[Computerbase]Radeon HD 5870, 6970, 7970, 290X and Fury X in comparison

[PontiacGTX]

Test system and test results

For all benchmarks the graphics card test system with an updated Windows 10 was used. Since AMD supports the graphics card from the architecture GCN (Radeon HD 77xx) with the current driver, the Crimson ReLive 16.12.2 comes only from the Radeon HD 7970 onwards. The Radeon HD 5870 as well as the Radeon HD 6970 based on the VLIW5 or VLIW4 architecture have to deal with the legacy driver Crimson 16.2.1, which is not quite a year old.

Old and new games as a benchmark

14 different games are used for the article, which like the graphics cards are different old. So the oldest games come from the year 2011, while with Battlefield 1 there are also very current titles in the test parcours.

Settings optimized for graphics cards with 2 GB of memory

In all benchmarks, the highest graphic details are used in the resolution of 2.560 × 1.440, while in the mid-range models, the games ran in Full HD. This is a tribute to the performance class of the graphics cards. Only in the current games were partly the details reduced, since otherwise also the current generation does not reach sufficient Frameraten. In addition, the texture details have been reduced in some games, so graphics cards with at least 2,048 MB of memory have to contend with only minor losses. Supports a game DirectX 12, decides the highest Framerate, which result is used.

Benchmarks from Skyrim to Battlefield 1

In addition to the Radeon HD 7970, the Radeon R9 290X also makes a clear difference. On average, the speed increases by 50 percent compared to the predecessor. This is due to the higher raw material output and the fine art of architecture. The Radeon R9 290X can also increase the performance quite consistently through all games. Slips are few.

The Radeon R9 Fury X has problems with the utilization

The Radeon R9 Fury X and thus the last high-end graphics card from AMD, on the other hand, has some misfires. Therefore, the average performance is just 24 percent. The reason is in some games where the Radeon R9 Fury X can not use their raw power. An example of this is Call of Duty: Black Ops III, which is only eight percent faster than on the Radeon R9 290X. Others are Alan Wake and Company of Heroes 2, also they scale with Fiji badly. But there is also a brake in other titles: considering the naked specifications, the card should be a maximum of 50 percent faster, while in most games it is only about 30 percent. The problem is that AMD, despite the double number of units, uses only the front end of the half-sized Tonga GPU. This can not overload the over 4,000 shader units.

Looking back, the high-end graphics cards from AMD have "only" grown by a factor of 5 since 2009, and they are six times as fast in mid-range. The next big leap could be with Vega but imminent.

Source

 

[crisium]

https://www.computerbase.de/2017-01...diagramm-the-elder-scrolls-v-skyrim-2560-1440

At least the cards chosen have a more complete theme this time - each the fastest by AMD at the time of release regardless of price.

I still would have liked further reduced settings for the VLIW cards out of curiosity.

And thanks for reminding us how poorly Fiji lives up to its theoretical performance, Computerbase. 7970 to 290X should be about the same as 290X to Fury X. Hopefully the successor chip is designed better.

 

[Face2Face]

I will say Hawaii/Grenada will go down as one of AMD's best GPUs; the staying power is incredible. I'm still impressed with the 7970 GHz in my second system. I don't feel the need to upgrade it for 1080p.

 

[raghu78]

yeah Hawaii / Grenada will probably go down as the GPU with the longest staying power ever. I would not be surprised to see this chip perform well even in 2018 / 2019 .

 

[SPBHM]

power usage evolution is probably going to look very different on the Nvidia version of this.

 

[Mercennarius]

My 390X still performs great in the newest games... i've noticed it's performance increase gradually over time with newer drivers.

 

[Headfoot]

Seems strange to do this retrospective now on the very tail end of the Fury generation when Polaris is out and Vega is about to drop...

 

[nathanddrews]

Between this comparison and the previous mid-range comparison, the gap between TeraScale and GCN performance is rather ridiculous (over 200%). Looking at the NVIDIA comparisons on the same site, no single architecture has had a leap that large (170% is the highest that I saw).

ALL ABOARD the Hype Train! Next stops, Vega, Volta, Navi!

 

[PontiacGTX]

https://www.computerbase.de/2017-01...diagramm-the-elder-scrolls-v-skyrim-2560-1440

Spoiler: Comment

At least the cards chosen have a more complete theme this time - each the fastest by AMD at the time of release regardless of price.

I still would have liked further reduced settings for the VLIW cards out of curiosity.

And thanks for reminding us how poorly Fiji lives up to its theoretical performance, Computerbase. 7970 to 290X should be about the same as 290X to Fury X. Hopefully the successor chip is designed better.

If games had similar programming oriented toward Asynchronous compute and intrinsic shaders under a low level API, with proper multithreading it would give the scaling you would expect from those GPUs

but DX11 most of times is holding back the game developtment in AMD GPUs

I will say Hawaii/Grenada will go down as one of AMD's best GPUs; the staying power is incredible. I'm still impressed with the 7970 GHz in my second system. I don't feel the need to upgrade it for 1080p.

It stays relevant because it had the power to be used for most DX11 games and then with the asynchronous compute and shader features werent enabled on DX11 API would show its true performance, also Maxwell GPUs for comparison had worse performance on DX12 that would show that GCN was a better architecture for compute scenarios when it is done under the right presets

Seems strange to do this retrospective now on the very tail end of the Fury generation when Polaris is out and Vega is about to drop...

These are GPUs released as Flagship If they had to include Ellesmere they had to also add Tonga and Pitcairn
and if they want to add Vega they would have to wait more, and it wouldnt show the driver evolution over time these cards have gotten