Bottlenecking, or a faulty cpu?

[PG]

I have an Asrock FM2A88X PRO+ and 8gb of 1886mhz ram (one stick).

I think this is part of your problem. With one stick of memory you are running single channel vs dual, and I bet there is a serious performance penalty.
Intel memory controllers are very efficient and you can probably get away with single channel memory for gaming, but not AMD.
Also all FM2+ cpus are slow compared to Intel. Sorry, but that's reality. Look at some BF1 numbers: http://www.gamersnexus.net/game-bench/2673-battlefield-1-cpu-benchmark-dx11-vs-dx12-i5-i7-fx/page-2
The A10-7870K is very similar to your 860K on the cpu side, and the 6600K gives slightly over double the average fps.

 

[DidelisDiskas]

I think this is part of your problem. With one stick of memory you are running single channel vs dual, and I bet there is a serious performance penalty.
Intel memory controllers are very efficient and you can probably get away with single channel memory for gaming, but not AMD.

Then i will put 2x 4gb ram sticks to see if there is any difference.

Also all FM2+ cpus are slow compared to Intel. Sorry, but that's reality. Look at some BF1 numbers: http://www.gamersnexus.net/game-bench/2673-battlefield-1-cpu-benchmark-dx11-vs-dx12-i5-i7-fx/page-2
The A10-7870K is very similar to your 860K on the cpu side, and the 6600K gives slightly over double the average fps.

You don't need to be sorry, i know that this is a slow cpu, i bought it because it was cheap and i did not have a pc at the time. I will replace it in the following months. However i am not playing titles like battlefield 1, i'm playing games like Team Fortress 2, pillars of eternity, wasteland 2 - all games that can run on a igpu (well wasteland 2 is streching it a bit). I started asking this question because i have an rx 470 and it did not perform all that much better than my r7 360 (20-40 fps on pillars of eternity with the rx 470!), so i am trying to narrow down where the problem stems from exactly, so that i would not need to spend a bunch of money on a new mobo+cpu+ram and still get shit performance.

 

[LTC8K6]

IIRC, you only see large gaming gains going from single to dual channel ram if you are using an IGP because the GPU is using system ram.

Overall you probably won't notice any difference just going from 8gb SC to 8gb DC ram if you are not using an IGP.

You may see some small gains in benchmarks, though.

 

[DidelisDiskas]

Well, i installed ubuntu 16.10 with the newest mesa drivers, 4.8 kernel and i also swapped in my rx 470 to take it for a ride. So i ran 3 benchmarks on stock with boost and power states, and then on 4.4 ghz OC (on a stock cooler , but since the cooler and case fans were set to manual full speed it never reached any troubling temperatures, i think i could have tried to go higher):

So Unigine heaven on rx 470 stock is 3 times faster than my r7 360, while xonotic and tesseract.... Well i have no idea what is happening here. The whole thing just confuses me. I will now try to install pillars of eternity again and see how that does on the newer drivers.

 

[DrMrLordX]

Well at least the clockspeed boost helped some. I would say that you may be looking at a driver issue here.

Tesseract is clearly more CPU-limited than Heaven 4.0, hence the outsized gains there.

Also, since you have ventured to 4.4 GHz on your CPU (nice, by the way), you may want to look at tweaking your NB some. There are gains to be had there.

 

[DidelisDiskas]

Well at least the clockspeed boost helped some. I would say that you may be looking at a driver issue here.

Tesseract is clearly more CPU-limited than Heaven 4.0, hence the outsized gains there.

It would be a terrible driver issue, as in some games my rx 470 get's almost equal performance to the r7 360, but as the unigine heaven benchmarks shows it's clearly outpacing it when the task is heavilly gpu dependent. I wonder if some games in linux are just super cpu hungry, or flat out terrible ports.

Also, since you have ventured to 4.4 GHz on your CPU (nice, by the way), you may want to look at tweaking your NB some. There are gains to be had there.

NB, as in northbridge? Could you elaborate a bit, as i'm not at all acquainted with overclocking procedures, this was the first time i overclocked antything (even though i had an unlocked amd phenom for 5 years in the past).

 

[DrMrLordX]

If you look at your UEFI, you'll see the CPU NB Multiplier at 18x, leading to a clockspeed of 1800 MHz.

The voltage being fed to this particular part of the APU is the CPU NB/GFX Voltage, 1.275v

Leave the voltage alone for now, and try 1900 MHz and then 2000 MHz CPU NB speed. It can be a little bit difficult to test those settings for stability, though Prime95 Blend, y-cruncher 2g, and the usual graphics benchmarks are okay ways to test this (Unigine Heaven picks up NB instability fairly well).

And there are some fairly bad Linux ports out there. You may also be surprised how poor some graphics drivers still are for Linux.

 

[SlickR12345]

My suggestion is to ditch Linux! There are free windows versions out there, like college ones and stuff and with older windows you can also get free win10 (even now). PM me if you want info to get a free legit copy of win10.

 

[DidelisDiskas]

If you look at your UEFI, you'll see the CPU NB Multiplier at 18x, leading to a clockspeed of 1800 MHz.

The voltage being fed to this particular part of the APU is the CPU NB/GFX Voltage, 1.275v

Leave the voltage alone for now, and try 1900 MHz and then 2000 MHz CPU NB speed. It can be a little bit difficult to test those settings for stability, though Prime95 Blend, y-cruncher 2g, and the usual graphics benchmarks are okay ways to test this (Unigine Heaven picks up NB instability fairly well).

And there are some fairly bad Linux ports out there. You may also be surprised how poor some graphics drivers still are for Linux.

How does this differ from the multiplier method? Is there a good source on the web that explains these concepts?

My suggestion is to ditch Linux! There are free windows versions out there, like college ones and stuff and with older windows you can also get free win10 (even now). PM me if you want info to get a free legit copy of win10.

I enjoy the freedom (and variety) in linux to do anything i want too much to go back to windows, even if that means having inferior gaming experience (though it's getting better).

 

[DrMrLordX]

How does this differ from the multiplier method? Is there a good source on the web that explains these concepts?

Actually I am suggesting you use the multiplier. That is the only way I know of to change the clockspeed short of messing with bclk which you do NOT want to do.

 

[DidelisDiskas]

Actually I am suggesting you use the multiplier. That is the only way I know of to change the clockspeed short of messing with bclk which you do NOT want to do.

What i mean is, what is the difference between the cpu frequency multiplier and the cpu nb frequency multiplier (what is the nb portion and why is it at 1800 mhz at the moment)?

 

[DrMrLordX]

What i mean is, what is the difference between the cpu frequency multiplier and the cpu nb frequency multiplier (what is the nb portion and why is it at 1800 mhz at the moment)?

Oh.

They are not linked. CPU multi and CPU NB multi are independent of one another.

The NB - NorthBridge - is a separate block within the CPU that contains the memory controller and whatever control logic handles the iGPU. It would also serve as the cache controller for the l3 if your chip had any.

The default NB clockspeed on Kaveri is 1800 MHz. If you increase that clockspeed, it generally improves the performance of the memory controller. The net effect is usually lower system memory latency. That is a good thing.

Most Kaveri CPUs can have stable NB speeds as high as 2000-2100 MHz.

 

[DidelisDiskas]

Oh.

They are not linked. CPU multi and CPU NB multi are independent of one another.

The NB - NorthBridge - is a separate block within the CPU that contains the memory controller and whatever control logic handles the iGPU. It would also serve as the cache controller for the l3 if your chip had any.

The default NB clockspeed on Kaveri is 1800 MHz. If you increase that clockspeed, it generally improves the performance of the memory controller. The net effect is usually lower system memory latency. That is a good thing.

Most Kaveri CPUs can have stable NB speeds as high as 2000-2100 MHz.

Ah, thanks. Does it increase energy consumption by much?

 

[DidelisDiskas]

Well, i increased the NB multi to x20 and:

I could not believe my eyes, but leaving the cpu multi at stock with boost and then increasing the cpu NB multi by 2 (to 2000mhz), has increased the average of xonotic result by 24 fps and the minimum by 7 fps on my r7 360. It even increased the ungine heaven by 3 fps, all of which was not possible with a 4.4 ghz overclock..

So is my memory controller the bottleneck here?

 

[DrMrLordX]

It is possible. Try a 4.4 GHZ CPU OC and a 2000 MHz NB OC together and see how they behave. Keep an eye on temps as always.

 

[DidelisDiskas]

So now the results are closer to the 6600k + r7 360 (branded as red-radeon-1 in this test, with older mesa drivers though):

My minimums are still low, but the average has improved by a solid 30 fps from stock in xonotic (in which i initially got only 83fps). I'm now currently downloading pillars of eternity to see if this will have any effect on that game.

 

[DrMrLordX]

Good luck! And now you are learning about one of the oddities of AMD microarchitectures since at least the Phenom II days: NB matters.

Though Vishera/PD owners might disagree with that.

In any case, chips like the 860k are made to be overclocked. They're okay-ish @ stock, but they really can't compete in anything until you jack up that clockspeed. Makes me feel kinda bad for A8-7600 owners. Kinda. They can OC too, a little.

 

[DidelisDiskas]

Good luck! And now you are learning about one of the oddities of AMD microarchitectures since at least the Phenom II days: NB matters.

Though Vishera/PD owners might disagree with that.

In any case, chips like the 860k are made to be overclocked. They're okay-ish @ stock, but they really can't compete in anything until you jack up that clockspeed. Makes me feel kinda bad for A8-7600 owners. Kinda. They can OC too, a little.

Well hopefully i will not have to deal with this all that much longer, as i'm waiting to see what zen has in store. If that flops i will get a new i5. However this was an interesting learning experience, and kind of fun too. Thanks for the help.

 

[DrMrLordX]

No problem.

I'm lining up for Zen/Summit Ridge as well. It's gonna be a brave new world . . .

 

[DidelisDiskas]

No problem.

I'm lining up for Zen/Summit Ridge as well. It's gonna be a brave new world . . .

Well i was going to end this disccusion here, but before i go off to reddit, i thought i'd post these results of my rx 470 with the x4 860k at stock:

Age Of Wonders (33 fps)

Divinity Original Sins Enhanced Edition (~20 fps)

Torchlight 2 (23 fps)

Metro: Last Light Redux (47 fps)

Day Of Defeat Source (The counter on screen is wrong, as the fps goes between 200 and 80~, never lower. It slows down when i exit from the windowed mode)

Talos Principle (from 30 to 50 fps, the counter on the screen has the same problem as day of defeat, where it lowers fps when i exit the window)

Borderlands 2 (42 fps)

Unigine Heaven (58 fps)

Unigine Valley (54 fps)

Any final thoughts on this looking at the gpu usage?

 

[DrMrLordX]

GPU clock is running very low on a lot of those titles. It only reaches any kind of high clockspeed in the Unigine tests.

That bears investigation.

Can you do similar tests with the CPU OCed to 4.4 GHz and your NB running at 2000 MHz?

 

[DidelisDiskas]

GPU clock is running very low on a lot of those titles. It only reaches any kind of high clockspeed in the Unigine tests.

That bears investigation.

Can you do similar tests with the CPU OCed to 4.4 GHz and your NB running at 2000 MHz?

I will try to redo all of these with the overclocks when i get back from my shift. I would also note that manually setting the gpu clock on max did not seem to have any effect on those games that perform poorly. However correct me if i'm wrong, but Torchlight 2 being cpu bound on this cpu seems like a crazy notion to me, considering Metro and Unigine are performing better.

 

[DrMrLordX]

A lot of Linux games are just Windows applications running in a wrapper, which can lead to all kinds of poorly-coded nonsense. But really there's no telling why the GPU usage is so low in so many of those titles . . . at least not without doing some more digging.

 

[DidelisDiskas]

A lot of Linux games are just Windows applications running in a wrapper, which can lead to all kinds of poorly-coded nonsense. But really there's no telling why the GPU usage is so low in so many of those titles . . . at least not without doing some more digging.

To give you an idea how poorly coded some of them are, i have to use a special hack where i compile a shim for divinity original sin to even launch. The developers were too lazy to include proper opengl compliance check, so the game refuses to run on anything, but the old deprecated fglrx driver.... It's a really simple fix and the community have raised the awareness on their forums, but they could not be bothered to spend 10 minutes updating the game so it could actually run on polaris cards by default...

Anyway, i oc'd the machine to 4.4ghz multi and 2ghz NB multi like you said. I also forgot to mention, but the previous results were not completely on stock, as i have lowered the boost clock to max 3.7 ghz, so the results are 3.7ghz boost vs 4.4ghz stable. Here are the 4.4ghz + 2.0ghz results (kernel 4.8.0-30, mesa 13.1):

Age Of Wonders (about a 14 fps increase, probably the most improved of the bunch, but still no where near where it should be)

Divinity Original Sin (still can't force it to show fps, still terrible, laggy performance):

Torchlight 2 (an amazing increase of 5 fps):

Shadowrun Dragonfall (an amazing 27 fps in this scene, hangs on around 30-40 eslewhere):

Trine2 (holds at around 45 fps):

Xcom Enemy Unknown (hangs at around 35 fps):

Metro Last Night (could not force the damn game to go into windowed mode this time, but not much change fps wise):

Day Of Defeat (no change, the game run pretty smoothly, not considering the frame drops of up to a 100, but it still does not go bellow 60. The counter is again wrong, since i exited windowed mode for screenshot):

Talos Principle (not much change):

Borderlands 2 (Not much change):

Unigine Heaven (a bit more stable, but pretty much the same):

Unigine Valley (pretty much the same as above):

Could not test Pillars Of Eternity, as the game refuses to run all of a sudden (it tries to change the resolution and when fails, exits), but it would be the same story i think.

So out of all the games of this bunch, the best performing ones are Day Of Defeat (really big frame drops though), Metro (gets from 40~ to 120~ fps), and Unigine benchmarks. Actually Unigine benchmarks also have the most stable frames (well they drop in the screen transitions, but for a very brief moment).

 

[DidelisDiskas]

I thought that maybe games would run better on wine, but trying out Pillars Of Eternity and Age of Wonders on wine, they ran even worse.