@lauriewired: The hardware in old Chinese cloud accelerator cards never fails to impress me. If you go on Chinese ebay (idlefish) you…
Summary
Old Chinese cloud accelerator cards contain Xilinx UltraScale FPGAs that can be bought for ~$50 on Idlefish, significantly cheaper than the ~$2,100 price on Mouser.
View Cached Full Text
Cached at: 05/18/26, 08:30 AM
The hardware in old Chinese cloud accelerator cards never fails to impress me.
If you go on Chinese ebay (idlefish) you can get a Xilinix UltraScale FPGA for ~$50 USD.
For perspective, the same raw chip is currently ~$2,100 on Mouser. https://t.co/JyZSmgpiky
Similar Articles
Buying AI accelerators/GPUs in China...
A user asks about buying Chinese AI accelerators/GPUs for inference, specifically looking for Huawei alternatives to Nvidia, with support for vLLM or Llama.cpp.
Tiny company steals AMD's thunder and challenges Nvidia with old-tech PCIe AI accelerator that runs 700B LLMs locally, sipping just 240W thanks to decade-old DDR4 and 28nm chips
Taiwanese startup Skymizer unveiled the HTX301, a PCIe AI accelerator that uses older 28nm chips and DDR memory to run 700B parameter LLMs locally at just 240W, challenging high-power GPU solutions from Nvidia and AMD.
People are making single-slot, half height pcie v100 with nvlink in China
Chinese GPU enthusiasts have created a custom single-slot, half-height PCIe V100 GPU with NVLINK, retaining full performance and aiming to sell for around $220.
AMD BC-250 and the search for Cheap Compute
A developer reverse-engineers AMD BC-250 boards (salvaged PS5 APUs) to enable all 40 compute units, achieving significant performance gains in LLM inference with custom HIP kernels. The project shows potential for low-cost AI compute.
Was my $48K GPU server worth it?
A former FAANG engineer recounts building a $48K GPU server with six RTX 6000 Ada cards for independent AI research, detailing the build process, power constraints, and a cost comparison against cloud GPU rentals.