Well, the first reviews of the new Radeon 6900 series are out… AMD is going for a new approach in their architecture. So, let me just sum up my thoughts based on these early reviews:
- The ‘standard’ graphics performance is good, the new 6970 is faster than the 5870 and 6870. It is however not the fastest GPU available, the GeForce GTX580 is in a league of its own. The GTX570 is a close match to the 6970 though.
- However, like the 6800 series, the 6900 series seems to be relatively less efficient at running compute tasks than the old 5800 series was. In fact, the 5870 appears to be as fast or even faster than the new 6970.
- Power consumption and die size have also gone up, compared to the 5800 series.
- The tessellation performance is not what AMD had promised. It still has the same exponential dropoff problems that plagued the 5800 and 6800-series.
I’d like to get into the tessellation bit some more, as I’ve been discussing the topic a lot already. AMD had promised us ‘scalable’ tessellation. Well, let’s look at their chart here:
Do you see the problem that I’m talking about? We already know that the 5870 has a significant performance dropoff with higher tessellation factors, making the use of tessellation limited to low tessellation factors. But look at the charts. The 5870 is used as the 1.0 scale, and effectively the 6900 series follows exactly the same line. It has some ups and downs in the low tessellation factors (which was never a problem for the Radeons anyway), but after that, it essentially is slightly over twice as fast as the 5870 at all higher tessellation factors. Which means it follows the same dropoff curve that the 5870 has! This is NOT the scalable tessellation that AMD promised us. In a way they only made the tessellation performance more extreme. Since it has a peak in the low factors, effectively the dropoff will be even worse, even though in the absolute sense it is still faster than the 5870. But it’s still only twice as fast as the 5870 in the problem areas, and that’s nowhere near good enough to catch nVidia. See for example how much faster then GTX480 was than the Radeon 5870 in TessMark. Even at ‘moderate’ settings it was already twice as fast, let alone the ‘normal’, ‘extreme’ and ‘insane’ settings (if I’m not mistaken, the tessellation factors in the TessMark tests are 8, 16, 32 and 64 respectively).
I would love to see what the 6970 makes of the Endless City demo, but I doubt that its ‘improved’ tessellation performance comes close to the GTX570/580. I suppose it will be similar to the 5970, since effectively it also has two parallel tessellation units, and about twice the tessellation performance of a 5870.
So I’m rather underwhelmed with the new 6900 series. Purely as a gaming card it is okay… but it does not appear to be a very forward-looking card, as it appears to trade off this gaming performance for compute performance, actually being slower than the old 5870 in some cases. It also still does not solve the tessellation bottleneck. The DirectX 11 and OpenGL 4.0 specifications allow for tessellation factors up to 64, and AMD still only performs well in a very small subset of that range, while its main competitor nVidia has hardware on the market that handles the full range adequately.
Another thing I don’t really like about it is that the power consumption has gone up quite a bit, compared to the 5870. For the past few years AMD has been all about making small, efficient GPUs, and the 6800-series were another nice example of that: nearly 5870 performance, but with lower power consumption, and cheaper cards as well. Speaking of price, AMD passed up the opportunity to price these card aggressively. The 6970 is priced about the same as the GTX570.
All in all, I’d say nVidia and AMD have gotten closer together in the past months, with price, performance and power consumption. I would say that nVidia has come out of this as the winner, since AMD failed to beat nVidia’s GTX580 performance in games, and nVidia maintains its lead in compute, tessellation and things such as PhysX and 3D vision. The power consumption between the two brands is now close enough that it is no longer an important factor in the decision process, I’d say. And with AMD not pricing their cards very aggressively, I’d personally go with nVidia for now. For the same price you get similar gaming performance and power consumption, but nVidia delivers a more complete package overall.