- Category: Z77X-UD5H
- Published on Thursday, 27 June 2013 19:14
- Written by Super User
First of all if you want to know what everyone else does about MVP, more than what I have to say, please read up on it through Lucid’s Whitepaper on MVP, it is really easy to read and explains everything much better than anyone else could.Here is the White paper. It isn’t technical, and all the pictures you see on articles like adnatech’s come from it.
Now is it just me or are we seeing huge boosts in iGPU technology recently? For me that points to the evolution of iGPUs increasing in performance at a much faster rate than the dGPU. It is leading to their actual usefulness for processing. As we saw Sandy Bridge’s iGPU is extremely good at video transcoding, however is it just as good at dealing with discrimination? The discrimination of frames to be processed and not to be processed is a huge task which has been offloaded to the iGPU with the introduction of Lucid MVP. It offers users that HUGE performance boost in their FPS, and it is what all the hype is about.
So what the heck is MVP? Well in short MVP is a program made by Lucid, the guys who made the iMode and dMode switching program we saw on Z68(Virtu) as well as the Lucid Hydra which was a multi-GPU PCI-E bridge which wasn’t really that good. This time however they have been at work with something special. Lucid says that by being able to take into account human responsiveness as well as screen refresh rates they are able to process things more efficiently, but not at the level at which you think. They don’t speed up rendering, instead they allow you to not render things you would never see and then they also move some of the dGPU’s tasks to the iGPU, this is called HyperFormance. They also have a new VSync option that allows VSync up to 120FPS, called Virtual Vsync. Basically MVP allows your iGPU to work in tandem with your dGPU, but not like in SLI. In fact increasing the iGPU frequency or memory allocation doesn’t help the MVP scores at ALL!! So they aren’t using the iGPU to process the actual rendering, however the iGPU is being given the output buffering task, and the program is allowing the total removal of frames in the processing que which the user will never see on the screen. MVP takes control of the information following to the dGPU as well as that which the dGPU sends back to the CPU and iGPU, so that the dGPu can be better managed.
Here is my own diagram on how it works:
The iGPU and CPU are first determining what not to process and what to process and in what order. Then those frames are sent to the dGPU where they are processed, and at this point the iGPU knows what frames it should get back and at what time since it has calculated all of this, it then processes the output buffer and the image is displayed on the screen. Since the iGPU and the CPU control the flow of information to and from the dGPU at all times, they are basically making the dGPU the slave and the iGPU and CPU the master, which in turn allows hierarchical organization of the information being sent up and down stream. This should lead to much more efficiency processing, however this “peer to peer”, more like “master-slave” communication is going to cause some microstutter. We will take a look at that later though.
HyperFormance, because of what it does, greatly increases the FPS measured, as while the frames aren’t being rendered faster, the scenes as a whole are being rendered at a faster rate. So instead of rendering all 2000 frames for a scene the GPU will only render 1000 or so, now if the GPU does this at the same rate as it did before then the scene can be rendered in half the time. MVP takes into account your responsiveness and it also uses a pretty good algorithm to do this as the picture quality doesn’t take much of a hit, at least I couldn’t see the difference. Below I have added a picture, please tell me if you can chose which is with MVP and which is without, but please note this is the first image I saw when the level started:
I will show you which are which later. I expect this program not to be perfect, I expect it to have flaws, however why would the picture be worse at a given frame? Is the GPU rendering worse than it would normally? The answer is no, however the way the frames are being put together would allow there to be issues such as when there is tearing, and that is where their Virtual Vsync comes into play. I left Virtual Vsync on and I guess it works in the background because my FPS in this game at this resolution I used are extremely high, like easily in the hundreds. I did however have no tearing whatsoever. In the benchmarks above we saw the huge performance gains from MVP, so I used one of the games that worked flawlessly with MVP, which was Call of Duty: MW4 for my analysis. In my mind the HyperFormance, the mode of the program which allows for this removal of frames, is like a broader implementation of Z-Culling which is a technology from NVIDIA which eliminates pixels that the user will never see from being rendered. In this case the elimination of whole frames is taking place.
This huge rise in FPS exploits vulnerabilities in many popular benchmarking programs such as 3DMark, which angers not only 3DMark, but it also makes SLI and CrossFireX looks un-needed since MVP doesn’t fully support SLI or CrossFireX. The pure FPS number gives the user a feeling of “OMG that is damn fast!!! “
However that FPS number isn’t always as accurate as it can be, and it doesn’t take into account when there is an issue with the removal of frames. Sometimes there will be an error and the previous frame will be displayed again instead of the frame that was supposed to display. Lucid says this is a bug.
Now there is a slight amount of micro-stuttering as we said before because of the master-slave communication, as I like to call it. There is a lot of it going on in the background, as you would expect because of the dual GPU (iGPU and dGPU) which means two things. First of all it means that there can be large spikes of performance followed by low drops, however we don’t seem to see this affect in the games, but it also means that there is actual master-slave communication going on between the dGPU and the iGPU+CPU. Which means the iGPU isn’t being used just for show, but is really dealing with sorting tasks in which a GPU is needed for but which the dGPU isn’t being bothered with.
So to test this we use a program made by a user on XS, and he wrote a small DOS script which takes the FRAPs output of game play, and gives you a % index of microstuttering, and a fixed average FPS! So it takes into account the microstutering and gives you a more normalized frame rate that you actually witness.
Here are the FPS graphed in excel:
I used a single GTX570, then turn on hyperformance, and then used SLI GTX570s.
Here are the results from his program:
Here is a table with them organized, as well as a percent increase over a single card:
So it seems that when the average FPS is basically normalized, it drops right below that of the FPS when I used SLI. That to me is exactly what MVP should be chiming in at, it shouldn’t be faster than dual GPU, but it should be a lot faster than single; however being way above both single and dual GPU is a bit unbelievable. This type of analysis makes the FPS of MVP more believable, so I would encourage 3DMark to make create something like this for when it detects MVP. To me that is more of a real FPS number, which in my opinion makes a lot more sense than one that is sky high. The program takes the difference and then averages it out, even though the effects of the microstutter aren’t noticeable, that index is pretty high. The peer to peer communication still exists, but not as peer to peer communication. However I did run the tests a bunch of times, and sometimes the index was as low as 25% and sometimes as high as 65%, the only difference is how I play and what I do in the game. Above was when I got it down to a steady execution of gameplay in which little variance could be observed.
So folks, if you want to see MVP’s impact:
Above is MVP without having its FPS altered by the program I used above.
BTW here are those pictures:
Lucid MVP isn’t a gimmick, it isn’t just a piece of software manipulating FPS, it isn’t just another broken piece of software from Lucid. MVP really is something special; it is part of the evolution of the iGPU, and is needed for further enhancements in such areas of computer engineering. Not because of the high FPS rating, but because of what it stands for and what is does. So please don’t be offended by its high FPS, don’t be mad that it can beat your 7970 in FPS flat out, don’t be sad, instead be glad that we are being provided software that has accomplished what no other GPU manufacturer has been able to accomplish. This type of software and thinking benefits us! It for once doesn't benefit the big GPU companies, but it does increase GPU performance. This type of software is what we should be seeing from small companies like LucidLogix, this is a great idea which they have executed very well, they have been upfront with issues, and asked users to report problems, as well as honestly made a program that does what it says. So try it out and see how you like it. It seems like Lucid is doing a good job to taking feedback, so if you have issues please report them!