Technology, Delays and Testing Setup

The gang at Lucid finally let us get our paws on some actually testing-ready hardware this week and I ran the configuration through some tests to see how it stood up compared to existing CrossFire and SLI systems. We also of course test cross-vendor performance and see how much scaling you can get when you combine the performance of a GTX 260+ and a Radeon HD 4890.
Lucid HYDRA Technology Background

Even in its short life, LucidLogix has gone through quite a few interesting twists and turns to get to where it is today.  We first saw Lucid back in September of 2008 at the Intel Developer Forum where the startup was showing off a revolutionary, and somewhat controversial, multi-GPU rendering scheme that claimed to support different generations of GPUs in the same system.  At the time, with Windows Vista at the helm, only a single vendor could be utilized at one time (NVIDIA or ATI) but the ability to pair a GeForce 9800 GTX with a GeForce 9600 GT (as an example) and see some performance scaling was appealing to a great many users, as you might expect. 

The methods Lucid is using in its HYDRA technology, that includes both a software / driver layer as well as a custom ASIC with PCI Express lanes and RISC processor, differs greatly from what both ATI and NVIDIA do with their own GPUs.  While the graphics card vendors use solutions like AFR (alternate frame rendering) and SFR (split frame rendering) in their driver, and require identical GPUs (in ATI’s case, MOSTLY identical), the Lucid HYDRA technology actually intercepts DirectX information and has the ability to break up the graphics processing workload by task, image or object at a very granular level. 

Lucid HYDRA 200 Multi-GPU Technology Performance Preview - Graphics Cards 21

The best way to describe it is to look at this picture:

Lucid HYDRA 200 Multi-GPU Technology Performance Preview - Graphics Cards 22

What you see here is the output of each GPU in a HYDRA-enabled system before the image is recombined in a single graphics cards frame buffer for output to the monitor.  You can see that individual items like the weapon, the foreground and the background are divided up between different GPUs based on the company’s internal load balancing algorithms. 

The promise of Lucid today though is that of complete freedom of choice in GPUs; including same-vendor acceleration and cross-vendor acceleration.  The ability to add in a new GPU, of almost any type, to your system and see tangible gaming performance benefit is something unprecedented in the world of visual computing.  While NVIDIA and AMD have their own stances on the world of multi-GPU computing (AMD is more open while NVIDIA tends to lock things down and license them) if the HYDRA technology comes to fruition then a truly open standard could exist.

If you want more technical background on the Lucid HYDRA technology and how it works, I would highly recommend you check out my two previous articles on the subject that dive into MUCH greater detail on the history and implementation of this exciting new offering.  Today we are focusing on the here and now: initial testing and performance. 

Also, e have a great video of about 20 minutes in length that looks over this new motherboard, shows a couple of Lucid’s demos on working systems and talks with the co-founder Offir Remez about Lucid’s technology and plans for the future.

Delays, delays…

Let’s get the political side of things out of the way up front.  We had expected to have final, retail ready hardware in our hands by now but obviously that hasn’t happened.  The MSI “Big Bang” motherboard based on the P55 and HYDRA chipsets was supposed to be launched on October 29th according to a counter on their own landing site but instead on that date a new, and completely unknown, motherboard featuring the P55 chipset and the NVIDIA nForce 200 bridge chip launched in its place.  This has led many to believe that pressure from NVIDIA on MSI has delayed the launch of the Big Bang motherboard as it represents a shift in the discrete graphics market that probably doesn’t line up with NVIDIA’s goals and vision for the future.  MSI stated to us that some driver issues on Lucid’s side of things were holding back the release while Lucid clearly told us that the driver was ready to go and that they didn’t know what the hold was.

Obviously something is amiss here – MSI has a lot of money invested in these motherboards and wouldn’t just hold back the board after some significant buzz had built around it for no reason at all.  Based on various conversations with others in the industry I do believe that some pressure from NVIDIA fell on MSI about the Big Bang motherboard and that likely had some influence.  If it’s true, shame on NVIDIA for attempting to “pull an Intel.”  There also might be some truth to the software complaints on MSI’s side of things. 

UPDATE (11/11/09):  I spoke with a couple of representatives from NVIDIA today and they had some interesting information on this subject.  NVIDIA flatly denies placing any type of pressure on Lucid or MSI in regards to any Lucid-based products, schedules or releases.  NVIDIA also wanted to point out that the nForce 200 based MSI Big Bang Trinergy motherboard had been in development for quite some and in fact was show at CeBit this past March; so that kind of goes against the argument that MSI and NVIDIA “snuck in” a new motherboard at the last minute.  We are hoping for more comments from all players in the coming week.

In my testing there were a couple of instances of questionable compatibility with some multi-vendor configuration testing though Lucid still firmly reiterates that they claimed multi-vendor driver support would be “ready in Q1.”  And that’s true – we quoted as much in September 2009 at IDF.

The real answer is likely in the middle here – some pressure from NVIDIA on one side and a desire to have multi-vendor support moved up in the schedule on the other resulted in MSI’s Big Bang launch getting pushed back to … well we don’t know yet. 

To help with their own PR, Lucid wanted to meet with us to show the technology was working, that it DID indeed scale in performance and that they still believe the driver is ready to go for same-vendor scaling operations.  I took an afternoon earlier in the week to work with the company’s CEO and get some hands on time with the goods.

The HYDRA Testing Configuration

Lucid, obviously sensitive to the fact that MSI remains a key partner going forward, did not want to simply upstage them by showing benchmarks running on an MSI motherboard – that would basically be a virtual finger-pointing.  Instead we were able to run some HYDRA tests on a development platform that gives us the same functionality in a somewhat more complicated form factor.

Lucid HYDRA 200 Multi-GPU Technology Performance Preview - Graphics Cards 23

What you are seeing here is a single system running in two cases; the bottom system houses a Gigabyte X58 motherboard and Core i7-920 processor (but only 2GB of memory for some reason) as well as a PCI Express 2.0 bridge card that connects the system to the top chassis.

Lucid HYDRA 200 Multi-GPU Technology Performance Preview - Graphics Cards 24

Inside this case is a Lucid HYDRA development board that consists of a HYDRA 200 chip and four PCIe 2.0 slots for graphics cards to be installed.  It is powered by a separate power supply that had to be turned on before the primary system was booted.  The GPUs installed on the development board were the ONLY graphics cards in the system and we had the monitor plugged directly into the primary one.  In this picture we are testing the multi-vendor graphics configuration with a Radeon HD 4890 up top and an EVGA GeForce GTX 260+ underneath it. 

Lucid HYDRA 200 Multi-GPU Technology Performance Preview - Graphics Cards 25

Here is a better view of the connection between the two chassis just as a clarification for users curious how this is actually working.  Note the circled areas – this is the high speed cable that runs between the development board and the PCI Express 2.0 bridge card on the Gigabyte motherboard.  The connection provides 95-97% of the bandwidth that a direct PCIe 2.0 connection would have so our results should be close to the final product. 

« PreviousNext »