Bottlenecking, or a faulty cpu?

[DidelisDiskas]

So for some time now i'm using a system with an x4 860k cpu (stock cooler, not overclocked) and an r7 360 gpu running linux. Because nobody seems to have this card (especially on linux) it was hard for me to find a point of reference as to it's questionable performance, so not long ago i got tired of the ultra low fps, which i thought was maybe a product of bad drivers and the card being slow, however when i bought an rx 470 the performance barely increased and actually got a bit worse in some games. After not believing my eyes to the abysmal performance i was getting, i tried to swap the r7 360 and a gtx 660 between 2 linux systems, one with my x4 860k and one with the i5 4430..... Every damn card performed very similarly on the x4 860k, getting terrible, jaggy fps and when i switched to the i5 4430, the fps on my r7 360 doubled (almost tripled in some heavier areas of the game) the gtx 660 also got almost double the performance (could not test the rx 470, as the intel system only had a 6 pin connector).

So now i'm not sure what to do with that thing (the 860k), can it be so much slower as to bottleneck cards of different tiers to the same performance level, or is something faulty in my system (i'm also getting coil whine like sounds from the motherboard or the cpu itself, can't really discern from which one)?

edit: the intel cpu is 4430, not the 4460

 

[VirtualLarry]

So now i'm not sure what to do with that thing (the 860k), can it be so much slower as to bottleneck cards of different tiers

YES!

The 860K is a dual-module chip, each module handles two threads. It's more comparable to an Intel i3 (dual-core with HyperThreading), than an Intel i5 (a TRUE quad-core).

Edit: And to address the question in the thread title - if the CPU were defective, it wouldn't just be running slower, you would see crashes and errors.

 

[bystander36]

Is the i5 4460 system also on Linux? Linux isn't a gaming platform, and any game that runs on it, is almost always going to perform worse than on Windows. It is also known that AMD's support of Linux is also poor.

All that said, it sounds like a CPU bottleneck. You have an AMD APU, which is quite slow compared to the i5 or even AMD's more powerful CPU's.

It is possible you are getting throttled by the APU, and also possible that you are using the built in GPU, rather than the discrete cards. Is the monitor plugged into the motherboard or the dGPU?

 

[frozentundra123456]

Classic symptoms of cpu bottlenecking, obviously. As to whether there is just that much difference in the cpus or if there is some problem with the 860k is harder to say. How are temps on the 860k---could it be throttling, or maybe there is a motherboard of ram problem with the 860k system. If you have a good cooler, you could also try overclocking the 860k to see if performance improves. I would expect the 4440 to be faster than the 860k of course, but not to the huge degree that you are seeing.

 

[daxzy]

Is the i5 4460 system also on Linux? Linux isn't a gaming platform, and any game that runs on it, is almost always going to perform worse than on Windows. It is also known that AMD's support of Linux is also poor.

It is possible you are getting throttled by the APU, and also possible that you are using the built in GPU, rather than the discrete cards. Is the monitor plugged into the motherboard or the dGPU?

OP said the GTX 660 performed similarly to the R7 on the x4-860 as well, ruling out possible AMD driver issue. Your last sentence makes no sense, as the x4's aren't APU's.

I think its a CPU bottleneck as well. The x4-860K is about the equivalent of a i3-2100 at stock or slightly slower than the Pentium G3258 at stock. That said, we also don't know enough info. What linux distro/kernel? What games are you running? I'm also going to make a blanket statement that unless you're highly invested in the linux environment, Windows 10 offers superior performance as well.

 

[AtenRa]

There is no way in hell that you get the same performance with R7 360 and RX 470 even if your CPU is the Athlon 860K at 3GHz. Even Ritchland 750K at 3GHz gains 10fps going from R9 285 to GTX 980 in heavy CPU bound games like GTAV.

Most likely its the GPU driver or anything else not optimized in Linux (game itself etc). What game are we talking about here ??

http://www.anandtech.com/show/10436/amd-carrizo-tested-generational-deep-dive-athlon-x4-845/10

 

[DidelisDiskas]

OP said the GTX 660 performed similarly to the R7 on the x4-860 as well, ruling out possible AMD driver issue. Your last sentence makes no sense, as the x4's aren't APU's.

I think its a CPU bottleneck as well. The x4-860K is about the equivalent of a i3-2100 at stock or slightly slower than the Pentium G3258 at stock. That said, we also don't know enough info. What linux distro/kernel? What games are you running? I'm also going to make a blanket statement that unless you're highly invested in the linux environment, Windows 10 offers superior performance as well.

I have distro hoped (on this particular machine) from distros with 4.1 to 4.8 kernel and from mesa 11 to 12.04, i also tried the amdgpu-pro, fglrx drivers, but it does not matter, the games that get very low fps continue to do so. This was the case for all three gpu's tested (gtx 660, r7 360, rx 470) all got to about 8-9 fps in pillars of eternity forge area.... Yes, that's 9 fps on a game that even a toaster could handle. What seems to always crush my fps are things like fire enhancements, candles, haze, mist, these kinds of effects. For example in Bioshock infinite, with my r7 360 runs fine at 40-50 fps (i'm not sure it let's you edit graphics settings, so i would assume medium, can't remember), when i look at some sun effects, or candles, the fps goes from around 50 to 15. The same happened with such games like shadowrun and wasteland 2, where some areas with some particular effects would just crush my fps to nothing and others would be good. Also smoke grenades and effects in Insurgency. The fps holds at 30-40, then when i look at some smoke it goes down to the teens. However not all games run terribly, as Borderlands 2 and Witcher 2 ran, well while not exceptionally, but at least without so many fps jumps and at around 30-40 fps.

What i don't understand is how can this cpu bottleneck rx 470 running games like pillars of eternity? The game is not at all intensive. And i have asked other people in reddit who ran this game on linux with similar setups (but all had i5 cpus) and they had no problems.

As for windows, i know it's superior for gaming, i have used windows longer than linux, and i'm not using it anymore because i like the feeling of owning my pc and having an almost fully open environment that doesn't bother me and i don't even have to pay for it. I have tried windows 10, but it makes me claustrophobic, plus i already have most of the stuff that i need on linux.

 

[DrMrLordX]

Try overclocking your CPU and see if adjusting the clockspeed helps performance at all. If it does, then you know you have a CPU bottleneck.

You should also look at increasing NB speed and look at your RAM speed/timings as well.

 

[DidelisDiskas]

Try overclocking your CPU and see if adjusting the clockspeed helps performance at all. If it does, then you know you have a CPU bottleneck.

You should also look at increasing NB speed and look at your RAM speed/timings as well.

It has a stock cooler, wouldn't that be dangerous? I could try it just to quickly check on the performance, although seeing as i have never overclocked anything i might end up melting the cpu . What would be a safe limit just to check if it does anything fps wise (as i mentioned the cpu has a stock cooler, but i can manually set the 2 120mm case fans to go max throttle all the time).

 

[DrMrLordX]

No, it won't be dangerous unless you touch the voltage. You should be able to get a few hundred extra MHz just at stock vcore. If it gets too hot, it'll shut down. That isn't likely though.

That is an 860k. It is made for overclocking. Unless you are using abnormally low voltages (undervolting), you should be able to clear 3.9 GHz easily, maybe 4.0 GHz. Some samples do better than others.

What motherboard are you using?

 

[DidelisDiskas]

No, it won't be dangerous unless you touch the voltage. You should be able to get a few hundred extra MHz just at stock vcore. If it gets too hot, it'll shut down. That isn't likely though.

That is an 860k. It is made for overclocking. Unless you are using abnormally low voltages (undervolting), you should be able to clear 3.9 GHz easily, maybe 4.0 GHz. Some samples do better than others.

What motherboard are you using?

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

 

[DidelisDiskas]

Here is how my settings page looks:

Am i supposed to increase the cpu frequency multiplier, or the boost? What is an nb frequency multiplier?

 

[DrMrLordX]

Raise the CPU frequency multipier to 40.0 and check to see if it is stable (I recommend Prime95; ask here for settings if you want them, or just use the Blend test for simplicity. You can use whatever test you see fit). If that works we can move on to the NB frequency multiplier next.

 

[DidelisDiskas]

Well i ran a blend test for 35 minutes with all 4 cores at 100% and the case fans spinning at idle. It seems to do well, although i also ran some tests with xonotic and unigine heaven, what i find particularly interesting, is that when i ran the unigine heaven benchmark with a 34 multiplier, it performed better than when i ran it with a 40 multiplier. The cpu usage was around 26% when the benchmark was running:

Also, here's the same r7 360 (Red-Radeon-1), only coupled with an i5 6600k (and older drivers) running xonotic 08:

The other games i tried like pillars of eternity, had no perceivable difference and Team Fortress 2 continues on having constant fps drops of up to a 100, oscillating between 200 and 30 fps...

This is the scene (with cpu usage) from pillars of eternity where i am getting down to 10 fps~:

Are those 100% spikes bad, or is that normal?

 

[DrMrLordX]

Yeah that is pretty bad. Also you aren't exactly getting high CPU utilization. I'm not sure CPU is the problem here.

I'm not sure why you would get any performance degradation with a higher CPU multi. Are you seeing any throttling? What kind of CPU temps are you holding during these graphics benchmark runs?

 

[DidelisDiskas]

Yeah that is pretty bad. Also you aren't exactly getting high CPU utilization. I'm not sure CPU is the problem here.

I'm not sure why you would get any performance degradation with a higher CPU multi. Are you seeing any throttling? What kind of CPU temps are you holding during these graphics benchmark runs?

It's hard to say. because the cpu temp at idle is reported as 0C (the gpu temps are reported correctly though), so i would think there is some threshold bellow which the sensors are reporting 0C. At the stress test with blender the cpu tries to keep 65C, or at least that is what the sensors are reporting, so if i would add something like 20-30C would that be a rational guess?

 

[TheELF]

65°C is pretty much the point where it starts to throttle,72 is the max safe temp, so don't look at temps but at clocks in real time,no idea how to do that in linux,

 

[DidelisDiskas]

65°C is pretty much the point where it starts to throttle,72 is the max safe temp, so don't look at temps but at clocks in real time,no idea how to do that in linux,

Will add that info after i'm done with some additional testing (currently ongoing).

 

[DidelisDiskas]

So, ok, what is interesting is that the system is reporting the cpu not ever hitting anywhere close to 4ghz, even though the multiplier is set at 40 (i checked twice), the temps were fine as the cpu fan didn't even speed up, and here is the cpu usage with xonotic and unigine heaven:

x4 860k + r7 360: xonotic 08

x4 860k + r7 360: Unigine Heaven

And i have also tested these on a system with an i5 4430 + GTX 650:

i5 4430 + GTX 650: xonotic 08

i5 4430 + GTX 650: unigine heaven

Now since these systems use different gpu's and drivers one might say that this is because of the gpu drivers, but this cannot be for Xonotic, because as we see bellow amd cards perform better on xonotic than nvidia, and my r7 360 performed worse by an average of 30 fps than the GTX 650, which is a much weaker card:

And as i have mentioned before, when i swaped my r7 360 into the system with the i5 4430, it performed much smoother and better not even looking at the fps and the fps have doubled and tripled in some cases.

 

[DrMrLordX]

I have had some issues with proper reporting of CPU speed in Linux before. I hate to say it, but it would be a lot easier to troubleshoot your problems in Win10 where the software support for tuning/overclocking is a lot better. That being said, the problems I had should not be an issue for you since you are not using an APU like my old 7700k or my current 7870k.

What kind of clockspeed governor are you using in Linux? CPUFreq?

 

[DidelisDiskas]

I have had some issues with proper reporting of CPU speed in Linux before. I hate to say it, but it would be a lot easier to troubleshoot your problems in Win10 where the software support for tuning/overclocking is a lot better. That being said, the problems I had should not be an issue for you since you are not using an APU like my old 7700k or my current 7870k.

What kind of clockspeed governor are you using in Linux? CPUFreq?

I'm note sure, as it's part of the phoronix test suite, however i did my own little benchmark, as i wrote this script to report the frequency of all 4 cores every 0.2 seconds while i was playing pillars of eternity for 11 minutes:

#!/bin/bash
while true; do
sleep 0.2
cat /proc/cpuinfo | grep "MHz" >> freq.txt
printf "\n" >> freq.txt
done

The command cat /proc/cpuinfo | grep "MHz" reports all 4 cpu frequency (or state, i'm not sure if there is a difference between the two, can a cpu be 3510 mhz, or does it just bounce from state to state?) at the instance it is called. What it reported is:

3227 instances = 4000 mhz -- > 25.58%
168 instances = 3500 mhz -- > 1.33 %
674 instances = 3000 mhz -- > 5.34 %
6164 instances = 2400 mhz -- > 48.86%
2383 instances = 1700 mhz -- > 18.89 %

I also sorted (by cpu core) and put that info into an excel file, but i'm not sure how to plot that data in a meaningful way, any ideas?:

http://s000.tinyupload.com/?file_id=36004585505662811646

 

[DrMrLordX]

If you have all the power saving stuff enabled in the UEFI, then your 860k will have different p-states that correspond to particular clockspeeds. One of the p-states is your max turbo (4.0 GHz) and those other clockspeeds look like lower p-states.

Did it record instances of the CPU running at 2.4 GHz and 1.7 GHz during a benchmark?

 

[DidelisDiskas]

If you have all the power saving stuff enabled in the UEFI, then your 860k will have different p-states that correspond to particular clockspeeds. One of the p-states is your max turbo (4.0 GHz) and those other clockspeeds look like lower p-states.

Did it record instances of the CPU running at 2.4 GHz and 1.7 GHz during a benchmark?

I will try to run the script on the benchmark when i will come back home tomorrow.

Also, so if i have p states enabled, does it make any sense to increase the frequency multiplier, or should i try to increase the boost frequency multiplier? Or is there a setting to disable the p states, to see if that makes any difference (though i suppose the cpu could not thorttle down in such a case, if it gets too hot)?

 

[DrMrLordX]

If you disable all the power saving features for the CPU, it will disable turbo and idle/low power throttling modes. The CPU will run at its base clock. That's the surest way to get the CPU running at a particular clockspeed. Then you adjust the base clock accordingly.

It can still throttle due to thermals though.

 

[DidelisDiskas]

Once i disabled the boost and cool 'n' quite settings in the bios, all cores were holding 4 ghz all the time (i did not test it thoroughly, but the cpu did not produce much heat and was staying on pretty good temps) though the benchmarks and my gaming tests were mixed:

Strangely Unigine Heaven had even worse cpu utilization, holding on to 20%~ most of the time, while it was about 25%~ without the overclock. Maybe just the r7 360 drivers at fault.

Pillars of Eternity seemed to run a bit better too, still it does not perform as i would expect, since it's maybe 5-10 fps better, but is not getting anywhere near 60.