Find cause for crazy high CPU frametime?

[Donglr]

Hi gents,

 

I just upgraded my GTX 1060 and Rift CV1 to a RX 6900XT and a Reverb G2 and was pretty bummed when I found that my fps improved f**k all. Yes, I have better visuals now, but I still cannot make it past ~25fps on a populated server. So I got fpsVR and found that my GPU frametime is about 20ms, which would barely be enough for 45fps but I can see that maybe I have to go down on SS. The real problem is the CPU frametime however. It is around 30-50ms! A friend of mine has a RTX 3090 so I asked him for his values. We setup the same graphics settings in DCS and same SS values in SteamVR. Then we joined the same server (Rotorheads). The results:

 

RX 6900XT, i7 6700K @4.5 GHz, 32GB

GPU: ~15 ms

CPU: 30-50 ms

 

RTX 3090, i7 6800K @4.2 GHz, 64 GB

GPU: ~11 ms

CPU: ~19 ms

 

Why on earth is my CPU frametime so bad compared to his? We have the same family of CPU and mine is actually overclocked higher! Both our regular RAMs are not full, btw, my VRAM (16GB) is full, his isn't (but he has 24GB).

 

Does anyone know a way to find out what process is eating up my ressources while gaming? I mean simply looking at the task manager is the obvious solution, but there is nothing that draws significant CPU %. I mean, my values are basically half as good as my buddies, like if I was running 2x DCS at a time.

 

Thx

 

 

[The_Nephilim]

Hi I was wondering what cooling solution you are using, it looks as if your CPU is being throttled.. perhaps time to do a repaste. have you tyried testing the cpu to see if it being throttled overheating?

[Donglr]

good idea. Been running my overclock for a year now and when I tuned it back then I did do measurements and it was all ok. But since then maybe my AIO water colling lost some liquid. Checking for throttling is an easy test, I will do that.

[Donglr]

11 hours ago, The_Nephilim said:

Hi I was wondering what cooling solution you are using, it looks as if your CPU is being throttled.. perhaps time to do a repaste. have you tyried testing the cpu to see if it being throttled overheating?

 

Nope, that's not it

I did a test flight and my CPU is cold, barely any readings above 70°C. Interestingly my core utilisation is only up to 60% on the most heavily used core, still not fast enough to get 45fps

[FETUS]

Congrats on the new video  card. I've been wanting to get a 6900XT for a bit. I had been running a 9900KF and 2080 Super and would get 35-45 FPS on medium settings. I upgraded my mobo and processor/GPU to a Ryzen 5900X and 3090 FTW. Cranked up my settings and now get 45-60 FPS. I've been really happy with the results.

 

In my experience, multiplayer is very taxing on DCS regardless of GPU. I had a 6700k back in 2015 when they came out. I think you should upgrade to a Ryzen 5600 or 5800X to get the most out of that video card.

 

Or check your bios settings for this : PCI-Subsystems Re-Size Bar

[Donglr]

6 hours ago, FETUS said:

I think you should upgrade to a Ryzen 5600 or 5800X to get the most out of that video card.

 

Or check your bios settings for this : PCI-Subsystems Re-Size Bar

 

Ah well, trying to avoid throwing more money at it, esp since my CPU according to HWInfo isn't even utilized fully. I get 60% max CPU load on any of my cores. So what the heck, DCS.exe, please go ahead and properly use the hardware you are presented!

 

Resizable BAR: no gonna happen with a Z170 chipset from 5 years ago. Asus did not bring a new BIOS with support for that.

[Taz1004]

On 6/25/2021 at 3:30 PM, Donglr said:

Hi gents,

 

I just upgraded my GTX 1060 and Rift CV1 to a RX 6900XT and a Reverb G2 and was pretty bummed when I found that my fps improved f**k all. Yes, I have better visuals now, but I still cannot make it past ~25fps on a populated server. So I got fpsVR and found that my GPU frametime is about 20ms, which would barely be enough for 45fps but I can see that maybe I have to go down on SS. The real problem is the CPU frametime however. It is around 30-50ms! A friend of mine has a RTX 3090 so I asked him for his values. We setup the same graphics settings in DCS and same SS values in SteamVR. Then we joined the same server (Rotorheads). The results:

 

RX 6900XT, i7 6700K @4.5 GHz, 32GB

GPU: ~15 ms

CPU: 30-50 ms

 

RTX 3090, i7 6800K @4.2 GHz, 64 GB

GPU: ~11 ms

CPU: ~19 ms

 

Why on earth is my CPU frametime so bad compared to his? We have the same family of CPU and mine is actually overclocked higher! Both our regular RAMs are not full, btw, my VRAM (16GB) is full, his isn't (but he has 24GB).

 

Does anyone know a way to find out what process is eating up my ressources while gaming? I mean simply looking at the task manager is the obvious solution, but there is nothing that draws significant CPU %. I mean, my values are basically half as good as my buddies, like if I was running 2x DCS at a time.

 

Thx

 

And both of you have G2?  And ALL values in SteamVR are identical including motion smoothing?

[The_Nephilim]

I was thinking have you checked for malicous software or miners or similar? does any other game tank that bad?

[Donglr]

5 hours ago, Taz1004 said:

 

And both of you have G2?  And ALL values in SteamVR are identical including motion smoothing?

 

Yes, he has a G2 as well, I explicitly selected same SS as him and we went thru the DCS settings together. For the purpose of the comparison I think we used as similar conditions as we could besides the obvious differences that he as has a better GPU, a different (but lower clocked) CPU, more RAM and a different mainboard.

[The_Nephilim]

well the SS in Steam VR is probally it. what are you set at rez wise?

[Taz1004]

19 hours ago, The_Nephilim said:

well the SS in Steam VR is probally it. what are you set at rez wise?

I think the point is that difference with everything else being equal.

Use AfterBurner hardware monitor and check each core's clock and temp.  And compare it to your friend's.

If that doesn't show what's wrong, only major difference is the graphic card so I'd try different drivers.  If he's near, I'd even try swapping.  Only way to troubleshoot these is just process of elimination.

Edited June 28, 2021 by Taz1004

[Thinder]

On 6/25/2021 at 8:30 PM, Donglr said:

Hi gents,

 

I just upgraded my GTX 1060 and Rift CV1 to a RX 6900XT and a Reverb G2 and was pretty bummed when I found that my fps improved f**k all. Yes, I have better visuals now, but I still cannot make it past ~25fps on a populated server. So I got fpsVR and found that my GPU frametime is about 20ms, which would barely be enough for 45fps but I can see that maybe I have to go down on SS. The real problem is the CPU frametime however. It is around 30-50ms! A friend of mine has a RTX 3090 so I asked him for his values. We setup the same graphics settings in DCS and same SS values in SteamVR. Then we joined the same server (Rotorheads). The results:

 

RX 6900XT, i7 6700K @4.5 GHz, 32GB

GPU: ~15 ms

CPU: 30-50 ms

 

RTX 3090, i7 6800K @4.2 GHz, 64 GB

GPU: ~11 ms

CPU: ~19 ms

 

Why on earth is my CPU frametime so bad compared to his? We have the same family of CPU and mine is actually overclocked higher! Both our regular RAMs are not full, btw, my VRAM (16GB) is full, his isn't (but he has 24GB).

 

Does anyone know a way to find out what process is eating up my ressources while gaming? I mean simply looking at the task manager is the obvious solution, but there is nothing that draws significant CPU %. I mean, my values are basically half as good as my buddies, like if I was running 2x DCS at a time.

 

Thx

 

 

 

You mentioned everything but not your RAM.

 

I don't know Intel CPUs that well but compared to AMD, in principle, they both have controllers that give you lower buses use when they throttle down, so if your RAM bound is average or mediocre, your latency is going through the roof as well because your CPU controller isn't able to cope under load*, and it does affect every bus, including CPU-to-GPU.

 

* This part is important because if you upgrade your PC you also can cause the problem by increasing the load on the CPU controller, so an issue which was not appearent before can creep up after the upgrade.

 

It can come from the number of ranks per stick plus high frequencies, CPU controllers have limits when it come to this combination, the more ranks, the lower the frequency limit, in the best case for your 1rst processor, the recommanded frequencies are DDR4-1866/2133, DDR3L-1333/1600 @ 1.35V, DDR4 2400/2133 for the second.

 

This can variate depending on how many ranks per stick this limit was computed for, I'd be surprised if it's 2 in the case of a 4 X sticks configuration, more likely 1 rank X 4 or 2 X 2 optimum, in any case you have to take into acount the Max Memory Bandwidth, for a  i7 6700K, 34.1 GB/s.

 

Compare this with an AMD 5600X 47.68 GiB/s which is optimized for low latency and still with a recommanded DDR4-3200 frequency, for this precise reason.

 

Intel users have forums just like AMD users, if you research and even ask Intel support, they can advise you on the subject of RAM/CPU bounding, the fastest RAM and higher number of ranks and/or sticks are not necessarily the best solution because of the limits of your CPU controller...

 

RAM latency of course is important, a CL14 will always be better than CL16 or above, this sort of details are much too often overlooked by players, as I previously said, I don't know (didn't research) Intel CPUs that well but I'm pretty sure that there are RAM kits which allows for a lower latency than others, you might want to have a look at them.

Edited June 29, 2021 by Thinder

[Thinder]

You can try this App, it's free and works.

https://timerresolution.soft112.com/

[Donglr]

Well, I ordered new components, a 5600X, so let's see if this is an actual step forward.

On the RAM: I have 4x8Gb 3200MHz memory, although my current mainboard/CPU only eats it a t 3100MHz. And it's CL16 unfortunately. But I will not upgrade this as well. On the benchmark I saw faster RAM was more of a cosmetic change over a better CPU.

[BitMaster]

4 hours ago, Donglr said:

Well, I ordered new components, a 5600X, so let's see if this is an actual step forward.

On the RAM: I have 4x8Gb 3200MHz memory, although my current mainboard/CPU only eats it a t 3100MHz. And it's CL16 unfortunately. But I will not upgrade this as well. On the benchmark I saw faster RAM was more of a cosmetic change over a better CPU.

Good move with the 5600X ! I dont think it will dissapoint you. It runs circles around my enthusiastically cooled 5GHz 8700k even w/o PBO2 engaged and with the stock AMD cooler.

It should give you a good boost forward.

 

For the RAM: Once you can run them stable at 3200/CL16 without any manual OC otherwise you can try to up the Volts to say like 1.385-1.40v and lower the CL to 14. It's a trial and error and may not be worth it but it's a good way to waste dozens of hours and maybe you can get them to 3200-14-14-14-34-1T @ 1.38-1.40 Volts That's what I would do if they run XMP/DOCP just fine. If they don't, you have to manually find out the correct settings or buy new RAM. 

 

I am also keeping my RAM, 3600-16-16-16-36-2T 32GB/4x8 for my planned 5900X. Only if they dont run properly I will buy new RAM...and if I have to buy I will go 64GB, likely 2x32GB if I can get those with the timings I want. RAM can be really tricky if it wont work out of the box as intended.

[Thinder]

On 6/30/2021 at 5:27 PM, Donglr said:

Well, I ordered new components, a 5600X, so let's see if this is an actual step forward.

On the RAM: I have 4x8Gb 3200MHz memory, although my current mainboard/CPU only eats it a t 3100MHz. And it's CL16 unfortunately. But I will not upgrade this as well.

 

As I specified previously, the number of ranks per sticks can cause the CPU to throttle back, OK, Cl16 is not the best but it is within the controller range, but if your RAM sticks have more than 1 rank each, since you have more of them you're already out of the controller limits and it will throttle down under load.

 

I'm surprised that your motherboard doesn't support 3200MHz. one of the reasons (you need to check your motherboard manufaturer specs) can be the quality of the RAM chips, which doesn't allow for tighter settings, reason why top AMD users like to use B-Die RAM, the quality of the material used to make their chips provide more stability and control over frequencies and settings you can use.

 

Quote

On the benchmark I saw faster RAM was more of a cosmetic change over a better CPU.

 

It is not a bout faster RAM but lower latency, which the Ryzen 5600X is optimized for, since you chosed to fit 4 X sticks, you can hit one or more of the issues I quoted, starting with number of ranks, here is what AMD support said about it...

 

 

I wrote to them before picking my RAM kit and it is obvious that not only they limit this CPU to 3200MHz because of the controller limit in terms of ranks, but if you don't take this into account and add twice this number (4 X 2 instead of 4 X 1), notr only you end up loosing the advantage of bank interleaving but you also lower the controller ability to cope with the number of ranks and lose the main quality of this CPU which is its ability to use lower latency, which you increase anyway when your CPU throttle back.

 

The result is increase latency under load but also about a loss of 5/6% + interms of CPU performances, you whole system is working out of its specifications and nowhere near to where it could be had you bounded a RAM kit more suited to the 5600X.

 

I could have made the same mistake by adding two RAM sticks to those I had previously, Cl16/2 ranks but I didn't simply because I wanted to optimize my ststem within the specifications of AMD and knew about the 5600X main qualities, which is what I decided to improve with a better RAM bound.

 

The other issue is that only adding another 2 RAM sticks can lead to more problems as manufacturers can use different chips from a kit to another even if the RAM are seemingly identical, so best buying a 4 X sticks kit because otherwise you are at risks of not being able to use your new RAM and P.C at all.

 

Here is what G.Skill said about it as reply to my request for information when I couldn't find a 4 X sticks kit.

 

 

So since I don't know what motherboard you're susing I only can tell you those things as thumbrules but can't offer an advise for a solution, in passing, your O.S settings and background use can also be a source of increase latency...

Edited July 2, 2021 by Thinder

[Thinder]

OK. So, I know that it will sound like a burden to you, but here are the reasons for upgrading your RAM and use the RAM kit I recommanded (or another Cl 14 4 X 1 rank 3200MHz kit).

 

1) Lower latency: A Cl14 kit witll take full advantage of the Ryzen CPU architecture, it is optimized for low latency.

 

2) Rank interleaving: A 4 X 1 rank kit will allow the controller a lot more flexibility and use all ranks to the best of its capabilities.

 

3) No RAM-to-CPU bootleneck from iddle to full load. Since the Cl 14 4 X 1 rank 3200 MHz is fully within the CPU recommanded RAM specs, you won't experience any issue of your CPU throttling back under load, I personally saw a gain of 6.04% at 4K using 3D Mark Pro 4K Gaming, 3840 x 2160 and 2 X MSAA.

 

Graphic scores also improved with this RAM kit, which is logical since the CPU-to-GPU bus wasn't limited by the CPU controller anymore.

 

Quote

Trident Z RGB DDR4-3200MHz CL14-14-14-34 1.35V 32GB (4x8GB); back-to-back tests in 3D Mark Pro 4K Gaming, 3840 x 2160 and 2 X MSAA, my previous RAM kit was a Crucial BALLISTIC GAMING MEMRORY 32GB DDR4-3200 CL16 2X 16GB.

 

Here are the results (Crucial RAM first. G.GKILL second)

Fire Strike Ultra 4K gaming with DirectX 11

3840 x 2160 (16:9) (4K) MSAA x 2

MSI Afterburner ON. AMD Ryzen Master ON; Game mode.

>>>>>>

Graphics score: 6 496 vs 6 583. +1.33%
 
Physics score: 23 894 vs 25 339. + 6.04%

Combined score: 3 605 vs 3 654. + 1.35%.

 

You might not want to spend an extra £80 for this sort of result but think about the 2.76% you can expect to gain with full water cooled CPU overclocking, running at 4.8 GHz, thus not under AMD guaranty especially when you know that their CPU RAM limit under guaranty is precisely 3200 MHz, you might find a better RAM bound more attractive.

 

I obtained more than TWICE this O.Ced CPU performance without splashing any money on water cooling and loosing manufacturer guaranty, because all I use is Ryzen Master boost but NO O.C, on ther other hand, the G.Skill kit cost <> £80 more than the Crucial kit it replaced.

 

Oh yeah, I forgot, the maximum temperature my CPU reached during this test was 73 °C.

Edited July 3, 2021 by Thinder

[Donglr]

Wow man, you are really getting at it.

Well long story short: I am not going to spend another 240 Euros on new 32GB of CL14 RAM. Maybe it's 6% in some benchmarks but that is not the bottleneck. The real bottleneck is DCS with its spaghetti code and no multi-threading. I have a CPU usage of maybe 40% and I am pretty sure you can throw as much hardware at this simulator as you want, the code base is just not up to par.

[Thinder]

14 hours ago, Donglr said:

Wow man, you are really getting at it.

Well long story short: I am not going to spend another 240 Euros on new 32GB of CL14 RAM. Maybe it's 6% in some benchmarks but that is not the bottleneck. The real bottleneck is DCS with its spaghetti code and no multi-threading. I have a CPU usage of maybe 40% and I am pretty sure you can throw as much hardware at this simulator as you want, the code base is just not up to par.

 

DCS has got little to do with it, if your RAM bound is mediocre, blaming the game won't solve your problem, and you skipped the point where those 6% are obtained without O.Cing CPU and/or RAM, which if you haven't get it yet reduce both life-span and cost money, short like long term AND, using the CPU low latency capabilities.

 

So at the end of the day you can be stuborn and try to blame the game for the poor performances of your system, it's your lot, personally I don't have this issue, my frame times are top noch, another point which you seems to have forgoten but complained about, again the game has little to do with it, it's not the source of system latency.

Edited July 5, 2021 by Thinder

[BitMaster]

I wouldnt buy new RAM either unless really forced to. 

 

Enjoy

[Lurker]

Just to be clear, have you eliminated the possibility that any potential malware\adware\bloatware was eating up your CPU cycles? 

Edited July 7, 2021 by Lurker

[Donglr]

10 hours ago, Lurker said:

Just to be clear, have you eliminated the possibility that any potential malware\adware\bloatware was eating up your CPU cycles? 

 

 

If the Windows task manager is anything to go by, yes. It's DCS using the biggest chunk, then for a long while nothing and the next process in line is at like 2% tops. According to HWInfo my most loaded core is at 60% max. Nothing like you see with other games where there is a core maxed out. It actually looks similar to my friend, his CPU usage is also low for a CPU bound game. But then a single thread can only run 100% of the time and more cores won't help.

[Thinder]

On 7/6/2021 at 1:29 PM, BitMaster said:

I wouldnt buy new RAM either unless really forced to. 

 

Enjoy

 

Bad CPU-to-RAM bound affects shader performances, increases latency and frame time, if that's not enough to look at it as an issue I don't know what does.

[Lurker]

12 hours ago, Donglr said:

 

If the Windows task manager is anything to go by, yes. It's DCS using the biggest chunk, then for a long while nothing and the next process in line is at like 2% tops. According to HWInfo my most loaded core is at 60% max. Nothing like you see with other games where there is a core maxed out. It actually looks similar to my friend, his CPU usage is also low for a CPU bound game. But then a single thread can only run 100% of the time and more cores won't help.

 

That's not accurate at all. It's only an indicator but can't represent anything other than core usage. If a core has to wait for example, while another core is doing something (or another system component) it will sit at idle. You could have a process that is slowing down your entire system without it being noticed as an overhead in core usage. Before your hardware upgrade I suggest you do a complete re-install of windows. It's what I would do if I had your symptoms. 

[Mars Exulte]

14 hours ago, Donglr said:

If the Windows task manager is anything to go by, yes. 

 

  Windows Task Manager is in fact NOT anything to go by, especially if you're trying to troubleshoot something nitpicky like CPU usage per core or something like this. It is a generic half-assed utility, and for example, does not account for how Windows itself shuffles the load between cores dynamically at the rate of dozens of times per second. It is useless for troubleshooting, and that bit about it showing 40-60% core usage is absolutely inaccurate (it basically displays ''averages''). This has been hashed out a few hundred times on these very forums.

 

  While DCS does not utilise more than two cores due to its age, it DOES utilise those cores and it has nothing to do with ''spaghetti code'' (at least in this context). In particular you're describing two similar systems with a noticeable performance disparity that probably shouldn't exist... by definition that's on YOU the operator's end, NOT the software. Also, you mentioned the ''higher overclock'', a common misconception is that ''moar hurtz'' is automatically better, and that's not necessarily the case.

 

  There are a great many things that can contribute to high CPU latency. While in general I agree replacing your RAM is itself probably not worth it compared to the cost, numerous ''little things'' like that can soon add up in combination together and that's most likely what you're encountering : a series of inefficiencies or minor issues, drivers, etc gradually accumulating until you've got a noticeable discrepancy.

Edited July 8, 2021 by Mars Exulte