NVIDIA’s upcoming “China-compliant”GeForce RTX 5090D faces significant changes that could limit its AI and crypto-mining capabilities, raising concerns for professionals in these sectors.
New Restrictions on NVIDIA’s GeForce RTX 5090D: Impact on Multi-GPU Configurations
NVIDIA remains committed to the Chinese market, showcasing its dedication by launching the GeForce RTX 5090D, the successor to the RTX 4090D, alongside its mainstream RTX 50 series. While early speculation suggested minimal downgrades to the 5090D’s specifications, recent findings from Chiphell indicate that critical aspects related to AI and mining performance have indeed been altered.
The first key issue identified is a “performance cut-off”that affects workloads related to AI and cryptocurrency mining. Reports suggest that after three seconds of intensive tasks, the RTX 5090D will throttle its performance, possibly reducing clock speeds via onboard software. In addition, power limits will remain confined similarly to its predecessor, the RTX 4090D, which poses challenges for users aiming for optimal performance in professional computing tasks.
Moreover, the GeForce RTX 5090D will prohibit users from utilizing multiple GPUs in a single setup, a move that removes the ability to aggregate power by stacking multiple WeUs. This limitation follows a trend set by the RTX 4090D, which allowed for modified configurations that could match the performance of mainstream AI accelerators. As a result, even software configurations like Linux won’t support multi-GPU setups with the new model, following NVIDIA’s enhanced restrictions.
For gamers, these new limitations are unlikely to hinder performance; they mainly target AI professionals and those engaged in computational workloads. As for technical specifications, the GeForce RTX 5090D boasts a cutting-edge GB202 GPU die featuring 21,760 CUDA Cores and 32 GB of GDDR7 memory clocked at 28 Gbps, yielding a remarkable bandwidth of 1792 GB/s. Early partner models from companies such as GALAX have already emerged in the market, signaling readiness for release.
Preliminary Specifications of NVIDIA’s GeForce RTX 50 Series:
| Graphics Card Model | NVIDIA GeForce RTX 5090 | NVIDIA GeForce RTX 5080 | NVIDIA GeForce RTX 5070 Ti | NVIDIA GeForce RTX 5070 | NVIDIA GeForce RTX 5060 Ti | NVIDIA GeForce RTX 5060 |
|---|---|---|---|---|---|---|
| GPU Name | Blackwell GB202-300 | Blackwell GB203-400 | Blackwell GB203-300-A1 | Blackwell GB205-300-A1 | Blackwell GB206? | Blackwell GB206? |
| GPU SMs | 170 (192 Full) | 84 (84 Full) | 70 (84 Full) | 50 (50 SM Full) | TBD | TBD |
| CUDA Cores | 21760 (+33%) | 10752 (+11%) | 8960 (+16%) | 6144 | TBD | TBD |
| Clock Speeds | 2.41 GHz | 2.62 GHz | 2.45 GHz | 2.51 GHz | TBD | TBD |
| L2 Cache | TBD | TBD | TBD | TBD | TBD | TBD |
| Memory Capacity | 32 GB GDDR7 (+33%) | 16 GB GDDR7 (0%) | 16 GB GDDR7 | 12 GB GDDR7 | 16 GB GDDR7 | 8 GB GDDR7 |
| Memory Bus | 512-bit (+33%) | 256-bit (0%) | 256-bit (0%) | 192-bit (0%) | 128-bit (0%) | 128-bit (0%) |
| Memory Speed | 28 Gbps | 30 Gbps | 28 Gbps | 28 Gbps | 28 Gbps? | 28 Gbps? |
| Bandwidth | 1792 GB/s | 960GB/s | 896 GB/s | 672 GB/s | 448 GB/s | 448 GB/s |
| TBP | 575W (+27%) | 360W (+12.5%) | 300W (+10.5%) | 250W (+14%) | TBD | TBD |
| Power Interface | 1 12V-2×6 (16-Pin) | 1 12V-2×6 (16-Pin) | 1 12V-2×6 (16-Pin) | 1 12VHPWR (16-Pin) | 1 12VHPWR (16-Pin) | 1 12VHPWR (16-Pin) |
| Launch Date | January 2025 | January 2025 | February 2025 | February 2025 | March-April 2025 | March-April 2025 |
| Estimated Price | $1999 US | $999 US | $749 US | $549 US | TBD | TBD |
Related Articles:
President Trump Issues Warning to Chipmakers: Avoid ‘Hefty’ Tariffs by Investing in U.S. Manufacturing as it Becomes a Top Priority
8:58
Lenovo Unveils LOQ Tower PC Featuring AMD Ryzen 8000HX CPUs and RTX 50 Desktop GPUs; Legion Pro 7 Laptop Boasts Up To Ryzen 9 9955HX3D and RTX 5080
8:53
Lenovo Launches Legion Pro 32UD-10, 27UD-10, and 27Q-10 OLED Gaming Monitors with Up to 240Hz UHD and 280Hz QHD Refresh Rates
8:51
Leave a Reply