Borderlands 2 GPU/CPU benchmarks [TechSpot/HardOCP/Others]

[Termie]

TechSpot's review of Borderlands 2 performance on the PC
http://www.techspot.com/review/577-borderlands-2-performance/

Here are the cliff's notes:

(1) Nvidia runs better in this game (no surprise, as they were very involved in the production of it). This is most evident at 2560x1600.
(2) It's a very CPU-bottlenecked game (EDIT: as others have pointed, out, it's actually CPU-limited) - at 1680, the top 10 cards all exhibit about the same fps on a 3770k, and even at 1920, there's very little differentiation at the top. It responds linearly to CPU overclocks - that's almost unheard of in modern games.
(3) AMD CPUs do not do well...ouch.
(4) The HD7850 is underperforming the HD5870. AMD has more work to do in getting the most performance out of that card.
(5) The 660Ti performs almost exactly like a 670, whereas the 680 is able to pull away. Clearly this game doesn't depend on memory bandwidth at all.

UPDATE: It appears that TechSpot has erroneously tested PhysX on AMD cards using the CPU, which defaults to low settings. There is no way to run PhysX directly on an AMD card, and there is no way to run high PhysX on a CPU without significantly compromised performance.

UPDATE - 9/24/12
For a more in-depth comparison of PS3 vs. Xbox vs. PC, see this EuroGamer review
http://www.eurogamer.net/articles/digitalfoundry-borderlands-2-face-off

For a more in-depth comparison of PhysX in Borderlands 2, see this article
http://physxinfo.com/news/9500/gpu-physx-in-borderlands-2/

UPDATE - 10/1/12
HardOCP's Performance and IQ Review
http://www.hardocp.com/article/2012/10/01/borderlands_2_gameplay_performance_iq_review

UPDATE - 10/4/12
PCPerspective's PhysX Comparison
http://www.pcper.com/reviews/Graphi...and-PhysX-Comparison-GTX-680-and-HD-7970/GPU-

UPDATE - 10/7/12
German-siite PhysX tests on GPU and CPU
http://www.pcgameshardware.de/Borderlands-2-PC-234034/Tests/Borderlands-2-Physx-Test-1026921/

 

[BenSkywalker]

Actually, the high end AMD cards are slightly edging out the high end nVidia cards when everything is cranked all the way(including PhysX).

This game actually seems like a really good benchmark in general. Not so much because of the results it shows us, but because it scales so well both with CPU and GPU speed depending on the settings.

 

[railven]

There are two mods, one was the old one developed by a guy named GenGL (or something like that) over at ngohq.com. He stopped working on the mod sometimes in August 2011 due to lack of nVIdia card.

Some other guy quickly stepped in with a different mod that is implemented per game (ie modifies the *.dlls directly.)

Anyways, both mods worked up to Batman: AC and benching Batman: AC with both mods using a 7970+9800GTX+ system yielded the same results.

I couldn't get the legacy mod to work with BL2, but the newer mod worked fine, offloading up to 60% load on my 9800GTX+ when Blood was involved.

Will be reading this article once I get the chance.

 

[railven]

Oh, seems I was wrong and must have done something wrong on my end:

Using the Hybrid PhysX mod v1.05ff developed by "GenL," we managed to use PhysX on Radeon cards. Note that we had to use version 9.11.0621 of the PhysX System Software, whereas the Nvidia cards could use version 9.12.0213, but this didn't seem to affect performance.

So the GenL (the guy who's name I couldn't remember) mod still works? I might give that a test when I get home.

 

[Termie]

Oh, seems I was wrong and must have done something wrong on my end:



So the GenL (the guy who's name I couldn't remember) mod still works? I might give that a test when I get home.

Yeah, but read what they wrote - they are running PhysX on the 7970, it seems. So maybe this GenL guy actually stepped up to the challenge of making his mod work without an nVidia card.

 

[railven]

Yeah, but read what they wrote - they are running PhysX on the 7970, it seems. So maybe this GenL guy actually stepped up to the challenge of making his mod work without an nVidia card.

While skimming the review, they didn't disclose what card is being used for the PhysX offloading.

With that mod, you can turn PhysX to medium/high, which I assume would attempt to run off the CPU because at least on my end, my FPS tanked to <30 FPS.

EDIT: Unless they aren't disclosing everything, or they pulled off something I didn't experience.

Here is what I saw on my system (in sig):
7970 PhysX High = <30 FPS during tons of PhysX effects
7970 + 9800 GTX+ PhysX High == ~40 FPS during tons of PhysX effects
7970 + GTX 460 = ~60 FPS during tons of PhysX effects

 

[Olikan]

maybe PhysX is done by the cpu in this game

it explains the cpu bottleneck, and the mod to amd cards

 

[railven]

maybe PhysX is done by the cpu in this game

it explains the cpu bottleneck, and the mod to amd cards

While possible that some of the PhysX is being done on the CPU, from my experience with the game - there is no way a Radeon is doing PhysX on high by itself with 60+ FPS.

My system in sig is running my card and CPU faster at a lower resolution and I was tanking below 30 FPS running max settings, No FXAA, and PhysX on high. My FPS didn't improve until I fixed the PhysX Mod, and then swapped out my aging 9800 GTX+ for my GTX 460.

 

[exar333]

Some interesting observations to be made in this article: http://www.techspot.com/review/577-borderlands-2-performance/

I debated posting this in the CPU forum or the PC Gaming forum, but I know a lot of people were going to buy new graphics cards to play this game.

Here are the cliff's notes:

(1) Nvidia runs way better in this game (no surprise, as they were very involved in the production of it). This is most evident at 2560x1600.
(2) It's an incredibly CPU-bottlenecked game - at 1680, the top 10 cards all exhibit about the same fps on a 3770k, and even at 1920, there's very little differentiation at the top.
(3) It responds linearly to CPU overclocks - that's almost unheard of in modern games.
(4) AMD CPUs do not do well...ouch.
(5) The HD7850, a card I'm seriously considering for my HTPC, is again underperforming the HD5870. I really think AMD has more work to do in getting the most performance out of that card.
(6) And finally the 660Ti performs almost exactly like a 670, whereas the 680 is able to pull away. Clearly this game doesn't depend on memory bandwidth at all.

Oh, and what the heck is that mod they're using to run PhysX on AMD cards? Does that mean without an nVidia card even in the system for dedicated PhysX???

Interesting to note that the 5870 is also better than the 6870 by a decent margin. Except the 560 variants, this game looks to like 2GB of VRAM.

AMD CPU performance here is abysmal, period. The i7-920 is 10% better than the 8150, which has double the 'cores' and an extra ghz in speed. Yikes. The 8150 at 4.5ghz is only 2 fps better than a 2.5 ghz 3770k (slower than stock!). I wonder what the 3960x scaling was? Curious.

I agree with some others here that the Physx performance looks suspect. More information about the configuration would have been helpful. That said, it was nice to see the 7970 see less of a drop in performance vs. the 680 with it enabled. That is assuming the data and methodology are solid.

Edit: Little disappointed that no FXAA/MSAA/etc. runs were included.

 

[wand3r3r]

I haven't even got to level 3 yet, but the first few minutes haven't been amazing yet as far as physx goes. Maybe it'll start to show more when you start getting elemental weapons etc. There was a blanket in a doorway and I swiped it and broke it as I saw the blanket effects in the Physx demo. I haven't checked the framerate but the load on the GTX 690 isn't too high. It's smooth as butter and probably doesn't drop too low ever.

I'll try play it more on the weekend.

Can I force some additional settings from the Nvidia driver being the 690 isn't even fazed by the game? There was only one AA related setting in game that I noticed, I can't remember what it was called. Playing at 1980x1200...

 

[boxleitnerb]

FXAA.
You can force SGSSAA through NV Inspector, makes the game look more smooth and less blurry.

 

[exar333]

I haven't even got to level 3 yet, but the first few minutes haven't been amazing yet as far as physx goes. Maybe it'll start to show more when you start getting elemental weapons etc. There was a blanket in a doorway and I swiped it and broke it as I saw the blanket effects in the Physx demo. I haven't checked the framerate but the load on the GTX 690 isn't too high. It's smooth as butter and probably doesn't drop too low ever.

I'll try play it more on the weekend.

Can I force some additional settings from the Nvidia driver being the 690 isn't even fazed by the game? There was only one AA related setting in game that I noticed, I can't remember what it was called. Playing at 1980x1200...

With a 690, you should be able to have max AA + Physx and still hit 60fps without an issue.

 

[railven]

With a 690, you should be able to have max AA + Physx and still hit 60fps without an issue.

You can do that with a GTX 680.

 

[JohnnyChuttz]

A 690 should be closer to 100fps

Avg ~120 FPS, minimum 59, everything maxed here.
670 FTW SLI

 

[BFG10K]

(2) It's an incredibly CPU-bottlenecked game - at 1680, the top 10 cards all exhibit about the same fps on a 3770k, and even at 1920, there's very little differentiation at the top.

I wouldn’t infer that from 1680x1050. That’s a prehistoric resolution with essentially no AA. Even 1920x1200 is a middling resolution and again, no real AA.

A hyper-threading dual-core Sandy/Ivy will likely run the game faster than the i7 920 (i.e. just fine).

 

[Termie]

I wouldn’t infer that from 1680x1050. That’s a prehistoric resolution with essentially no AA. Even 1920x1200 is a middling resolution and again, no real AA.

A hyper-threading dual-core Sandy/Ivy will likely run the game faster than the i7 920 (i.e. just fine).

1920x1200 is still above what the average gamer is playing at, so it's an entirely relevant test of CPU-bottlenecking. And I highly doubt a 2120 would beat the 920, but even if it did, that's still not disproving my point. It would only show that the game wants 4 fast logical cores. If there were no CPU-bottlenecking, then the two CPUs would perform identically to each other and to every other CPU, which obviously they won't.

 

[BFG10K]

1920x1200 is still above what the average gamer is playing at...[snip]

So are all of those high-end quad/hex CPUs, and those top end graphics cards. I’m not sure where you’re going with this?

A 6850 - more in line for an average gamer - struggles even at 1680x1050.

Ironically, the dual-core processors I mentioned would comfortably fit into an average gaming system based on pricing alone. When paired with a mid-range GPU, they'd cause the GPU to be the primary bottleneck, even at lower resolutions.

And I highly doubt a 2120 would beat the 920, but even if it did, that's still not disproving my point. It would only show that the game wants 4 fast logical cores. If there were no CPU-bottlenecking, then the two CPUs would perform identically to each other and to every other CPU, which obviously they won't.

The point is that I wouldn’t call it an “…incredibly CPU-bottlenecked game…” based on the AA-less resolutions you quoted.

We already know games get CPU limited at low resolutions with no AA, and this one is no different.

 

[Ferzerp]

The graphics all said MAX, and that would include the only natively supported AA (FXAA).

If you zoom in on the tiny picture, you can see the FXAA option has two letters beside it. That would denote On as Off is three letters.

 

[BFG10K]

The graphics all said MAX, and that would include the only natively supported AA (FXAA).

Yeah, but it's not real AA. By that I mean it's a very cheap (in terms of performance) post-filter compared to real AA like MSAA/SSAA, and barely loads the GPU at all.

 

[WiseUp216]

I haven't even got to level 3 yet, but the first few minutes haven't been amazing yet as far as physx goes. Maybe it'll start to show more when you start getting elemental weapons etc. There was a blanket in a doorway and I swiped it and broke it as I saw the blanket effects in the Physx demo.

I'm around Level 11 and I've seen a few pretty cool PhysX implementations. I was fighting my way through some kind of warehouse and I saw a long banner hanging from the rafters and I shot one of the ropes and it crashed to the ground. Doesn't sound like much, but it looked amazing in all the chaos.

Also, you'll get a vehicle pretty early on that shoots saw blades and they look great bouncing all over the place.

My i5-2300 @ 3.5 and GTX 670 is running this game maxed beautifully. I use the framerate smoothing option (22-60 FPS or something like that).

 

[Ferzerp]

The game does not natively support any other AA.

I don't see how you can argue that the game isn't cpu limited more than GPU limited when the max settings show just that.

It seems you're saying "well, if the game put more of a load on the gpu, it wouldn't be cpu limited". Well great, but that's not what this game does (and somewhat of a tautology). I agree, if this were a different game that is gpu limited, it wouldn't be cpu limited....

 

[Arkadrel]

(1) Nvidia runs way better in this game (no surprise, as they were very involved in the production of it).

1920x1200 + max quality:

Nvidia Geforce 680 = 74 fps avg.
Amd Radeon 7970 = 72 fps avg.

Thats a ~2,7% differnce in performance.
Hell a 7970 1ghz edition might even beat out that 680's score.

(2) It's an incredibly CPU-bottlenecked game - at 1680, the top 10 cards all exhibit about the same fps on a 3770k, and even at 1920, there's very little differentiation at the top.

Yeap looks very demanding on the CPU.

(3) It responds linearly to CPU overclocks - that's almost unheard of in modern games.

Looks like its one of those, better overclock your CPU before you try it.
Doesnt look like GPU matters nearly as much as what CPU you have.

(4) AMD CPUs do not do well...ouch.

Amd FX 4170 = 120$ = 47 fps. ( 0.39 fps pr $ ) (higher is better)
Intel i5-2500k = 220$ = 61 fps. ( 0,27 fps pr $ ) (higher is better)

Looks like you get more "value" out of a Smaller FX-4170 than you do a i5-2500k.
That said... the 29% differnce in performance (fps-wise) is kinda noticeable.

(5) The HD7850, a card I'm seriously considering for my HTPC, is again underperforming the HD5870. I really think AMD has more work to do in getting the most performance out of that card.

Yep looks like AMD hasnt had a chance to optimise drivers for this game yet.
Probably performance updates to come for this game with newer drivers.

 

[BFG10K]

The game does not natively support any other AA.

And?

I don't see how you can argue that the game isn't cpu limited more than GPU limited when the max settings show just that.

Run any other modern game at 1680x1050 with FXAA on a top-end GPU, and many will show the same thing (i.e. primarily CPU limited).

People are acting like this is something new, but it isn't. That's the point.

 

[Silverforce11]

I am shocked i dont even know what to make of that. Identical fx seen from running with Physx on an AMD card. Seriously what the hell is going on and its crazy that it even works on AMD gpus at all. WOW.

 

[railven]

I am shocked i dont even know what to make of that. Identical fx seen from running with Physx on an AMD card. Seriously what the hell is going on and its crazy that it even works on AMD gpus at all. WOW.

Don't put too much value in that, there are things amiss. They don't state if they used a second card for offloading (which I'm confident they did) and they don't state what PhysX effects they tested.

Debris and flags run fine off the CPU with just an HD 7970, but once you hit some blood/goo, good bye reliable FPS.