The HD 7970 vs. the GTX 680 – revisited 3 years later

The GTX 680 versus the HD 7970 was a classic confrontation that began in 2012 that we are now repeating 3 years later.  Codenamed Tahiti, the then new AMD flagship Radeon HD 7970 was announced on December 22, 2011, and it launched the next month at $550 with disappointing drivers which only placed it originally about 20% faster than the Nvidia Fermi flagship, the GTX 580.  However, the primary competitor to the HD 7970 became Nvidia’s GeForce GTX 680 which launched at $499 in March, 2012 as Kepler architecture, with a strong emphasis on performance per watt and efficiency.

The GTX 680 is a 2GB vRAM-equipped mid-sized GPU on a 256-bit wide bus compared with the larger and more expensive to produce HD 7970 which is equipped with 3GB of GDDR5 memory on a 384-bit bus. Although the HD 7970 has higher bandwidth and better specs on paper, the GTX 680 was originally judged faster overall than the HD 7970, mostly due to better DX11 optimized drivers.  This is what we had to say in our conclusion of the GTX 680’s launch review on AlienBabelTech in April, 2012:

No matter how you add it up, the GTX 680 is generally faster overall than the HD 7970.  Both overclock superbly so neither card has much of an advantage over the other with headroom on stock clocks and with the stock fan profile.  As the resolution increases, the performance advantage of the GTX 680 diminishes a bit over the HD 7970 and the gap narrows.  AMD would have to significantly increase the clocks of the HD 7970 to catch the GTX 680.

Here is our performance chart from this 2012 evaluation:

Just as we predicted shortly just before the HD 7970 launched, it turned out to be a short-lived SKU at 925MHz/1325MHz which was subsequently replaced by the AMD-overclocked HD 7970 GE (Gigahertz Edition – 1000MHz/1500MHz), and then respun the following generation into the nearly identical R9 280X with the same GE core speeds.  In fact, our reference PowerColor HD 7970 and our VisionTek R9 280X will CrossFire with each other and will perform almost identically when set to the same clocks.  Although the brand new R9 380X has nearly identical performance to the 280X/7970-GE, it is based on the similar but slightly more efficient Tonga architecture, not on Tahiti.

Our game benchmark suite has also changed from 2012.  We have also added minimum frame rates in addition to averages, and we are now testing at resolutions up to 4K, with a strong emphasis on the very latest games.  We want to see how these two formerly flagship GPUs stand now in relation to each other by playing the latest games with updated drivers at demanding settings that we usually bench at.

Over the past 2 years, Nvidia has reworked their Kepler architecture into the even more efficient Maxwell, and the 4GB vRAM-equipped GTX 970 has replaced the 2GB GTX 680 as a faster and more power-efficient GPU by Nvidia in their line-up.  We want to test how well our original HD 7970 performs after also being degraded from AMD’s flagship, and after being replaced by the 290X, the 390X, and then by the Fury X in turn.  We want to see how it can perform now against our same reference GTX 680 about 3 years after their respective launches.

It will be interesting to see which GPU was actually more “future proof”, and if the 2GB of vRAM that the GTX 680 is equipped with has any current disadvantage to the 3GB of vRAM the HD 7970 is equipped with.  There has been a major console platform update in the meantime, and many new PC games require more vRAM  than previously.  So let’s compare today’s performance of the HD 7970 versus the GTX 680 after we look over our test configuration on the next page.