Horrible FPS with NVIDIA GeForce GTX 680 on 10.9.4

[xav8tor]

Can I ask what your x-plane set up is? I just upgraded my mac pro to a dual x5680, 24g ram, but I haven't set up a gpu or PCIe SSD yet. Any advice?

Well, two now with a third to be built for those times OS X or oMP just won't do.

1) 4.1 MP flashed to 5.1, 12 GB 1333 RAM, CPU (single) upgraded to 3.46 3690, and EVGA 670 SC 4 GB GPU with XP on dedicated 256 GB SSD. Standard FPS results posted over on the XP org forum (Hardware) running Nvidia web drivers. Run XP under BC in Win 7. Pretty good for my needs but no 60 FPS at 1080p and extreme textures, distance, AA, etc. Maybe 45 isn FPS steady.

2) New MP standard 3.5 GHz CPU hex off the shelf with D-500 GPU running stock drivers from Apple. Beats #1 above by 10 to maybe 15% on standard tests and in "real life" maybe 20%. Running XP off G-Raid TB in RAID 0. With AMD GPU, running Mavericks only. Forget Win in this config. AMD GPU and Windows on a Mac are mutually exclusive terms in single threaded OpenGL apps.

3) To be built for those special high end projects. Win 7 or 8.1 SFF box with OC'd 4790k CPU, GTX 780 6 GB VRAM or 780ti with 3 GB (TBD). 16 GB RAM with 256 GB SSD for OS/Apps and another SSD with 500 GB or 1 TB for XP and captured vids. That machine will double to triple either of the above. Recent controlled XP benches prove that. For XP state of the art, #3 is it...by FAR!

Fastest CPU you can afford, ditto for the GPU with sufficient VRAM for you textures, don't worry about core count (not relevant in XP), and don't get carried away with rendering settings or weather (cloud/viz) settings. Experiment to find the best combo that works for you. If you are HEAVILY into XP with tons of objects and eye candy, go the custom PC route. If you're into it for realistic flying or other professional use, the MP will do nicely when properly configured under the right OS (depends on the GPU choice). SSD will help load times, which are long in XP, but until you reach a completely new area, it won't make a difference...theoretically. I've found a fast RAID or SSD eliminates rare momentary stutters, despite the gurus who say all loads are in the background. All I know is, before I had them on occasion, now, with the RAID and SSD, I don't.

 

[CASLondon]

I have a 4,1>5,1, I did dual X5680s (3.33ghz) as the 3.46 is expensive. I assumed 3.33 to 3.47 isn't too different, but maybe I'm wrong. If it is, I can always buy a single cpu tray and get the w3690. I have 24g of RAM.

I'm going to put ssds on PCIe slots, as soon as that new apple flash storage pcie adapter is available I'm going to do that. And a GTX780 6G for video card. I'm assuming this will perform pretty well, but I'm a novice.

I wanted to know more about your experience of GPU as I assumed that makes most of the difference.

 

[xav8tor]

I have a 4,1>5,1, I did dual X5680s (3.33ghz) as the 3.46 is expensive. I assumed 3.33 to 3.47 isn't too different, but maybe I'm wrong. If it is, I can always buy a single cpu tray and get the w3690. I have 24g of RAM.

I'm going to put ssds on PCIe slots, as soon as that new apple flash storage pcie adapter is available I'm going to do that. And a GTX780 6G for video card. I'm assuming this will perform pretty well, but I'm a novice.

I wanted to know more about your experience of GPU as I assumed that makes most of the difference.

Since XP doesn't make much use of cores, and the clock speed is so close, the difference between the 3680/90 is negligible, especially when Turbo kicks in. PCI-E flash vs. a conventional SSD won't make a difference in XP. However, with the nMP, in boot times, and read/writes to my limited stock (256GD) storage, the speed is crazy fast. It pegs the needle on the BlackMagic speed test.

Whether you are CPU or GPU limited all depends upon your app. In X-Plane most of the time, I'm CPU limited but there are times when it's the GPU and a 780 might help. One things is for certain: With OpenGL apps that are primarily single -threaded, if you have Nvidia GPU's, the single greatest perforce gain will come by booting into Windows. Double your FPS for the price of the OS. YMMV...for instance, with HD video editing/rendering. Then, cores make a difference, as does GPU compute.

 

[5noopy]

Same here, I got a mac pro, with Xeon E5645, 12Cores, EVGA GTX680 2G, flashed to the Mac Edition. Have the latest Nvdia webdriver and CUDA. Now the performance is even lower than my macbook pro with GTX650m.

Cinebench r15 OpenGL only 40FPS, my original ATI HD5770 was 57FPS. My laptop was 52FPS... Oceanware Opencl is only 61FPS..

The TDP of GTX680 is 195W, and Macpro should be fine with a card within 225W. I think the power is enough. And the CUDA performance was also pretty bad. I don't feel any improvements after updating to GTX680 from 5770.

I tried to switch to windows 7 64bit with the card, and it got 75 FPS in Cinebench r15, the windows version, and the CUDA performance was even 50% faster than the Quadro 4000.

What's wrong with mackintosh? The card is plugged in the PCIE*16 port with all the correct status check in CUDA-Z, system profile, OpenGL view... and apps like that.. I'm running OSX 10.10.1 by the way.

 

[MacVidCards]

Same here, I got a mac pro, with Xeon E5645, 12Cores, EVGA GTX680 2G, flashed to the Mac Edition. Have the latest Nvdia webdriver and CUDA. Now the performance is even lower than my macbook pro with GTX650m.

Cinebench r15 OpenGL only 40FPS, my original ATI HD5770 was 57FPS. My laptop was 52FPS... Oceanware Opencl is only 61FPS..

The TDP of GTX680 is 195W, and Macpro should be fine with a card within 225W. I think the power is enough. And the CUDA performance was also pretty bad. I don't feel any improvements after updating to GTX680 from 5770.

I tried to switch to windows 7 64bit with the card, and it got 75 FPS in Cinebench r15, the windows version, and the CUDA performance was even 50% faster than the Quadro 4000.

What's wrong with mackintosh? The card is plugged in the PCIE*16 port with all the correct status check in CUDA-Z, system profile, OpenGL view... and apps like that.. I'm running OSX 10.10.1 by the way.

All in flawed tests.

Cinebench tests nothing at all, at least not for Nvidia cards. i have demonstrated that pretty much any Nvidia card will give same score, it is 100% determined by CPU.

Ocean wave has been broken for several OS iterations.

61 fps is....your monitor refresh rate. You have to use dev tools to turn off v-sync to get a real score.

Try Lexmark or Valley if you want real results.

 

[CuamckuyKot]

Cinebench is testing via OpenGL which relies on single core performance.

That's why it doesn't benchmark normally a lot of videocards on cMP.

Heaven and Valley — are most normal tests.

 

[5noopy]

Hey, thanks for the information! I downloaded Vally and tested it out. Attached are the screen shots.
With 1280*720, Medium quality, AA Off, It's 39.1 FPS, 1634 Score. It was only running around 80Watts according to my monitoring app.

WIth 1920*1080, Ultra quality, 8XAA, 32.4 FPS, 1357 Score. Running around 170Watts.

I feel like there's a lock that sets a limit to the graphic card.. What should I do?

All in flawed tests.

Cinebench tests nothing at all, at least not for Nvidia cards. i have demonstrated that pretty much any Nvidia card will give same score, it is 100% determined by CPU.

Ocean wave has been broken for several OS iterations.

61 fps is....your monitor refresh rate. You have to use dev tools to turn off v-sync to get a real score.

Try Lexmark or Valley if you want real results.