"Lucid's technology represents a new technological approach that can enable multi-GPU processing independent of the vendor. Through this approach, consumers will be able to upgrade to multiple GPU processing and load balancing, with less complexity and without being locked into a particular vendor...
...The HYDRA engine analyzes the frame before rendering and intelligently distributes the rendering tasks between the GPUs on board. The frame decision mechanism resolves bottlenecks and inter-frame dependencies prior to rendering, in real time, such that there is near zero latency."
What they are saying is that a board with the Hydra chipset will allow us to combine two types of graphics cards in one system for enhanced performance. This means we could for example choose to use a Radeon 5700 series graphics card then, rather than buy a second for CrossFire, we could use an NVIDIA card as our second GPU to boost performance in games. In addition to this manufacturers also seem keen to push this solution as a way to get SLI working on AMD chipsets.
On paper this sounds an attractive idea but in practice the results are far from impressive.
Regarding the benefits of multi-GPU systems, we have spent a huge number of hours over the past few months testing with various configurations of AMD and NVIDIA cards. In every case, for real world performance, the results were terrible. To give an indication of what we mean here are some results for Crysis Warhead on a system with the most recent Lucid/Hydra based board, latest BIOS, OS patches, DirectX, game patch, Lucid drivers and video drivers.
Crysis Warhead 1920x1080 - Enthusiast
It is clear from the above table that there was absolutely no scaling in SLI mode. It just didn't work at all, despite using supported hardware, the latest drivers and a well-established game. Using mixed graphics cards with similar features and performance had a 2fps advantage over the single NVIDIA card but nothing like using two of the AMD cards in standard CrossFire.
The same results trend was shown across multiple products on multiple boards with multiple Lucid/Hydra/AMD/NVIDIA drivers. We know the technology was active as a watermark is added to our games (it can be disabled) when everything is working together but despite this, there was little to no performance benefit.
So our testing tells us that the technology currently does not work well. What are the chances of it being fixed? In our opinion they are not good unless there are changes made to the way that Hydra works. Currently the technology relies on a set of profiles within the driver. In theory if a game has a profile then it should scale/work and if there is no profile we can either wait on official support or add a profile manually and hope for the best.
In practice profiles are no bad thing, NVIDIA and AMD use them for their drivers but they also have more experience, a more mature architecture and more resources than Lucid. In addition to this, they are not trying to add an additional layer into an already complex scenario. The sad thing is that the game tested above, Crysis Warhead, is actually a game with a profile. If that doesn't work well then what hope do we have for new releases?
To answer that point, take Mafia 2 as an example. The game was released on the 24th August and the most recent driver was made available on the Lucid site from September the 7th, 2 weeks after the game was released there was still no profile and to this date (27th September) there is still no driver with a Mafia 2 profile. People will have finished the game before they will get any benefit from their Hydra technology. Looking to one of the more recent games in the list of supported titles, StarCraft 2, it comes with a note that in some cases "Textures are missing". 2 months after launch those with a Hydra setup are still experiencing issues with one of the biggest selling games of the year. This is unacceptable.
It is also worth noting that when we use Hydra rather than SLI we lose the ability on NVIDIA cards to enable 3 screen desktops (and 3D Vision Surround etc.) due to NVIDIA locking them out.
To resolve this situation a few things would need to change.
The most ideal fix would be for Lucid to figure out a way that Hydra could work without the need for profiles. Get it to a stage where we plug in two cards and they just work. As much as we would like this, it is unlikely to happen.
Therefore it has to be fixed on a software and process level. Lucid need to get their foot in the door of every developer; get their staff in these studios working on the games as they are created to ensure that any bugs are ironed out before release. In addition to this, Lucid need to get employees working directly with ATI and NVIDIA, ensuring that their video drivers are optimised to work with Hydra. This will be a costly process but if they are serious about making Hydra a success then it will be one which is necessary.
In addition to this, Lucid also need to bump up the number of developers working on their drivers to ensure driver updates are released on a very regular basis... probably weekly... and some significant work needs to be done on making the technology easy for consumers to understand. We have lost count of the times we have seen conflicting information on the Lucid site when compared to motherboard manufacturers own site. It gives the impression that even those making the boards don't actually know how a major feature works. Further to this, someone in marketing needs to think of some user friendly terms for the technology... telling end users that one of the multi-GPU setups available on Hydra is "Raid Mode" will confuse more people than it will help.
There would have been one saving grace for Lucid and their Hydra technology. If we could have plugged in an AMD card as the primary display and then, through Hydra, enabled a 2nd GPU based on an NVIDIA card for PhysX we would have a killer feature. Consumers would love to be able to use PhysX on AMD Radeon based systems but even with Hydra this is not possible. Considering how protective NVIDIA are about their technology we fear that even if Lucid were to find a way to enable this, NVIDIA would disable it in every driver from then on.
That is one of the major issues Lucid have in trying to establish Hydra in the marketplace... if their technology is successful it will cause other, larger companies to lose money. As things stand, NVIDIA make money from PhysX only being available on their cards. Licence fees are paid from one company to another for use of multi-GPU solutions on chipsets. AMD have a unique selling point in having their own cards be the only multi-GPU solution on their chipsets. The list goes on and it is unlikely that any of the major players will want to take a decision which will cost them in the area that matters, finances.
So where do Lucid go from here? In our opinion all is not lost. Plug some cards into the Hydra chipset and fire up benchmarks like 3DMark or Heaven 2.0 and we do see performance scaling. This means that in a fixed, closed environment there is a use for the technology. For this reason it could well be something which is best applied to the professional space. Allowing business users to increase their performance by using Hydra for very specific applications could well be beneficial to productivity and finances. In the consumer marketplace though, for actual real world use rather than fixed benchmarks, it is our opinion based on actually using the technology that Hydra has no place on any consumer product
now, or in the foreseeable future. The sooner that motherboard manufacturers realise that 3DMark scores are not important we may see them move away from this technology but as things stand they are blinded by irrelevant percentage figures and clearly haven't actually tried to use the products in real world situations which is hugely disappointing. It shows a complete disregard for their customers and a lack of understanding about the market, hugely surprising for some of the largest motherboard manufacturers around.
As things stand Hydra wastes your money and time and that we cannot support.