frametime: GPU vs CPU?

[Donglr]

I gents,

just upgraded my GTX1060 and Rift CV1 to a RX 6900XT and a Reverb G2. I played with the settings and got a bit disappointed, tbh. I was expecting more. I hoped for stable 45 fps with nearly full DCS graphics setting and ~80% SS.And even though this works nicely on Nevada ground in a basically empty single player mission, it is very much different in multiplayer. When I went on Hoggit I didn't event get a stable 30 fps on the ground. So I bought fpsVR to get my analysis going and low and behold: my GPU frametime was actually lower than my CPU frametime. GPU seems to be actually able to handle what I request and CPU is struggling.

 

Now to my question: How do I have to interpret these frametimes, .i.e. does each of them need to be below my target fps (22.2ms for playing at 45fps) or does the SUM need to be below that value? Because fpsVR told me my CPU frametimes are about 30++ms at a populated multiplayer server. There is no getting 45fps with such CPU load.

 

my specs: i7 6700K @4.5GHz, RX 6900XT, 32GB @3100MHz, Reverb G2 (SS tbd), 256GB+1TB SSD

[ouseler]

I have an RX 6800XT paired with a i9-9900K and TrackIR.  I saw great improvement in my FPS, but I don't have VR either which loads the system down.  My issue was a frame rate stutter which I finally figured out to be my monitor refresh rate (60 Hz) and GPU frame rate clashing.  I found that if I adjust my GPU FPS to just shy of 60 (say, 59), the stutter disappeared.  May not be the same issue you are having, but it was the bain of my existence for a long time.  

[Scorpion_knight]

I may be sticking my nose in here, so apologies because you may be a complete PC genius... but something I tried recently after hours and hours of changing one setting after another (and I do mean hours and hours) was to essentially turn OFF all anti aliasing  and lighting modes within DCS and ramped up all settings within my AMD Radeon card letting the card do all of the work. I went from 35-45 ish FPS to 60-70 and the game looks and plays amazingly...who knew??...this appears to go against all I have seen to do/ try but thought I might share this with someone. I currently have a AMD Radeon 5700XT running on an i7 10700k 32gb RAM running the game at 1440p. The reason for this was that I had an odd lighting shimmer particularly around dials and models but that has now completely gone.

[fab.13]

As Scopion says, you should start from 'DCS VR Preset' ,

and see what is your basic performance in a free flight with ni cloud

 

such 'minimum' graphics setting must give you good performance before to change option.

If not, you have to solve this problem first.

 

then after you change setting one by one.

 

Another thread of the forum explain wich graphics settings

have the most impact in FPS performance ( AA, PD & SS )

[Donglr]

well gents,

I appreciate your comments, but none of you have actually answered my question It was very specific on the meaning of GPU and CPU frametime.

Because my assumption is right now, that actually I can change all the DCS settings I want to the lowest values, the CPU is what limits we and I will never get good fps because of CPU frametime. I have a setting that runs nicely in Nevada in an empty single player mission, but that is not where I spend my play time.

[speed-of-heat]

they both need to be below the 22.1 ms value if one is above it you are done 

[pimp]

Have you tried overclocking your 6700K? I used to use that chip but I was able to OC it to 5.2 watercooled.

[Donglr]

5 hours ago, pimp said:

Have you tried overclocking your 6700K? I used to use that chip but I was able to OC it to 5.2 watercooled.

 

Well, it already is overclocked. I run it at 4.5GHz, it's only 4 GHz stock. More I cannot get out of it, it already has quite a high core voltage stock, so not really a silicon lottery winner.

[pimp]

1 hour ago, Donglr said:

 

Well, it already is overclocked. I run it at 4.5GHz, it's only 4 GHz stock. More I cannot get out of it, it already has quite a high core voltage stock, so not really a silicon lottery winner.

That's too bad. Well if you can't OC it anymore, than a newer cpu might be the next step.

[RED]

11 hours ago, Donglr said:

 

Well, it already is overclocked. I run it at 4.5GHz, it's only 4 GHz stock. More I cannot get out of it, it already has quite a high core voltage stock, so not really a silicon lottery winner.

Here is an example with a 5600X. Data from SP and MP on the Syria map:

CPU frametime is around 12ms in MP and 9ms in SP. You have a good CPU bottleneck with 30+. I think it's time to upgrade your CPU, too. If you want to stick with it make sure you follow guides for reducing the windows overhead. Thud has some stuff: https://vr4dcs.com/2020/07/19/os-optimization-for-win-10-2004-key-tuning-for-dcs-world-and-vr/

I also upgraded from an older Intel CPU and it's a game changer.

[Donglr]

Really nice graph, how do you get it?

 

Well, throwing money at it is always a viable route. But before I go there I first would like to understand this:

 

This is my CPU usage during MP. Once I am done starting the game and stuff my processors are basically below 60%. Now I know that DCS is horrible and makes no use of multithreading, but at least I would expect for one core to be maxed out, like here: https://forums.eagle.ru/topic/220017-cpus-and-dcs-cpu-usage/?do=findComment&comment=4092430 . Why is this not the case? Is it because the DCS task gets all the time it wants but my system is so un-loaded that the task gets assigned different cores because they are all idle a lot? So like DCS.exe runs 100% of the time, but 50% on core 1 and 50% on core 2 so both cores seem only half used? If this was the case then only increasing clock would help. Or changing to another CPU architecture that gets done more per cycle? But is that really the case? Will a AMD 5600X at 4.6GHz be faster than a i7 6700K at 4.6GHz? And by how much?

 

Or is there actually something wrong with my scheduling? How can I measure process uptime, not per core but in total across all core? Like "DCS.exe is executed 95% of the time, 45% of that on core 1 and 55% on core 3"?

 

[RED]

Not an expert on this but I'll try to answer some questions.

 

The graph is actually FPSVRS's "History Viewer". 

 

Also seeing around 50% "usage" on the graph. It's normal that the most active cores change. The CPU is trying to spread the heat. On my end it's always moving between the two best cores.

To compare CPU's a benchmark like cinebench helps. Here are some stock values from that benchmark:
https://www.cpu-monkey.com/en/cpu_benchmark-cinebench_r20_single_core-9
This is the single core run and just what we want to look at with DCS.

 

You can't compare frequencies between AMD and Intel. Currently the AMD CPUs get more done per cycle.

[Donglr]

Hmm, unsatisfactory that DCS is so full of spaghetti code that it does not use the provided resources and hence does not perform. But after talking to a wingman of mine with a similar CPU he pretty much gets the same CPU usage. So I guess indeed no way around then buying a better CPU. Waiting for ED to provide a modern game core would be like waiting for Godot.

[speed-of-heat]

if you have a older cpu , you might want to check out some of the hygine steps in my 3090 guide they might help

[Willie Nelson]

As a side note, keep in mind a shift to the Vulkan API which is at least slated for Q3 this year may change things quite a bit. Just a thought. 

[Burt]

That’s true but if I had to flip a coin right now for Vulcan or a Tomcat freeze issue I would pick the Tomcat fix.

[C3PO]

On 7/5/2021 at 12:28 AM, Willie Nelson said:

As a side note, keep in mind a shift to the Vulkan API which is at least slated for Q3 this year may change things quite a bit. Just a thought. 

We're in Q3 now

 

[Lurker]

1 hour ago, C3PO said:

We're in Q3 now

 

 

You must be new around here, add 3-6 months to most Eagle Dynamics estimates to get a realistic time frame for release. That is, unless it has been delayed or has run into implementation issues which might delay it even more, in which case implement "Everything is subject to change" logic.

[Willie Nelson]

1 hour ago, C3PO said:

We're in Q3 now

 

Yep, I get it and I have even less idea than the next guy about when it may land or whether or not it will even be any good but it would seem to be a potential waste of money to go and spend several thousand dollars on a new rig only to find that the old one would have sufficed and you could have spent it on something that was a real improvement elsewhere such  as the HP Reverb Gen 3

 

[C3PO]

11 hours ago, Willie Nelson said:

Yep, I get it and I have even less idea than the next guy about when it may land or whether or not it will even be any good but it would seem to be a potential waste of money to go and spend several thousand dollars on a new rig only to find that the old one would have sufficed and you could have spent it on something that was a real improvement elsewhere such  as the HP Reverb Gen 3

 

Once the campaign engine lands, I’d imagine they’ll need as many cores as they can grab hold of.

[VampireNZ]

On 6/29/2021 at 5:20 AM, Donglr said:

Hmm, unsatisfactory that DCS is so full of spaghetti code that it does not use the provided resources and hence does not perform. But after talking to a wingman of mine with a similar CPU he pretty much gets the same CPU usage. So I guess indeed no way around then buying a better CPU. Waiting for ED to provide a modern game core would be like waiting for Godot.

 

Looks like you have sussed it out - I had the same CPU with a 1080ti and Reverb G2 and it couldn't hack it at all. You would be doing yourself a LOT of favours getting a 5900X CPU to go with that GPU as the core power is substantially better than Intel options (5950 not worth it for a few extra cores - 5800 overheats badly).

I upgraded to a 5900X & 6800XT and saw a nice improvement in DCS (but not great - all other VR games are amazing now thou) - but more importantly the CPU generally sits around 5-6 ms, maybe up to 10 if on heavy maps. You don't want your CPU anywhere near 22ms as it has no way to 'handle' the shortfall - unlike the GPU that can drop frames etc.

So I ended up adding in some shadows to the cockpit (with the 'shadow range' mod to limit to immediate area of cockpit)for a little 'eye candy' after all these years of potato graphics lol.

 

Some of the code in DCS is so suspect and dodgy thou - I remember when testing out settings with new setup there was a IA mission on Nevada with some warbird flying around in the distance - frame times were absolutely in the toilet if that dot was on the screen at all, even if it was 20 miles away! Was good frame times...look at warbird speck....performance of sim cut in half! lol.

 

Also just FYI, 4 sticks of RAM work better than two with those CPU's as well.

Edited July 8, 2021 by VampireNZ

[C3PO]

My 5800X doesn’t overheat … in fact in full extended use in DCS doesn’t go much above 49C, according to FPSVR.

Edited July 9, 2021 by C3PO

[Morat]

Just for comparison, I run a 3080 and a G2. With 109% SS in SteamVR (damn that slider!) I'm able to hold 45FPS flying over Guam with the default VR settings and some increases for things like Tree radius to avoid the "Genesis Effect" which is very distracting on that map.

My CPU is a 9700K and it is now holding me back as CPU Frametime is higher than GPU Frametime. I went to overclock and realised my temps are too high so I'm messing with fans and AIOs at the moment.

Long story short, I'm also looking for better CPU performance but the good news is that I'm sure you'll get a lot more out of your 6900XT than you are at the moment.