graphics card comparison test

[granite07]

I'd like to comparison test the Nvidia 8th generation G80 and the new 11th generation GF100 - as I understand these are two distinctly different designs. Last year I tested some cards so I have a baseline test procedure, test machine, 70MB corridor model, draft measurable metrics, and a few cards laying around.

To validate the test and help with the research, we should complete this with several people. Who is interested to join this discussion? I will purchase a Quadro 2000 for the test. This is what I have now http://cife.stanford.edu/wiki/doku.php?id=granite:3dmodelling:performance#video_card_test I envision a preliminary discussion to share our experience with video cards, a day of testing in the next couple weeks, and then a day or two of discussion followed by retests as necessary. Just a simple round of tests for a pragmatic comparison specific to Civil 3D.

I think the customization forum is the place for this discussion, it is not customization but also it does not fit with the general C3D forum.

[Sinc]

In my experience, something like a GeForce 560 outperforms a Quadro 2000 for C3D, by a significant margin.

[granite07]

What do you think is the significant factor? 1) the core design [guessing not sinced it is same as Quadro 2000], 2) the core speed, 3) the memory speed, 4) the bandwidth

[wang890]

i got an awesome card, can do 3d work and gaming lol.

wonder if there will be a kepler quadro. probably will. wait for that one then.

[granite07]

I tested a Firepro 3D v7800, it has a 700mhz 1440 core core and DDR5 memory - this is comparable to the GTX 560 and 590. For civil 3D the v7800 performed just as good as a 4-core 400mhz DDR GeForce 6500? Unless I rendered something or ran a 4D simulation there was no noticable difference; maybe the linewights and colors had more definition.

[wang890]

at work i have a 500 series as well. quadro 580 lol. super entry level. the cpu is also worth less than 100 bucks. and the tell me oh this is new computer. anyways. yeah if you do some research there are people asking why 3dmax can use new cpu fully with ram and video card compared to autocad can't even rotate a stairway. pretty old engine driven autocad and civil 3d. you don't know how many brain cell i lose from getting mad at the software daily. lucky no white hair yet.

last job i worked the contractor end up getting gaming video cards i think its the 295 for their new builds. worked out fine.

you see this machine i have, if i have a large corridor and try to mess with it in object viewer it will still freeze. what does that tell you? lol.

[granite07]

Do you have time for a runtime test - your 4.5Ghz sandy bridge and gtx 590 is a good comparison with my 4Ghz  deneb and FX 5600 - other then these they are comparibly configured with 16GB 1600Mhz RAM, ASUS mb, and SSD.

I posted a test corridor model here test_corridor_file (turn off and freeze all layers and then turn on render layer group) and test metrics are here metrics_wiki , use the Test #9 table (these links are to the same page but different subsections). This corridor model is made 'all wrong' so it is a good benchmark test since nothing should be able to run it well - therefore pushing performance into measurable differences in seconds.

Note: a suggestion for autodesk; place a corridor file with baseline performance metrics and benchmark score for their test machine - like this - in the installation files or on their website, so we have a benchmark for comparison. I looked in the install files and the sample files are for simple corridors that I can run in nanoseconds.

[wang890]

i love benching marking. lol. was not gonna bother before but then i see why when people get a new machine they want to compare numbers. so yeah i will run this tonight. right now at work and i will try to run it using crappy work computer first. xeon 3505, 2.53ghz super slow computer.

i see you are doing a research on hardware and civil 3d? that's cool job.

[wang890]

dude can you give more instruction which drawing to open? is it this one?

MER99_RGW_CIFE_09.4.dwg

[granite07]

yes - instructions are nice, but who would read them 🙂 MER99_RGW_CIFE_09.4.dwg is the correct file, it is in the dwg folder.

[wang890]

oh so to time it i just grab a watch and eyeball the time?

i thought you have some built in program in civil 3d that's linked to a website that update that table. that would be cool. pull in every one's system spec and time then compare do some stastical analysis. totally possible just a lot of work.

so in the property you want me to set the view to whatever it's in the table yeah? like 2dframewire and realistic then run the corresponding command? i think i got it

oh for the orbit you talking about 3do command or just use c3d object viewer?

oh you using 2012? don't have that i just got a trial and will install it at home on that fast computer.

just so you know i have a amd x4 9950Black edition and that only got 3pt on cinebench and this sandy bridge - E got 13 points. clock to clock amd can't compare to intel. still pretty nice computer though my amd. served me well.

you have x4 II or the first version? got it to 4.0 is not bad.

[granite07]

• yup, just eyeball, close enough is good enough for this.
• And, some built in program in civil 3d that's linked to a website that update that table - would be cool, but will not be built by me anytime soon. I have another program that requires a lot of time to write.
• For orbit, I was using the icon, but now will have to retest with 3do
• 2011 should be close enough, I have tried 2011, 2012 and 2012 SP1
• cinebench tests rendering; this is were this benchmark will help since scoring 3 or 13 on cinebench may mean nothing for how Civil 3D will perform, mostly because C3D runs on one core and a GeForce 6500 is all you really need for wireframe.
• the x4 Phenom II X4 975BE with c3 stepping - I have had it up to 4.2Ghz but seems more stable at 3.9Ghz, skype screenshare crashes when over 4.0Ghz

[wang890 · Accepted solution]

did the best i could

1. fire up civil 3d=5s

2. 3do command start 2dwireframe = 11s

3. 3do rotate then press escape til respond= 33s

4. regen=5s 2dwireframe

5. change surface name in RW4>06 then hit ok til modelspace responsive=50s

6. save=5s

not sure that alignment hierchy one that depend on how fast you click. but mine opens right away no delay.

fatal error many times maybe because of custom subassemblies.

you can give this a try. open the autodesk sample file:

C:\Program Files\Autodesk\AutoCAD Civil 3D 2012\Help\Civil Tutorials\Drawings\Corridor-5c.dwg.

set interval to 1m dont' touch anything else. do that in corridor property. then hit ok then start timing til become responsive. took me .34 seconds. i think that's not bad.

but what i realized from this is again emphasize how much time we waste on waiting for the program to responde. only if it can use more cores even 4 cores we would see a marked improvement. imaging cutting down all these time by 75%.

if you don't have a ssd go get one. don't bother with magnetic drives. they are garbage.

have a nice weekend!

[granite07]

thank you, you conformed that these are typical runtimes for intel and amd cpu, and that excluding rendering or simulations the older nvidia g80 core is comparable to the newer gf110 series.

For the site alignment hierarchy sampleline - the lag is the first time it is expanded, after that it is nearly instant until the file is closed and reopened.

I agree with your ssd advice - I use a pci-e revodrive (raid0 ssd), two or more ssd in raid0 is nearly a mandatory suggestion.

[wang890]

how's the revodrive?

have you try ramdisk? i am gonna dump civil 3d and visual studio on that thing and see what happens. maybe can start the program in 1 second lol.

[granite07]

I have the revodrive 1x -  OCZ has a newer version; it works good. Compared to a Muskin SSD, day-to-day I do not notice the difference. In C3D there is a slight noticeable difference but not enough to benchmark, 2 SATA SSD in RAID0 is just as good. The 'drive' visualization in C3D is where there is a difference, with the SATA SSD the feature is unusable in realistic, the PCI-e SSD can flow through the simulation in realistic.

I tried RAM disk, not sure if my setup was correct. The results are posted on my experiment wiki, it is Test 10 and Test 11.

[granite07]

14 seconds

to open the autodesk sample file:

C:\Program Files\Autodesk\AutoCAD Civil 3D 2012\Help\Civil Tutorials\Drawings\Corridor-5c.dwg.

set corridor property interval to 1m did not touch anything else. then hit ok and start timed until responsive

.34s is very fast