Anybody with a Rift and GTX Titan or AMD equivalent

Thanks for all the great feedback.

I am finding that the vehicle models I loaded were in .obj format which Unity obviously hates after reading about importing. I swapped to .fbx and now they run brilliantly, if slowing a little as they are fairly complex models.

I think I will upgrade my desktop to quad core and gtx680 4gb, and then buy a replacement Macbook Pro when the refresh happens in June, fingers crossed for faster GPU but I feel that will be my best bet as I need to take this around and demo it. For the record I can run this demo on my existing 3ghz dual core, Nvidia 6800m Macbook Pro but at around 5fps :-)

I will be doing some work on the actual avatar control next followed by some app store control by avatar. I have uploaded a fuller physics version and the vehicles are back in.

Thanks for your help and to all the devs allowing me to put there apps on my app store.

All the best

"amoceri" wrote:
I have a 690 and I love it. I have not had any issues with stuttering due to SLI. I would be happy to test it out for you.

You need two or more cards to use SLI, so of course there CANNOT be any SLI stuttering with only one card (and therefore no SLI).

"

"geekmaster" wrote:

"amoceri" wrote:
I have a 690 and I love it. I have not had any issues with stuttering due to SLI. I would be happy to test it out for you.

You need two or more cards to use SLI, so of course there CANNOT be any SLI stuttering with only one card (and therefore no SLI).

"http wrote:
Scalable Link Interface (SLI) is a brand name for a multi-GPU solution developed by NVIDIA for linking two or more video cards together to produce a single output.

A 690 is a dual gpu card and must run in SLI mode to get the performance of the full card. If you do not run it in SLI mode, you get the equivalent speed of (1) gtx670

A 690 has (literally) two gpu dies on it (both equilavent to the speed of a gtx670 separately, so when in SLI, it allows both of them to function, giving then the power of two 670's, but the RAM of only one [SLI only permits accessible VRAM from one die or card in SLI mode, the other cards vram is mirrored), and 2GB's of RAM per die. It is substantially longer than a single gpu card, and uses up quite a bit of power.

The 670, 680 and Titan are all single gpu cards.... however even though the titan is only single gpu, it is equivalent to having one 680 and 3/4's of another in performance. with a full 6GB's of VRAM to use. So it is about 75% (on average) more powerful than a single 680, without having to run in SLI. A 690, however, must run in SLI even with one card and the game/drivers have to support SLI to get the performance gain, BUT it can run about potentially 90% faster than a single 680, or about 100% faster than two 670's. Having (2) 680's and then running them in SLI, however, will give you 5% to 10% faster than a single 690.

So this is why the Titan is a huge game changer... its single gpu that competes with dual card or dual gpu solutions (like the 690 and amd's radeon 7990 [which is also dualgpu and must be run in crossfire]) and does not have all the woes of running SLI or size, power and heat issues of dualgpu cards or losing VRAM because of SLI or Crossfire.

"captaintrips" wrote:
A 690 is a dual gpu card and must run in SLI mode to get the performance of the full card. If you do not run it in SLI mode, you get the equivalent speed of (1) gtx670

A 690 has (literally) two gpu dies on it (both equilavent to the speed of a gtx670 separately, so when in SLI, it allows both of them to function, giving then the power of two 670's, but the RAM of only one [SLI only permits accessible VRAM from one die or card in SLI mode, the other cards vram is mirrored), and 2GB's of RAM per die. It is substantially longer than a single gpu card, and uses up quite a bit of power.

The 670, 680 and Titan are all single gpu cards.

SLI is for connecting CARDS together, not GPUs. If what you say is true, then this Nvidia page contains a logical conundrum:

"
Whether or not a GTX 690 uses SLI to connect its GPUs is irrelevant anyway. Micro-stutter is not a property of SLI, but rather the method chosen to use the GPUs for parallel processing. Micro-stutter is an inherent property ONLY of AFR (Alternate Frame Rendering). The alternative methods of parallel rendering do NOT suffer from micro-stutter. The "safe" methods include Scissors Interleave, Checkerboard Interleave, and Scan-Line Interleave. In fact, "SLI" was originally the acronym for "Scan-Line Interleave", so it could be said that micro-stutter is a side-effect of using SLI in a way outside its original design specifications.

So whether or not any multi-GPU combination has micro-stutter is a matter of how it is used, as determined by the device drivers. So one could say that it is really a software problem than a true hardware problem.

I know 110% fact that the 690 has to run in SLI to get the full card performance. I know someone that use to have one, and I know quite a bit about the hardware. Just as the 7990 will have to run in crossfire.

All they've done is instead of 2 separate cards, they've literally taken the (2) GPU's, made the card longer with a bigger heatsink, and put them both parallel. That's how dual gpus cards work. They are literally (2) cards condensed to one PCB, using one controller, and running in scalable link.

Now, you are partially correct. There will be more microstutter on (2) separate cards, because there is some technology they can implement on a PCB with two gpu's that reduces the microstutter slightly... HOWEVER, your average framerates will be rollercoaster on a dual gpu pcb, such as a 690 or 7990 more than two separate cards running in SLI, because while it still has dual gpu's, there's data bandwith limitations of having both both physical gpu's on a single PCB vs, two separate PCB's.

While (2) 680's in SLI will have slightly more microstutter, it will have a better average framerate. The 690 is an opposite.... less microstutter, but a framerate that fluctuates more. A single card with a single gpu, however, will have almost zilch microstutter (if any), and a far steadier constant framerate.

Sure, a 7990 or a 690 might be 10 or 15% faster than a titan when a game supports SLI, but a single titan will have a higher average framerate most of the time, while the other two skyrocket and drop and rollercoaster all over the place when it comes to framrate. Technology will get better for dual gpu cards, and SLI all together, and in the years to come, it will be reduced with issues in microstutter, cooling, power, framerate averages, etc..., but a single gpu is now and will always be the more stable solution.

For example, a 690 might hit 110fps in a game under max settings, and hit a low of 50fps, and average around 75fps.
A titan might hit 105fps in a game under max settings. and hit a low of 60fps, and average around 90fps.That's the rollercoaster framerate difference I'm talking about.


REGARDLESS... if you have a 690 or a Titan, you are in gaming heaven, and both are ferocious cards. There's just multiple paths one can take, each with their own tradeoffs. Even if you have a 670 or 680, you still have a behemoth of a gaming machine, and should be mighty proud of that beast.

The only question left that hasn't been asked is: "If Titan is single gpu, and the 690 is dual gpu card, then why is the Titan $200 more?"

Simple... the technology in the Titan is far more expensive, especially costs to cover the research and development. It took ALOT of effort to get nearly the high performance of the 690 on a single gpu. And this tech was mainly reserved not for gaming or enthusiasts, but for supercomputing... in their Tesla gpu line. So they basically took what they had learned with the Tesla technology, applied it to a GTX-brand, added non-ECC RAM (and lots more of it) since you don't want ECC ram for gaming, and that is why it costs $1200 a card. (a Tesla card will cost about $4000 a card)

With a 690, its far cheaper for them to take an already existing gpu, make a longer board, slap another gpu on it, and send out to sea.


The reason I run two titans instead of two 690's is that the titan is almost as fast as a 690, and it keeps better average framerates. So when I run two titans in sli, im getting almost the equivalent of two 690's. "So what?" you ask? Here's the difference. A single 690 already has microstutter and whackier framerate averages because its already running in SLI mode. Add another 690, and now you have 4x the amount of microstutter, framerate whackinees, power requirements, possibility of driver or game incompatibility between 4 gpu's (2 on each card). And even though each 690 card has 4gb's of RAM, and you would think you are getting 8GB's of vram... your not. Your only getting 2GB's, and the remaining 6gb's are mirrored of the first gpu. That's because the RAM is independent to each gpu as it has to be, and thus, in SLI mode, you only get the VRAM of one gpu (and as I stated earlier, each gpu on a single 690 is allocated 2GBs of vram). On the Titan, I end up with 6GB of VRAM in sli.... 6gb's per card. So the other cards 6GB's of VRAM is just mirroring data of the first.

Don't believe a word I'm saying? you can start here...

http://forums.guru3d.com/showthread.php?t=372630

and

https://forums.geforce.com/default/topic/542290/gtx-690-or-titan-or-2-680s-in-sli-which-to-buy-btw-prolly-most-common-question-on-web-at-moment-s/

"captaintrips" wrote:
...
Don't believe a word I'm saying? you can start here...

http://forums.guru3d.com/showthread.php?t=372630

and

https://forums.geforce.com/default/topic/542290/gtx-690-or-titan-or-2-680s-in-sli-which-to-buy-btw-prolly-most-common-question-on-web-at-moment-s/

Oh, I believe you. You obviously know a lot more about this than I do, having invested both time and money on it. I only went by what Nvidia had to say, which is obviously obsolete...

I was not aware that SLI was also used inside the GTX690.

And thanks for the detailed reply. I learned something useful from it, and I appreciate it.
:D

Their ultimate goal years ago was to one day have a more unified architecture for dualgpu cards (or multigpu machines in general), where all the GPU cross talk was done directly on the PCB and less through the drivers. You'd expect years later, that would have happened. It still has a long way to go, as does SMP with processors (now they just toss on more cores, though it causes similar crosstalk latency as would SLI, but its far more heavily optimized both in hardware and software when it comes to CPU's).

It's true, dualgpu cards that run in SLI and having multiple cards that run in SLI have made leaps and bounds with improvements over past generations, but still lack ability to stack VRAM into one physical allocation due to technical limitations, and still have driver restrictions, software restrictions, cooling and power concerns, and size problems.

I remember when I had SLI'd 7800gtx's back in 2005... and you want to talk about first gen Nvidia SLI woes... no wait, scratch that... let me tell you about the original SLI method used by 3dfx on my Voodoo 2's that I had linked together, lololol, ill tell you buddy... sometimes it wasn't worth the hassle. Now we've come far enough were SLI is stable, widely adopted and supported, and works a lot better.

You would think it would work as you suggested by now... in time it probably will.

However with the Titan and Tesla series.... it shows Nvidia is more concered with going forward of far more powerful CUDA cores and more capable single GPU architecture VRAM than focusing on a future of SLI. SLI will exist for a great many years to come, wheter it be dual cards or dual gpu cards.... but it's losing it will quickly lose its supremacy to powerhouse single GPU cards, and we've only really seen the first generation of those in the everyday consumer market...the Titan. They may even someday largely favor that for extreme high end gaming cards as the tech becomes cheaper to sell to the masses, and phase out SLI... who knows. I doubt it... SLI still serves a place, especially in cluster supercomputing and number crunching with the Tesla line.

However, there's also the new front of APU's making a sting... while CPU/GPU die intergration have existed for sometime, it's becoming a bigger wave. While it has it drawbacks right now, it seems to be picking up a particular interest in the mobile and console market, where you can't fit a gigantic video card in there. It also greatly reduces the time processes and calls are sent to/from the cpu and gpu, giving large performance boosts. It does have its limitations and drawbacks like anything, but I think what you may see going forward is Tesla/Titan-like Technology running more in unison with the CPU via die integration. How and when that will work, hell if I know... probably a few years down the road yet when it comes to High end gaming PC's. But you can already find that tech in the upcoming nextgen consoles, lets them have a little more competive technological edge against high end gaming rigs. Of course, a highend gaming rig would still runs laps around a PS4 or "next"-box technically, but those APU's keep the consoles up to snuff for where it needs to be. Ever wonder why your Hd43000/4000 intel integrated graphics mid-range laptop can play some decent games with only the hardware it has? That's a bit of the cpu/gpu integration magic working for ya... if more developers used specific driver development just for that chipset, it'd work even better than it does. The APU's take out all the driver overhead in CPU and GPU crosstalk (well not all, but a lot of it)

Quite simply "The right hardware in the wrong place can make all the difference.". By that I mean supercomputing gpu technology in the general consumer market... and gpu's directly integrated into the CPU die. And now, VR tech in your home.

Though I'd prefer not to "wake up and smell the ashes". That would mean my computers on fire. Kudos to you if you got the past couple of quotes and their references.

I could talk about this all night. And, I probably will.... in my sleep.

Time for bed :lol:

"cybereality" wrote:

I run the 690 at work, and I have GTX 470's in SLI at home. I haven't noticed any major issues. I know people complain a lot about SLI, but I think it has improved greatly from what it used to be.

^this, 100%.

It's made a lot of great comealongs since it's early days. It's only on occasion you find some odd issue that is pretty soon fixed in another driver release, or the odd game that doesn't have support out of box for it.

Other than that, it's a great way to double performance as long as you know what your getting into.

Ok, so there is a lot of chat in this thread about GTX690, if they are SLI and compare to Titans.

But can somebody please test my app on the 690 and Titan and tell me the framerates :-)

If you turn left, got to the end, turn through 180 degrees and look back to the church and TRex, I drop to about 17fps here obviously due to the amount of data to process thats in view, however, I average 30fps at other areas.

FPS reports would be cool thanks.