NVIDIA's RTX 2080 Ti and RTX 2080 graphics CARDS use the Turing architecture and integrate the Tensor Core in addition to ray tracing technology. Therefore, they are more powerful in AI than the current Pascal graphics card. If the RTX 2080 Ti graphics card is purchased for the purpose of deep learning and the price is 70% higher than that of the GTX 1080 Ti graphics card, which means the cost is increased by 70%, is the performance improvement in deep learning worth the price increase?

As for the in-depth learning performance of RTX 2080 Ti graphics card, lambdalabs have been tested recently. If anyone concerned about the performance of RTX 2080 Ti graphics card in this aspect, they can first refer to their test results and compare whether the lower value is worth.

Single precision resnet-152 test

Semi - precision resnet-152 test

In the resnet-152 test, the performance of RTX 2080 Ti in the single precision of FP32 was 27% to 45% higher than that of GTX 1080 Ti, and the performance was 60% to 65% higher in the precision of FP16.

In this regard, lambdalabs said that if you need to run the FP16 half-precision test, the performance of RTX 2080 Ti can be directly proportional to the price. If you do not need FP16, you should consider the problem of the cost increase of 71% and the average performance increase of 36%.

Speed comparison of different models of RTX 2080 Ti and GTX 1080 Ti video card

The above part is actually their test summary, and the following is the original data of the test.

FP32 precision image processing speed

FP16 half-precision image processing speed

Performance and cost analysis

In this part of the comparison, the RTX 2080 Ti price base is $1,200, and the GTX 1080 Ti graphics card is $700, then converted to performance per selling price. In general, GTX 1080 Ti is still the best choice for FP32 precision. If you are not sensitive to the selling price, most people obviously want to get the RTX 2080 Ti card.