Nvidia GeForce 337.50 Beta

As always lots of claims that they're improved but as always zero for me.

We should not expect anything from Nvidia. Any optimization and better performance should come from BIS.

Unfortunately the performance today is worse than 6 months ago. Talking about "improvement" :rolleyes:

Bullshit ;)

Care to elaborate?

Only because there's a difference in other games and synthetic benchmarks doesn't mean there has to be any difference in ARMA.

True.

The Combined Arms showcase has enough AIs (20? 30?) to cause significant slowdown (all the way down to about 20-30 fps a lot of the time) and didn't show any significant improvement.

As long as player interaction is present, measuring the avg.fps means nothing.

Hundreds of AI fighting would be untestable.

Because of..?

Thank you Captain. Indeed. I do empty map tests to see if there's any difference what so ever because when there is it usually shows up in an empty editor even when it doesn't in the showcases.

You're welcome, what's your CPU load in empty map?

sneakson, please learn simple calculation of percentage: 145fps to 160fps isn´t similar to 29.5fps to 30fps. Correct is: 26.9 to 30...

I meant helo´s benchmark http://www.armaholic.com/page.php?id=19802

Try testing settings related to draw calls, viewdistance and number of objects on screen would be excellent tests for this driver.

This driver is meant to reduce CPU overhead. You didn't test CPU-heavy situations with hundreds of AI fighting. Star Swarm has thousands of spaceships on the screen fighting each other(CPU heavy), hence the FPS boost with the new drivers.

How can the driver help here? Doing CPU jobs?? I would be so lucky if the tragedy would be over... Seeing Fps dropping with one core on full power while all others idle.

What about the Gtx-5xx?

Sneakson

You are on a single card right?

"First this, this driver is not going to do magic things for people with a more mainstream single GPU based graphics card"," Basically this driver is GOLDEN for high-end SLI setups like the GeForce GTX 780 Ti" guru3d.com

I have 3 gtx 680 1300/1750 and a 3930k @4.7 on a 2560x1440. thats why i got more performance than you did. The driver is better.

Care to elaborate?

True.

As long as player interaction is present, measuring the avg.fps means nothing.

Because of..?

You're welcome, what's your CPU load in empty map?

I will elaborate some.

No, player interaction does not make measurements useless. Contrary to what you seem to believe everything mustn’t happen in the exact same every time. In some cases it would be ideal however not always and it is actually never necessary. If you do a test manually several times and you have the same result every time then you can assume that your interaction does not significantly affect the test values.

Measuring a test with the maximum amount of AI that ARMA supports would be useless since the framerates would be very low and when framerates are low it takes a big change for us to see any difference and besides the AI would do different things every time based on chance and that could cause quite significant differences in comparison to the small differences any driver would.

All in all I believe it would be very difficult to determine any small driver differences using such a scenario. It would be too uncertain.

I’m not sure what my CPU load in an empty map is.

sneakson, please learn simple calculation of percentage: 145fps to 160fps isn´t similar to 29.5fps to 30fps. Correct is: 26.9 to 30...

I meant helo´s benchmark http://www.armaholic.com/page.php?id=19802

JumpingHubert, please consider the fact that fps is an exponential and not a linear scale… d’oh.

I’m also not a fan of any benchmark scenarios because in my testing I’ve seen massive fps differences in benchmarks but then I’ve started Showcase: Infantry or something such and there’s been no difference at all.

Makes me believe there’s a significant difference between the benchmark and common gameplay. It was the Altis ARMABench or whatever and it zooms around the island a lot which probably isn’t similar to any true gameplay situations unless maybe when you’re flying around a lot.

Currently I don’t have a lot of time to play ARMA but in a few months I should have time and by then I’ll try a lot more missions and search for some good “average†CPU-usage missions that may be a bit heavier than the showcases that I use because they’re convenient.

Sneakson

You are on a single card right?

"First this, this driver is not going to do magic things for people with a more mainstream single GPU based graphics card"," Basically this driver is GOLDEN for high-end SLI setups like the GeForce GTX 780 Ti" guru3d.com

I have 3 gtx 680 1300/1750 and a 3930k @4.7 on a 2560x1440. thats why i got more performance than you did. The driver is better.

Single card yes. If Guru3D says this is golden for high-end SLI then sure.

However you still only got a quite small difference.

Also mostly all users don't have multi 780 Tis.

I will elaborate some.

No, player interaction does not make measurements useless.

Comparing two runs that were executed different ways makes no sense in this case.

Measuring a test with the maximum amount of AI that ARMA supports would be useless since the framerates would be very low and when framerates are low it takes a big change for us to see any difference and besides the AI would do different things every time based on chance and that could cause quite significant differences in comparison to the small differences any driver would.

All in all I believe it would be very difficult to determine any small driver differences using such a scenario. It would be too uncertain.

Yet you are certain you played the showcases identically enough to make a conclusion this driver does not help, except in top-tier SLI setups. :confused:

I’m not sure what my CPU load in an empty map is.

How do you know this driver doesn't help with CPU overhead?

Just made a quick test(Helo Benchmark) and I got +-0 FPS. I got the GTX 570 ASUS.

JumpingHubert, please consider the fact that fps is an exponential and not a linear scale… d’oh.

if you have a gain from 145fps to 160fps then you will have the same gain in percentage with lower fps, for example from 26.9fps to 30fps. Please think about it a short while instead of pressing repeat button and write something about exponential nonsense. Sorry, thats really a basic thing. :rolleyes:

Mr authority on universe, barbie and hellokitty (sorry, can´t resist)

@Nikiforos

thanks for your testing.

Comparing two runs that were executed different ways makes no sense in this case.

Yet you are certain you played the showcases identically enough to make a conclusion this driver does not help, except in top-tier SLI setups. :confused:

How do you know this driver doesn't help with CPU overhead?

Why doesn't it make sense to compare two runs executed in different ways if I proved that exact execution doesn't matter?

Yes, I am certain I played the showcases identically enough to make the conclusion that the driver didn't help because the scenarios in question play out basically the same every time, because I proved they did.

I've never said anything about the driver doesn't helping with CPU overhead. Just doesn't do any good to ARMA framerates.

if you have a gain from 145fps to 160fps then you will have the same gain in percentage with lower fps, for example from 26.9fps to 30fps. Please think about it a short while instead of pressing repeat button and write something about exponential nonsense. Sorry, thats really a basic thing. :rolleyes:

Mr authority on universe, barbie and hellokitty (sorry, can´t resist)

@Nikiforos

thanks for your testing.

First of all non-linear is what I meant, not exponential. If that makes more sense to you.

Anyways in my original quote:

"Even if the drivers did anything 145 to 160 fps is a very small difference comparable to 58 vs 60 fps or 30 vs 29.5 fps."

What I mean is that the visual differences between the framerates mentioned are as described above.

In other words the visual difference between 26.9 to 30 fps is much greater than the visual difference between 145 and 160 fps is.

145 to 160 fps is a very small upgrade on the user end.

Next is what the update did to performance. You believe that everyone would be able to draw frames 10% quicker. I believe everyone would be able to draw frames x milliseconds quicker, a constant value.

Unfortunately this may not have an answer as we're generalising quite a lot here but my statement about 145 to 160 being a small visual difference (and smaller than 30 vs 26.9) still holds.

Edit: everything would had been a lot easier if Grillob had benchmarked a bit more performance heavy situations too lol.

Why doesn't it make sense to compare two runs executed in different ways if I proved that exact execution doesn't matter?

Are you really asking why doesn't it make sense? I fail to see how you proved it.

Yes, I am certain I played the showcases identically enough to make the conclusion that the driver didn't help because the scenarios in question play out basically the same every time, because I proved they did.

Even looking in the different direction affects FPS, drops and peaks affect the avg.fps.

I fail again to see how you proved that.

I've never said anything about the driver doesn't helping with CPU overhead. Just doesn't do any good to ARMA framerates.

Ok, maybe I misunderstood this:

This driver is meant to reduce CPU overhead. You didn't test CPU-heavy situations with hundreds of AI fighting. Star Swarm has thousands of spaceships on the screen fighting each other(CPU heavy), hence the FPS boost with the new drivers.

Bullshit ;)

Only because there's a difference in other games and synthetic benchmarks doesn't mean there has to be any difference in ARMA.

The Combined Arms showcase has enough AIs (20? 30?) to cause significant slowdown (all the way down to about 20-30 fps a lot of the time) and didn't show any significant improvement.

Hundreds of AI fighting would be untestable.

Are you really asking why doesn't it make sense? I fail to see how you proved it.

Even looking in the different direction affects FPS, drops and peaks affect the avg.fps.

I fail again to see how you proved that.

Ok, maybe I misunderstood this:

I proved it by doing the runs several times and comparing results lol... already said this.

If you measure 60 seconds of Showcase: Combined Arms, then do it again you and the AI won’t do the exact same thing unless you’ve scripted everything however if your measurements both end up on 42 fps then all things considered it is likely that it doesn’t matter SIGNIFICANTLY if you turn a few milliseconds later or walk some degrees in the wrong direction.

Everything you do in a video game doesn’t severely affect your framerates. In one direction there’s a bunch of trees but in the other direction there’s also a bunch of trees and the game does AI calculations no matter if they’re on screen or not. In some cases such as if you’re on a completely empty map and looking straight into the sky or into the ground you can usually affect framerates a lot but that didn’t happen in my testing.

Framerates are only measured once every second anyways.

So I did Combined Arms twice, got 42 fps average both times and that tells me that maybe my execution doesn’t affect framerates significantly. Out of experience I’m assuming that this is true and that the framerates didn’t vary wildly and by chance happened to end up on 42 every time.

Naturally everything that happens differently will affect things slightly differently compared to a scripted mission however that doesn’t mean it’s significant. If you do things differently and the average framerates end up the same within 1 fps then you can consider it insignificant.

Even in synthetic benchmarks there is a certain variation if you do a test several times, often.

In the computer world exact tests are possible however in the outside world they are not. In neither world are exact tests necessary though because of statistics that let us determine the odds that something was caused by chance in our imperfect testing. In the outside world 95% is a common certainty that scientific experiments try to achieve. In my testing here I didn’t do anything as accurate based on experience.

I could test the Combined Arms mission again since I did so few tests on each driver and maybe there would be a 1-2 fps difference in that one mission of all the missions but I don’t really care because the claim we’ve been hearing from a lot of users that they’re suddenly experiencing 10% more everywhere has been falsified on my machine anyways.

It’s not worth the trouble for everyone to go update their drivers manually.

And yes, I didn't really mean the driver is bullshit for what it does to other games. I only believe it fails to bring any improvement to ARMA.

Just made a quick test(Helo Benchmark) and I got +-0 FPS. I got the GTX 570 ASUS.

I had a 3 gig 580 (hoover edition:p) I wouldn't be expecting any gains from anything under a 7 series, going on what people are writing here anyways.

I had a 3 gig 580 (hoover edition:p) I wouldn't be expecting any gains from anything under a 7 series, going on what people are writing here anyways.

Click

:)

Btw, what do you mean by "synthetic"? Do you mean that it's scripted? Star Swarm is the most unscripted benchmark ever made (thousands of ships spawning almost randomly)

Btw, what do you mean by "synthetic"? Do you mean that it's scripted? Star Swarm is the most unscripted benchmark ever made (thousands of ships spawning almost randomly)

I guess it means generic, as in not exactly any real gaming scenario but created just as a benchmark.

Btw, what do you mean by "synthetic"? Do you mean that it's scripted? Star Swarm is the most unscripted benchmark ever made (thousands of ships spawning almost randomly)

What Greenfist said. Star Swarm is (I'm assuming) not an actual video game, only benchmark software made to work something similar to a game.

However unless the results in a synthetic benchmark are the same as actual games, it’s useless.