*PROOF* that a 3930k bottlenecks SLI Titans.

[Cacao]

After getting completely eaten up on a topic I made on the subject a while ago (http://linustechtips.com/main/topic/40495-3930k-bottlenecking-2x-gtx-titans/) I thought I would show some actual benchmarks to backup my claim. So currently I have only benchmarked BF3, however I'll be doing more games soon.

 

Here are the results:

 

3930k at stock speeds, 1080p, Ultra, 64 players:

 

 

3930k at 4.6ghz, 1080p, Ultra, 64 players:

 

 

While running at stock speeds my GPU usage never went above 60%, staying mostly at around 40-50%. When I overclocked my GPU usage stayed between 60-70% and didn't really go below 50%. I tested both on the same server, with the same map and the same amount of players. I think I could get even better results with a higher overclock but unfortunately 4.6ghz is my limit. I'll show some more results as soon as I can get to testing other games, but for now have I won you over @TechSage @ @CreepingMoth @WoodenMarker ?

 

Once again let me stress, the ONLY thing that changed between these two tests were the CPU clock speeds. 

 

----------------------------------------------------------------------------------------------

 

Here are some more crazy results, I decided to test with SLI disabled (on the same server with same settings etc.) and here's what I got:

 

 

The EXACT same average FPS as with SLI and an even higher minimum and max. The GPU usage here stayed above 95%, so pretty much I get 0% SLI scaling (in BF3) if I don't overclock, how is that NOT a bottleneck?

[SSOB]

Well time to get a 4930K when it's released isn't it

[GrievingJosh]

After getting completely eaten up on a topic I made on the subject a while ago (http://linustechtips.com/main/topic/40495-3930k-bottlenecking-2x-gtx-titans/) I thought I would show some actual benchmarks to backup my claim. So currently I have only benchmarked BF3, however I'll be doing more games soon.

 

Here are the results:

 

3930k at stock speeds, 1080p, Ultra, 64 players:

 

 

3930k at 4.6ghz, 1080p, Ultra, 64 players:

 

 

While running at stock speeds my GPU usage never went above 60%, staying mostly at around 40-50%. When I overclocked my GPU usage stayed between 60-70% and didn't really go below 50%. I tested both on the same server, with the same map and the same amount of players. I think I could get even better results with a higher overclock but unfortunately 4.6ghz is my limit. I'll show some more results as soon as I can get to testing other games, but for now have I won you over @TechSage @ @CreepingMoth @WoodenMarker ?

 

Once again let me stress, the ONLY thing that changed between these two tests were the CPU clock speeds. 

Well, if you feel it's that massive of a problem, upgrade to an Ivy-E CPU when they're released. Though if I'm honest with those kind of FPS I don't see how you're going to gain that much benefit.

[Lamb]

Maybe dual titans are actually worth it?

[Cacao]

Maybe dual titans are actually worth it?

What do you mean?

[Lamb]

What do you mean?

 

I don't know. I have always heard people say that dual titans are not worth it because it is not much of a performance increase from 1 titan. Maybe it is a big difference in performance, we just need a better CPU to test it.

[Dravic]

I don't know. I have always heard people say that dual titans are not worth it because it is not much of a performance increase from 1 titan. Maybe it is a big difference in performance, we just need a better CPU to test it.

 

no heck sherlock. If you wired up an old good 580 or 480 to some CPU from 10 years forward only then you probably would max it out.

 

A card can work as fast as it has the data going through. If you have a beast CPU overpower versus the card the card is working at 100% load and that is very good for FPS even with worse cards, not to mention the newest and most powerful ones, because then the only thing GPU does is calculate.

 

Now with our current gen high-end GPUs we are bottlenecking them as long as they are not geting all the data out there ALL the time without any waiting and it aint possible even with an intel i7 as of now to supply a Titan so fast.

 

Yes you will see some small performance increases with any CPU performance gains overally but at some point it stops being worth it. I've seen a guy with really fast DDR3 memory + intel i7 OC-ed heavily combined with 9600GT maxing out BF3 at playable framerates (30+ FPS minimum and going average of a bit higher) rock solid.

[Overl0rd]

What was the amount of CPU usage during both tests?

[Cacao]

What was the amount of CPU usage during both tests?

Crap! I should've watched that, I feel kind of stupid for not looking at that. I'll redo the tests as soon as I can.

[Cacao]

I don't know. I have always heard people say that dual titans are not worth it because it is not much of a performance increase from 1 titan. Maybe it is a big difference in performance, we just need a better CPU to test it.

Check out single card result, it pretty much backs it up.

[Overl0rd]

Crap! I should've watched that, I feel kind of stupid for not looking at that. I'll redo the tests as soon as I can.

Okay, cause I think that there isn't really a bottleneck, but that Battlefield just benefits from a higher clockspeed. Plus the fact that it's optimized for quad core cpu's.

[Hunter259]

no heck sherlock. If you wired up an old good 580 or 480 to some CPU from 10 years forward only then you probably would max it out.

 

A card can work as fast as it has the data going through. If you have a beast CPU overpower versus the card the card is working at 100% load and that is very good for FPS even with worse cards, not to mention the newest and most powerful ones, because then the only thing GPU does is calculate.

 

Now with our current gen high-end GPUs we are bottlenecking them as long as they are not geting all the data out there ALL the time without any waiting and it aint possible even with an intel i7 as of now to supply a Titan so fast.

 

Yes you will see some small performance increases with any CPU performance gains overally but at some point it stops being worth it. I've seen a guy with really fast DDR3 memory + intel i7 OC-ed heavily combined with 9600GT maxing out BF3 at playable framerates (30+ FPS minimum and going average of a bit higher) rock solid.

Stop insulting people, leave that for another forum. Also a 9600gt cannot max out bf3 at 1080p, let alone 720p.

[Overl0rd]

Crap! I should've watched that, I feel kind of stupid for not looking at that. I'll redo the tests as soon as I can.

Also, is it maybe possible to monitor all the 12 cores (task manager)? I'd be very interested in the results! Thanks in advance

[Dravic]

Stop insulting people, leave that for another forum. Also a 9600gt cannot max out bf3 at 1080p, let alone 720p.

 

What. Insult? Where?

[mrthuvi]

If for gaming, 4770k is better than 3930k.

[Cacao]

If for gaming, 4770k is better than 3930k.

Ok.... Do you have some benchmarks or something to show?

[mrthuvi]

Ok.... Do you have some benchmarks or something to show?

http://www.youtube.com/watch?v=TkoK8UmJnpU

And 4770 is a little better than 3770.

[Guest D1G1T4L1Z3R]

My 3570K bottlenecks two Reference Design Nvidia GeForce GTX670 (made by MSI) and factory overclocked to 1110MHz - 1124MHz

 

So I decided to overclock it to 4500,00MHz @1250mV and gulf of oman and noshahr canals runs smoother in 64 players and in ultra in 1080p (when you wait to respawn, the blue screen) 

 

not 54-57fps, but 68-72fps now ===> in blue screen "lobby"

[Jackbe]

That seems like an extremely slow frame rate for dual sli titans on battlefield at 1080p. Something very strange is going on there.

[Guest D1G1T4L1Z3R]

That seems like an extremely slow frame rate for dual sli titans on battlefield at 1080p. Something very strange is going on there.

 

This is totally normal because this game is even not optimised for TITAN and all GTX6XX series and GTX7XX series

DICE and EA need the money, not optimizing anything

[Jackbe]

This is totally normal because this game is even not optimised for TITAN and all GTX6XX series and GTX7XX series

DICE and EA need the money, not optimizing anything

Battlefield 3 is pretty well optimised. I get 70-80 FPS at 1080p on ultra with an overclocked GTX 670. If I had sli 670s, I would be getting around 120 FPS which is close to the OP's results and he has 2 titans! Other people have had much better results than the OP with similar specs, I don't think this is an optimisation or a bottleneck problem. I think there is something wrong with the OP's system.

[Cacao]

Battlefield 3 is pretty well optimised. I get 70-80 FPS at 1080p on ultra with an overclocked GTX 670. If I had sli 670s, I would be getting around 120 FPS which is close to the OP's results and he has 2 titans! Other people have had much better results than the OP with similar specs, I don't think this is an optimisation or a bottleneck problem. I think there is something wrong with the OP's system.

As you can see without SLI my system performs as it should, when I add the second card I get almost no scaling and my GPU usage stays under 60%, that can only lead me to believe that there's a bottleneck, especially because overclocking betters the situation. So either we don't have powerful enough CPUs to run multiple titans, or our games are terribly optimized for multiple titans.

[Jackbe]

As you can see without SLI my system performs as it should, when I add the second card I get almost no scaling and my GPU usage stays under 60%, that can only lead me to believe that there's a bottleneck, especially because overclocking betters the situation. So either we don't have powerful enough CPUs to run multiple titans, or our games are terribly optimized for multiple titans.

There may be poor optimisation for titans because they are not a common card. I saw someone on the post on the topic you linked in your first post who said he got much better performance than you and he had similar specs. Something is not quite right, whatever the issue is.

[brownninja97]

Everything in the world bottlenecks, its as simple as that

[Cacao]

There may be poor optimisation for titans because they are not a common card. I saw someone on the post on the topic you linked in your first post who said he got much better performance than you and he had similar specs. Something is not quite right, whatever the issue is.

Well honestly anyone could say 'similar performance', I wish I could see some actual test results from people with the same setup as me.