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

[blastingcap]

This game appears to have quite a few bugs and is causing slowdowns and lags regardless of video hardware. See here and here as examples.

bugfixes should come quickly w/ so many users generating error logs, etc.

http://store.steampowered.com/stats/

bl2 - most-played steam game right now by far

i'm lol'ing at the three counterstrikes hogging 3 of 10 top-10 slots though. and wonder where starcraft ii would rank

 

[Jaydip]

Sorry for a bit off-topic but this reminds me of the performance numbers from Diablo 3.In the reviews 680 was getting >100 fps @1080P while my game play experience was totally different.It seems reviewers are too busy publishing the charts while lacking the time to actually play the game for some time.Also raw numbers are useless for this sort of games what we need is a fps graph.For e.g. I dip to 35-40fps in Diablo 3 while playing with other friends in inferno(the witch doctors seem to cause the most lag along with the cm wizards)

 

[Avalon]

My GTX 460 bit the dust the night before I finished downloading BL2 on steam. Been playing it at 1680x1050 on mostly medium settings on my onboard HD 4000, gets about 25 fps lol.

 

[Keysplayr]

This game is so much fun. On a single GTX680, all setting maxxed at 1920x1080, PhysX High, I seldom see less than 50fps. I haven't kept a solid eye on the fps counter (eVGA Precision) but when action gets hairy, I take a look over and it is usually somewhere in the 60's. Most of the time with less action I get around mid 70's to mid 80's. And when there is nothing around, it breaks 100. One 680 doing it all.

 

[wand3r3r]

Yeah physx is running on the gpu (gtx 690).

What is a good fps benchmark tool? I can benchmark something and throw it up here. The game is generally very nice, but physx seems insanely demanding period.

For those of you getting worried if your systems will work, they will, you just have to turn a setting or two down (physx probably).

What's with the bickering about it. The game may be buggy and is certainly demanding. My buddy plays with a single 680, I'll see if I can have him benchmark it too just cuz I'm curious if the same scene drops are 1/2 or does SLI drop as far as a single card (or further). As I mentioned earlier I mostly notice 60-200 fps. It's hard to see the fps drops on the counter because you're usually next to a boss with fireworks so intense you can't even see anything else then the explosions etc.

 

[KingFatty]

This game is so much fun. On a single GTX680, all setting maxxed at 1920x1080, PhysX High, I seldom see less than 50fps. I haven't kept a solid eye on the fps counter (eVGA Precision) but when action gets hairy, I take a look over and it is usually somewhere in the 60's. Most of the time with less action I get around mid 70's to mid 80's. And when there is nothing around, it breaks 100. One 680 doing it all.

I've wondered how benchmarking websites are able to get those fancy charts that claim to show super-precise frame information, is that some utility that can capture the FPS over time? Wouldn't the very act of collecting that information cause an impact to the FPS, especially if you are CPU-bound?

But I bet if you could examine a chart of the FPS over time, and see where dips occur, maybe there would be some pattern that revealed any oddities or whatever, or if performance just scales according to how many particles are on the screen.

 

[blackened23]

Yeah physx is running on the gpu (gtx 690).

What's with the bickering about it. The game may be buggy and is certainly demanding. My buddy plays with a single 680, I'll see if I can have him benchmark it too just cuz I'm curious if the same scene drops are 1/2 or does SLI drop as far as a single card (or further). As I mentioned earlier I mostly notice 60-200 fps. It's hard to see the fps drops on the counter because you're usually next to a boss with fireworks so intense you can't even see anything else then the explosions etc.

Only the physx high part is demanding. If you put it on low or turn it off the game is ridiculously fast with framerates well past 100 - Physx has always incurred a performance drop. Not sure why it surprises anyone. You can still run it with physx high but it can slow down during boss fights or heavy combat...just turn it down a notch...

 

[SPBHM]

Only the physx high part is demanding. If you put it on low or turn it off the game is ridiculously fast with framerates well past 100 - Physx has always incurred a performance drop. Not sure why it surprises anyone. You can still run it with physx high but it can slow down during boss fights or heavy combat...just turn it down a notch...

judging by the benchmarks, even with PhysX in low the game can be demanding for the CPU in some parts, not nearly as much as with high, but I wouldn't expect 100fps locked all the time with any CPU!?

 

[Lonbjerg]

I like where NV is going with this, making Physx run well on CPU but runs better on their GPU.. widespread usage/acceptance = better mindshare for their GPU products.

PhysX has always been able to run via the CPU, and the name of the game has always been that there were better hardware than your CPU for physics calculations:
http://www.youtube.com/watch?v=Z2xeqYoqEkc

(In that example the uploader isn't aware that PPU cloth had a finer mesh, thus doing more calculations)

It's not that NVIDIA has fixed it so it runs better om the GPU vs the CPU...it's just the nature of physics and the associated calculation that favours the PPU/GPU architechture over the CPU architechture.

It's the same thing today:
http://www.youtube.com/watch?v=EohV5YfBR8w

GPU vs CPU...

But I hope no one ever thought their CPU are just fine for physics...ecause then your CPU is also just fine for graphics...😉

 

[Lonbjerg]

^This.

If 2xGK104 can't manage it then PhysX in this game is too demanding to run on current hardware. I wonder if OpenCL or Direct Compute is better?

Find me one test were DirectCompute or OpenCL runs better than CUDA.
Otherwise...lets not go there :whiste:

 

[Lonbjerg]

Yeah physx is running on the gpu (gtx 690).

What is a good fps benchmark tool? I can benchmark something and throw it up here. The game is generally very nice, but physx seems insanely demanding period.

For those of you getting worried if your systems will work, they will, you just have to turn a setting or two down (physx probably).

What's with the bickering about it. The game may be buggy and is certainly demanding. My buddy plays with a single 680, I'll see if I can have him benchmark it too just cuz I'm curious if the same scene drops are 1/2 or does SLI drop as far as a single card (or further). As I mentioned earlier I mostly notice 60-200 fps. It's hard to see the fps drops on the counter because you're usually next to a boss with fireworks so intense you can't even see anything else then the explosions etc.

So you have one GPU(GK104) running graphics and one GPU(GK104) running PhysX.

Both your GPU's are clocked lower than eg. a GTX680 SLI setup...what is your problem?
(Besides SLI, PhysX and how they interact?)

If you compare running SLI with no/low PhysX (via CPU) to running WITH PhysX on one GPU and graphics running on the other, I can understand your confusing...but not at the technical point, it tick me off.

Try disabling 1 GPU and running with PhysX low/off...and then with PhysX set to high.
You are sorely lacking in datapoints...:whiste:

 

[wand3r3r]

So you have one GPU(GK104) running graphics and one GPU(GK104) running PhysX.

Both your GPU's are clocked lower than eg. a GTX680 SLI setup...what is your problem?
(Besides SLI, PhysX and how they interact?)

If you compare running SLI with no/low PhysX (via CPU) to running WITH PhysX on one GPU and graphics running on the other, I can understand your confusing...but not at the technical point, it tick me off.

Try disabling 1 GPU and running with PhysX low/off...and then with PhysX set to high.
You are sorely lacking in datapoints...:whiste:

No, it's running in SLI with one gpu in use to ALSO run physx. Exactly the same as 680 in sli. The picture shows SLI with physx allocated to gpu2.

You'd best not be "ticked off", you didn't drop the 1000 on the gpu. 😉

Anyways, they are barely lower then a 680 on clocks. I can easily oc higher then 680 stock. But regardless, the point here is discussing how demanding BL2 is on CPU/GPUs. I am merely relaying my experiences playing on basically the higher(/est) end hardware that probably only 5% of people use. Even if I overclock to 680 stock speeds what do I gain, 10% at most to 15 fps + 1.5 fps = 16.5 fps. 😛

 

[3DVagabond]

Find me one test were DirectCompute or OpenCL runs better than CUDA.
Otherwise...lets not go there :whiste:

Apples and oranges. I don't need to find anything and if you don't want to go there nobody asked you to.

 

[Lonbjerg]

No, it's running in SLI with one gpu in use to ALSO run physx. Exactly the same as 680 in sli. The picture shows SLI with physx allocated to gpu2.

You'd best not be "ticked off", you didn't drop the 1000 on the gpu. 😉

Anyways, they are barely lower then a 680 on clocks. I can easily oc higher then 680 stock. But regardless, the point here is discussing how demanding BL2 is on CPU/GPUs. I am merely relaying my experiences playing on basically the higher(/est) end hardware that probably only 5% of people use. Even if I overclock to 680 stock speeds what do I gain, 10% at most to 15 fps + 1.5 fps = 16.5 fps. 😛

http://www.nvidia.com/object/physx_faq.html#q4

How does PhysX work with SLI and multi-GPU configurations?
When two, three, or four matched GPUs are working in SLI, PhysX runs on one GPU, while graphics rendering runs on all GPUs. The NVIDIA drivers optimize the available resources across all GPUs to balance PhysX computation and graphics rendering. Therefore users can expect much higher frame rates and a better overall experience with SLI.

A new configuration that’s now possible with PhysX is 2 non-matched (heterogeneous) GPUs. In this configuration, one GPU renders graphics (typically the more powerful GPU) while the second GPU is completely dedicated to PhysX. By offloading PhysX to a dedicated GPU, users will experience smoother gaming.

Finally we can put the above two configurations all into 1 PC! This would be SLI plus a dedicated PhysX GPU. Similarly to the 2 heterogeneous GPU case, graphics rendering takes place in the GPUs now connected in SLI while the non-matched GPU is dedicated to PhysX computation.

What this mean in action is this:
http://physxinfo.com/wiki/Configuration_types

In SLI mode, only one GPU (main, connected to monitor) is calculating PhysX. Frame processing goes like this (in case of 2-Way SLI, as example):

• GPU1 and GPU2 start rendering their frames
• GPU1 is doing the PhysX calculation
• GPU1 and GPU2 start rendering their frames

You are breaking up the graphics rendering on GPU1 to do PhysX with your current settings...try all configurations before talking about what is doing what...that is my point.

 

[Lonbjerg]

Apples and oranges. I don't need to find anything and if you don't want to go there nobody asked you to.

You got nothing, hence the false claim "I don't need too..."
You raised the "question" so let me inform you:

There has never been a benchmark showing any diffrences worth talking about on a NVIDIA GPU performance when running OpenCL/DirectCompute compared to when running CUDA.

It always end up like this:

http://blog.accelereyes.com/blog/2010/05/10/nvidia-fermi-cuda-and-opencl/


Now let it die.

 

[wand3r3r]

http://www.nvidia.com/object/physx_faq.html#q4


What this mean in action is this:
http://physxinfo.com/wiki/Configuration_types


You are breaking up the graphics rendering on GPU1 to do PhysX with your current settings...try all configurations before talking about what is doing what...that is my point.

From your quote:

When two, three, or four matched GPUs are working in SLI, PhysX runs on one GPU, while graphics rendering runs on all GPUs. The NVIDIA drivers optimize the available resources across all GPUs to balance PhysX computation and graphics rendering. Therefore users can expect much higher frame rates and a better overall experience with SLI.

Using a gtx 690 with physx is using both GPUs for graphics, and part of one for physx. I don't have an additional card to run physx, a person would assume with the nvidia advertising for physx that a gtx 690 would suffice wonderfully, but I am merely pointing out that isn't the case. A gtx 690 is brought to it's knees in BL2 with everything on high.

try all configurations before talking about what is doing what...that is my point.

Is there something you want me to try? I tried using the graphics power from both with physx being strictly on one gpu. I can try dedicate one gpu to physx but that's just like any othere 680 + 2nd card for physx setup and I'm sure someone who has such a setup will chirp in if they are getting much better fps. I'm not sure where this is going but I'm happy to try something out if it will clarify anything.

 

[3DVagabond]

You got nothing, hence the false claim "I don't need too..."
You raised the "question" so let me inform you:

There has never been a benchmark showing any diffrences worth talking about on a NVIDIA GPU performance when running OpenCL/DirectCompute compared to when running CUDA.

It always end up like this:

http://blog.accelereyes.com/blog/2010/05/10/nvidia-fermi-cuda-and-opencl/


Now let it die.

I'm not pretending to be a programmer. Thus why I said I wonder if it would run better using OpenCL/D3D. i do understand English though.

The results indicate that there is an overhead when using OpenCL with smaller data sizes, which seems to disappear at larger data sizes. Currently it is unknown whether the overhead is due to the time taken to launch a kernel in OpenCL or something else within the API.

The benches in your link show OpenCL running faster as data size increases and eventually being faster than CUDA. The quote above gives a possible reason.

This is our report of the current status of OpenCL relative to CUDA on the new NVIDIA hardware.

Also these benches were run on nVidia hardware with nVidia's OpenCL API, which was new at the time. Your link is over 2 years old.

Before we present the benchmarks, we should comment on the programmability of OpenCL versus CUDA. OpenCL is notably more difficult to program and debug than CUDA since OpenCL documentation, tools, and scientific computation libraries are still very limited. Considering these handicaps, only a few matrix / vector operations were considered for this benchmark. All the vector operations are modified versions of the SDK examples provided by NVIDIA. All the tests were for single precision numbers.

If you read this^ you'll understand that they are explicitly noting that these tests are limited. Therefore they don't prove your claim that CUDA always runs faster. Note that I made no claim. I just wondered if OpenCL or D3D might be more efficient than PhysX. You were the only one who made any claims and these benches in fact disprove your claim that CUDA is always faster.

Considering there's evidence that GCN executes OpenCL and D3D far faster than Kepler, Fermi, or previous AMD architectures, what would those benches look like on current AMD hardware (Assuming AMD has an API to run these tests with.)? Far faster, most likely. There is a reasonable possibility that GCN could run the same physics calculations that are being run with PhysX faster using D3D or OpenCL. So, I ask again, "I wonder if it would?" That's all. No claim by me that it would indeed be faster.

If I ask a question and you know the answer, feel free to respond. Other than that why don't you at least try talking reasonable to people and let the attitude be what dies? :whiste:

 

[toyota]

Only the physx high part is demanding. If you put it on low or turn it off the game is ridiculously fast with framerates well past 100 - Physx has always incurred a performance drop. Not sure why it surprises anyone. You can still run it with physx high but it can slow down during boss fights or heavy combat...just turn it down a notch...

there is no off. its low, medium or high. putting it on low is turning off the hardware effects though.

 

[Lonbjerg]

From your quote:


Using a gtx 690 with physx is using both GPUs for graphics, and part of one for physx. I don't have an additional card to run physx, a person would assume with the nvidia advertising for physx that a gtx 690 would suffice wonderfully, but I am merely pointing out that isn't the case. A gtx 690 is brought to it's knees in BL2 with everything on high.


Is there something you want me to try? I tried using the graphics power from both with physx being strictly on one gpu. I can try dedicate one gpu to physx but that's just like any othere 680 + 2nd card for physx setup and I'm sure someone who has such a setup will chirp in if they are getting much better fps. I'm not sure where this is going but I'm happy to try something out if it will clarify anything.

Go read my post again...it's really simple..you havn't tried out all settings ffsand stop using only the parts of my posts you think suuport your stance...and ignore the parts that dont:

Single card
Single NVIDIA card, performing both graphics rendering and PhysX calculations.

2-Way SLI
Two NVIDIA cards, performing SLI graphics rendering and PhysX calculations (on main card).

3-Way SLI
Three NVIDIA cards, performing SLI graphics rendering and PhysX calculations (on main card).

Dedicated PhysX card
One NVIDIA card, performing only graphics and additional card, dedicated for PhysX calculations.

2-Way SLI + dedicated PhysX card
Two NVIDIA cards, performing SLI graphics rendering and additional card, dedicated for PhysX calculations.

3-Way SLI + dedicated PhysX card
Three NVIDIA cards, performing SLI graphics rendering and additional card, dedicated for PhysX calculations.

You have only tried 2 settings out of:
Single GPU, CPU physX
Single GPU, GPU PhysX
Dual GPU, CPU physX
Dual GPU, dedicated GPU PhysX
Dual GPU, split Physx/Graphics GPU.(Auto setting)

You have to few data point to conclude anything....get it now?

 

[Lonbjerg]

there is no off. its low, medium or high. putting it on low is turning off the hardware effects though.

Like always people opposed to hardware physics don't have all the details....not surprising 😉

 

[Keysplayr]

From your quote:


Using a gtx 690 with physx is using both GPUs for graphics, and part of one for physx. I don't have an additional card to run physx, a person would assume with the nvidia advertising for physx that a gtx 690 would suffice wonderfully, but I am merely pointing out that isn't the case. A gtx 690 is brought to it's knees in BL2 with everything on high.


Is there something you want me to try? I tried using the graphics power from both with physx being strictly on one gpu. I can try dedicate one gpu to physx but that's just like any othere 680 + 2nd card for physx setup and I'm sure someone who has such a setup will chirp in if they are getting much better fps. I'm not sure where this is going but I'm happy to try something out if it will clarify anything.

How is this even conceivably possible when I'm kicking this games arse with a single GTX680 doing everything?
SLI is disabled and I choose GPU1 for PhysX and check "dedicated". eVGA monitoring tool shows 0% utilization for GPU2 and the temp stays put at idle. So I know that card is doing zip. GPU1 is chugging away.

I am owning this game with every single setting on it's highest/farthest/PhsX High/1920x1200. No matter how intense firefights get or how many characters on screen. I'm trying to understand what possible issue you could be having. Your CPU is plenty fast enough. I can only think that something is amiss with your system somehow.

 

[railven]

Perhaps there is a scaling issue for SLI users or Wand3r is using an outside configuration to push his system further?

On my end, I can vouch for Keys. My GF's rig isn't as elaborate as other's and the lowest we've seen her system go to was 47 FPS. It's possible it went lower, but she hasn't mentioned it.

Hybrid PhysX isn't a cure all either, there are instances where my system has sporadoic stops as if something is loading. It would make the game unplayable, however, the GTX 680 system experiences no issues or slow downs as severe.

Either way, the PhysX effects are a great addition in this game. Sure you don't need them, but after playing the game with them I wouldn't want to play without them. Kudos nVidia on adding something of value, for once 😉

 

[Lonbjerg]

:whiste:

Perhaps there is a scaling issue for SLI users or Wand3r is using an outside configuration to push his system further?

On my end, I can vouch for Keys. My GF's rig isn't as elaborate as other's and the lowest we've seen her system go to was 47 FPS. It's possible it went lower, but she hasn't mentioned it.

Hybrid PhysX isn't a cure all either, there are instances where my system has sporadoic stops as if something is loading. It would make the game unplayable, however, the GTX 680 system experiences no issues or slow downs as severe.

Either way, the PhysX effects are a great addition in this game. Sure you don't need them, but after playing the game with them I wouldn't want to play without them. Kudos nVidia on adding something of value, for once 😉

PhysX has had value since day 1.
But first it was NVIDIA and ATi fans disliking it...because it didn't run on their hardware.
The NVIDIA aqquired AGEIA.
Then it was AMD fans disliking PhysX...and NVIDIA fans suddenly liking it.

I have liked it since day 1, got a PPU in 2006 and have enjoyed hardware PhysX ever since...a classic example of "the grapes are sour" if I have ever seen one

 

[Ferzerp]

Perhaps there is a scaling issue for SLI users

Works perfectly fine for me. No slowdowns at all. Certainly not like he's reporting.

 

[Lonbjerg]

Works perfectly fine for me. No slowdowns at all. Certainly not like he's reporting.

He should try running a DCP checker...to make sure it's not his system that is borked...before screaming up on forums with flawed results.
http://www.thesycon.de/deu/latency_check.shtml