Skip to main content
RTX 4090 Ti Engineering Sample Has Surfaced On A Chinese Marketplace With 14% Higher Memory Bandwidth, 11% More CUDA Cores Than Regular RTX 4090 - Wccftech

RTX 4090 Ti Engineering Sample Has Surfaced On A Chinese Marketplace With 14% Higher Memory Bandwidth, 11% More CUDA Cores Than Regular RTX 4090 - Wccftech

NVIDIA GPUs
16 lis 2026, 10:07Hardware News
Oryginał
1 min read

An alleged engineering sample of NVIDIA's unreleased RTX 4090 Ti has surfaced on a Chinese marketplace, revealing 14% higher memory bandwidth and 11% more CUDA cores than the standard RTX 4090.

The engineering sample, discovered on a Chinese secondhand marketplace, showcases significant improvements over NVIDIA's current flagship GPU. The listing, first spotted by VideoCardz, reveals a GPU-Z readout indicating 18,176 CUDA cores—1,792 more than the standard RTX 4090's 16,384 cores. This represents an 11% increase in shader hardware, positioning the GPU alongside the professional NVIDIA RTX 6000 Ada Generation.

Enhanced Performance

The prototype features 24GB of Micron GDDR6X memory across a 384-bit bus, with memory chips clocked at 24Gbps—faster than the RTX 4090's 21Gbps. This results in a memory bandwidth of 1,152GB/s, a 14.3% improvement over the standard model's 1,008.0GB/s. The card also includes 192 ROPs and 568 TMUs, utilizing the full AD102 silicon NVIDIA originally intended for the RTX 4090 Ti.

Prototype Details

The Chinese seller reports the prototype operates without stability issues during testing, though it requires an internal NVIDIA driver not available to the public. This isn't the first leak of the RTX 4090 Ti, as a previous 2024 prototype featured a quad-slot design with an unusually large heatsink. Despite these developments, NVIDIA never brought the GPU to mass production.

Future Outlook

With NVIDIA confirming its focus on gaming GPUs, attention now turns to the anticipated RTX 6000 series, which could launch as early as 2028. The company's GeForce Evangelist Jacob Freeman has indicated no plans to revisit the previous generation's unreleased hardware, leaving enthusiasts to speculate about future performance improvements.

Z.ai
Article written by AI
Z.ai GLM-4.5-FlashFree
Free AI articles150 of 150
No limit after signing in with Telegram