Jump to content
feathers632

Ryzen versus 8700k IL2 and VR

Recommended Posts

That only applies if you are GPU limited, if CPU limited, auto super sampling does nothing.

Share this post


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

That only applies if you are GPU limited, if CPU limited, auto super sampling does nothing.

 

It maintains smooth framerate and when your gpu can handle it thr SS is pushed up.   Elite D looks amazing since the the new steamvr update and IL2 remains smooth. 

Share this post


Link to post
Share on other sites

any of you had a chance to try il2 with the 2700x yet ? I cant find any VR performance reviews

 

 

Share this post


Link to post
Share on other sites
8 minutes ago, =FEW=Herne said:

any of you had a chance to try il2 with the 2700x yet ? I cant find any VR performance reviews

 

 

A comparison test was performed on that other flight sim yesterday.

I know that is like comparing apples and oranges but it does give an indication of performance.

https://forums.eagle.ru/showthread.php?t=206341&page=3

Share this post


Link to post
Share on other sites

So, I finally had a chance to play a little bit with the new setup. I've purcheased Ryzen 2700X, Asus X470 Crosshair VII Hero, G.Skill 3600 Mhz Cl15 16 Gb set and Samsung 960 Evo M.2 SSd. Assembly took me some time but then startup and simple setup were easy. I managed to overclock Ram with no effort to 3600 Mhz with tight CL 15 timings (used default Stilt preset for 3466 Mhz) which caused a massive leap in performance. This CPU needs fast RAM due to the way Infinity Fabric works. I may try to approach 3800 or perhaps a 4000 Mhz at a later date.

 

Second thing is that you actually do not OC the CPU but instead use onboard tools to relax the limits so that XFR can boost longer and with more cores to 4.35 Ghz (which is the speed limit). I'm currently running PE3 state which seems to be fine for gaming. The last thing is that I could try a bclk setting, its to set bclk of no more than 105 as it turns of XFR. For this motherboard a most likely stable setup would be 103.4 bclk with +25 mV offset. But on default settings this CPU has a lot of short spikes and bursts even at PE2 (which is non OC) to 1.5 Volts so I'm not going to do that. Perhaps I will later try 101 bclk but Im not sure its worth it. There certainly is a lot of work on BIOS for this board so I might wait and see if it becomes easier to play with features without that much voltage.

 

Either way, using testing setup from a VR thread I got following results which honestly I find inconclusive. So for the record, my current rig is as follows : GTX 1080Ti, Ryzen 2700X at 4.25 to 4.35 Ghz, 16 gb G.Skill 3600 Mhz Cl 15, Samsung 960 Evo 512 Gb. 

First is Passmark and Cinebench:

OUVH5w.png

The 2600k is from my previous system, I made sure to keep the records. 177 pts was for 5.1 Ghz, 168 for 4.8 Ghz. As you can see even at stock settings 2700X beats in Cb 15 overclocked to the limits 2600K. On the other hand passmark results are lower for Ryzen, since 2308 is what I got here while 4.8 Ghz 2600k would score near 2600 pts. Seems each benchmark favors different set of instructions.

 

Here is 1920x1080 with standard setup, balanced power mode in windows:

2018-04-28 14:31:16 - Il-2
Avg: 114.990 - Min: 69 - Max: 200

 

Here is 3440x1440 with settings on ultra, balanced power mode in windows:

2018-04-28 15:07:34 - Il-2
Avg: 106.533 - Min: 76 - Max: 146

 

Keep in mind those results may change around as I learn more about the cpu. I havent overclocked AMD cpu since 9 years and a lot has changed, this is what I could get trying simple things after 2 or so days. Also, what Ive discovered is that unlike some people claim, this game is single core only. Both my HWinfo and MSI Afterburner indicated that out of 8 cores only one was loaded to about 70-82 % and two to three others were between 10 - 20 % which I think is more related to working in the background software like trackir, afterburner, etc.

 

This cpu is certainly an upgrade for me, both in DCS and Il-2 game became more smooth and there are less FPS fluctuations. In DCS I no longer "enjoy" game stutters :) On average I'd say I have 20-30 FPS more, depending on situation and load. Perhaps if Il-2 follows DCS trend and makes an effort to implement multi core support with a change of API to Vulkan (or really any other way, its 2018, utilization of one core only seriously limits hardware members have this days) we will see a serious performance gains.

Share this post


Link to post
Share on other sites

Nice! Great to see that AMD caughtup to Intel in single thread performance in a many core CPU. It is very close to my i9-7900X. I have 2 additional cores, putting multithread score to about 24000. But this is not really needed here. Cool. 

Share this post


Link to post
Share on other sites

Yup. Quad core cpus became mainstream at least since first i7 series which was over 8 years ago, back in 2011 first eight core chips entered market but only now we move to those as mainstream. With more and more people getting above 4 cores its a big waste not to utilize so many available resources. I hope this will finally happen during BoBp development cycle. 

Share this post


Link to post
Share on other sites

Prepar3D v4 uses all cores available already. This mainly for autogen, the most load intensive task. The 2700X will be a great CPUfor that. On my system, I can use 10 cores / 20 threads almost 100% when balancing CPU tasks with GPU overhead. The 7900X is about twice as fast as a higher OC‘d 7700K. Plus you have less microstutters and Outlook in the background is still no issue.

 

What we really need with DX11 is the possibility to heavily OC one or two cores and bind the main simulator threads to those exclusively. Many other tasks (such as autogen, air and road traffic, systems,...) scale well over many cores. There is no requirement to OC them, just get many cores.

 

Once the sim is DX12 or Vulcan, you can multithread the main thread as well and your 36 core CPU will be very, very useful...

Share this post


Link to post
Share on other sites
On 4/29/2018 at 10:11 AM, =362nd_FS=Hiromachi said:

So, I finally had a chance to play a little bit with the new setup. I've purcheased Ryzen 2700X, Asus X470 Crosshair VII Hero, G.Skill 3600 Mhz Cl15 16 Gb set and Samsung 960 Evo M.2 SSd. Assembly took me some time but then startup and simple setup were easy. I managed to overclock Ram with no effort to 3600 Mhz with tight CL 15 timings (used default Stilt preset for 3466 Mhz) which caused a massive leap in performance. This CPU needs fast RAM due to the way Infinity Fabric works. I may try to approach 3800 or perhaps a 4000 Mhz at a later date.

 

Second thing is that you actually do not OC the CPU but instead use onboard tools to relax the limits so that XFR can boost longer and with more cores to 4.35 Ghz (which is the speed limit). I'm currently running PE3 state which seems to be fine for gaming. The last thing is that I could try a bclk setting, its to set bclk of no more than 105 as it turns of XFR. For this motherboard a most likely stable setup would be 103.4 bclk with +25 mV offset. But on default settings this CPU has a lot of short spikes and bursts even at PE2 (which is non OC) to 1.5 Volts so I'm not going to do that. Perhaps I will later try 101 bclk but Im not sure its worth it. There certainly is a lot of work on BIOS for this board so I might wait and see if it becomes easier to play with features without that much voltage.

 

Either way, using testing setup from a VR thread I got following results which honestly I find inconclusive. So for the record, my current rig is as follows : GTX 1080Ti, Ryzen 2700X at 4.25 to 4.35 Ghz, 16 gb G.Skill 3600 Mhz Cl 15, Samsung 960 Evo 512 Gb. 

First is Passmark and Cinebench:

OUVH5w.png

The 2600k is from my previous system, I made sure to keep the records. 177 pts was for 5.1 Ghz, 168 for 4.8 Ghz. As you can see even at stock settings 2700X beats in Cb 15 overclocked to the limits 2600K. On the other hand passmark results are lower for Ryzen, since 2308 is what I got here while 4.8 Ghz 2600k would score near 2600 pts. Seems each benchmark favors different set of instructions.

 

Here is 1920x1080 with standard setup, balanced power mode in windows:

2018-04-28 14:31:16 - Il-2
Avg: 114.990 - Min: 69 - Max: 200

 

Here is 3440x1440 with settings on ultra, balanced power mode in windows:

2018-04-28 15:07:34 - Il-2
Avg: 106.533 - Min: 76 - Max: 146

 

Keep in mind those results may change around as I learn more about the cpu. I havent overclocked AMD cpu since 9 years and a lot has changed, this is what I could get trying simple things after 2 or so days. Also, what Ive discovered is that unlike some people claim, this game is single core only. Both my HWinfo and MSI Afterburner indicated that out of 8 cores only one was loaded to about 70-82 % and two to three others were between 10 - 20 % which I think is more related to working in the background software like trackir, afterburner, etc.

 

This cpu is certainly an upgrade for me, both in DCS and Il-2 game became more smooth and there are less FPS fluctuations. In DCS I no longer "enjoy" game stutters :) On average I'd say I have 20-30 FPS more, depending on situation and load. Perhaps if Il-2 follows DCS trend and makes an effort to implement multi core support with a change of API to Vulkan (or really any other way, its 2018, utilization of one core only seriously limits hardware members have this days) we will see a serious performance gains.

 

whats your FPS like in VR with the 2700x ? been waiting to see some feedback before I pull the trigger

Share this post


Link to post
Share on other sites
16 hours ago, =362nd_FS=Hiromachi said:

I just wish that game would use 4+ cores and that would be already a big leap. No need for 36 ;) 

If it can take 4+ cores instead of 2, it can take 36. And then you can have 256 AI bombers in one mission at the time and you rather kill yourself than scaling back.

Share this post


Link to post
Share on other sites

In C++:

thread AIThread0(aiUnits, attributes, moreAttributes, ThreadID);

...

 

To close the thread when finished:

AIThread0.join();

...

 

Do this as many times as you want. Want to split AI across 128 threads, copy paste 127 times. As for how this exactly works in code, the command "thread" creates a new program thread for the CPU to handle. "AIThread0" is the name of that new thread, aiUnits is the function called in this new thread (you can only apply threads to functions, not individual lines of code). The aiUnits() function can then call additional functions itself which will remain within its own thread. The attributes, moreAttributes and ThreadID are all attributes of the aiUnits() function. If called without creating a new thread, it would look this:

aiUnits(attributes, moreAttributes, ThreadID);

 

Obviously you can put as many or as few attributes within the function as needed.

 

Since timing does matter in a game, some sort of synchronization will need to occur. The simplest way is to force all the threads to reference the same clock time for each frame. This is trivial and not worth describing as even single threaded games from 30 years ago do this. Nothing is changed except some possible gating depending on how exactly it's coded.

 

 

The catch with all of this is the game must use functions instead of being one giant monolithic thing inside main(). Additionally, this can only be done with game logic (AI units, physics, sound, netcode and so on) in a DX11 game. The graphics pipeline cannot be effectively multi-threaded.

Edited by BeastyBaiter
  • Thanks 1

Share this post


Link to post
Share on other sites

One thing I forgot to mention, the idea in the above is that you would split AI units across multiple iterations of the above command. How exactly they would be split would take some thought. A naive approach would be to separate them by unit type, such as all Axis fighters on one thread, all axis bombers on a different thread, all allied fighters on a third, all allied bombers on a fourth and all ground units on a fifth. This is certainly not an optimal approach but it's massively better than no threading at all. This is doubly true if that AI stuff is done on the rendering thread. A more complicated but better alternative would be doing something along the lines of all Ju-87B's and Pe-2 s.87's on one thread with all Ju-87D's and Pe-2 s.35's on a different thread. That approach would better balance load if thought out, but would be more work. An optimal approach would look at each unit spawned and try to balance across all available threads. This is a far more complicated algorithm and I have never attempted anything like that.

Share this post


Link to post
Share on other sites

Hmpf, his scores are on the lower end of results I've seen. Some of it due to moderate ram speed and some due to overclocking variables. What he means by autooverclock is probably built in Performance Enhancer, but there are four states. One and two that are within AMD limits, PE3 that carries some OC and PE4 for those who won silicon lottery and have a custom cooling loop.

Dont know which one he uses.

 

But neither pushes CPU beyond 4.35 Ghz as for that you need a bclk overclock, which messes up ram speeds and stability. Probably the best right now to have stable oc and reasonable temperatures/voltages is PE2 with 3466 mhz low latency ram. As high speed ram with low latencies really does matter for Ryzen:

 

 

Edited by =362nd_FS=Hiromachi

Share this post


Link to post
Share on other sites

The consensus on the 2700x from the tech tubers I trust most seems to be that trying to overclock actually reduces performance in many cases since XFR 2.0 is better at overclocking than most humans. The proper way to overclock it seems to be using a beefier cooler so that XFR can push harder.

Share this post


Link to post
Share on other sites

Well yes, though Wraith Prism can do the trick quite well. Its a well designed cooler with heatpipes and decent fan. Furthermore they even put a good quality paste on its base. What a contrast to what Intel offers.

Anyway, if you want to let Precision Overboost go wild then most likely you will need AIO or one of the larger tower coolers like ones Noctua manufactures. But its also about sufficient voltage, load line calibration and few other tricks. Personally I find it interesting the new ways of pushing this cpu beyond specs. So refreshing from this repeatable multiplier oc. 

Share this post


Link to post
Share on other sites

Aha, only just found this!

 

Went for a Ryzen 2700X myself, with a X470 MB, a 1060 and 16gig (but 24mhz ram)... Haven`t had a chance to run BOS with it as I`m waiting to replace my Tracker clip.

 

So far it`s thrown around everything I`ve tried on it no probs and while slower than the high I7s it certainly don`t feel like it. The Wraith  Cooler is impressive and hasn`t even started heaving yet, very quiet. This pc is for work so I need something that can be a workhorse, not just gaming.

Share this post


Link to post
Share on other sites

 

"The consensus on the 2700x from the tech tubers I trust most seems to be that trying to overclock actually reduces performance in many cases since XFR 2.0 is better at overclocking than most humans. The proper way to overclock it seems to be using a beefier cooler so that XFR can push harder."

 

That might be the case for average overclockers but those who know what they're doing well get better performance manually.

 

 

Edited by feathers632

Share this post


Link to post
Share on other sites

It's not quite true. Or well, it requires both inputs.

XFR 2.0 and Precision Boost will boost up to 4.35 Ghz but that happens rarely due to termal and power limits imposed on default settings. All core boost will probably stop at 3.95 Ghz.

Manual overclocking of all cores, like most do by simply increasing multiplier is also not the best way since it increases performance of all cores equally and beyond certain point it turns off XFR. Thus, you will get best results in multithreaded aplications if you manage to push the cpu to 4.15, 4.2 or if you have balls - up to 4.3 Ghz. But you will achieve worse results in those applications which do not require 6, 8, 10 and 16 threads but only 1-4. 

 

Il-2, DCS or similar games lacking multithreading optimization are this kind of software. And here to achieve best results you should take advantage of XFR and PBO while doing some manual OC. What I'm running right now is exactly that. I'm boosting on few cores up to 4.4 Ghz and all cores to 4.1 Ghz due to my motherboard function (Precision Boost Overdrive Level 3) which drops some of the termal, power and other limitations (still leavin cpu within safety limits). I've also bumped slightly base clock to 101.2, insteand of default 100 since this allows precisely to get higher clocks. I've seen some people going for 103.4 to 104 even, but I'm using four sata drives and base clock is a very tricky thing to work with, especially with numerous sata, usb, pcie and other things. I've tried it and could run few benches at 103, which provided me with almost 4.5 Ghz, but system froze eventually and I'd probably have to move from negative voltage offset to positive one in order to make it stable. Not worth it for me, especially since temperatures raised substantially and cpu starts climbing to 1.51 V. Again, not worth it for me.

So currently I'm running up to 4.4 Ghz with following setup: PBO Lvl 3 + 101.2 bclk + Core Enhancement enabled + Load Line Calibration 3 + Voltage in offset mode with - 0.0375 V. My RAM is running at 3508 Mhz with Cl 14 and very tight timings. Some people state that you should leave LLC on Auto, but it gives me too big Vdroop and hence I get freeze on cpu tests. LCC3 fixes that and allows to run greater negative voltage offset.

There are some videos on youtube showing people on Asrock Taichi, Gigabyte X470 Aorus or other high end boards pushing cpu that way (with PBO + bclk OC) to 4.45 Ghz - 4.5 Ghz. Two days ago I saw a guy who managed to get on one core 4.62 Ghz, that must have been a well binned cpu. So there is a lot of potential in those cpus.

 

I'd say there is some work to be done with improving things but I will wait with that for more matured bios (current Asus Crosshair VII bios is not that great) and probably better cooling. Current AIO proved to be an upgrade over previous SilentiumPC Fortis 3, but pump is loud and not very efficient (even with 360 mm radiator) so I've started preparing for a switch to a custom loop with proper 360 mm radiator from Hardware Labs and XSPC Raystorm Pro cpu block. Probably wont get there any time soon, but I always like to plan things in advance. In the meantime I can fully recommend Ryzen 2xxx series as a good alternative to (boring) Intel cpus.

  • Upvote 1

Share this post


Link to post
Share on other sites

Did anyone managed to overclock Intel or Rzyen over 1.2 Ghz over stock, anyone? I mean for example 8700k 4.3Ghz + 1.2Ghz = 5,5Ghz or 2700x 4.0Ghz + 1.2Ghz = 5.2 Ghz with reasonable Vcore between 1.35 - 1.4V? Nobody? I remember the time where I overclocked my Q6600 G-0 Stepping from 2.4Ghz to 3.6Ghz with a Vcore of 1.35V using air cooling where it was worth to overclock. Nowadays you have Turbo Boost or XFR whatever that overclock GPU and CPU automatic - the chips run already in their limits. Who care that a 8700k can do 5Ghz that is +700Mhz over stock ram him 1.42V and almost every 8700k run 5Ghz........and Ryzen ram him beyond 1.42V and almost every Ryzen runs 4.4 Ghz that is +400Mhz over stock.

 

The overclock results are me too low to even play with or even waste money for better motherboards , higher clocked RAM just to play silicon lottery for really marginal improvements.

 

 

 

 

 

 

 

 

 

Share this post


Link to post
Share on other sites

VR renders at either 22ms per frame or 11ms per frame. That equates to 45 and 90 fps respectively. With modern Intel cpu's (Haswell or later), the jump from 45 to 90 fps under low CPU load situations in BoX occurs between 4.5 and 4.7 GHz. Everything beyond 4.7GHz helps stabilize the frame rate at the higher 90 fps mark as the action increases. So yes, the difference in clock speeds is fairly small, but the end result is double the framerate in VR provided  you have a GPU to match.

 

As for max overclocking, I've taken my 8700k to 5.0 GHz but backed it off pretty much immediately due to thermal and long term voltage concerns. That's 700MHz over the out of the box 4.3GHz all core turbo, for an increase of 16%. My daily setting is 9% faster than stock while also running cooler due to lower voltage and power draw.

Edited by BeastyBaiter

Share this post


Link to post
Share on other sites

As BeastyBaiter, I also have backed down my OC for 24.7 use. I run now at 4.8GHz for daily use, but with also better temps and lower voltage. It does indeed seem unimpressive but still the gains are evident, and with lower temps and wattage it's better long term.

Share this post


Link to post
Share on other sites
6 hours ago, 335th_GRMultiDoc said:

As BeastyBaiter, I also have backed down my OC for 24.7 use. I run now at 4.8GHz for daily use, but with also better temps and lower voltage. It does indeed seem unimpressive but still the gains are evident, and with lower temps and wattage it's better long term.

Using offset vcore voltage?

Share this post


Link to post
Share on other sites

Do you play 720p to enjoy 15% gains with an 5.0Ghz 8700k and call it already evident - Years ago i had easy more than 30% gains what I had with my overclocking experience......................

No SLI no gains, plain and simple............................nowadays

 

Spoiler

https://www.youtube.com/watch?v=BjgyuzSDkn8

 

 

I think I wait a little more for the Intel Core i7 8086K take look to the new chipset Z390 (intel), B450 (AMD) and their price point. But at the moment the 2700x cost me +$100 more and is slower than the 8700k - RiP Ryzen at the moment.......

 

My overclocked i7-2600k @ 4.5 Ghz still do his job fine so i can wait a little to see if an Rzyen 2800x or an Core i7 8086K comes out

 

 

Share this post


Link to post
Share on other sites

I play at 2560x1440 and always use the max settings possible (noty only in the game itself but from the Nvidia control panel also to override game settings if need be), so yes even a 10% gain on performance is noticeable (at least for me it is). And in any case it doesn't make much sense (for me) not to try and push the hardware at max.

As a side note: I never understood why someone would play at 1080p having hardware that would give good performance at 2k resolution. I don't play competitive FPS so no need to chase hundredths of fps.

 

As for my hardware setup, my current system was build after years on my previous system with a 3570k and a GTX970 (both overclocked since day 1 for years with no problems).

Share this post


Link to post
Share on other sites

Well, in my case R 2700X price was substantially lower than 8700k. Motherboard was also cheaper than comparable model for 8700k. The only expensive thing was RAM, since I purcheased the best binned RAM on the market. But I'd do the same for Intel, I just wanted to have fast RAM no matter what. And I've upgraded from 2600k running at 5.0 Ghz. Personally I consider it a noticeable upgrade, in Il-2 and DCS I've gained 15-20 FPS and less stuttery experience.

Point is that I did not have to fight with poor IHS design, toothpaste under it and need to delid (and void warranty). I will most likely swap it for Zen 2 as soon as it drops, but I am less and less convinced that Intel has much to offer beyond their Skylake arch running on 14++ nm process.

Share this post


Link to post
Share on other sites
4 hours ago, =362nd_FS=Hiromachi said:

Well, in my case R 2700X price was substantially lower than 8700k. Motherboard was also cheaper than comparable model for 8700k. The only expensive thing was RAM, since I purcheased the best binned RAM on the market. But I'd do the same for Intel, I just wanted to have fast RAM no matter what. And I've upgraded from 2600k running at 5.0 Ghz. Personally I consider it a noticeable upgrade, in Il-2 and DCS I've gained 15-20 FPS and less stuttery experience.

Point is that I did not have to fight with poor IHS design, toothpaste under it and need to delid (and void warranty). I will most likely swap it for Zen 2 as soon as it drops, but I am less and less convinced that Intel has much to offer beyond their Skylake arch running on 14++ nm process.

 

That was pretty much my reason for ditching intel.  I'd only ever built intel systems for myself and others since 486dx days.  Last intel was i5 2500k.  Last year ditched it for my first ever AMD.  Was nervous about it and months later I came close to running back to intel but instead went back into the bios and set manual advanced timings for low latency.  Gave a massive boost in performance noticeable from windows boot to windows desktop and games.  My 1700 was at 3.85ghz and 4000mhz ram clocked at 3466.  So to clarify... even though I'd had that ram and cpu overclock... the full performance was locked away until I manually set advanced ram timings.  I'm now happy with my AMD and will switch to 2700x at some point.

 

 

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

×
×
  • Create New...