–
This is the weakest implementation of the GA102 chip so far, thanks to which this card can handle 200W TDP, which is 30W less than in the case of the RTX A5000. In addition to the number of CUDA cores reduced to 7168, it also has fewer Tensor cores compared to other versions, namely 224 and further uses 20 GB of GDDR6 memory on 16 GT / s and a 320-bit bus with a resulting throughput of 640 GB / s.
–
In the case of the powerful RTX A6000, NVIDIA has already exchanged memory throughput for capacity, because in order to be able to offer 48 GB of memory, NVIDIA only had to use the older GDDR6 instead of the new GDDR6X, which were not available in the required capacities. And it is possible that they are now, but the significantly weaker RTX A4500 is not quite logical.
The RTX A4500 offers 23.7 TFLOPS in FP32, 46.2 TFLOPS in RT cores and 182.2 TFLOPS in Tensor.
–
In addition, we learn that the card will also come to the market RTX A2000, which we already know, but in the new version it gets not 6, but 12 GB of GDDR6 memory. However, it is still a model with a GA106 chip equipped with 3328 CUDA cores and with an output of up to 8 TFLOPS in FP32, while maintaining a low TBP at 70 W. The only difference is that the new version has twice the memory, so even that can only power from the PCIe x16 slot and does not need an additional power connector, while the RTX A4500 uses one eight-pin.
–
Prices of related / similar products:
–
