Separate names with a comma.
Discussion in 'General Discussion' started by garyjpaterson, Jun 23, 2014.
Stock Ryzen 5 3600 with stock cooler, and 3200mhz CL 14-14-17-14 ram
Feels so good. Benchmark done with Teamspeak and Chrome running in the background but i don't think that matters too much.
I'm in the process of replacing my old i7 4790k that got around 175 MBeams at 4.5Ghz.
Replacing it with my new and shiny 3900X lead to 447 MBeams. That's an improvement.
Precision : I'm still running it full stock, with the included cooler. It's loud AF, but it get the job done till I receive my watercooling stuff
Max temp during this test was 72°, frequency was 4.1 on all cores after 20 cars.
I'm using 3600Mhz CL16 sticks.
my non k oc I7-6700 seems to perform pretty well, paired with a GTX-1070 it even runs quiet smooth.
I know it probably isn't much, but i just wanted to compare it when i upgrade to a Ryzen 3600
I don't really understand these numbers, but i tried the game and the difference in performance is huge.
Just built a 3900X 12c/24t system for a buddy.
Can't remember the exact numbers but shot that things flies.
So now that my 3900X is under a custom watercooling loop, I tried out to OC it one CCX at a time, at a fairly conservative 1.32V. Each core was getting 1.28V final because of load line calibration not set to the max. I did it in an hour or so, I'm still far from the limit with a 61°C max and 110W, with 4.45 to 4.3 Ghz fixed across CCXs.
Bananabench said 472 in this case. Now I really want to see the 3950X and new threadrippers in action .
For some reason Ryzen seem to have a drop after numOfCores -1, like you have high result with 11 vehicles and with 12th vehicle there is a drop.
I have seen same with 2700X too, it's numOfCores -1 with that too.
With my CPU though drop happens with 7 vehicles, so with 6 cores that is one vehicle later than numOfCores.
I wonder what could be reason for that, would think it being something in CPU design. Probably it does not matter much, but it is bit odd to see such constant difference.
Your 4.4Ghz is probably equivalent of around 4.6Ghz 8th gen intel, so with 4.6-4.7Ghz you would get around same performance per core as 8th gen intel with around 5Ghz, maybe.
That paired with 12 cores is really good for this game.
At a guess, it's probably because one core has to handle the rendering, inputs, and possibly synchronisation of the other cores. So as soon as you ask it to also process a car, you are asking for more than one core can do.
Your CPU also has to run an operating system and any background services, but I have never looked into how they are allocated between cores. Its possible that it allocated them to the spare core, and when that was taken decided to spread them out across all core evenly, pushing them over the limit.
I would expect to see the same behaviour on Intel, but it's possible that their per core strength is enough to get around that.
I think he knows that, maybe the drop off is larger than intel equivalent.. (possibly something to do with ryzen schedulers and cpu design like he said)..
But i don't have a ryzen so can't comment really yet
Yeah, in bananabench there is no those other things running, but consistently I have seen Ryzen's having a drop 1 vehicle before same core number Intels, but I don't think it has much meaning in performance wise, it is more of odd thing that makes curious mind curious.
Who knows maybe it is even within Windows 10, I have mostly examined my CPU compared to these new CPUs in order to figure out where they stand in various usage cases as mostly everyone just looks how they run with all cores fully loaded.
This is 4.8Ghz and as you can see from Patate_Folle's results, his CPU is pretty close to this up to 6 vehicles even running at lower frequency and from 7 vehicles onwards Ryzen really excels, it is very good CPU.
With 6 cores Ryzen has usually lower result for 6 vehicles than for 5 vehicles and with 12 cores it seems 11 vehicles is higher than 12 vehicles, while Intel seems to be higher score with 6 vehicles than with 5 vehicles.
That is when we look MBeams/s, if we look % realtime story is different as can be seen.
So obviously it is interesting why such happens, is it Win 10, CPU or just something random. Random would be unlikely as same thing I have been seeing when I compiled lot of these results together, there was a thread about those results too, which I think I should update when I have time for that.
Also I hope nobody attacks me from pointing this out, that Ryzen model most definitely is the one to get for BeamNG, IPC is great, can get enough clock speed and has 12 cores that matters.
Hyper threading / SMT does not much for number of vehicles one can run, at least with my testing, so more cores while still having IPC and clock speed is way to go for best possible performance.
A small 2% update on this, I bought a better cooler (Snowman T6):
Maybe that is something to do with the way the OS loads in priority the "real" cores, even if it has to do OS stuff as well on them and then start to use SMT cores when the "nbCore + 1" th thread pops up.
Maybe on intel side it starts to use HT before that?
This is just a theory I may try out later along with higher clock speeds...
My first thoughts was because i'm still running stuff in the background but as you pointed this out, it's clearly not the only cause.
Another test i made was to pop as many cars as possible on a map (pickups - Italy), checking the fps between each. I went up to 16 cars but the FPS drop was very consistant across all of them, without these dips at coreCount-1 ... Weird? maybe. Didn't made it to 24 because of GPU though.
Added a quick graph of the perf on italy map
It might be just way Bananbench is or something to do with Win8.1 (my OS) vs Win10, also if there is some software running on backround like browser, it already affects Bananbench score and with Win10 there is always something happening, so many variables and it indeed is not showing up in game at all.
Did you observe CPU load per core when you did test FPS at Italy? That is not so huge issue nowdays, but spawning several different cars vs spawning copy of one was different, but haven't tested much with latest version, LUA and drawcalls have been requiring fast single core speeds, I presume that you might also hit limit of that one highest usage thread at some point during to test.
Depends a lot from graphics settings of course. Performance in this game is pretty challenging, it can change so much depending from place, graphics settings etc. From CPU to GPU limited you get by changing camera angle if at limit of performance.
This view angle at Norte spawn is pretty tough for any system at high graphics and everything enabled, also it shows how difficult it is to see if one is CPU or GPU limited, 4.3Ghz but task manager shows it wrong most of the time:
GPU is not maxed out, CPU is not maxed out by task manager, but what really is happening is very quick spiking CPU load that task manager just is not picking up, same happens with Cities Skylines too.
With SSAO on, dynamic reflections on and with shadows set to all, I think drawcall count gets high, because when drawcall count gets really high this phenomenon happens and it is so with all games where drawcall count gets very high.
CTRL+F twice shows you drawcalls, usually if numbers is changing really fast without movement in scene, that tells it has gone over board and CPU limitation is likely to happen, or that is what I have observed at least, might be useful, interesting or already known by you, either way I share what I have learned so it is easier for people to understand what is limiting performance:
Anyway, that is really sweet numbers that you are getting with Italy, solid 60fps with 12 cars is so close that with little squeezing that might happen. I think I need to test Italy again too, can't remember how many cars I could run before dropping under 60fps, no way 11 traffic cars for sure
What interest me greatly is the point where one core becomes limiting as that is how well game can currently use all cores, some aspects are single threaded by design so it is likely that there will be X number of cores that Y amount of single core performance can support.
What I can see from your test is that it is probably quite bit more than 4 real cores with my IPC and 4.8Ghz it used to be in older version of game.
I'm just obsessed with performance numbers
Yesterday I did run Bananabench with 4.3Ghz locked, got 285 Mbeams/s Might need to test with higher clocks and see what happens.
Between 4.3Ghz and 4.8Ghz I gain whole 10% CPU performance improvement, so it is not really a lot.
Copying single vehicle I can run 12 with 60fps, but this is not at all same as your vehicle test, just to show how much performance can change and which makes it quite difficult to say anything certain, just so much variables:
Here is one car to compare this 4.8Ghz to earlier 4.3Ghz screenshot, might be interesting for some, gained whole ~8fps, so I mostly run with 4.3Ghz as it is much more quiet and cooler.
Can't run 5Ghz constantly because weak VRM in z370 motherboard:
Also as CPU power is increased 10% that would be 10fps, but I got around 8fps improvement, would GPU start to limit around there is anyone's guess, but pretty much CPU limited in Italy with 1 pickup truck as fps increases so much with overclock.
Got my new CPU / Mobo / RAM installed today:
And just like that, a ryzen CPU that doesn't do that. Thanks aljowen for proving this wrong lol.
Turns out I am as well...
These numbers were on Italy at full stock with the original cooler, so loud and hot. A new instance of the pickup from the "spawn new" menu each times aligned neatly on the highway entrance at Fastello. I will redo that later while looking drawcalls (wasn't aware ) and with higher clocks too.
Yeah, it is interesting how Aljowen's results show that there is no drop, I'm really thinking it must be something in Bananbench or windows doing, because if it would be hardware based, then his result should also have that drop.
I think that your fps in perfectly same conditions with 12 vehicles should be higher than mine. Afaik, it should not matter if you spawn individually same config or if you clone the vehicle, graphics optimizations should allow much more simultaneous vehicles then than if you spawn 12 completely different vehicles, where drawcalls increase much more and CPU single core performance becomes limiting factor earlier.
We should probably have pretty much same single core performance, but you have more cores so your performance should be higher than mine or equal.
Really great times for this game though with these kind of performance levels game starts to run way nobody could even dream 3 years ago.
In Highway entrance of Fastello with single D15, I got 110-120fps depending which way I was looking, using 4.3Ghz
My results were on a fresh windows install with no updates installed. The Windows installation media was created today though, so it is pretty fresh, but no platform specific updates. No motherboard included software has been installed, such as LED control software etc. No peripheral software has been installed for mice etc. Soundcard software has been installed (Xonar Uni Drivers). Otherwise, just Firefox, Steam, BeamNG at that point.
I ran the test with no other software open (other than explorer, Xonar would have been in system tray too come to think of it).
Updated the spreadsheet, still no R9 3950X... Who'll be the first one?
Also, I made some graphs (still not sure if they're right) for the market share in the BeamNG community of CPUs;
Blue is Intel and red is AMD
Don't mind the central graph as it was a transitory period of the Banana Bench. It seems AMD is becoming more popular (the total tested cpus are a bit more - around 100 more samples - in the first graph) now than before