More Crysis 3 CPU benchmarks

[c22]

And if you play at higher resolutions, CPUs still bottleneck GPU performance so its still a fair method of analysing CPU performance.

Testing at these low resolutions is stupid, as are the reasons that people want me to believe as to why it is a valid way of testing CPU performance.

No one puts a GTX 680 and a current 4-8 core CPU together only to play games at 1024x768. Show some real results from at least 1080p.

So we're clear on that one. PCLab alone done a test in 3 places, 2 resolutions and PCGH made test in ONE location, ONE resolution which is still useless. Now tell me more about "PCGH being no. 1" Ridiculous

EOT

[El Mano]

More threads = AMD is better
One main thread that do most work = Intel is better

FX-8350 can handle up to eight threads (kind of).
Intel?s 3770K can handle eight threads.
Care to explain why is AMD better when it comes to handle several threads?

It's all about the coding. Crytek just coded the game for AMD's CPUs.
As stated before, a patch for Intel CPUs might well turn the graphs.

Altough it would be cool if AMD was better, just for the sake of competence.

[Hornet331]

And if you play at higher resolutions, CPUs still bottleneck GPU performance so its still a fair method of analysing CPU performance.

Testing at these low resolutions is stupid, as are the reasons that people want me to believe as to why it is a valid way of testing CPU performance.

No one puts a GTX 680 and a current 4-8 core CPU together only to play games at 1024x768. Show some real results from at least 1080p.

Man... do we have to go over this in every god damn benchmark thread.

Both low and high resolution test have a place, as said a low resolution test is a pure cpu benchmark test and has its right to exists. So lets see what happens when a new gpu comes out, one is already close to the limit the other can still provide more and put the new gpu to better use.

[Mungri]

as said a low resolution test is a pure cpu benchmark test

I disagree. Its really not.

If you want to purely test CPUs, then you use synthetic benchmarks not games. If you want to test game performance, then you run settings that people would actually run.

[Darakian]

I disagree. Its really not.

If you want to purely test CPUs, then you use synthetic benchmarks not games. If you want to test game performance, then you run settings that people would actually run.

I believe he meant that it was pure as far as the game in concerned. If you reduce the graphical load you ensure that the GPU will not be slowing your test.

[Mungri]

As far as games are concerned, 1024x768 testing is meaningless as it doesnt show real world performance that gamers will be getting at 1080p and up.

And anyone thats running lower than 1080p isnt going to have specs like those in the test rig.

In this one case for Crysis 3, I dont think that any of the results are valid because it seems like another case of early release performance bugs and poor optimization, and the amount of difference between the tests is just too random to pinpoint a single reason.

It's all about the coding. Crytek just coded the game for AMD's CPUs.
As stated before, a patch for Intel CPUs might well turn the graphs.

Exactly, there doesnt seem to be as much optimization for Hyper Threading. The results are not conclusive of which CPU is better, only which CPU certain parts of the game are currently better optimized for.

[gosh]

FX-8350 can handle up to eight threads (kind of).
Intel?s 3770K can handle eight threads.
Care to explain why is AMD better when it comes to handle several threads?

3770K is better for threads compared to 3570K and it is more expensive. The cache is also bigger compared to 3570K but not as large as FX-8350. Cache is important if games are using a lot of memory

[Darakian]

As far as games are concerned, 1024x768 testing is meaningless as it doesnt show real world performance that gamers will be getting at 1080p and up.

And anyone thats running lower than 1080p isnt going to have specs like those in the test rig.

In this one case for Crysis 3, I dont think that any of the results are valid because it seems like another case of early release performance bugs and poor optimization, and the amount of difference between the tests is just too random to pinpoint a single reason.



Exactly, there doesnt seem to be as much optimization for Hyper Threading. The results are not conclusive of which CPU is better, only which CPU certain parts of the game are currently better optimized for.

Testing at 1024x768 is not supposed to show real world performance, it shows variable limited performance. It's showing the effect of the CPU alone (as best they can) on frame rates. Looking at the results it just seems like crysis 3 loves it some threads.

[R101]

Come on guys, we've been over this 123314 times before. If your CPU has 100 fps in lower resolution it might have similar fps to that if you GFX is strong enough. On the other hand, if it can only do 60 fps in lower res, what do you think it will have in higher res..

If you are using GF8400GS you might as well have Celeron.. As some guy called Keiichi Tsuchiya said - it's all about the ballance .

[Hornet331]

Testing at 1024x768 is not supposed to show real world performance, it shows variable limited performance. It's showing the effect of the CPU alone (as best they can) on frame rates. Looking at the results it just seems like crysis 3 loves it some threads.

Come on guys, we've been over this 123314 times before. If your CPU has 100 fps in lower resolution it might have similar fps to that if you GFX is strong enough. On the other hand, if it can only do 60 fps in lower res, what do you think it will have in higher res..

If you are using GF8400GS you might as well have Celeron.. As some guy called Keiichi Tsuchiya said - it's all about the ballance .

well at least some people get it, before whining "b-b-but its not real world"

[Mungri]

Even if you're right, as many people have asked - WHERE are the 1080p+ CPU comparison results so they can at least be compared to the low resolution results?

Without the comparison, these tests are meaningless and serve zero purpose to people playing at higher resolutions. I want to see how different CPUs compare at higher resolutions when they potentially start bottlenecking the GPU, not at lame low resolutions where the GPU isn't even being stressed.

It makes a HUGE difference when your GPU is high end and potentially being bottlenecked by weak CPUs at high resolutions.

[kromosto]

FX-8350 can handle up to eight threads (kind of).
Intel?s 3770K can handle eight threads.
Care to explain why is AMD better when it comes to handle several threads?

It's all about the coding. Crytek just coded the game for AMD's CPUs.
As stated before, a patch for Intel CPUs might well turn the graphs.

Altough it would be cool if AMD was better, just for the sake of competence.

actually not just coding. having native cores is always better then having virtual cores. intel implementing ht tech really well but as the cores are not physical ht has it's flaws (also advantages). also it is not a rare situation that disabling ht increases the gaming performance.

[STEvil]

Even if you're right, as many people have asked - WHERE are the 1080p+ CPU comparison results so they can at least be compared to the low resolution results?

Without the comparison, these tests are meaningless and serve zero purpose to people playing at higher resolutions. I want to see how different CPUs compare at higher resolutions when they potentially start bottlenecking the GPU, not at lame low resolutions where the GPU isn't even being stressed.

It makes a HUGE difference when your GPU is high end and potentially being bottlenecked by weak CPUs at high resolutions.

CPU scaling benchmarks with relation to resolution and cpu importance on GPU bottlenecking? Been done, gets ugly.

[boxleitnerb]

Tell me something I don't know.




Of course C3 is GPU limited. That's why pclab made a test in 3 different places using 2 different resolutions.

PCGH is showing us only 1/3 of a truth, but it doesn't matter cause "PCGH is the no. 1 in CPU benchmarks.", right? Time to look around and verify some vortals, because they're apparently not even close to being to "no. 1". Far from that.

You trolling? Didn't it enter your mind that due to the multithreaded nature of Cryengine 3 AMD could gain on Intel under some circumstances? I've read comments of a programmer working at Crytek regarding this. Both results are perfectly fine, theirs and yours. Again: different level, possibly different outcome. While I agree that they could have chosen more levels to bench, they have a long standing expertise in selecting demanding and yet representative benchmark scenarios for their CPU tests.

And you have yet to change your antiquated resolution for benchmarking. No one plays at 5:4 or 4:3 anymore.

As far as games are concerned, 1024x768 testing is meaningless as it doesnt show real world performance that gamers will be getting at 1080p and up.

No, they are not meaningless. What counts is framerate. 1080p and up would decrease performance further unless you are CPU bottlenecked. What purpose do 1080p benchmarks serve if the fps in those settings are too low for comfort? You can and should always lower resolution or settings to achieve the fps you want/need. What else are these options for? People always forget that because reviews mostly benchmark with maximum graphic settings.

Just look at the quoted benchmarks above. The FX4300 gets 22fps. You can cram 4 graphics cards in there, overclock them to kingdom come, but in this scene you will never ever get more than those 22fps. At 1080p the 3570K and FX4300 would likely seem on par at 20fps or so, but in reality they are not. The 3570K can push at least 30fps, the FX4300 only 20fps. You would not have that information if those benchmarks were conducted at 1080p or above. And more information is always better.

So in conclusion:
Low res benchmarks (16:9 or 16:10) can provide valuable information and are perfectly fine and in tune with "real gameplay" as long as the gamer is smart enough to adjust game settings to suit his/her needs.

[Mungri]

I only just noticed that there were some 1080p benchmarks right there.

Its a very similar case to BF3, the game hits GPU limitation very quickly, its hardly CPU limited at all.

I cant believe the claims I'm hearing that this low resolution testing is a valid method of testing CPU limitation in a graphics hungry game running with a GTX 680. If you want to test CPUs then do it in an application that doesn't rely so heavily on the GPU. Any game is going to be biased based on which hardware it is coded to run best on. And its very obvious, at 1080p neither Intel nor AMD are winning. Both their CPUs are too weak for such a graphics limited game.

And no, you should NEVER lower the resolution below a current monitors native resolution. The amount of blurring an loss of sharpness and IQ is never worth moving away from your monitors native resolution. Reduce AA first, then graphics settings. The only exception being if you are playing very old games that don't support, or look too uncomfortable at high resolutions.

[boxleitnerb]

I only just noticed that there were some 1080p benchmarks right there.

Its a very similar case to BF3, the game hits GPU limitation very quickly, its hardly CPU limited at all.

I cant believe the claims I'm hearing that this low resolution testing is a valid method of testing CPU limitation in a graphics hungry game running with a GTX 680. If you want to test CPUs then do it in an application that doesn't rely so heavily on the GPU. Any game is going to be biased based on which hardware it is coded to run best on. And its very obvious, at 1080p neither Intel nor AMD are winning. Both their CPUs are too weak for such a graphics limited game.

It doesn't matter if a game is graphics hungry or not. You act like ingame graphics options don't exist. Look at these GPU benchmarks:


The 680 gets 38fps avg and 32fps min. Would you want to play at such low fps? I certainly wouldn't. I would turn down settings to achieve at least 50fps avg, better yet 50fps min, and exactly then will the difference between different CPUs become more apparent. It seems you haven't read my explanation at all. It's about fps, not bottlenecks. fps...if CPU A can do only 40fps, but CPU B can do 60, that is valuable information right there, no matter if at default settings you are GPU bottlenecked or not. Because settings aren't set in stone, they are changeable to cater to your demands. That is why they exist.
Only when all CPUs can achieve highly playable fps, let's say 100+, differences become irrelevant.

[Krautmaster]

the problem of Crysis 3 seems to be that one thread is continually on max - no matter how much clock or IPC on this thread. This means on every Intel / AMD CPU one Thread is durably not available for graphics in general. You can see in benchmarks that clock doesn't matter that much, means if there could be a FPS limit regarding to that one cpu limited thread, it should scale differently...
Of course, when a dualcore like the i3 is pushed to max on one core (what ever causes this 100% single thread) - there is only one core left to handle the other stuff which is shared on other threads quite well.

See here:





EIST off 6x 4,32 Ghz


Settings:


Graphics + FPS:




By the way. U need to disable vsync else the load on gpu is less due fix on 30 FPS.

[Mungri]

It doesn't matter if a game is graphics hungry or not. You act like ingame graphics options don't exist.

Erm WHAT?

And no, you should NEVER lower the resolution below a current monitors native resolution. The amount of blurring an loss of sharpness and IQ is never worth moving away from your monitors native resolution. Reduce AA first, then graphics settings. The only exception being if you are playing very old games that don't support, or look too uncomfortable at high resolutions.

I said that you should reduce settings to maintain comfortable smoothness, but without changing from your monitors native resolution.

if CPU A can do only 40fps, but CPU B can do 60, that is valuable information right there, no matter if at default settings you are GPU bottlenecked or not.

I agree with you, but thats not the case here at all. The difference in the results at 1080p with the latest CPUs is 33 FPS vs 31 FPS, theres virtually no clear cut winner between Intel or AMD CPUs. Overclocking the CPUs by 1 Ghz hardly makes any difference either.

Also in such a CPU comparison, why were they afraid to also add in an Intel hex core CPU?

[boxleitnerb]

You don't seem to understand
33 or 31fps, for many people both are unplayable. If you lower graphics quality so that you can achieve more fps, the difference becomes 34 vs 48 in the "Post-Human" level and 35 vs 48 in the "The roof of all evil" level (FX8350 vs 3770K). Quite seizable, isn't it. It doesn't really matter if you get these results by lowering resolution or by lowering graphics settings - the end result is that with reduced graphics workload, the Intel systems are much faster and provide a better gaming experience than the AMD systems at actually playable framerates.

How often do I have to say it?
1080p doesn't or above matter if the fps achieved at this resolution are too low.

[Mungri]

Oh right, I get it now... So we need to see 1080p with reduced graphics settings as well then.

[Manicdan]

@bhavv

there is no 1 perfect cpu gaming test, we probably agree on that.
the point of low res testing it to show the max fps you will get for each cpu.
but that is still limited by the settings. so they should do a few, like max physics and effects, medium, and lowest settings.
some of us have super powerful cpus and want to see what speed is needed to get 60fps at max settings for the cpu AND gpu side, so they have to split that.
others have limited resources and should see what settings are needed to enjoy the game on their cpu. there is no way i would be playing crysis under 30fps, so i would sacrifice settings if its cpu limited, or resolution and AA if its gpu limited.

its hard to test "real world" because no configuration is the same. for me i like to have 1920x1200 and will give up view distance and AA to keep it. someone else might prefer 1280x800 and keep 4xAA on, and each has their benefits. typically though, cpu speed needed is not affected heavily by resolution.

make senes now?

[gx-x]

@c22

Would you mind telling me why PCL used very high setting as well? Aren't CPU test supposed to be done with low res + low settings? VH in C3 is very demanding of GPU even on 10x10 pixel resolution :d You are still very much GPU bound in those tests. I play C3 with i3 2100 @1080p and have fps limited to 60 and use low/high combo settings. Very rarely does fps go under 40, or even 55 tbh (except for the first levels due to that rope bug). In pcgh and pcl tests anything bellow i5 cpu is pointless at VH settings unless the goal was to just show which processors cannot handle VH at all o.O

[c22]

@gx-x

But do you play at low res + low settings? Nope, so this test is pretty much pointless, but we do that anyway, to show the difference IF title is bottlenecked by GPU at 1920x1080.

We're testing mostly in full HD, because it's actually how it should be done. It's the simple rule: you're not playing at 1024x768 nor 1280x720. You're playing at 1680x1050 and higher. So you need to check CPU performance at 1920x1080 anyway. You can almost always find a place in the game that's CPU dependant, even at 1920x1080 max settings - it's just a matter of time and.. knowledge.

Look at there: http://pclab.pl/art50000-14.html

Pages from 14 to 28 - CPUs at stock settings.
Pages from 51 to 71 - OCed settings.

Battlefield 3, Max Payne 3, Metro 2033. They all say those titles are GPU dependant... Sure they're unless you know how to find a proper place, where CPU testing has actual sense. And most of those review sites out there simply CAN'T find the place, and on the other hand: we CAN

Another example: Far Cry 3 being GPU bottlenecked? That's funny, because it's not. It's CPU limited.




It's a matter of finding a good place - just saying.

[gx-x]

@gx-x

But do you play at low res + low settings? Nope, so this test is pretty much pointless, but we do that anyway, to show the difference IF title is bottlenecked by GPU at 1920x1080.

Maybe I wasn't clear enough: when testing/comparing CPUs, why very high details? I also don't play in 720p very high settings, I'd rather go native res and lower details, BUT, when comparing CPU capabilities in a game shouldn't you just eliminate some shoddy codding that often happens in very high settings (they don't care to optimize for that kind of setting) and just pair lower res with some lower or mid settings? I mean, who goes for gtx680 with Athlon x4 or i3 540 and uses lower res with ultra high settings? Remember, we are testing CPUs, not GPU. I am Ok with 1080p tests but you should have included lower details so the lower end lot in the test actually shows some "numbers" that can be compared. From your CPU test in C3 one would conclude that game is not playable on something like i3...which is far from truth. I didn't pair my i3 with TITAN, but with 560Ti 2gb...and I play it with 50-60fps (locked at 60) with low/high combo.

[boxleitnerb]

Games can be CPU and GPU bottlenecked, depending on the scene/level. If you disable MSAA in FC3 (a low setting ), of course you will be more CPU bottlenecked.

But do you play at low res + low settings? Nope, so this test is pretty much pointless, but we do that anyway, to show the difference IF title is bottlenecked by GPU at 1920x1080.

Wrong. I thought we had that covered already. But have fun with your low fps...