Lucid HYDRA 200 Multi-GPU Technology Performance Preview
HYDRA Driver and Performance
The HYDRA Driver
When presented like this, the hardware part seems easy enough
(though we know there are some impressive interactions going on under
the hood). But anyone familiar with the world of multi-GPUs since the
technology’s revival in the GeForce 6-series of graphics will be able
to recount the horrors of driver incompatibilities, profiles,
performance drops and instability that buffeted the multi-GPU solutions
from both NVIDIA and ATI for quite a while. The HYDRA driver, though
unique in how it works, will have to stand firm in the same winds.
option-less for now. You can see here that a simple check box either
enables or disables the HYDRA technology and there is no delay, reboot
or screen flashing involved in the process. This is how SLI/CrossFire
should be working!
The option to show the logo in the game allows you
to have a HYDRA icon in the corner to prove HYDRA technology is working
– though I imagine it would off by default in the final revision of the
can also choose to enable HYDRA on specific games or disable it if
there happens to be some kind of software glitch or incompatibility
that needs to be addressed down the road.
desktop tray icon allows you to easily enable/disable the HYDRA engine
though we are told the entire face of the software will likely change
in the next major revision.
this slightly different icon indicates that multi-vendor rendering is
at work here. Obviously this will be cleaned up for the final retail
release – that is really an awful font for this purpose!
Scenarios to Watch For
With the Lucid HYDRA technology there are going to be a metric ton
of combinations to watch out for when it comes to performance testing
and evaluation. Luckily for us, because we had very limited time with
the hardware, we didn’t have much more than a few tons of testing and
data to go through. Here is what you should be on the lookout from us
- HYDRA scaling with two identical GPUs
- HYDRA scaling with two non-identical GPUs from the same vendor
- HYDRA scaling with two GPUs from different vendors
I detail every test below with some additional information so let’s just plow ahead.
Same Vendor Scaling (GTX 260+)
In this test we are going to be looking at same-vendor scaling
using the GeForce GTX 260+ as the basis for our testing. You will see
results from a single GeForce GTX 260+, a pair of GTX 260+ cards
running HYDRA (not SLI, it was not an SLI-ready motherboard ironically
enough) and the GTX 260+ paired with a GTX 285.
with the 3DMark Vantage results, you will see that the HYDRA scaling
method with the pair of GTX 260+ cards pushed performance up by 83% -
definitely a competitive solution to SLI! Considering the fact that
this motherboard, in its theoretical construction, didn’t have to pay
for any type of SLI licensing, I would say that THIS is the reason
NVIDIA might have put pressure on MSI to delay the Big Bang
In Call of Juarez the scaling going from a
single card to dual cards was around 80% while on Far Cry 2 it was just
over 51%. Both results are again pretty impressive.
When we take a look at the GTX 260+
combined with the GTX 285, the results are not what I initially
expected. As most of you, I would assume that a configuration with a
single GTX 260+ and a much more powerful GPU (the GTX 285) would
produce a higher combined framerate than the pair of GTX 260+ graphics
cards but that was not the case. Instead, the performance of the
260/285 combination happened to be nearly identical to the 260/260
results. Why is this?
Lucid tells us that the software based
algorithms for separating workloads across identical GPUs differs
greatly from the one required for load balancing with non-identical
GPUs and thus scaling will in fact differ. There is some software
overhead and load balancing overhead that has to take place and that
costs us some performance. We had to assume this would be the case but
it just kind of goes against standard logic: 2 + 2 = 4 but also 2 + 3 =
4 in the case.
Generational Scaling (GTX 285 -> GTS 250)
This series of tests expands on the results seen above by including
individual results for the GTX 285 as a single card configuration as
well as adding in a GTX 285 / GTS 250 combination in order to see
scaling across more GPUs.
we see here is that in both games the GTX 285 + GTX 260 card
combination outperforms the GTX 285 + GTS 250 combination – proof that
the HYDRA engine is working as promised to at least some degree. That
difference is much more noticeable in the Call of Juarez DX10 testing:
adding a GTX 260+ to the GTX 285 nets us a 52% performance gain over
the single GTX 285 while combining it with the GTS 250 (a much lower
powered card) produces a 15% performance boost. In Far Cry 2 those
scaling percentages fall to 31% and 24% respectively.
So while the CoJ testing impresses us from
a performance perspective and the Far Cry 2 performance does not, the
fact is that the technology is indeed working in some titles.
This is where things can get even more interesting – combining an
NVIDIA graphics card and an ATI graphics in a single system and seeing
them work together, as nature (but not corporate executives) originally
It works! In these two games you can see that simply put, the
combined scores of the HD 4890 and GTX 260+ card are better than either
individually. In the case of Call of Juarez the GTX 260+ adds another
40% or so to the gaming performance of the HD 4890 while in Operation
Flashpoint the combined cards run about 33% faster than either
What is important to note there is that the HYDRA technology will
take the processing power of either GPU on any game and work to the
best of its ability. Because on CoJ we saw that the HD 4890 was the
better card individually, the GTX 260+ supplemented it. And in
Operation Flashpoint the GTX 260+ was the better card going solo but
the combination still allowed the HD 4890 to aid for better overall
performance. In both cases, the “faster” card was able to save about
50% of the total performance of the “slower” card and add that into the
gaming experience for the user.