Since the 1995-2005 era of VGA cards, and the release of Nvidia’s GeForce 256 GPU in 1999, there have been a lot of technological advancements in the world of graphics cards. New generations of cards are released every year, with improved efficiency and benchmark performance.
To study the evolution of GPUs, we create a GPU hierarchy that gives us a powerful comparison of different chips.
The hierarchy of graphics cards is established regarding their performance, compared to their peers.
GPU performance is evaluated by running multiple tests that simulate the actual usage of the card. This includes average GPU Framerates in the most demanding AAA titles, G-Flops and T-Flops testing, bandwidth testing, and rendering testing.
Some factors such as card value, date of issue, and TDP are not of great importance since we are evaluating cards for initial performance. Benchmarks from reliable sources will be studied and used for comparison.
Generations and geometry of GPU issues under evaluation
The most important factor in evaluating this GPU is keeping track of the generations and technologies used in the card. A higher or more V-Ram bit rate does not automatically translate to “better GPU”.
There are now multiple generations of GPUs, both from AMD and Nvidia. There are architectural differences between these generations (such as Nvidia’s Ampere).
It’s true that new technologies improve overall performance, but that certainly doesn’t mean that an older-generation card won’t be able to compete with newer ones.
Some examples of this are 1080 vs. the 1660 Super, the RTX 2080 Ti, and the RTX 3060 Ti, where the older GPU outperforms the newer by a respectable margin.
Benchmarks are our benchmark for measuring GPU performance
Performance benchmarks provide us with insight into what a GPU is capable of, and how it weighs against its peers.
In this GPU hierarchy, only newer GPUs are included in tabular form, while older cards (such as Nvidia’s old GTX 1000 series or GTX 900 series) are excluded with some exceptions.
One thing to note is that there can be conflicting benchmark results from different sources.
The reason for this is the difference in performance measurement methodologies. Geek bench’s OpenCL benchmarks GPU performance with Sobel, Canny, Particle Physics, Depth of Field, etc.
While PassMark (G3D Mark) runs on-card simulations as well as general-purpose computing tests. To improve accuracy, almost all benchmarking websites collect performance data from their users. This turns into millions of results, which can then be rounded down to the final GPU score.
Note, standardized scores should be taken with a pinch of salt. While they can provide insights, they may or may not represent the ultimate capability of the GPU.
GPU HIERARCHY 2022 – Standard graphics card performance rating
We can divide the GPU performance category into several levels.
Platinum GPU: Best of the Best – Worst performance per dollar
These are the best GPUs. While the performance/dollar value isn’t exactly befitting of these GPUs, they are for enthusiasts who can’t settle for anything less than the best.
It’s the epitome of performance in the GPU commercial hierarchy.
At an MSRP of $1,999 and a G3DMark score of 30,105, the RTX 3090Ti is worth 15 points/$. Hence not the best value actually.
The NVIDIA RTX 3080 has an MSRP of $699 with a G3DMark score / USD = 36 points / USD which makes it more cost-effective compared to the RTX 3090Ti.
Gold Series GPUs: Advanced – Best performance per dollar value
These GPUs are designed for most high-performance gaming hardware. The average professional gamer will aim for these graphics cards.
This graphics card often represents the best value for money in the GPU hierarchy.
NVIDIA RTX 3070Ti G3DMark Value / USD 39 points / USD (with MSRP of $600).
With an MSRP of $399, the RTX 3060Ti has a G3DMark score / $value of an impressive 50 points / $!
You can see that the Gold GPUs offer a better value compared to the Platinum GPUs.
Silver Series GPUs: Mid-Range – Average performance per dollar value
Silver GPUs are mostly older-generation graphics cards that are still very relevant in the GPU hierarchy and gaming today.
With the MSRP of these GPUs, they don’t always have the best value.
Take for example the MSRP for the RTX 2060 Super. It’s the same price as the 3060Ti Gold Series at $399.
Thus, unless you can find a great deal on Silver-class GPUs, buying older generic graphics cards at their MSRP value will not be the best course of action.
The only gen GPU on this list is the AMD Radeon RX 6600. With an MSRP of $330 and a G3DMark score of 14009, this is worth 42 points/$—which is fairly decent but still not as great as Gold-class GPUs. .
Bronze Series GPUs: Low end
This is the minimal graphics card of choice for any gamer worth their salt.
While it’s certainly not the best in the GPU hierarchy, it has excellent performance/dollar value and is almost as good, if not better, than Gold-class CPUs.
Take the NVIDIA GeForce RTX 3050, with an MSRP of $250, this GPU has a G3DMark/USD value of 51 points per dollar.
Likewise, take the AMD RX 6500XT. With an MSRP of $200 and a G3DMark IF score of 9324, this GPU has a G3DMark / USD value of 46.62 points / USD.
But then again, it’s not for QHD, let alone 4K gaming or gaming at above 60fps in FHD.
There are other variants of some GPUs (Max-Q design chips etc.) that are not included in this list. This is because these variants do not necessarily represent the GPU and have significantly lower performance. Basically, all portable variants were left out as well as GPUs designed for industrial use.
There are a few GPUs, such as the 1080 Ti, that have a good score with one benchmark (in this case PassMark: 18349) but not with the others.
This is due to the testing methodologies of the Benchmarking Service. In any case, the GPUs with the most balanced score on all websites are included in the list.
This list also does not represent the in-game FPS these cards will produce. Some game setups will benefit more from an RTX-enabled GPU, while others will benefit more from a graphics card with higher VRAM (eg TITAN RTX).
Hence, the exact performance of the GPU varies from game to game.
The current trend and improvements in the GPU market
The latest GPUs are now being built on 7nm and 8nm architectures by AMD and Nvidia respectively. This has resulted in a lower TDP for these cards compared to the older generations.
We’ve also seen a significant jump in performance (benchmarks) for each generation. For example, the RTX 3080 is 58% faster than its predecessor, the RTX 2080. (userbenchmarks.com).
This improvement can be attributed to better architectures, additional cores, improved teraflops, AI integration, etc.
AMD has also shown incremental improvement with each release. The RX 5700 XT which had a score of 71,091 on Geekbench, was replaced by the RX 6700 XT which had an effective speed improvement of 36% and had a score of 102,988 on Geekbench.
All in all, with innovations happening in the tech industry every single day. We can safely assume that there is more to come. According to rumors, Nvidia is set to launch its own RTX 4000 series, while AMD is preparing the RX 7000 series.
Each of these new generations of cards will have upgraded core technologies and fonts. AMD’s RX 7000 series is expected to be based on the new RDNA 3 technology, along with the new flagship Navi 33 processor.
It is expected that the AMD RX 7800 and 7800XT and the upper-tier RX 7900 XT will occupy solid positions in the upper hierarchy.
On the other hand, Nvidia is set to release its own RTX 4000 series later this year. The release time was supposed to be around mid-July of 2022, but has been delayed.
This series is expected to be based on Nvidia’s new Ada Lovelace architecture. The most interesting thing about these cards, however, is that they are expected to be based on a 5nm architecture which will reduce TDP and heating issues, as well as provide improved performance.
The more anticipated RTX 4080, 4080 Ti and 4090 Ti will be the standouts of Nvidia’s offering, landing firmly at the top of the hierarchy alongside the AMD 7000 series.