Jump to content
old_Sneakers

CPU performance monitor on 6 core @ 4.7 ghz - SQUAD PROBLEM IN A NUTSHELL

Recommended Posts

AFAIK Devs are limited in how much they can (or even should) multithread stuff due to UE engine. If they go ahead and start doing it themselves they may end up duplicating effort with some changes that may be done by the UE itself in the future, similar to the audio bottleneck issue, which devs were about to tackle but had to hold because of major changes in UE in this area coming in the near future. Anyway devs efforts on their own code was very good indeed, and future UE optimizations should only add to that, thus I think the future is bright for most, but no miracle should be expected for old HW specs.

Share this post


Link to post
Share on other sites
1 hour ago, Zenrique said:

AFAIK Devs are limited in how much they can (or even should) multithread stuff due to UE engine. If they go ahead and start doing it themselves they may end up duplicating effort with some changes that may be done by the UE itself in the future, similar to the audio bottleneck issue, which devs were about to tackle but had to hold because of major changes in UE in this area coming in the near future. Anyway devs efforts on their own code was very good indeed, and future UE optimizations should only add to that, thus I think the future is bright for most, but no miracle should be expected for old HW specs.

Indeed.  we could.. be we aren't going to start multithreading everything when we can optimize what we have an introduce enough scalability to let AMD systems play..  its sad but UE4 simply isn't built well for AMD systems..  we will do what we can. (we own to AMD systems in the studio now for profiling)  

 

 

Share this post


Link to post
Share on other sites
2 hours ago, IrOnTaXi said:

Indeed.  we could.. be we aren't going to start multithreading everything when we can optimize what we have an introduce enough scalability to let AMD systems play..  its sad but UE4 simply isn't built well for AMD systems..  we will do what we can. (we own to AMD systems in the studio now for profiling)  

 

 

 

It is a dilemma for sure.

 

Scalability to me is synonymous to writing it heavily threaded. But there is no reason to manhandle the engine into doing things the UE devs themselves will make native further down the line.

 

The obvious problem i see is in 2-3 patches down the line, especially when small arms ballistics with penetration are being introduced we will see exponential growth in draw calls and be back at v7 situation again.

 

You can write pop music songs that sound good in the car and with ear plugs, or you can write a concerto that scales  all the way from the ear plug device down to Albert Hall.

 

ABBA vs Mozart.

Share this post


Link to post
Share on other sites
6 hours ago, moppel said:

Well i tested Multicore enhancement by deactivating 4 of my 6 Cores so basicly a 4,4 GHz I3 with some benifits ( Quadchannel, bigger Cache )

In the past i didnt note a Performancedecrease but with V 8 i get lower FPS and GPU usage (60-90) an a slight Performance impact . But even at this setup I3 on steroids its playable with 60-70 FPS.

So iam with you on this good improvment in V8 But still heavly dependent on IPC

 

i3simulated2lazr.jpg

 

For these tests it is better to deactivate the physical cores and have HT turned off.

You do not want OS logic to become an unnecessarily large variable in this test.

 

Turn off 3 cores and observe cpu utilization behavior with windows monitor logging. The fps isnt interesting whats interesting is how the cpu is being used by the app under certain conditions.

 

For example a cpu with 100 Gflop but with 1.5× ipc at same clock will out perform a cpu with 300 Gflop with 1.0x ipc. This is a problem in squad at the moment. The scalability of hardware is focused on one thing, brute force single thread.

 

As dev pointed out this could be something the UE devs will solve on their own.

Share this post


Link to post
Share on other sites
3 hours ago, IrOnTaXi said:

Indeed.  we could.. be we aren't going to start multithreading everything when we can optimize what we have an introduce enough scalability to let AMD systems play..  its sad but UE4 simply isn't built well for AMD systems..  we will do what we can. (we own to AMD systems in the studio now for profiling)  

 

why not try some extreme temporal solution, let them turn off the animation (only static stances like standing\crouching\prone), will that help amd users? 

 

Share this post


Link to post
Share on other sites
1 hour ago, rincewind said:

why not try some extreme temporal solution, let them turn off the animation (only static stances like standing\crouching\prone), will that help amd users? 

 

Hmm if u make a parallel animation system with fewer variables u could turn down message rates in theory and reduce draw calls.

But i think it would induce a sense of lag since the draw call reduction would come from reduced message rates to sub systems?

For everyone else on the server using the regular animation system it would look like the amd users were lagging since they pool with fewer ticks.

If u keep message rate to sub system the same then the effect on amd sys would be local only and give no drawcalls benefits since client server messaging would be the same.

 

Dice tried this (kind of) in bf3 using prediction algorithms. This resulted in "dead behind corner" and "killed behind cover" when the algorithm failed to predict actual player client position.

Share this post


Link to post
Share on other sites

UE4 is yet to become multi threaded, that's a fact and Epic is working on it. The great mistake Epic did is not making the engine multi-threaded at day 1 of release. Even after 10+ releases, they are yet to introduce it fully. Just day before yesterday I asked on their forums on when people can expect optimum usage of all cores and one of them replied that DX12/Vulkan will provide massive increases in performance across all platforms and games since these would allow all cores to function. Theres no ETA available though.

Share this post


Link to post
Share on other sites
On 11/3/2016 at 0:08 AM, rincewind said:

why not try some extreme temporal solution, let them turn off the animation (only static stances like standing\crouching\prone), will that help amd users? 

 

actually yes.. that is exactly what we are working on as it doesnt exist natively in the engine.. we are in the testing phase now. 

 

10 minutes ago, shiv1990 said:

UE4 is yet to become multi threaded, that's a fact and Epic is working on it. The great mistake Epic did is not making the engine multi-threaded at day 1 of release. Even after 10+ releases, they are yet to introduce it fully. Just day before yesterday I asked on their forums on when people can expect optimum usage of all cores and one of them replied that DX12/Vulkan will provide massive increases in performance across all platforms and games since these would allow all cores to function. Theres no ETA available though.

we don't ask.  I trust they have a plan as they have shown us in the past. 

 

I think we are finally at the point where we have much more control of both the client and server side to the point where I am comfortable saying.. we can deal with it sooner rather than later. 

 

It took a lot of time, learning and profiling to come to this point. 

 

 

Share this post


Link to post
Share on other sites
On 11/3/2016 at 2:09 AM, old_Sneakers said:

Hmm if u make a parallel animation system with fewer variables u could turn down message rates in theory and reduce draw calls.

But i think it would induce a sense of lag since the draw call reduction would come from reduced message rates to sub systems?

For everyone else on the server using the regular animation system it would look like the amd users were lagging since they pool with fewer ticks.

If u keep message rate to sub system the same then the effect on amd sys would be local only and give no drawcalls benefits since client server messaging would be the same.

 

Dice tried this (kind of) in bf3 using prediction algorithms. This resulted in "dead behind corner" and "killed behind cover" when the algorithm failed to predict actual player client position.

 

We have introduced a whole bunch of very small under the hood systems to scale various functions..  when they all finally accumulate into the completed product we should have very good control of the outcome on different systems. Some parts by tweaking,, some parts by varying update rates dynamically and in addition each time we refactor a core system written early on the gains are enormous. 

 

We are also working on a "roll our own" instancing system which should mesh very well with UE4's already nicely functioning native vegetation system. Remains to be seen how far we can push right now but we are keenly aware of the draw limits of certain manufacturers systems. ;) 

Share this post


Link to post
Share on other sites

@IrOnTaXi

 

Exactly. We have full trust on Epic and they'll deliver. After all, UE4 is powerful and it needs enough power from all the cores. I believe some support DX12/Vulkan will be seen in 4.14 or 4.15 releases.

Edited by shiv1990

Share this post


Link to post
Share on other sites

I didnt want to open a new thread or hijack this, but speaking of single-core performance would there be an advantage from switching to a 4790k from a 4670k (4ghz OC) ?
There is no chance in getting a higher clock, tried everything but it capps out at stable 4 ghz, everything beyond gets a bsod over time.

Share this post


Link to post
Share on other sites
35 minutes ago, BretterBerni said:

I didnt want to open a new thread or hijack this, but speaking of single-core performance would there be an advantage from switching to a 4790k from a 4670k (4ghz OC) ?
There is no chance in getting a higher clock, tried everything but it capps out at stable 4 ghz, everything beyond gets a bsod over time.

4.8Ghz stable  is almost a given from 4790K's at reasonable voltages and temps so yes I would say it's worth it.

Share this post


Link to post
Share on other sites
24 minutes ago, Shoshin said:

4.8Ghz stable  is almost a given from 4790K's at reasonable voltages and temps so yes I would say it's worth it.

Well statistics say otherwise (https://www.3dcenter.org/news/umfrage-auswertung-wie-hoch-laesst-sich-skylake-der-praxis-uebertakten)last generations seems to clock around the same  and 4,8Ghz at reasonalble( whats reasonable here 1,3V 1,35V ,1,4V ? ) Voltages is above avarage CPU quality , its allways a big gamble ,temperaturwise you need to be lucky too ,to get a Sample with decent applied TIM. but when aiming @ 4,8 GHz i would delid anyway.

I would stick to the 4670k Squads benifit from more cores is very limited , Maybee you can realise 4,4 ~ 4,5 Ghz .

Share this post


Link to post
Share on other sites
4.8Ghz stable  is almost a given from 4790K's at reasonable voltages and temps so yes I would say it's worth it.


4.8 GHz is more uncommon than you think.

Share this post


Link to post
Share on other sites
25 minutes ago, LookAss said:


4.8 GHz is more uncommon than you think.

ah right, I am lucky then as mine goes to 5Ghz game stable!

Share this post


Link to post
Share on other sites

4,4 or 4,5 are pretty much unreachable. cant put that much voltage on my cpu. Thats why i think about changing to a 4790k, it's the last quick 1150 cpu aswell.

 

Share this post


Link to post
Share on other sites
ah right, I am lucky then as mine goes to 5Ghz game stable!


Yes, you are VERY lucky! Mine goed up to 4.8GHz, but it's difficult to keep the temps under control with a Deepcool Maelstrom 240.

Share this post


Link to post
Share on other sites
1 hour ago, BretterBerni said:

4,4 or 4,5 are pretty much unreachable. cant put that much voltage on my cpu. Thats why i think about changing to a 4790k, it's the last quick 1150 cpu aswell.

 

Whats your Core Voltage for 4 Ghz ?

Share this post


Link to post
Share on other sites

Well thats the silicium lottery got a low grade sample this time :-) but a 4Ghz Haswell should still be okay to play it after V8 ,would keep it .

Well remember the gloriuos Sandybridge generation when every forum guy told you they all reach 4,6 and more Ghz easy ? My 2500k needed 1,210 V for 4 Ghz and allmost 1,4V for 4,5 GHz but had still some hickups in Prime with 800 ish FFt size .

 

 

Share this post


Link to post
Share on other sites

Yes, kinda disappointed that some clusters are that bad in production, but i'm glad i didnt even payed more money for a z-chipset instead of the h-set im running now.
Still thinking about going for a 4790k, this should atleast give me the room up to 4,6 GHz and the last hoorah for the 1150 socket.

Share this post


Link to post
Share on other sites
3 hours ago, BretterBerni said:

Yes, kinda disappointed that some clusters are that bad in production, but i'm glad i didnt even payed more money for a z-chipset instead of the h-set im running now.
Still thinking about going for a 4790k, this should atleast give me the room up to 4,6 GHz and the last hoorah for the 1150 socket.

 

I think thats the root cause of your OC Troubles--> H Chipset  is unable OCing over the unlocked Multi the 4670K has , so you are forced to raise BCLK , most systems get unnstable @ around 105 Mhz . 

Share this post


Link to post
Share on other sites

From memory I need to set my voltage to 1.412v for 5Ghz game stable, it runs with HT off at around 80c (in testing with CPUz 4 x100%) on a i7 4790K.

 

I generally run at 4.8Ghz with voltage to auto and HT off these days, seems a nice balance.

Edited by Shoshin

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×