With the launch of the Super update of 40-Series NVIDIA GPUs, the company’s product lineup has become quite complex. In the 4070 series alone, four models of the NVIDIA GeForce RTX are available: the original 4070, 4070 Ti, and now also 4070 Super, and 4070 Ti Super. Understanding the differences between these cards and learning which models offer the best price/performance ratio in password recovery jobs are crucial considerations for IT professionals.

Why GPU for password cracking?

In certain tasks, the massively parallel compute units of today’s GPUs deliver performance far surpassing that of your computer’s CPU, all while maintaining similar power consumption. Our products support a hardware acceleration technology, employing the available computational power of modern GPUs to speed up password recovery attacks. Our tools have support for GPUs built with AMD, Intel, and NVIDIA chips, yet we consistently recommend NVIDIA cards for their maximum stability and reliability in 24×7 environment.

Our acceleration technology offloads some of the complex computations onto fast and scalable compute units (NVIDIA CUDA cores), which results in password recovery speeds going up by a factor of 25 to 250 compared to the central processor.

NVIDIA RTX 40 series

The first GPUs released in the Ada Lovelace powered 40-Series were the powerful NVIDIA GeForce RTX 4090 and 4080 boards, announced at the end of 2022. These boards were followed by the lower-performance models. The original lineup included the RTX 4090, 4080, 4070 Ti, 4070, 4060 Ti (in 8GB and 16GB versions), and 4060.

We benchmarked some of these GPUs in NVIDIA RTX 40 Series Graphics Cards: The Faster and More Efficient Password Recovery Accelerators, and concluded that the mid-range NVIDIA GeForce RTX 4070 Ti was faster than the last-gen champion RTX 3090, all while consuming less power. The new RTX 4070 gets about 40% higher performance and about 30% lower power consumption at the same time compared to the older but similarly priced RTX 3070 Ti board. This made a strong point in recommending the Ada Lovelace generation over Ampere for password recovery tasks.

In January 2024, NVIDIA refreshed the lineup, adding several models labeled as “Super”. The “Super” versions included the 4080 Super, 4070 Ti Super, and 4070 Super; the lower-end models of GPUs did not receive an update. NVIDIA then discontinued some of the older models (4080 and 4070 Ti), although they can still be found for sale. Meanwhile, both the 4070 and 4070 Super cards continue to be part of the company’s current lineup, while the RTX 4070 received a price cut. The new lineup is not very straightforward, especially considering that discontinued models are still sold in retail alongside with refreshed cards. Understanding the variety of GPU models and the performance they offer is important when choosing a GPU for password recovery jobs.

For password recovery, the amount of VRAM is largely irrelevant. All else being equal, there will be no difference in sheer attack speed between the NVIDIA GeForce RTX 4060 Ti (8GB) and GeForce RTX 4060 Ti (16GB) variants, yet the later costs more than the former. However, the 16GB 4070 Ti Super not only doubles the amount of VRAM compared to the now discontinued 12GB 4070 Ti but also expands the interface width from 192 to 256 bits, all while upping the number of execution units. As a result, the “Super” version of the 4070 Ti delivers a 10% performance increase in double precision computations.

Speaking of performance, one can benchmark video cards in many different ways, while expressing the results in different units. We don’t have the ability to test all GPU models in our setup, so for apples to apples comparison we rely on the data in the spec sheets as well as the data obtained from independent sources. Specifications list numbers for Single precision, Double precision, Half precision, FP16, and TFLOPS; for password recovery tasks, only the Double precision metric is relevant. Comparing different models (importantly, within the same architecture) based on this metric alone allows us to fairly accurately predict their relative performance. With this information, you can assess their performance in certain tasks; we suggest using the numbers from our test as a base.

Note: Additional test data is available in the full article.

Our price/performance numbers are based solely on U.S. MSRP prices. Street prices may and do differ; please bear that in mind when shopping for the most cost effective board. Finally, make sure to understand the effect of the boards’ TDP to build an efficient, well-cooled and adequately powered setup.

Note: The specifications listed above are for reference only and may vary based on the OEM and specific model.