I experience half the FPS of one card with a crossfire R9 290 setup.

[purecain]

make sure you've selected high performance in power options, the cpu could do with a small bump in clock speed. im sure that cpu will do 4ghz on stock volts... the fact your cpu was maxed out gives you the answer so overclock it.

if you have problems overclocking and getting your system stable at 4ghz then I would say that your psu would be the next port of call... I don't think you will have a problem. go into bios and set multi at x40 and let us know how you get on...

ps.my 2700k ran 4.0ghz at stock iirc and it never went above 75% usage in most games... don't worry about pcie3. you are not losing performance because of that. and 290x is the top card, 290 has less shaders.

good luck with the oc...

 

[GhostRyder]

you people are beating a dead horse
both battlefield 4 and crysis 3 scale like utter shit with crossfire and with the recent bf4 patch dice broke it again ..

No they are not I just played last night on 3 R9 290X with Mantle enabled..............At 4k no less and at 1080p...

Hi, Everyone.

Thanks for all of your replies.
I managed to get some money and I went ahead and bought the i7-3770k earlier today so it should arrive by Wednesday/Thursday. I should be able to tell you by then if it helps so that we are able to settle this.
I had been wanting a new CPU anyway so this kind of gave me an excuse to buy one, even if it turns out that my CPU isn't the cause of the problem.

I had not been ignoring any of your posts, if that's what any of you thought; I just merely wasn't aware that there were any new ones. After page 2, I received no more emails notifying me that someone had posted on my thread so I had assumed that there were no more messages. I came on today and to my delight, there were another 2 pages of replies which I am very grateful for. It's nice to know others are trying to help you.

For those of you whom were wondering what my CPU usage was(in-game*) whilst only using 1 GPU, then here it is:

Also, for those of you whom are talking about whether it's my PSU, then here is a snapshot of the figures whilst in-game*:

I have tried a couple of drivers (I did make sure the previous ones were uninstalled before installing new ones), and they didn't really help. 14.7 just seemed to be absolutely disastrous in terms of multi-gpu gaming, so I stuck with 14.6 which seems to be stable.

In terms of windows updates, I have all the necessary ones installed.
In terms of software, then I'm not sure.... anti-virus was mentioned so I turned it off and it didn't make a difference (as I expected), but I was just trying out what people had said : )
In terms of the airflow, I keep my side panel off even though it has a 1x120mm fan because the temperatures are quite a bit lower when it is like that. My tower is quite packed(picture shown on page 1) so there isn't much room for sufficient airflow, so I chose to take the side panel off.

*in-game refers to Crysis 3 Multiplayer

Cool, that should help out significantly especially if you overclock it a bit.

 

[OneMoar]

No they are not I just played last night on 3 R9 290X with Mantle enabled..............At 4k no less and at 1080p...

Cool, that should help out significantly especially if you overclock it a bit.

290X != 290
also you have 3 cards the op has TWO and yes the scaling is horribly broken( with a single monitor) test it your self
your system is a completely different beast then the ops you have triple monitors and triple cards the op has a single monitor and 290's
don't believe me disable one card and unplug all but your primary monitor and run at 1080 or 1440p

 

[Aquinus]

290X != 290
also you have 3 cards the op has TWO and yes the scaling is horribly broken( with a single monitor) test it your self
your system is a completely different beast then the ops you have triple monitors and triple cards the op has a single monitor and 290's
don't believe me disable one card and unplug all but your primary monitor and run at 1080 or 1440p

Well, if the game is CPU bound of course it's not going to scale with more GPUs, that's a given. Higher resolutions benefit from it because more time is spend rendering but in his case, the CPU is pegged out. There is no headroom for gaining more performance regardless of the GPU setup. His CPU was pegged out with just one GPU, forget two.

make sure you've selected high performance in power options, the cpu could do with a small bump in clock speed. im sure that cpu will do 4ghz on stock volts... the fact your cpu was maxed out gives you the answer so overclock it.

if you have problems overclocking and getting your system stable at 4ghz then I would say that your psu would be the next port of call... I don't think you will have a problem. go into bios and set multi at x40 and let us know how you get on...

ps.my 2700k ran 4.0ghz at stock iirc and it never went above 75% usage in most games... don't worry about pcie3. you are not losing performance because of that. and 290x is the top card, 290 has less shaders.

good luck with the oc...

He has a 2500 not a 2500k, the multiplier isn't unlocked as you suggest.

 

[GhostRyder]

290X != 290
also you have 3 cards the op has TWO and yes the scaling is horribly broken( with a single monitor) test it your self
your system is a completely different beast then the ops you have triple monitors and triple cards the op has a single monitor and 290's
don't believe me disable one card and unplug all but your primary monitor and run at 1080 or 1440p

I have already as I mentioned...I said I was gaming at 4k on 3 290X's and I had tested in a previous thread recently that 1080p and two cards works fine. Also 290X is not going to magically scale significantly better than its 290 counterpart...There is always room for error but its not like 290 does not scale and 290X does. I have tested it in multiplayer and I maintain beyond a shadow of a doubt 60FPS in multiplayer except when significant levolution happens (Like the building collapsing in Seige of Shanghai) on two cards. The third almost removes any changes in FPS at 60 completely minus still a random occurrence every now and then...

I use of course mantle API but even in DX11 it ran smooth and fine...CFX is not the issue here which has already been established...

 

[newconroer]

At what point did I say either setup i5 or i7 overclocked would bottleneck? I never stated that and said either perform exactly the same in most cases (with very rare exceptions).

The original comment was that the i5, whether overclocked or not, can NOT be a bottleneck (that's so severe) that would cause the type of issue(s) he is facing.

Dent apparently agrees with this :

But the important question is should he be getting bottlenecked?

And can a bottleneck explain 2 cards performing 50% worse than 1 card? Have you seen any previous examples of this?

And you responded to my comment about the relevance of it being an i5, with ..

2 or 3 years ago being the key word there. Were talking about recent games and much higher power GPU setups and resolutions. The Sandy-Bridge architecture is a few years old, but maintains good performance when overclocked very far as it is excellent for that. The problem here is the i5 2500, its base and boost clocks are not extremely high mixed with the fact that its on an old architecture (Which would mean about a 20% increase give or take on new chips clock to clock) and its clocks are pretty low. An i5 2500k clocked at 4.5ghz would still be beyond excellent for gaming and a single GPU on the base clocks would be fine, however multi-gpu setups require a bit more CPU power in this day and age and that chip is just showing a bit of age.

What ARE you saying then? If the i5 is good for a single GPU on base clocks(?), and a multi-gpu requires a bit more CPU power, are you then suggesting the i5 CPU is not sufficient for more than one GPU?


I will reiterate my previous comment(s). An i5, with no overclock should not significantly bottleneck 3d applications (universally speaking), on a mulit-GPU setup. If that was the case, New Egg never would have sold any. Furthermore, Sandy Bridge is not that old, that any current GPU architecture is either incompatible, less optimized or substantially too powerful that it causes such adverse results as those found in this scenario.

Correct me if wrong, and hopefully he figures it out, though it feels the fault lies elsewhere.

 

[GhostRyder]

The original comment was that the i5, whether overclocked or not, can NOT be a bottleneck (that's so severe) that would cause the type of issue(s) he is facing.

We have already established that with one GPU the i5 he has is hitting near 100% usage constant while gaming. Also multi-player has significant more requirements than single player does on games in general.

Dent apparently agrees with this :

People are agreeing the CPU is the bottleneck?

And you responded to my comment about the relevance of it being an i5, with ..
What ARE you saying then? If the i5 is good for a single GPU on base clocks(?), and a multi-gpu requires a bit more CPU power, are you then suggesting the i5 CPU is not sufficient for more than one GPU?

Newer i5 on newer architecture have yielded higher clock to clock in performance throughout the years. We are talking about an i5 2500 (Non unlocked) from a few years back which is suffering and causing a bottleneck. A newer i5 even at stock speeds would not yield as much of a problem but most people why they have high end GPU's or reviewers with high end (or any) gpu tend to have the CPU overclocked which is why we do not see a bottleneck (Especially when they test for multi-gpu setups).

I will reiterate my previous comment(s). An i5, with no overclock should not significantly bottleneck 3d applications (universally speaking), on a mulit-GPU setup. If that was the case, New Egg never would have sold any. Furthermore, Sandy Bridge is not that old, that any current GPU architecture is either incompatible, less optimized or substantially too powerful that it causes such adverse results as those found in this scenario.

Correct me if wrong, and hopefully he figures it out, though it feels the fault lies elsewhere.

Not everyone who buys a CPU is gaming...Locked CPU's except on budget builds normally do not yield such high end GPU's which is the problem. We have seen already the OP showed us while gaming a high yield of usage on the i5 in a single GPU, Dual GPU have high requirements in general and on top of that the computer still tries to split the load between them but hits a bottleneck because of the processor. Hence why framerate returns are not as high because multi-GPU setups are not perfect scaling (There is no such thing essentially as a perfect scaling). If it was a 2500k, we would all just say punch it to 4.5, but the OP can't and hence he bought an i7 3770k to replace it.

 

[OneMoar]

so once again we have the know-enough-to-be-dangerous crowd here on TPU costing somebody time and money
the new cpu isn't gonna fix the problem there is no reason that a i5 2500 would bottle neck to 50% try more like 15% assuming the absoulute worst FFS hes running at frigging 1080p NOT 4k infact with these new Hawaii cards the cpu overhead is LOWER then previous generations and the final nail in the coffin is that there is less then TWO-FPS difference between a 2500 and a 4670

http://www.ocaholic.ch/modules/smartsection/item.php?itemid=1129&page=6

 

[Frag_Maniac]

so once again we have the know-enough-to-be-dangerous crowd here on TPU costing somebody time and money
the new cpu isn't gonna fix the problem there is no reason that a i5 2500 would bottle neck to 50% try more like 15% assuming the absoulute worst FFS hes running at frigging 1080p NOT 4k infact with these new Hawaii cards the cpu overhead is LOWER then previous generations and the final nail in the coffin is that there is less then TWO-FPS difference between a 2500 and a 4670

http://www.ocaholic.ch/modules/smartsection/item.php?itemid=1129&page=6

There's quite a few differentiating factors in that test though. Older drivers, older GPU, single GPU vs Crossfire.

The minute differences you see in one bench can become significantly bigger in a bench that uses quite a different test setup.

 

[OneMoar]

I am aware but the on paper the new generation cards should have LESS cpu overhead due to xDMA witch if its buggered like I believe it is currently might be the cause of the cpu spikes infact I would't be shocked at all if it was some ODD memory management bug or something throwing a interrupt every cycle

 

[Frag_Maniac]

I am aware but the on paper the new generation cards should have LESS cpu overhead due to xDMA witch if its buggered like I believe it is currently might be the cause of the cpu spikes

Things don't always work out that way though. Lately AMD are cramming to improve Crossfire performance. Often times cramming means better performance for some, worse for others.

My first reaction was a 2500 should handle it, but then I know that the more powerful GPUs get, the more powerful a CPU it takes to Crossfire them, and even the non X version of the 290 is pretty powerful. That and consistent reports of the same problems with similar setups makes it kinda clear.

 

[Aquinus]

Things don't always work out that way though. Lately AMD are cramming to improve Crossfire performance. Often times cramming means better performance for some, worse for others.

My first reaction was a 2500 should handle it, but then I know that the more powerful GPUs get, the more powerful a CPU it takes to Crossfire them, and even the non X version of the 290 is pretty powerful. That and consistent reports of the same problems with similar setups makes it kinda clear.

Bingo. Me too, but 100% CPU load is proof to the contrary, not to say that can't be a driver issue or an issue with that patch on a particular game but I think we can say for this instance that it's probably CPU bound even if that may not be the case every time for every game or even between patches and drivers for that matter. As it stands right now, this screams CPU bottleneck.

For those of you whom were wondering what my CPU usage was(in-game*) whilst only using 1 GPU, then here it is:

Also, for those of you whom are talking about whether it's my PSU, then here is a snapshot of the figures whilst in-game*:

I have tried a couple of drivers (I did make sure the previous ones were uninstalled before installing new ones), and they didn't really help. 14.7 just seemed to be absolutely disastrous in terms of multi-gpu gaming, so I stuck with 14.6 which seems to be stable.

In terms of windows updates, I have all the necessary ones installed.
In terms of software, then I'm not sure.... anti-virus was mentioned so I turned it off and it didn't make a difference (as I expected), but I was just trying out what people had said : )
In terms of the airflow, I keep my side panel off even though it has a 1x120mm fan because the temperatures are quite a bit lower when it is like that. My tower is quite packed(picture shown on page 1) so there isn't much room for sufficient airflow, so I chose to take the side panel off.

*in-game refers to Crysis 3 Multiplayer

 

[Frag_Maniac]

Bingo. Me too, but 100% CPU load is proof to the contrary, not to say that can't be a driver issue or an issue with that patch on a particular game but I think we can say for this instance that it's probably CPU bound even if that may not be the case every time for every game or even between patches and drivers for that matter. As it stands right now, this screams CPU bottleneck.

It occurs to me I never checked the whole thread to see if you guys had Mantle running and verified it to be working, since it's supposed to use less CPU.

 

[JaredzzC]

Hi, Everyone.

I received my i7-3770k today and it runs like an absolute charm! (I haven't yet overclocked it)

To put it into perspective:
i5-2500 + 2x R9 290 was 30-50fps in Crysis 3 Multiplayer with all settings on the lowest.
The CPU usage was 100% with just 1GPU, so obviously it was 100% with 2GPUs.

i7-3770k + 2x R9 290 is 80-115fps in Crysis 3 Multiplayer with all settings on the highest(no AA).
The CPU usage peaks at around 90% with 2GPUs.

BF3's performance is very good, too. No where near as bad as what it used to be when using crossfire.

I am just as shocked as many of you are about the performance of my i5-2500. I thought it would be more than capable, but obviously not. It's hard to believe that the CPU would make that much of a difference in this case, but it did.
Before I bought my 2nd GPU I even asked around on multiple forums (Tom's Hardware etc) and I was told that the i5-2500 would be fine.

Maybe there was something wrong with the CPU? I've used it since March 2011, but still, its performance shouldn't have degraded.

I'd honestly like to thank everyone for their on-going help in trying to figure out the cause of my problem. It's greatly appreciated : )

 

[Final_Fighter]

great to hear that its working. i think the problem with your old cpu was it could have degraded but even if it ha-dent i dont think it would have been able to handle it anyway. i guess if you wanted with your new cpu you could overclock 200-300mhz sense your cooling is good enough.

so by peaks you mean its running more like in the 50~60% normal game play then the 90% on heated and intense graphics moments?

again thats great that it worked out it also should teach people something new with your experience. honestly i havent seen any i5 reports about sandy lately with your kind of gpu setup. ive seen single gpu but not double.

its a good thread and covered some good info so thanks for sharing this with us.

 

[OneMoar]

CPU usage is WAY high for a 3770K you have another problem the new cpu is just masking it crysis 3 is not that CPU intensive
install proces explorer and logg the cycles used
http://technet.microsoft.com/en-us/sysinternals/bb896653.aspx

 

[GhostRyder]

Hi, Everyone.

I received my i7-3770k today and it runs like an absolute charm! (I haven't yet overclocked it)

To put it into perspective:
i5-2500 + 2x R9 290 was 30-50fps in Crysis 3 Multiplayer with all settings on the lowest.
The CPU usage was 100% with just 1GPU, so obviously it was 100% with 2GPUs.

i7-3770k + 2x R9 290 is 80-115fps in Crysis 3 Multiplayer with all settings on the highest(no AA).
The CPU usage peaks at around 90% with 2GPUs.

BF3's performance is very good, too. No where near as bad as what it used to be when using crossfire.

I am just as shocked as many of you are about the performance of my i5-2500. I thought it would be more than capable, but obviously not. It's hard to believe that the CPU would make that much of a difference in this case, but it did.
Before I bought my 2nd GPU I even asked around on multiple forums (Tom's Hardware etc) and I was told that the i5-2500 would be fine.

Maybe there was something wrong with the CPU? I've used it since March 2011, but still, its performance shouldn't have degraded.

I'd honestly like to thank everyone for their on-going help in trying to figure out the cause of my problem. It's greatly appreciated : )

Glad it resolved the issue!

To all the haters, cough cough...

 

[RCoon]

crysis 3 is not that CPU intensive

Crysis 3 and BF3 are some of the very VERY few games that are heavily multithreaded. As I recall, Crysis 3 can use up to 16 threads.

 

[Batou1986]

Crysis 3 and BF3 are some of the very VERY few games that are heavily multithreaded. As I recall, Crysis 3 can use up to 16 threads.

This is true, scale form and cryengine 3 do not like each other much.
You should see how poorly MWO performs because the devs have no idea what they are doing
Turning the scaleform hud off doubles FPS.

Additionally BF3/4 is a cluster@$^& of who knows what the hell is wrong with the jenga code this patch.

 

[OneMoar]

http://www.overclock.net/t/1362591/lightbox/post/19332690/id/1302786
so once again all this proves is that AMD needs to fix there drivers and it is INDEED a software issue and not a hardware
a 2600k clocking at 3.4 witch is 100Mhz over the 2500's stock NON turbo clock of 3.3 is sitting at about 45-60FPS on very high going up to the TOP OF THE LINE extreme edition its 50-75

now the 290's are about equal on paper to the GTX690 with the 290's in theory pulling ahead by a fair margin

secondly multi player is FAR LESS intensive then the SP campaign so any threading metrics fly out the window as the MP binary only uses a fraction of the CPU time that the SP binary does ( no ai pathing less complex physics calculation and smaller maps witch means less LOD and culling calculations as well)

I ran crysis 3 MP on a frigging Athlon II @ 3.6Ghz at medium`ish settings and a 5770 and MP played at a steady 50 to 60FPS AT 1280X1024

Something is afoot here the op should easily be pushing a consistent 100FPS in MP unless hes running on ULTRA and not "highest"

 

[RCoon]

at a steady 50 to 60FPS AT 1280X1024

Go figure...

Also, OP had ~50FPS with his old chip. That FPS has DOUBLED since he got the new chip, and you're still banging on about crossfire software?

 

[OneMoar]

Go figure...

witch shows that MP isn't nearly as CPU bound as people in this thread think it is I was running up against the GPU's limits before the CPU
as you lower the resolution the cpu becomes the limiting factor in most cases I didn't really test my 6870 with the Athlon II as the phenom II was here the next day

 

[RCoon]

witch shows that MP isn't nearly as CPU bound as people in this thread think it is I was running up against the GPU's limits before the CPU

Then explain the OP's sudden miraculous double of FPS, if not to do with the change in CPU.

 

[OneMoar]

Then explain the OP's sudden miraculous double of FPS, if not to do with the change in CPU.

the underlying problem is most likely still present hes just thrown so much CPU power at it that it no longer renders a crippling impact thats why I asked for some info from process explorer as to what was sucking the cpu cycles down if hes got some background process using 15 or 20% at idle thats a problem and enough to put the 2500 in a bad spot
90% cpu load on a 3770k at ~80FPS strikes me as high indicating some other factor at play

 

[RCoon]

the underlying problem is most likely still present hes just thrown so much CPU power at it that it no longer renders a crippling impact thats why I asked for some info from process explorer as to what was sucking the cpu cycles down
90% cpu load on a 3770k at ~80FPS strikes me as high indicating some other factor at play

I believe he said "peak", unless he clarifies that it runs like that permanently. Also, doubling the FPS sounds like a good solution to me. Perhaps his original CPU was roshambo'd, and that made this upgrade perform far better than it should have.