Processing Speed - Continued

[Loover]

I have built a computer, hoping it would perform like my computer at work, but it is nowhere close. This is the original thread. http://forums.anandtech.com/showthread.php?t=2255264 I now have a little more info.

The program I am running is called Tekla Structures 18.0. It is a 3d modelling / drafting program.


Here is my work setup. Computer is 3 years old.

NVidia GeForce GTX 460
Intel(R) Core i7 CPU 860 @ 2.80GHz
Intel(R) 5 Series / 3400 Series Chipset Family 6 Port SATA AHCI Controller
74.4GB Intel Solid State
372 GB Hard Drive
16GB Ram


My new home setup.

Motherboard - Gigabyte Z77X-UD3H
Processor - Intel Core i5 3570k 3.4GHz 6Mb
Hard Drive - Crucial M4 128GB SSD Sata 6GB/s
Hard Drive - 1TB 32MB 3.5" SA
Video Card - HIS Radeon HD7870 IceQ 2GB GDDR5
Ram - Kingston 16GB 1600MHz DDR3


The program is installed on the solid state drive on both computers.

One command that takes 20 seconds on my work computer will take 5 minutes on my new home computer. The command involves using a program written by me in c++.

Anyone have any ideas why?

 

[Loover]

Some people from the previous thread said a quadro card or a proffesional video card could be the difference.

Is the NVidia GeForce GTX 460 a good video card? Is it better that my video card?

 

[Smoblikat]

Are you sure its configured properly? Your home computer is MUCH MUCH better than your work one.

 

[Loover]

Well, I built it myself because I want to know how it works. But I don't know alot about computers, only how to use it.

Really the only configuring I did is making sure it is AHCI. I barely know how any of these bios settings work.

Any ideas on what I can set things up different?

 

[Loover]

Should I take my computer to someone who knows what they are doing, or can I figure this out with your guys help?

 

[notty22]

Right of the bat, I see 2 differences in the computer specs listed above.

The work computer has a Nvidia card and the Intel cpu has hyperthreading.
For the large discrepancy, I'd say the program makes use of Nvidia hardware?

 

[Denithor]

Well, the only significant difference between your work & home systems is the video cards. In all other cases your home system is definitely faster.

Video card - the 7850 is absolutely faster than the GTX 460. However. It might not be correctly supported by the software, resulting in much, much reduced performance if the software is taking advantage of the GTX 460 for its workload.

These cards feature very different architectures and are supported differently in software. The nVidia card may be fully supported where the AMD card is not, resulting in seriously different performance on the two systems. Would it be possible to try out the GTX 460 in your home system? Or even just buy one to test out? Those go for like $100 these days, much cheaper than the 7850 you currently have.

 

[Mr. Pedantic]

Right of the bat, I see 2 differences in the computer specs listed above.

The work computer has a Nvidia card and the Intel cpu has hyperthreading.
For the large discrepancy, I'd say the program makes use of Nvidia hardware?

The thread OP linked to said the software was single-thread. Not sure if this is still true, but my bet would be on either the GTX460 or if something's not right with the SDD/HDD setup in the home computer.

 

[Loover]

Thank you Denithor, I think you may have found something.

Someone in the previous thread had pointed out that Tekla Structures has only been tested using Nvidia video card (according to release notes). So as so stated, maybe Tekla is only supported by Nvidia cards?

Just to confirm, this means processing speed can be affected by the video card? This doesn't make any sense to me, seems like the video card would only affect graphics, but I guess not.

I will start looking for a used Nvidia card to see if this is the case. Thanks everyone.

 

[Loover]

And yes the software is single thread.

 

[Denithor]

Thank you Denithor, I think you may have found something.

Someone in the previous thread had pointed out that Tekla Structures has only been tested using Nvidia video card (according to release notes). So as so stated, maybe Tekla is only supported by Nvidia cards?

Just to confirm, this means processing speed can be affected by the video card? This doesn't make any sense to me, seems like the video card would only affect graphics, but I guess not.

I will start looking for a used Nvidia card to see if this is the case. Thanks everyone.

In a single word, yes.

http://www.nvidia.com/object/cuda_home_new.html

It's called CUDA by nVidia. They created it with the intent of having various software programs take advantage of the very wide, very fast parallel capabilities of their video cards to accelerate certain tasks. Of course, many tasks are not parallel enough to use this acceleration, but when it works it's really really fast. And I know they focused a fair amount of their efforts on professional/CAD software so it sounds like this could easily be the case with Tekla Structure.

Like I said, try to find an nVidia card of a recent generation (GTX 4x0, GTX 5x0) to try out. Don't bother with the latest GTX 670/680 - they actually neutered the compute capability of these cards to force people into buying their much more expensive "professional" line of Tesla cards.

EDIT: Just so I don't get flamed, AMD also has their version of GPGPU compute - OpenCL. And as an open standard it is actually starting to collect more support than CUDA. But at this point, CUDA has made some serious inroads into various professional and commercial software. And these programs that use CUDA do not support AMD cards for this acceleration.

 

[ShintaiDK]

Again?

Go ask the people at your work.

Either the software uses the GPU if Quadro to accelerate. Or its done at work via distributed server offloading. Since the software also supports this feature. So maybe at work 20 PCs are doing the calculation and not 1.

 

[Soulkeeper]

You pay tekla for this proprietary software right ?

call them and ask, chances are they are the largest wealth of knowledge for their own closed software ...

It might be worth asking if it uses cuda, and if it does ask them if they have plans for opencl support.

 

[Charles Kozierok]

Graphics card

Tekla Structures rendering uses OpenGL technology, and graphics cards with good hardware support for OpenGL give the best performance. Tekla does not have resources to test all cards on the market, so we have chosen cards based on NVIDIA graphics processor to be our test platform. In 3D software the importance of good graphics card is highlighted, but up-to-date display drivers are equally as important. You can install the updates for your graphics card driver from the manufacturer’s
website.

Graphics card manufacturers have slight differences in their OpenGL implementation, and there might be differences in the picture quality even between cards using NVIDIA chips. Therefore it's good to evaluate and test the cards before purchasing. Tekla has developed a special application for testing and evaluating graphic cards for Tekla Structures purposes. The application is called Steelmark and you can download it from Tekla Structures Extranet > Product > Hardware info.

If you want to share your experiences of different graphics cards, feel free to post your experiences on Tekla Extranet Discussion Forum: Hardware and Operating System.

http://www.tekla.com/SiteCollection...s/TeklaStructures_Hardware_recommendation.pdf

 

[soccerballtux]

To find out if multiple PCs are helping with the rendering-- I would open up Task Manager on other PCs and watch their load percentage when you run one of these C++ programs. Do all the PCs jump up to 25% or 50% load? If 4 cores and max out single threaded program, it would be 25%. 2 cores would be 50%. Etc.
Alternatively move the network files from the server onto your work desktop. Then unplug the PC from the network and run the program that takes 5 minutes at home but 20 seconds at work. Does it take 5 minutes? Other PCs might be helping with the rendering. If it still happens in 20 seconds then it might be the video card that you have at home.
So, isolate if it's because of network computers helping, or the video card. If it looks like it's the video card, see if you can borrow the 460GTX from work. Uninstall the drivers for the AMD card, take it out, then put in the Nvidia card and run the test again.

Also, can you try downloading the latest drivers explicitly from AMD's website? If you didn't before. Sometimes the included display driver that Windows 7 automatically installs is not the "full thing".

Finally, considering they only test on Nvidia, there may be some special Nvidia-specific features (CUDA) that they make use of in the software, or that your c++ programs are making use of. AMD uses OpenCL not CUDA, so if the software wants CUDA that would slow down performance.

What libraries are you including for the C++ files?