how much cpu bottlenecking is happening w DCS?

[skypickle]

Der8auer made a video in July where he put a 3080ti into several different systems ranging from a 4770k up to a 12900k. 

His testing was in games that were modern. Does anyone have any data (or experience) with a cpu upgrade and how it affected your frame rate in DCS?

[Mr_sukebe]

If a move from a i5 2500 to a 9700 counts, then yes.  That resulted in roughly a 50% improvement in frame rate.

The CPU bottleneck is primarily is you have more complex missions, with a lot of AI.  An example being pathfinding for the AI.

There is a way to offload some of that i.e. by running a dedicated server, and then logging into it.  That results in the AI pathfinding and logic being completed by the CPU core running the dedicated server, and NOT the client.  Helps to spread the load a little.

[Mars Exulte]

Yeah, it's not just "CPU in general". It's that everything is crammed onto one thread. If you're environment is relatively uncomplicated, then CPU binding will be minimal. It's when you start loading it up with complex environments, VR, etc, that you start running into issues.

[Thinder]

12 hours ago, skypickle said:

Der8auer made a video in July where he put a 3080ti into several different systems ranging from a 4770k up to a 12900k. 

His testing was in games that were modern. Does anyone have any data (or experience) with a cpu upgrade and how it affected your frame rate in DCS?

You seriously can't pretend to run a reliable bottleneck test with this RAM kit, it is obvious it has one and he doesn't even mention the reason why...

I find amazing that people can pay so little attention to the bound between CPU and RAM, because the ability of the CPU to deliver maximum performances under load is directly depending of the RAM Timings and this mean that if the CPU controller can't access the timings it needs to run at full throttle it will strugle and eventually throttle down.

With Intel CPUs and older generation AMD Ryzen, frequencies are what matters most, that's why competitive players used B.Die kits to O.C those kits.

Intel techies run their tests with GSkill Trident Z RGB for this reason, it posseses the range of timing the CPU can handle to support higher frequencies but lso, ONLY one bank per stick and 3200 MHz Cl14, within the CPU frequencies limit without OC.ing it and low latency.

With the Ryzen 5 Zen 3 architecture it is a totally different matter, it favours low latency instead of high frequencies and this kit provide those Ryzen with the best stability and timing range possible while keeping within the controller's limits, same here, maximium of 4 X 1 bank, 3200 MHz.

That's what the actual CPU in use by the majority of players controllers can handle, out of those limitations; number of rank per stick + higher freequencies and if you try to play 4K you'll lose in performances because your controller doesn't have access to the timings it needs to handle the load.

So if you want to get rid of those RAM-to-CPU bottlenecks, you'll need a B.Die RAM kit, which are the only ones using the SAMSUNG die.

B.die is much more stable and allowing the RAM to provide your CPU with a much wider range of timings, and because they only have one bank per stick, a 4 X 1 combo also allows for interleaving for better spread of memory addresses across memory banks.

That's how you eliminate those bottlenecks, no other way because it is the way those generations of controllers are designed and what they requier to tun at full throttle under load.

For Intel users... How to Overclock RAM

Beware of what RAAM manufacturers sells you, you might well end up with wooden chariots instead of Ferraris, do your home work.

Edited November 9, 2022 by Thinder

[skypickle]

@Thinderthank you for your reply but cannot see what you are saying. Each system had different ram, cpu, chipset, etc representing different generations. So perhaps a more accurate question would ‘how much is the gpu held back by different systems in DCS’. This is an important question that has to be answered before spending.

[BitMaster]

There is one counter, one employee doing the whole show. If only 2, 3 or 4 guests are in it's all OK. Everybody gets their Soda and Food in time...just when Joe and his Football Team come through the door and every table and seat is full it would be SO MUCH BETTER to have 3 Girls serving the crowd, 2 guys doing the drinks and such and a great cook that has nothing to do but cooking.

One day we will get there, be patient.

[Thinder]

3 hours ago, skypickle said:

@Thinderthank you for your reply but cannot see what you are saying. Each system had different ram, cpu, chipset, etc representing different generations. So perhaps a more accurate question would ‘how much is the gpu held back by different systems in DCS’. This is an important question that has to be answered before spending.

It's simple. ALL CPUs have similar limitation, a look at Intel or AMD specs says it all and yours is not different.

What runs the show here, is not your perception of how "unique" a player system might be, it is the CPU controllers and to give you a clue both manufacturers give CPU limits, which are for the previous generations 3200 MHz for the Ryzen 5 3600X and 2666 MHz for an Intel I9 990K as used for the Intel test, I explained why by giving each CPU main characteristics.

Here are the Zen 2 and Zen 3 controllers specs.

In short before Zen 3, both AMD and Intel beneficiated from higher frequencies that a B.Die RAM kit would provide their controllers, so top gamers would use those kits and O.C them because they were the only RAM kits allowing for it, so did the Intel technicians. Zen 3 is biased toward lower latency so a Ryzen 5 5600X will take advantage of a Cl14 kit at 3200 MHz.

Right now, the rare CPUs designed to handle DDR5 like the 12900KS has a limit of 3200 MHz for DDR4 as well, 4800 MHz running DDR 5, meaning it does not support faster DDR 4 RAM using the non-B.Die solution than a Ryzen 5 3600X, to run at higher frequencies than those DDR5 or DDR 4 you'll need B.Die RAM and their range of timings, that's Controler limits for you.

Controllers manage timings, channel bandwidth and their own clocks, it's NOT the RAM or the GPU you fit in, it's the CPU controller and what they can handle or not, they aren't unique to each player or system as opposed to what you suggest.

If you wander out of those limits by fitting RAM kits that go beyond it, you keep the bottleneck I mentioned, if you provide your CPU with the RAM that let the controller work to its full potential by using the timings it needs to do so, be it with higher frequencies (Intel) or lower latency (AMD Zen 3) you remove it and AGAIN, there is no going around that, that's how it works.

Things controllers CANT handle are: Too high frequency. Too high number of banks. Meaning try to fit a Cl 16 4 X 2 banks 3600 MHz to any of those DDR4 CPUs and you will see your CPU throttling down under load. No 4K for you dude...

This is why Intel technicians are using B.Die Cl14 to O.C their RAM with their own CPUs, since their CPU is not designed to take advantage of lower latency, those RAM kits at least can be O.Ced because the material they are made of is much more stable and allow the controller to use the timing it needs to run at full speed at higher frequencies.

And there is NO difference between the last two DDR4 generations, from Ryzen 5 3600X to 5600X the controllers are the same apart for ECC support (question of speed vs safety here), pretty much the same story with Intel, the i9-9900K 2666 MHz has even lower frequency limits than a 3600X, simply because the rest couldn't cope with more load, there was no need for increasing those controllers limits, only the next generation of CPU might bring a progress at this level but right now it is not the case.

Ryzen DDR5 5200 MHz Intel DDR5 DDR5 4800 MHz

 

Edited November 9, 2022 by Thinder

[Mars Exulte]

6 hours ago, Thinder said:

You seriously can't pretend to run a reliable bottleneck test with this RAM kit, it is obvious it has one and he doesn't even mention the reason why...

I find amazing that people can pay so little attention to the bound between CPU and RAM, because the ability of the CPU to deliver maximum performances under load is directly depending of the RAM Timings and this mean that if the CPU controller can't access the timings it needs to run at full throttle it will strugle and eventually throttle down.

With Intel CPUs and older generation AMD Ryzen, frequencies are what matters most, that's why competitive players used B.Die kits to O.C those kits.

Intel techies run their tests with GSkill Trident Z RGB for this reason, it posseses the range of timing the CPU can handle to support higher frequencies but lso, ONLY one bank per stick and 3200 MHz Cl14, within the CPU frequencies limit without OC.ing it and low latency.

With the Ryzen 5 Zen 3 architecture it is a totally different matter, it favours low latency instead of high frequencies and this kit provide those Ryzen with the best stability and timing range possible while keeping within the controller's limits, same here, maximium of 4 X 1 bank, 3200 MHz.

That's what the actual CPU in use by the majority of players controllers can handle, out of those limitations; number of rank per stick + higher freequencies and if you try to play 4K you'll lose in performances because your controller doesn't have access to the timings it needs to handle the load.

So if you want to get rid of those RAM-to-CPU bottlenecks, you'll need a B.Die RAM kit, which are the only ones using the SAMSUNG die.

B.die is much more stable and allowing the RAM to provide your CPU with a much wider range of timings, and because they only have one bank per stick, a 4 X 1 combo also allows for interleaving for better spread of memory addresses across memory banks.

That's how you eliminate those bottlenecks, no other way because it is the way those generations of controllers are designed and what they requier to tun at full throttle under load.

For Intel users... How to Overclock RAM

Beware of what RAAM manufacturers sells you, you might well end up with wooden chariots instead of Ferraris, do your home work.

 

It's not that we don't know. We know. We also know there is ALWAYS a bottleneck somewhere, it's literally unavoidable. We also recognise, 99% of people are not going to buy ONLY the one RAM stick that meets your performance criteria. We know you like Gskillz.

[SkateZilla]

DX11 CPU Overhead and Core Thread Overhead.

[skypickle]

On 11/9/2022 at 1:47 PM, SkateZilla said:

DX11 CPU Overhead and Core Thread Overhead.

I appreciate your brevity but do not understand it. After all the words from these replies, i still do not have a sense of how DCS would fare in a study like der8auer’s

[skypickle]

@AirMeisterthank you for your reply. How did you determine headroom? Looking at individual cpu utilization in resource manager? 

[SkateZilla]

4 hours ago, skypickle said:

I appreciate your brevity but do not understand it. After all the words from these replies, i still do not have a sense of how DCS would fare in a study like der8auer’s

https://forum.dcs.world/topic/293072-multi-core-discussion/?do=findComment&comment=5081811

[skypickle]

Thank you. If this is true “going Vulkan also removes DX11's Hard Limit on resolution, MSAA Type, Single Channel Instruction paths, HDD->CPU->GPU Memory Access Overhead,and about 20-30 other things ” Then why are modern games (MSFS2020) slower on dx12 (actually the latest comparisons show dx12=dx11 in fps as dx12 continues to be updated). Is vulcan specifically optimized for dx12 and MSFS is using something else?

Edited November 14, 2022 by skypickle

[Nick_17]

6 hours ago, skypickle said:

Thank you. If this is true “going Vulkan also removes DX11's Hard Limit on resolution, MSAA Type, Single Channel Instruction paths, HDD->CPU->GPU Memory Access Overhead,and about 20-30 other things ” Then why are modern games (MSFS2020) slower on dx12 (actually the latest comparisons show dx12=dx11 in fps as dx12 continues to be updated). Is vulcan specifically optimized for dx12 and MSFS is using something else?

 

Vulkan is not something that can be optimized for dx12, its a direct competitor / replacement of dx11 / dx12. They are different GFX APIs.

[SkateZilla]

On 11/14/2022 at 2:56 PM, Nick_17 said:

Vulkan is not something that can be optimized for dx12, its a direct competitor / replacement of dx11 / dx12. They are different GFX APIs.

What he Said, Vulkan is not Part of DX12, it's a competing API

On 11/14/2022 at 8:21 AM, skypickle said:

Thank you. If this is true “going Vulkan also removes DX11's Hard Limit on resolution, MSAA Type, Single Channel Instruction paths, HDD->CPU->GPU Memory Access Overhead,and about 20-30 other things ” Then why are modern games (MSFS2020) slower on dx12 (actually the latest comparisons show dx12=dx11 in fps as dx12 continues to be updated). Is vulcan specifically optimized for dx12 and MSFS is using something else?

 

The issue with DX12 FPS/Performance Matching DX11 is because the developer of said engine used MS's easy conversion path, which simply runs the Graphics engine in DirectX11_1 Mode using the DirectX 12 API Libraries, and doesn't get any of the benefits of DirectX 12.