Crysis 3 CPU performance test

[SPBHM]

Originally Posted by guskline

Why post findings with a 1024x768 resolution?

making the game less GPU limited helps to show the CPU performance difference...

you can see the 1080p tests here http://pclab.pl/art52489-9.html
it's just to easy to hit the GPU limit with a single 680...
multiple GPUs could make things more interesting.

[guskline]

Originally Posted by UaVaj

simply so gpu is not the limiting factor.

this is a cpu thread.

OK but who plays Crysis 3 at those resolutions?

I'm posting real numbers with a GTX670 FTW at either 1920x1200 on the FX8350 or 2560x1440 on the 2500k.

Admittedly, this is a very demanding game and I decided to use High settings and Antialiasing with FXAA so consider that when looking at my numbers. My point on the lower resolution is that most gamers are NOT at that resolution so using those graphs is not very useful. (Perhaps they used such a low resolution for the slower cpus)

[csbin]

1080P VH

[guskline]

Originally Posted by SPBHM

making the game less GPU limited helps to show the CPU performance difference...

you can see the 1080p tests here http://pclab.pl/art52489-9.html
it's just to easy to hit the GPU limit with a single 680...
multiple GPUs could make things more interesting.

Thank you for the link.

[Borealis7]

That's a GPU bottleneck right there in all 3 graphs. not very helpful in a CPU discussion.

[SPBHM]

Originally Posted by guskline

OK but who plays Crysis 3 at those resolutions?

I'm posting real numbers with a GTX670 FTW at either 1920x1200 on the FX8350 or 2560x1440 on the 2500k.

Admittedly, this is a very demanding game and I decided to use High settings and Antialiasing with FXAA so consider that when looking at my numbers. My point on the lower resolution is that most gamers are NOT at that resolution so using those graphs is not very useful. (Perhaps they used such a low resolution for the slower cpus)

they also included 1080p testing, they did the right thing I suppose, included both.

the test don't need to be an exact representation of how you are going to play the game, I don't use Cinema4D, but I'm still interested in seeing how CPUs perform on cinebench, it can still give me a clue about performance in general and other applications

as you show, people can lower settings, or use a faster GPU, so the lower res testing is useful to have a better idea about the CPU performance limits with less GPU limitation... the fact that you can see one CPU at 40 while the other is at 30 is pretty significant, when the GPU limit could hide this difference and show both at 25, even if lowering a few things, or using multiple GPUs you could probably hit again those 40 with one CPU at a higher res.

[Ferzerp]

Originally Posted by Borealis7

That's a GPU bottleneck right there in all 3 graphs. not very helpful in a CPU discussion.

It's as helpful as benchmarking CPUs based on throughput over a 100Mbit ethernet link.


For those who still don't get it, increasing the resolution doesn't stress the CPU more. In most games, a CPU is capable of putting out almost as many fps at 1080p as they are at 1024x768. The problem is, the gpu isn't, and when you up the resolution, you end up just measuring your GPU. If you're trying to compare processors, it ends up telling you absolutely nothing. Exactly like my benching cpus via ethernet throughput example.

[guskline]

Originally Posted by SPBHM

they also included 1080p testing, they did the right thing I suppose, included both.

the test don't need to be an exact representation of how you are going to play the game, I don't use Cinema4D, but I'm still interested in seeing how CPUs perform on cinebench, it can still give me a clue about performance in general and other applications

as you show, people can lower settings, or use a faster GPU, so the lower res testing is useful to have a better idea about the CPU performance limits with less GPU limitation... the fact that you can see one CPU at 40 while the other is at 30 is pretty significant, when the GPU limit could hide this difference and show both at 25, even if lowering a few things, or using multiple GPUs you could probably hit again those 40 with one CPU at a higher res.

I did some shifting around of video cards and monitors. Put the PNY GTX680 together with the Achieva Shimian 27" 2560x1440 and the FX8350. Moved the EVGA GTX670 FTW to the 3 monitor setup. Didn't lose as much as I thought I would. Picked up a few fps (30 to 36) on the FX8350 setup and Crysis3.

[AtenRa]

Originally Posted by csbin

1080P VH

Why i cant find those slides in the review ??? They only have 1024x768 slides for the CPU scaling.

http://pclab.pl/art52489-9.html

Edit: heh never mind, i found them

[AtenRa]

For those wondering,

The settings for the 1920x1080 slides in CPU scaling is at Very High with MSAA OFF and 16x aniso.

[crashtech]

I don't know if this would be of any relevance or interest, but I did a little informal examination of CPU utilization in Crysis 3. I don't have enough knowledge of game benchmarking to publish findings, but what I did was to run the game on my dual Xeon X5472 (GTX 460) for a while, first with both CPUs installed, then with just one.

Having the Performance tab open on Task Manager during this exercise allowed me to get a rough idea of CPU utilization. With one CPU installed, the graphs showed very heavy utilization of all 4 cores, not quite saturated but close. With 8 cores in play, the load was distributed over all 8. Although the distribution did not appear even, it was obvious that even the cores that were loaded the heaviest were much farther away from 100% than with only one CPU.

I'd like to gain more knowledge on in-game benchmarking, get to the right save point(s) in the game that are the hardest on the CPU, and do some more checking. It doesn't seem to me like there was a much of a performance benefit going from 4 to 8 cores in the part of the game I tested, which was very near the beginning. I haven't had time to play lately. So far all I can say is that Crysis 3 can definitely load down 4 cores at the very least.

[inf64]

Originally Posted by guskline

I did some shifting around of video cards and monitors. Put the PNY GTX680 together with the Achieva Shimian 27" 2560x1440 and the FX8350. Moved the EVGA GTX670 FTW to the 3 monitor setup. Didn't lose as much as I thought I would. Picked up a few fps (30 to 36) on the FX8350 setup and Crysis3.

Pretty nice performance jump from 670 to 680,20% to be exact. Looks like 8350 is doing just fine with 680 monster

[UaVaj]

Originally Posted by guskline

I did some shifting around of video cards and monitors. Put the PNY GTX680 together with the Achieva Shimian 27" 2560x1440 and the FX8350. Moved the EVGA GTX670 FTW to the 3 monitor setup. Didn't lose as much as I thought I would. Picked up a few fps (30 to 36) on the FX8350 setup and Crysis3.

r u testing for gpu limitation or u testing cpu limitation??

at 1440p and cyrsis3 - you are obviously not testing cpu.

how about repeat that test at 720p?

[Yuriman]

Originally Posted by UaVaj

r u testing for gpu limitation or u testing cpu limitation??

at 1440p and cyrsis3 - you are obviously not testing cpu.

how about repeat that test at 720p?

Even at 2560x1440 you *can* be CPU limited. An i3 that can only put put 25fps will top out at 25fps whether you're running 1024x768 or 2560x1440. Having a slow CPU basically prevents you from improving your framerates past a certain point by lowering graphical settings.

[crashtech]

Originally Posted by Yuriman

Even at 2560x1440 you *can* be CPU limited. An i3 that can only put put 25fps will top out at 25fps whether you're running 1024x768 or 2560x1440. Having a slow CPU basically prevents you from improving your framerates past a certain point by lowering graphical settings.

That's an important distinction, imo, if just for the fact that many users, even if they won't upgrade their CPU or mobo, eventually will want to upgrade to a faster video card. If a given CPU can be shown to limit a game less than others at lower resolutions today, it will be less likely to limit game performance at higher resolutions on future graphics hardware.

For those who still think this concept is just an attempt to slight AMD, please read up on "scientific control."

[2is]

Originally Posted by crashtech

I don't know if this would be of any relevance or interest, but I did a little informal examination of CPU utilization in Crysis 3. I don't have enough knowledge of game benchmarking to publish findings, but what I did was to run the game on my dual Xeon X5472 (GTX 460) for a while, first with both CPUs installed, then with just one.

Having the Performance tab open on Task Manager during this exercise allowed me to get a rough idea of CPU utilization. With one CPU installed, the graphs showed very heavy utilization of all 4 cores, not quite saturated but close. With 8 cores in play, the load was distributed over all 8. Although the distribution did not appear even, it was obvious that even the cores that were loaded the heaviest were much farther away from 100% than with only one CPU.

I'd like to gain more knowledge on in-game benchmarking, get to the right save point(s) in the game that are the hardest on the CPU, and do some more checking. It doesn't seem to me like there was a much of a performance benefit going from 4 to 8 cores in the part of the game I tested, which was very near the beginning. I haven't had time to play lately. So far all I can say is that Crysis 3 can definitely load down 4 cores at the very least.

Task manager is not a good judge of utilization unfortunately, unless you know exactly what to look for. You can have 16 cores and it will look like all 16 are being utilized at the same time even if a game can only use 4 cores, when it reality, all that's happening is that you're still only using a maximum of 4 cores, but the windows task scheduler is switching affinity between the cores so fast that graphically in task manager, it looks like all cores are being utilized just a little bit simultaneously.

[guskline]

Originally Posted by UaVaj

r u testing for gpu limitation or u testing cpu limitation??

at 1440p and cyrsis3 - you are obviously not testing cpu.

how about repeat that test at 720p?

Not testing at all. Just playing the game!

[crashtech]

Originally Posted by 2is

Task manager is not a good judge of utilization unfortunately...

Yeah, I figure it's just playing around at this point. Definitely more CPU utilization than older games I have looked at, though. So what IS a good tool to monitor utilization?

[Deders]

Just re-installed my OS and found that my CPU usage is much higher, now reaching 100% at some points. The same with Arkham City. Am getting better framerates in Arkham City but the same in Crysis 3.

[SPBHM]

from Tom's Hardware

http://www.tomshardware.com/reviews/...ng,3451-8.html

[crashtech]

Gonna need to read the article, thanks! Looks like my informal results might have some grounding in reality. Might be able to keep my old 8-core rig going with just a graphics card update, just maybe.

[crashtech]

I wonder what accounts for the wide variance between min. and avg. fps on the Vishera?