Sign in with
Sign up | Sign in
Your question

PCI-E bandqith Physx and Borderlands 2

Last response: in Graphics & Displays
Share
November 4, 2012 12:18:29 PM

Hi,

I've seen that you can get a serious bump in FPS in BL2, during intense Physx scenes by adding a dedicated Physx card, results for minimum FPS improve about 20% with a gtx640 paired with a primary gtx 670 (which I own). I was thinking about getting one, since water physics maps with great battles get me down to the 25-20 FPS mark.

Here is the thing tough, my motherboard only supports two graphic cards in 16x 1x mode. I was wondering if that would actually cripple the FPS I got due to the limited bandwith... Please help =)

Specs:

i5-3570k
gtx 670
benq xl2420t
8 gb ram.
a b U Graphics card
November 4, 2012 12:26:37 PM

or just disable physx. and you didn't tell us what the motherboard is.
a b U Graphics card
November 4, 2012 12:44:38 PM

@ 1x mode are you serious please share you mobo specs with lowest i5 3570k mobo you will have atleast a x4 pcie lane
Related resources
November 4, 2012 2:21:06 PM

GIGABYTE GA-Z77M-D3H LGA 1155 Intel Z77 HDMI SATA 6Gb/s USB 3.0 Micro ATX Intel Motherboard

It says in the manual that when 2 cards are installed, the second one runs @ 1x. Maybe U read it wrong but I don't think so.

Regarding disabling Physx, I really think BL2 is the best Physx implementation to date and IMO it really adds to the gaming experience, so I don't believe I have to sacrifice that experience, if I am willing to invest... which I am.

If you could please clarify the issue of the 1x or 4x I would very much aprecciate it.

BTW sorry any mistakes, english is not my mother tongue.
November 4, 2012 3:20:00 PM

Actually you are right! It is 4x. So it is enough right?
a b U Graphics card
November 4, 2012 3:29:47 PM

Where did you see these results? I heard a few reports saying that adding a dedicated card didn't really boost performance with a 670/680..

A little concerned that you are dropping as low as 20fps? shouldnt get below 40 with a 670
November 4, 2012 3:37:51 PM

Can I post links to other forums? Was at techpowerup i think, gonna search.
a b U Graphics card
November 4, 2012 3:41:40 PM

Yea, there's no rule against it.... fire away...

I would be interested to see, not doubting you or anything was just what I heard..
November 4, 2012 3:47:14 PM

So the link is this:
http://forums.guru3d.com/showthread.php?t=369008

but there are a lot of topics on this issue. Physx on high is reaaaaaaally taxing, and you get better results with a second gtx 670 dedicated to Physx than with it in SLI
November 4, 2012 3:56:13 PM

I know I saw some benchmarks with different cards, I'll post them as soon as I find them
November 4, 2012 3:57:44 PM

GTX 680
2012-10-01 17:16:17 – Borderlands2
Frames: 127101 – Time: 1211191ms – Avg: 104.939 – Min: 0 – Max: 189
GTX 680 + GTX 580 SC (dedicated PhysX)
2012-10-01 01:18:30 – Borderlands2
Frames: 96895 – Time: 768648ms – Avg: 126.059 – Min: 13 – Max: 239
GTX 680 + GT 640 (dedicated PhysX)
2012-10-01 09:01:23 – Borderlands2
Frames: 420723 – Time: 3230437ms – Avg: 130.237 – Min: 23 – Max: 279

from: http://1pcent.com/?p=135
November 4, 2012 4:09:35 PM

Other results:
Results
GTX 690
93.597 FPS average
2012-10-08 16:38:49 – Borderlands2
Frames: 28809 – Time: 277058ms – Avg: 103.982 – Min: 56 – Max: 349
2012-10-08 17:33:51 – Borderlands2
Frames: 25823 – Time: 290833ms – Avg: 88.790 – Min: 46 – Max: 330
2012-10-08 17:40:29 – Borderlands2
Frames: 23553 – Time: 267588ms – Avg: 88.020 – Min: 47 – Max: 310
GTX 690 + GT 640
102.273 FPS average
2012-10-08 16:57:12 – Borderlands2
Frames: 24783 – Time: 260179ms – Avg: 95.254 – Min: 50 – Max: 345
2012-10-08 17:14:48 – Borderlands2
Frames: 28443 – Time: 257339ms – Avg: 110.527 – Min: 63 – Max: 325
2012-10-08 17:20:58 – Borderlands2
Frames: 32049 – Time: 317197ms – Avg: 101.038 – Min: 63 – Max: 334

source: http://www.evga.com/forums/tm.aspx?m=1762273&mpage=1

But anyway... Can my board handle the dedicated physx card without crippling the gtx 670 ? Sorry but no one actually answered my question directly so far.
November 4, 2012 4:16:48 PM

Caspase said:
GTX 680
2012-10-01 17:16:17 – Borderlands2
Frames: 127101 – Time: 1211191ms – Avg: 104.939 – Min: 0 – Max: 189
GTX 680 + GTX 580 SC (dedicated PhysX)
2012-10-01 01:18:30 – Borderlands2
Frames: 96895 – Time: 768648ms – Avg: 126.059 – Min: 13 – Max: 239
GTX 680 + GT 640 (dedicated PhysX)
2012-10-01 09:01:23 – Borderlands2
Frames: 420723 – Time: 3230437ms – Avg: 130.237 – Min: 23 – Max: 279

from: http://1pcent.com/?p=135


I am not an expert on PhysX, but just looking at those results, they don't seem very conclusive. The test with a 680 + 580 Superclocked is getting worse framerates than the tes with a 680 + GT 640? The guy says in his post that he just ran those benchmarks in uncontrolled, extended gameplay. He is honest about that fact, but does not seem to understand fully that this lack of control means the results are basically uncomparable, at least for a sample size of ONE run each. Look at his results from his more controlled test above (titled: "Repeated Test in a Similiar Area"). It is still not a very scientific test, but it is at least attempting some control. He still only does one run with the single 680 though.... :/ 

Anyway, those results show ~75 FPS min vs ~80 FPS min, with avg FPS in the 120s for all. The 75 FPS min for the single 680 could, realistically, simply be within the margin of error for this test. Are you gaming on a 120Hz monitor? I personally am not seeing an advantage here that is worth even 50 bucks for an additional card and the electricity to power it.
a b U Graphics card
November 4, 2012 4:17:56 PM

Caspase said:
GTX 680
2012-10-01 17:16:17 – Borderlands2
Frames: 127101 – Time: 1211191ms – Avg: 104.939 – Min: 0 – Max: 189
GTX 680 + GTX 580 SC (dedicated PhysX)
2012-10-01 01:18:30 – Borderlands2
Frames: 96895 – Time: 768648ms – Avg: 126.059 – Min: 13 – Max: 239
GTX 680 + GT 640 (dedicated PhysX)
2012-10-01 09:01:23 – Borderlands2
Frames: 420723 – Time: 3230437ms – Avg: 130.237 – Min: 23 – Max: 279

from: http://1pcent.com/?p=135


I wouldnt trust those benchmarks, look at the minimums.... 0fps??? very random amount of frames too

Had a look through the guru3d link you posted me, got up to about page 8 and didnt really find anything conclusive, just seemed to be a lot of people contradicting each other...

Edit: You should get away with dedicated PhysX on x4 lane

I have 2 570's, I tried my own experiments with this running the other card as a dedicated PhysX didnt really seem to have much effect if any...SLI was really temperamental when the game was 1st released but seems to be working out great now after Nvidia released new drivers so that may be your best option, if keplers are the same

November 4, 2012 4:34:31 PM

Yes, I do. The Benq xl2420T. And usually I don't complain too much about FPS, I play metro 2033 with avg 45 and it's okish, but getting to 20-30 is somewhat annoying, and despite what people say, I can notice the difference when FPS drop suddenly from like 90 to 50.
I read a ton of topics on this yesterday, unfortunately I didn't save them, but the general consensus seemed to be that it stabilized the FPS, and gave a small increase too. I can get a used gt640 for like 50 so it's not really a big deal but wanted to know for sure if you think is will solve the problem and if it is doable!
a b U Graphics card
November 4, 2012 4:42:57 PM

Im not saying that it won't have any effect as I'm yet to find a reliable source with regard to kepler and using a 640...

I played the majority of the game with a single 570 because of the SLI issues and never experienced dips below 40fps really...

I would look more deeply into that...20fps?
November 4, 2012 4:43:57 PM

Other interesting result... Seems that people with 3930k can handle the Physx in high very well!! source:http://www.overclock.net/t/1307142/borderlands-2-with-h...
It's a long thread I know...

I actually thought that Nvidia locked the Physx to one thread only. But it seems the game actually likes 6 cores ... So maybe the whole argument of i5 to i7 means nothing in gaming argument could actually be changing?
November 4, 2012 4:47:00 PM

paddys09 said:
Im not saying that it won't have any effect as I'm yet to find a reliable source with regard to kepler and using a 640...

I played the majority of the game with a single 570 because of the SLI issues and never experienced dips below 40fps really...

I would look more deeply into that...20fps?


were you playing max settings at 1080p? That's odd... Don't take it the wrong way, but I should have better scores than yours xD. Maybe it is because of computation deficiencies in Kepler, though my gtx 670 has like double cuda cores they are clocked to about half, so maybe speed is more important...

Just speculation, I know, but it has its interests :bounce: 
Anonymous
a c 117 U Graphics card
November 4, 2012 4:56:53 PM

Caspase said:
Other results:

But anyway... Can my board handle the dedicated physx card without crippling the gtx 670 ? Sorry but no one actually answered my question directly so far.


yes it can. i have read on the 3D guru forums that a x4 slot works just fine. if you need more info:
http://bit.ly/SJcV4B
a b U Graphics card
November 4, 2012 5:01:11 PM

Thats what I was thinking because kepler seem to be getting similar minimums to Fermi's usually better top end though....20-25 is still way too low, I have done pretty much everything in the game and have never seen it go as low as that and yea everything is maxed...

PhysX is really strange, BL2 isnt a very taxing game without the PhysX so I think it really boils down to how well the chip can handle it as running the game itself doesnt put much stress on the card if that makes sense, but if that were true then I would expect a 640 to decrease your performance if anything...and Im not fully convinced thats the case

I'm not 100% sure tbh on how PhysX works though, but I take comfort in the fact that most others don't either :lol: 

You might as well give it a blast if you think its worth it, if it doesnt work out then you can probably sell the 640 and break even if you are buying 2nd hand... Just thought I would encourage you to look a little deeper as it seems to be quite a mixed bag of results... and definitely big question marks over your minimum frame rates...

Edit: If you do go for it definitely post you results up here, interested to see what effect it has in your situation
November 4, 2012 5:10:31 PM

Just try any big map co-op with a lot of depth of view agains bleeding targets with water around xD see what that does to your FPS. I could just lower it to medium, but honestly I am curious now!

I do know that physx works with the cuda cores, and the gt640 seems to have enough.

What I find odd is that people who tried dedicated cards only see loads up to 45%, which makes me think that the system has to wait for the Physx calculation, and that causes the FPS drops I experience.
November 4, 2012 5:15:00 PM

Anonymous said:
yes it can. i have read on the 3D guru forums that a x4 slot works just fine. if you need more info:
http://bit.ly/SJcV4B


Thanks for googling that for me -.- The thing is there is absolutely no game that uses as much Physx as this (no, not even Mafia II nor Batman, this takes it to a whole other level), so there is a case to be made that this might be different. So, I guess the real question is if 4x was enough due to low Physx or due to Physx not requiring much bandwith. If you find this information regarding Borderlands 2 please share :D 
a b U Graphics card
November 4, 2012 5:30:47 PM

Well if thats the case it more of a reason to give it a shot, my load is typically around 60% again that might be a diiference between kepler and Fermi though...

Worst case scenario it dipped into the late 30's and it was pretty rare, cant understand your fps...
Anonymous
a c 117 U Graphics card
November 4, 2012 5:37:40 PM

Caspase said:
Thanks for googling that for me -.- The thing is there is absolutely no game that uses as much Physx as this (no, not even Mafia II nor Batman, this takes it to a whole other level), so there is a case to be made that this might be different. So, I guess the real question is if 4x was enough due to low Physx or due to Physx not requiring much bandwith. If you find this information regarding Borderlands 2 please share :D 

sorry man i am not exactly motivated to find your answer for you since i do not play BL2.

if you want some reliable info on physX and how it works:
http://physxinfo.com/news/

scroll down for a bunch of stuff with BL2.


cheers.
November 4, 2012 6:00:06 PM

Thanks for the help, I am going to try it miself, the card should be here by thursday.

I would also like to try something like overclocking shader cores specifically even if I had to underclock memory or something... Is that possible?
Anonymous
a c 117 U Graphics card
November 4, 2012 7:52:40 PM

Caspase said:
Thanks for the help, I am going to try it miself, the card should be here by thursday.

I would also like to try something like overclocking shader cores specifically even if I had to underclock memory or something... Is that possible?

yes you can but it wouldn't be advisable to downclock the memory.
here are some overclocking on a zoltac card (keep in mind all cards achieve different results!)
Overclocking
Quote:
Maximum stable clocks of our card are 1102 MHz core (22% overclock) and 1111 MHz Memory (25% overclock).

that is pretty good considering i doubt the card is on anything but a reference PCB. you may get either precisionX or afterburner to overclock the card. then you can get PhysX FluidMark to test the physX performance.

also a gpu can be finicky when overclocking. ( i am just pulling numbers out of the air) you may find a good increase in performance with a 12% overclock but a decrease with a 15% but then a not so great of a bump with 18%. now the difference between 12% and 18% may not be worth it so *i would* run the 12% knowing i am getting the best performance without stressing the gpu unnecessarily.

it will take time and patience to figure out what is the best configuration. try not to take short cuts by jumping on a 15%+ overclock immediately. you'll learn a lot more by taking it easily then jumping on a % figure blindly.
November 4, 2012 8:07:03 PM

But wouldn't that just up clock for core? I would like to fiddle with the shaders only because those are the ones that do the calculations and see if it improves fps drops significantly, bringing it closer to Fermi level.

Maybe I am being a noob about this... OC the core also OC the shaders? Is that it?
Anonymous
a c 117 U Graphics card
November 4, 2012 8:11:33 PM


yes, core/shaders same thing when setting the clock.
November 4, 2012 8:16:21 PM

Always saw different values, so figured they were different... Thanks for the trouble of writing all that up !
!