First hinted in a driver, Nvidia has finally launched the RTX 5880 Ada, a graphics card explicitly tailored for the Chinese market and conforms to the latest U.S. export rules. As you can quickly tell by its model name, the Ada Lovelace graphics card slots between the RTX 6000 Ada and RTX 5000 Ada.
Nvidia utilizes the AD102 silicon for the RTX 6000 Ada and RTX 5000 Ada; therefore, the RTX 5880 Ada is, in all likelihood, using a variant of the same silicon due to the number of CUDA cores. AD102, which measures 609 mm², houses 18,432 CUDA cores, giving Nvidia plenty of freedom to carve out a die for the RTX 5880 Ada. The U.S. export restrictions dictate that manufacturers cannot ship graphics cards with a Total Processing Power (TPP) rating higher than 4,800 TPP to China. The RTX 6000 Ada, which has a 5,828 TPP, is a no-go, leaving a space that the RTX 5880 Ada will fill.
The RTX 5880 Ada has 14,080 CUDA cores, equivalent to 110 Streaming MultiProcessors (SMs). The complete AD102 silicon has 18,432 CUDA cores (144 SMs). Therefore, the RTX 5880 has 23% fewer CUDA cores than the RTX 6000 Ada and only 10% more than the RTX 5000 Ada. As a result, the performance gap between the RTX 5880 Ada and the RTX 5000 Ada is small; however, the difference between the RTX 5880 Ada and RTX 6000 Ada is pretty notable.
While Nvidia didn’t list the clock speeds for the RTX 5880 Ada, the chipmaker did provide some performance metrics, which can be used to deduce the clock speeds and compare the performance between the three Ada-powered graphics cards.
The RTX 5880 Ada offers 69.3 TFLOPS of single-precision performance, 6% higher than the RTX 5000 Ada but 24% lower than the RTX 6000 Ada. Similar margins apply to the RT core and Tensor core performance. Given the RTX 5880 Ada’s single-precision figure, the graphics card likely has a 2,461 MHz boost clock, putting it slightly below the RTX 6000 Ada and RTX 5000 Ada’s boost clock speeds of 2,505 MHz and 2,550 MHz, respectively.
Nvidia RTX 5880 Ada Specifications
| Header Cell – Column 0 | RTX 6000 Available | RTX 5880 Yes | RTX 5000 Available |
|---|---|---|---|
| GPU memory | 48GB GDDR6 | 48GB GDDR6 | 32GB GDDR6 |
| Memory interface | 384-bit | 384-bit | 256-bit |
| Memory bandwidth | 960 GB/s | 960 GB/s | 576 GB/s |
| Error correcting code (ECC) | Yes | Yes | Yes |
| Nvidia Ada Lovelace architecture-based CUDA Cores | 18,176 | 14,080 | 12,800 |
| Nvidia fourth-generation Tensor Cores | 568 | 440 | 400 |
| Nvidia third-generation RT Cores | 142 | 110 | 100 |
| Single-precision performance | 91.1 TFLOPS | 69.3 TFLOPS | 65.3 TFLOPS |
| RT Core performance | 210.6 TFLOPS | 160.2 TFLOPS | 151.0 TFLOPS |
| Tensor performance | 1,457.0 TFLOPS | 1,108.4 TFLOPS | 1,044.4 TFLOPS |
| System interface | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 |
| Power consumption | Total board power: 300W | Total board power: 285W | Total board power: 250W |
| Thermal solution | Active | Active | Active |
| Form factor | 4.4” H x 10.5” L, dual slot, full height | 4.4” H x 10.5” L, dual-slot | 4.4” H x 10.5” L, dual slot |
| Display connectors | 4 x DisplayPort 1.4a | 4 x DisplayPort 1.4a | 4 x DisplayPort 1.4a |
| Max simultaneous displays | 4 x 4096 x 2160 @ 120 Hz, 4 x 5120 x 2880 @ 60 Hz, 2 x 7680 x 4320 @ 60 Hz | 4 x 4096 x 2160 @ 120 Hz, 4 x 5120 x 2880 @ 60 Hz, 2 x 7680 x 4320 @ 60 Hz | 4 x 4096 x 2160 @ 120 Hz, 4 x 5120 x 2880 @ 60 Hz, 2 x 7680 x 4320 @ 60 Hz |
| Power connector | 1 x PCIe CEM5 16-pin | 1 x PCIe CEM5 16-pin | 1 x PCIe CEM5 16-pin |
| Encode/decode engines | 3 x encode, 3 x decode (+AV1 encode and decode) | 3 x encode, 3 x decode (+AV1 encode and decode) | 2 x encode, 2 x decode (+AV1 encode and decode) |
| VR ready | Yes | Yes | Yes |
| vGPU software support | Nvidia vPC/vApps, Nvidia RTX Virtual Workstation | Nvidia vPC/vApps, Nvidia RTX Virtual Workstation | Nvidia vPC/vApps, Nvidia RTX Virtual Workstation |
| vGPU profiles supported | See the Virtual GPU Licensing Guide | See the Virtual GPU Licensing Guide | See the Virtual GPU Licensing Guide |
| Graphics APIs | DirectX 12, Shader Model 6.6, OpenGL 4.6, Vulkan 1.3 | DirectX 12, Shader Model 6.6, OpenGL 4.6, Vulkan 1.3 | DirectX 12, Shader Model 6.6, OpenGL 4.6, Vulkan 1.3 |
| Compute APIs | CUDA 11.6, OpenCL 3.0, DirectCompute | CUDA 11.6, OpenCL 3.0, DirectCompute | CUDA 11.6, OpenCL 3.0, DirectCompute |
| Nvidia NVLink | No | No | No |
While Nvidia has slashed some of the CUDA cores on the RTX 5880 Ada, the graphics card retains the identical memory subsystem of the RTX 6000 Ada. 48GB of 18 Gbps GDDR6 ECC memory is still present. The 384-bit memory interface remains intact, permitting the RTX 5880 Ada to deliver a memory bandwidth of 960 GB/s, the same as the RTX 6000 Ada.
Fewer CUDA cores mean the RTX 5880 Ada consumes less power than the RTX 6000 Ada. In that aspect, the RTX 5880 Ada has a 5% lower total board power (TBP). It’s not something to marvel at, though, since the RTX 5880 Ada loses 23% of the CUDA cores of the RTX 6000 Ada. When compared to the RTX 5000 Ada, however, the RTX 5880 Ada has a 14% higher TBP.
Given the slight TBP difference, the power connector layout on the RTX 5880 Ada is the same as the RTX 6000 Ada and RTX 5000 Ada. The new Ada-powered graphics card draws external power from a single 16-pin (12VHPWR) power connector. As expected of a Nvidia professional graphics card, the 16-pin power connector is located on the rear, so there won’t be any issues with 16-pin meltdowns, like on mainstream models, such as the GeForce RTX 4090.
Nvidia didn’t share the pricing for the RTX 5880 Ada. The RTX 6000 Ada retails for $9,999, while the RTX 5000 Ada typically sells for $6,999. We want to think that the RTX 5880 Ada’s price tag should be between those two numbers; however, this may not be the case. Nvidia recently rolled out the GeForce RTX 4090D for the Chinese market. Despite being a cut-down version of the GeForce RTX 4090, the GeForce RTX 4090D launched at the same price ($1,599) as the vanilla version. So, it wouldn’t be unexpected if the RTX 5880 Ada sells for close to the RTX 6000 Ada’s price.
