Graphics cards and memory

[olihar]

I see the results but I don't believe it... can they fix this with better drivers or is it fubar from a hardware perspective in relations to compute capabilities?

Well what it looks like is Nvidia is crippling the gaming cards in the Computational calculations in their drivers... One reason sell more Quadro and Tesla cards. I think its the most stupid move on their behalf.

If that is true, about it being software, as in on the driver side. Then there is allways a change of someone making hacked drivers. Like these. http://www.omegadrivers.net/

Sadly the guy behind the drivers can not keep developing, due to the recession.
http://www.omegadrivers.net/index.php?option=com_content&view=article&id=408:so-long-and-thanks-for-all-the-fish&catid=41:omegadriversnet-news&Itemid=74

[Wishgranter]

as far as i know CUDA should be opened even for AMD - http://www.electronista.com/articles/11/12/13/nvidia.drops.proprietary.lock.on.cuda.tech/

for the 680 in LUXmark propably or most propably use a Floating Point calc - 64 bit numbers and that i wrote are by DRIVER slowed down to just 1/24 of 32 Integer numbers. in 580 it was jus 1/8 of FP operations.

Photoscan use probably the 32 bit INTEGER operations not the FP, so with theoretical 3,7 Tflop should be faster than 680 3,1 Tflop. BUT as seen in benchmark in some areas - my own deductions says theoretically it can be faster with AMD...... we need a real benchmark in this situations..

AMD drivers are not so hindered in FP operations, that mean IF Photoscan use a 64 FP operations you need the proffesional line of quadro or tesla for speed bump up. if it uses integer then 680 can be eventuely faster. AMD cards are beast in raw power and as far as i know OpenCL is better processed on AMDs......

some other questions ?? 
   

[Wishgranter]

OK last thing on this matter up to real benchmark BUT there is a good explanation whereabouts im writing

http://vr-zone.com/articles/nvidia-kepler-gpus--back-to-games-away-from-compute-/15332.html

[Wishgranter]

What is clear at this time though is that NVIDIA is pitching GTX 680 specifically for consumer graphics while downplaying compute, which says a lot right there. Given their call for efficiency and how some of Fermi’s compute capabilities were already stripped for GF114, this does read like an attempt to further strip compute capabilities from their consumer GPUs in order to boost efficiency. Amusingly, whereas AMD seems to have moved closer to Fermi with GCN by adding compute performance, NVIDIA seems to have moved closer to Cayman with Kepler by taking it away.

[olihar]

So has anyone tried the new 680 card?

[Mark Florquin]

I think Agisoft is much happier with a sound investment in memory and CPU. But I'd love tto see speed improvements by NVIDIA's recent graphics cards...

[Wishgranter]

Oli, you must be a first tester. from what i know from Agisoft, reconstruction routines use single precision arithmetic, that mean the 680 could speedup the process. But still need someone with the card..

[olihar]

Oli, you must be a first tester. from what i know from Agisoft, reconstruction routines use single precision arithmetic, that mean the 680 could speedup the process. But still need someone with the card..

I think I will stick to the 580 card for now, I might just get another one of them as they should get cheaper now being an "old" card now.

What about the Nvidia Quadro cards, are they outperforming the gaming cards in Photoscan or is it the other way.

[Wishgranter]

Quadros are lower clocked, and they are not slowed down in FP64 with drivers, but for overall performance the "game" GPUs are better in performance /price ratio......

[Alexey Pasumansky]

Hello olihar,

Quadro 6000 has nearly equal PhotoScan performance to GeForce 480 GTX, while GeForce 580 GTX is twice as good as Quadro 4000.

Unfortunately we have not tested GeForce 680 GTX yet.

[akemono]

Alexey did you already have a chance to test the new ATI 7970? Im quite interested to see what kind of speedup that would provide vs the current Geforce 580 gtx.

Best,
Mike

[Alexey Pasumansky]

Hello Mike,

We have performed several PhotoScan tests with HD7970 and currently performance results are almost the same as 580 GTX, but depending on the object type performance could slightly increase or even decrease.

So currently we do not see much sense in changing GeForce 580 GTX to HD 7970 in respect of PhotoScan performance.

[akemono]

Hmm oke, thanks for the quick reply.

To bad that it is hardly any quicker then the 580 GTX, but thanks for testing it!

[Kjellis85]

Just wanting to check if anyone have run any tests with the 680, or even compared it to the AMD?

[Infinite]

Just wanting to check if anyone have run any tests with the 680, or even compared it to the AMD?

I'm also interested to know.

Also has anyone run any tests with dual cards? whether it be Quadro 5000/6000's or GTX580? does this do anything with Agisoft during Align stage?