Video card again - GTX disaster

Not sure if the plug-in is intended for Quadro but tried it anyway on my K2000.

Also I split the spheres in half to see the speed difference on open surfaces to add to the knowledge base:

Sorry, but it doesn’t work that way… besides, this plugin isn’t trying to solve problems on Quadros, it’s trying to discover why GeForces are so much slower on such a simple test.

All it does is disable specific rendering features that NVidia seems to be keying off of to throttle back their GeFroce drivers. It does not detect open or closed surfaces and act differently… Sorry, I should have been more clear.

-J

Hi Jeff,
will the object management enhanced for v6?
Can you do something that the Quadro 6000 run so fast like a K2000? Why are this Quadros so different? During my projects I have seen frame rates of 0.5 fps and a three time faster fps would be great.

Hi Jeff,

bad news:

Befehl: Testmaxspeed
Time to regen viewport 100 times = 3.89 seconds.
Befehl: GeForceTestEngine
GeForce Test Engine is now ENABLED
Befehl: Testmaxspeed
Time to regen viewport 100 times = 3.78 seconds.

So no effect here.

(5x5x4 spheres shaded)
GPU: GTX680, Driver 334.89

Thanks,
-C-H-A-R-L-E-S-

GTX 650(2gig ram)
x2 anti-alias
TestMaxSpeed = 6.41secs
GeForceTestEngine = 2.73secs

Charles,
you should probably state which GPU you are using for this one.

Good point!

Hi Jeff,

FWIW in case it you did not find it already it might be of use.
In this article (http://doc-ok.org/?p=304) linkend earlier by @Charles there is mention of a shader used to do the double sided rendering…fixing the speed quite a bit

-Willem

Hi Charles,

This doesn’t make any sense… I’m using that exact same card in my testing…it’s my benchmark. The only thing I can think of is that you’re using an older driver. I’m using 340.52…although your initial time does seem to line up with mine so I dunno.

I can see this isn’t going to be as black & white as it started to look (ugh, it never is).

-J

LOL, ya I read that after Brian James pointed me to it during our meeting yesterday. I felt pretty stupid having just rediscovered the wheel…but it was nice to see confirmation of my findings.

It’s not the shader that makes it faster, it’s turning OFF the double-sided polygon rendering feature. He uses a shader to offset the loss of that feature (i.e. he implements double-sided rendering within the shader). I find it very odd that drivers seem to be keying off specific features, and then somehow throttling back. This means that cases like this won’t be fixed by going to DirectX, since I’m pretty sure the drivers would do the same thing for the same reason. I’m just wondering (and trying to figure out) what other features they might be keying off of, since my GeForce times are still higher than a cheap Quadro…that tells me there are probably other features out there that can be fixed the same way.

-J

1 Like

Hi Jeff,

better news:

Befehl: Testmaxspeed
Time to regen viewport 100 times = 12.55 seconds.
Befehl: GeForceTestEngine
GeForce Test Engine is now ENABLED
Befehl: Testmaxspeed
Time to regen viewport 100 times = 2.67 seconds.

This is on a notebook with GT640M LE.

Fantastic result!

I can’t test again on the GTX680 today, but tomorrow I will.

Definitely you are on the right rack.

-C-H-A-R-L-E-S-

P.S.:
2500 sheres
Time to regen viewport 100 times = 242.78 seconds.
Befehl: GeForceTestEngine
GeForce Test Engine is now ENABLED
Befehl: Testmaxspeed
Time to regen viewport 100 times = 51.52 seconds.

1 Like

Not odd at all, just an application of the differentiating products for distinct markets willing to pay very different prices.

I can think of lots of scenarios where what you described seems to fall far short of honest and honorable.

i was recently on newegg looking at a quadro card and on the page was a youtube video of one of the newegg guys with his guest the nvidia quadro dev mgr or something of that sort. right in the video the guy actually admitted that the only real difference is in the driver/firmware. i thought to myself…did he just say that, as he was sitting next to a $5000 k6000.

This sort of behaviour I have seen Apple do for a long time. They used to only allow screen spanning on the pro laptops and mirroring only on the consumer. A firmware hack used to make this feature available on the consumer machines.

The only difference between my base model Asiga 3d printer and the next version up that has a bigger build area is a $1500 firmware upgrade :frowning:

OK, I did a 5x5x5 sphere test with my 580GTX

3.64 normal, 1.97 With TestEngine

I did not see much dif when rotating an actual architectural model.

Just a thought… Since Geforce cards are aimed at Gaming, and Nvidia CAD crippled the Geforce cards by removing support for CAD specific Opengl functions. It would be safe to assume that all the OpenGl commands used in game creation can be elliminated from the difference list. (This sounds alot easier me saying it than it probably is). Thanks for giving this issue attention. +1 for CAD team.

Hi Jeff,

I rebooted the PC with the GTX680.
No change.

Then I investigated the shaded display mode (while it was not blue).
Advanced GPU lighting was cheched, I unchecked it.
Current result:

Time to regen viewport 100 times = 3.97 seconds.
Befehl: GeForceTestEngine
GeForce Test Engine is now ENABLED
Befehl: Testmaxspeed
Time to regen viewport 100 times = 1.25 seconds.

-C-H-A-R-L-E-S-

A German magazine tested a bunch of Quadros/FirePro against the GTX680.
When we look e.g. at the Maya column, then we see the GTX680 is able to deliver high performance.

-C-H-A-R-L-E-S-

1 Like

Maya can have has double sided turned off by default.
I read a similar topic from 2012 about Blender, they got equal speedup by turning it off by default.

-Jørgen