Nvidia has started sales of its tiny RTX 4000 SFF Ada Generation graphics card that promises GeForce RTX 3070-like performance at 70W of power and will fit into virtually all desktop PCs. The low-profile, dual-wide board is not cheap — it costs more than the RTX 4080, for example, as it's aimed at professional users — but nothing is stopping you from installing it on a regular gaming computer.
PNY's Nvidia RTX 4000 SFF Ada Generation graphics card is available now for $1,444 from ShopBLT, a retailer known for landing hardware ahead of its rivals. This is why the board is being sold at a price that is higher compared to its official MSRP of $1,250. Keep in mind that the board is equipped with four Mini-DisplayPort connectors, so you'll also need to add the price of an mDP-DP or mDP-HDMI adapter to the cost of this miniature solution.
The Nvidia RTX 4000 SFF Ada Generation board features the company's AD104 GPU with 6,144 active CUDA cores out of a total of 7,680, as well as 20GB of GDDR6 ECC memory connected to the GPU through a 160-bit interface. The GPU has a a capped boost frequency of approximately 1560 MHz to reduce overall board power consumption and is rated for just 70W of power, which means it can be installed into almost any desktop computer, even those without an auxiliary PCIe power connector.
| Row 0 - Cell 0 | GPU | FP32 CUDA Cores | FP32 TFLOPS | INT8 TFLOPS | Memory Configuration | TBP | MSRP |
| GeForce RTX 4070 Ti | AD104 | 7680 | 40 TFLOPS | 160/320 TFLOPS | 12GB 192-bit 21 GT/s GDDR6X | 285W | $799 |
| GeForce RTX 4070 | AD104 | 5888 | 29 TFLOPS | 116/233 | 12GB 192-bit 21 GT/s GDDR6X | 200W | $599 |
| RTX 4000 Ada Generation | AD104 | 6144 | 19.2 TFLOPS | 153/307 TFLOPS | 20GB 160-bit 16 GT/s GDDR6 ECC | 70W | $1,250 |
| GeForce RTX 3090 Ti | GA102 | 10,752 | 40 TFLOPS | 160/320 TFLOPS | 24GB 384-bit 20 GT/s GDDR6X | 450W | $1,999 |
| GeForce RTX 3070 | GA104 | 5888 | 20.31 TFLOPS | 81/160 TFLOPS | 8GB 256-bit 14 GT/s GDDR6 | 220W | $499 |
From a performance point of view, Nvidia's GA104 graphics processor in this configuration chip delivers a peak FP32performance of 19.2 TFLOPS, making it theoretically similar to the GeForce RTX 3070. Yet, with 20GB of memory onboard, this card is a little more future-proofed than the RTX 3070, and also potentially more useful for professional and AI researchers. The memory configuration likely utilizes 2GB GDDR6 chips on both sides of the PCB, as otherwise the 160-bit interface would limit maximum memory to just 10GB.
The nearly 20 FP32 TFLOPS are overshadowed by the superior performance of the recently launched GeForce RTX 4070 (29 FP32 TFLOPS). The board also boasts a peak RT performance of 44.3 TFLOPS and a peak FP8/INT8 tensor performance of 153/306.8 TFLOPS/TOPS (without and with sparsity). FP8/INT8 performance of course has nothing to do with games, but it's an added bonus for the professional market. In fact, 153/306.8 TFLOPS/TOPS is comparable to the more expensive and power-hungry Nvidia GeForce RTX 3090 Ti (that's if you don't care about precision, as the 3090 Ti only has native FP16 support).
Other advantages of the RTX 4000 SFF Ada include Nvidia's professional drivers and support for professional software ISVs. Furthermore, it comes with a 3-pin mini-DIN connector for stereoscopic 3D output (e.g. Nvidia 3D Vision), and supports Frame Lock capability for multi-display applications.
We should see pricing and availability improve in the coming weeks as the card becomes more widely available. For now, this is a workstation part designed for compact systems, targeting low power rather than maximum performance.
