Overclock.net banner
1 - 17 of 17 Posts

shreduhsoreus

· Registered
Joined
·
18 Posts
Discussion starter · #1 ·
Before I get to my actual question, let me clear a few things up:

-I don't want to hear "It's not worth it". You're wasting your time because I simply do not care, sorry.
-I'm well aware that hardware accelerated Physx is supported in a very limited number of games.
-I understand that Physx support going forwards will probably be even less than it is now, if at all.
-I play/plan to play around a dozen Physx supported titles(that allow for dedicated Physx cards).
-I currently run 3 1440p monitors in surround(7788x1440) @ 165Hz with a single 1080 Ti and a 3GB 1060 for Physx.
-I am content with the limited performance boost my current Physx card gives me. Any strain removed from my 1080 Ti is a benefit to me.
-I have enough EVGA Bucks to get a replacement card for dirt cheap, and given that I spent over 4 grand on my current setup, spending an extra 30-80 is nothing to me.

With that out of the way, here's my situation: at most, my 3GB 1060 will see 25% utilization in hardware accelerated Phsyx titles and only boosts to 1569 MHz(while the 1080 Ti is pegged at 100% boosting over 2GHz). I plan to give the 1060 to my nephew for Christmas and want to replace it with a cheaper card that doesn't need to connect to the PSU. The GT 1030's die is essentially 1/3 of a 3GB 1060(then paired with less/slower memory) and would have no problem reaching the same core clock(and probably boost higher since it will have much higher utilization). Given that my 1060 only hits 25% utilization at low clocks, would a GT 1030 give me the same/very similar results or would I be much better off going with a 1050 or 1050 Ti? And if the GT 1030 should be enough, would I be alright running it in my bottom PCIe x4 slot(that uses chipset lanes) to keep my 1080 Ti running in x16 mode, or do I need to have the Phsyx card in my middle x8 slot? I have a Z170A Krait 3X, if that helps any.

I know this is a fairly niche setup and it's hard to find tests for specific cards(especially newer ones), so any insight(other than "Phsyx card not worth, don't buy") would be greatly appreciated. Thanks!
 
Last time I played a PhysX game was Borderlands 2 and a 9800GT was enough to run all of the PhysX calculations with no perceivable lag. In fact, I think I even ran PhysX on my FX 8320's additional cores. I don't think a 1050Ti is necessary for your PhysX needs, a 1030GT would be more than enough in my opinion.

EDIT: The setup I used when playing Borderlands 2 was my 7970 hybridized with the 9800 GT PhysX card.

EDIT2: I just remembered that I upgraded the PhysX card to a GTX 650 shortly after using the 9800 GT because I got the 650 for free, so my previous statement about the 9800 GT may be incorrect.
 
Back with BL2 I ran my PhysX card in the chipset lanes without a noticeable performance hit (GTX580 with GTX560) but with modern cards and games it might have more of a hit. You'd have to try it. I'd recommend not going with less than the 1050Ti. PhysX doesn't need to handle a lot of data but it does need to handle it fast. If it can't get the calculations back into the pipeline in time it'll hold up the frame on the 1080Ti.
 
Quote:
Originally Posted by Meisgoot312 View Post

Last time I played a PhysX game was Borderlands 2 and a 9800GT was enough to run all of the PhysX calculations with no perceivable lag. In fact, I think I even ran PhysX on my FX 8320's additional cores. I don't think a 1050Ti is necessary for your PhysX needs, a 1030GT would be more than enough in my opinion.
I have a post somewhere on the forums running my own tests in BL2. A 2600k tanked my perceived performance (minimum framerate in particular) running PhysX calculations with a GTX580, while the 560 gave a sizeable performance bump with a much smoother experience.
 
Discussion starter · #5 ·
Quote:
Originally Posted by MightEMatt View Post

PhysX doesn't need to handle a lot of data but it does need to handle it fast. If it can't get the calculations back into the pipeline in time it'll hold up the frame on the 1080Ti.
I plan to overclock whatever I get as high as I can. Do you think that if I manage to push a GT 1030 to ~1900MHz it will be fast enough to keep up or do you think it's more a matter of it not having the number crunching capability due to much fewer cores?
 
Quote:
Originally Posted by shreduhsoreus View Post

I plan to overclock whatever I get as high as I can. Do you think that if I manage to push a GT 1030 to ~1900MHz it will be fast enough to keep up or do you think it's more a matter of it not having the number crunching capability due to much fewer cores?
I honestly can't be sure unfortunately, you'd have to find somebody who has done it to be sure. There's a video with it running as a PhysX card for a GTX680 in Fluidmark and it was at around 60-70% usage for the majority of the test. PhysX calculations don't scale directly with framerate so it's not a lost cause, but it will definitely see an increased load. Games probably aren't going to be as demanding as Fluidmark either. Just approach it keeping in mind that its all about beating the frametime. The fact that you run such a large resolution is in your favor in that regard.

Edit: A GTX1050 would definitely be worth considering with all of this in mind. It'd give you nearly double the performance for 50% more cost, and wouldn't be far off from the 1050Ti.
 
Before you hand that 1060 off, do some testing and see how much of its memory is being utilized with PhysX.

If you find that it does not use much memory, you will have a good option of lowering the memory clock speed on the next card, whatever you grab, in order to allow more power draw from the gpu core before any sort of power limit kicks into place.

That alone will allow precious more overclocking margins to be available.

EDIT keep in mind also, grab MSI afterburner and use its voltage and frequency curve to your advantage. you can basically force a higher clock than GPU boost would normally call for in low or medium usage scenarios
 
Discussion starter · #8 ·
Quote:
Originally Posted by MightEMatt View Post

I honestly can't be sure unfortunately, you'd have to find somebody who has done it to be sure. There's a video with it running as a PhysX card for a GTX680 in Fluidmark and it was at around 60-70% usage for the majority of the test. PhysX calculations don't scale directly with framerate so it's not a lost cause, but it will definitely see an increased load. Games probably aren't going to be as demanding as Fluidmark either. Just approach it keeping in mind that its all about beating the frametime. The fact that you run such a large resolution is in your favor in that regard.

Edit: A GTX1050 would definitely be worth considering with all of this in mind. It'd give you nearly double the performance for 50% more cost, and wouldn't be far off from the 1050Ti.
I've actually seen that video you mentioned. I was just a little concerned since the 1080 Ti is so much more powerful than a 680 that the 1030 might hold it back(until I saw how little the 1060 is being used). I just wish I could find a single slot 1050/1050 Ti here in the USA, seems they only exist in European/Asian markets.

Quote:
Originally Posted by mattliston View Post

Before you hand that 1060 off, do some testing and see how much of its memory is being utilized with PhysX.

If you find that it does not use much memory, you will have a good option of lowering the memory clock speed on the next card, whatever you grab, in order to allow more power draw from the gpu core before any sort of power limit kicks into place.

That alone will allow precious more overclocking margins to be available.

EDIT keep in mind also, grab MSI afterburner and use its voltage and frequency curve to your advantage. you can basically force a higher clock than GPU boost would normally call for in low or medium usage scenarios
Memory usage on the 1060 is minimal. I haven't seen it even come close to 500MB. From my understanding, Physx is more about calculations and requires very little memory. I hadn't thought about lowering the memory to allow more power for the core, that's a pretty good idea, but I'm all about custom curves. Had the 1080 Ti running a hair over 2.1Ghz but I got lazy during stability testing and just dropped it to a flat +100 to hit 2050.
 
I tried a 1030 for physx with my Titan x Maxwell and it was too slow. I had higher fps using the Titan x by itself.
 
Discussion starter · #11 ·
hi. so how this turned out. im in the same boat than you. trying to find a cheap 1030 so my 1080 ti is free of some calculation while playing batman ak
I never ended up following through. Stock was atrocious by the time I had some cash to spend on a card(was aiming for the 1050 at the bare minimum based on benchmark tests I've seen, 1030 is just too slow) and prices were outrageous(was not about to pay $150+ for a 1050 just to use for Physx), so I just said screw it and turned down/off Physx in the games that was using it.

I will say though that having the dedicated 1060 for Physx was awesome in Borderlands 2/TPS. Cranked every setting up, still 100-120 FPS at 7800x1440. Before Warframe removed Physx I was getting heavy utilization of the Physx card in that game though, upwards of 50%. So if you're considering a 1030 I'd recommend going up at least a tier or 2. Or find a used 1060 once the new cards come out.
 
Also there was no difference in fps running it at 4x vs 16x for the 1030.
 
The only game I got into that works with Physx cards is the Borderlands series and my 1060 got basically the same performance with or without a dedicated 950.
 
physx cards are not going to help in any way. today gpus are really powerful and they do physx without a performance hit.
 
I repeatedly tested Metro Last Light using a dedicated physX video card every time I upgraded my nvidia cards. The worst performance was always observed using the GPU for PhysX and while the dedicated physX cards (a 780 dedicated physX card w/a 980Ti and the 980Ti as a dedicated physx card w/my 1080Ti) leveled the framerate out at 60FPS there were many more brief framerate spikes than when physX was set to CPU.

I'd be interested in seeing how Killing Floor 2 does w/a dedicated physX card and whether or not FleX effects are supported w/a dedicated physX card.
 
I never ended up following through. Stock was atrocious by the time I had some cash to spend on a card(was aiming for the 1050 at the bare minimum based on benchmark tests I've seen, 1030 is just too slow) and prices were outrageous(was not about to pay $150+ for a 1050 just to use for Physx), so I just said screw it and turned down/off Physx in the games that was using it.

I will say though that having the dedicated 1060 for Physx was awesome in Borderlands 2/TPS. Cranked every setting up, still 100-120 FPS at 7800x1440. Before Warframe removed Physx I was getting heavy utilization of the Physx card in that game though, upwards of 50%. So if you're considering a 1030 I'd recommend going up at least a tier or 2. Or find a used 1060 once the new cards come out.
I think I saw an article/benchmark a while back that the basic thrust of was you need a card of about the same speed as your main card to gain any benefit.

The idea, for whatever reason, is with your 1080ti you would need a 1080 in dedicated physx to get any benefit and it still is not a big benefit.
 
1 - 17 of 17 Posts