 4500 Ada and 5000 Ada, in other words, using AD103. However, Nvidia hasn't announced such an offering yet, and there's no room for it within the standard naming structure (i.e. RTX 5500 Ada would be this card). Perhaps Nvidia will simply bypass using AD103 on professional cards entirely.