Ryzen 3000

Has anyone already some experience with the new ryzen setup? I have sadly bought a new i9 9900k a couple of months ago and would like to know how the Ryzen 9 3900X compares to my i9 in beamng.drive.... but i am also curious how the other ryzen 3000s perform?

 

I would think that the 3900x would be a lot better for BeamNG considering that it has 12 cores, but I don't have much experience with Ryzen as my personal rig has an i5 8600k

 

The general CPU rule for Beam appears to be one car per thread minus one. eg A CPU with 4 cores/8 threads could handle 7 vehicles. As long as it's a 'decent' CPU, and not an Intel Atom or something. If you try to do more than that then 2 vehicles have to share a thread, and performance suffers.

Is that right?

 

3900X

 

Hello,
Thank you for your publication!
What frequency and timing does your RAM have? And is it the X570 motherboard?

 

I think the 3900x would be better because it has more cores/threads, and that can help alot with spawning in more vehicles and such.

 

It is currently running at 3200MHz, CAS-16-18-18-36 dual channel (2*8gb) in an x570 motherboard
--- Post updated ---

That was a 3900X

 

That score is around 3x higher than my i7 4790

Say I start dropping frames with 4 cars with my current chip, would that Ryzen allow me to have 12? Bearing in mind that the Ryzen has higher single core performance. This is making the assumption that my RAM and my GPU will not cause a bottleneck.

Just curious as to whether it scales linearly as games often don't work that way

 

I ran 14, didnt try any higher

 

You would need new RAM anyway as new systems require DDR4 and you have DDR3.

 

Imagine the 3950X with a few AGESA updates down the line.. holy shit.

Letting 16 Covets or Pessimas crash at the same time anyone?

 

Made it to 20 covets, a T65, a miramar, sunburst, burnside, roamer, hopper and bolide spawned at once, 40fps, grid map

 

BeamNG seems to be one of the games which scales nicely with Cores/Threads.
Would be nice if one of the i9 would load also 20 covets, a T65, a miramar, sunburst, burnside, roamer, hopper and bolide spawned at once, 40fps, grid map as comparison.

Also SMT Off is helping in ~50% of all games.
Also core affinity helps to increase performance again.
Microsoft is still doing it wrong even after the Ryzen Update last month.

 

Curiously my 3950X hits 288.XX MBeams/s
While SixSixSevenSeven's 3900X is hitting 380.XX Mbeams/s

My best guess is my 64GB ram @ 3000Mhz with much looser timings is holding the CPU back from performing quite that well.

It still runs 12-16 cars without much of a sweat, on a side note I do still run a 4yr old GTX 1080 @ 2560x1440.




Now worst case scenario I bought this ram as a place holder until next year when I can afford to get better parts and actually select things.. I got this ram for cheap as I really need the capacity personally.

 

I mean... its not the newer ryzen 3900x, but i still had a shitton of fun nonetheless.
I recently build myself a desktop with a 2700x and an rtx 2060 super.

16 cars, with the infamous big rig that melts cpu's for breakfast, while i was also recording.
i was still playing in the 20 fps ish. id say maybe higher 20's when i wasnt recording maybe.
for about 400bux cheaper then a 3900x. its still a good one id say, depending on where your budget stands.
Nice to see some numbers tho.

Apart from high core count, what should improve the fps in beamng?
what if we wanted to run beamng thru a multi cpu server, could we even do that? (offtopic sorry)
Im also curious to know how a Threadripper 3990X would handle beamng. (more core better, no? what about 64 core/128 threads?)
Does anyone on the forum even have one of those cpu? if so, mind to share numbers?

Cpu's are getting more affordable id say. i was about to go for intel until i noticed id have to pay up to 500$ minimum to get an equivalent of the 2700x. amd ftw.

 

Triple check your bios is actually set at 3000MHz, most sticks that fast get defaulted to much lower speeds. But yes, Ryzen and particular the multi die ryzens like 3900/3950 really benefit from higher speed memory in general, the infinity fabric used for inter core communication is synced with it so if that slows down, everything slows down.
I was also using a custom loop water cooler so also didn't hit the extremely premature thermal throttling that 3000 series chips can exhibit, even at 70c they will stop fully boosting
--- Post updated ---
Oh also quite possibly you're running against the wattage budget but I'm not sure how that would make an otherwise like for like 3950 Vs 3900 run differ, and it probably wouldn't account for such a huge difference in performance

 

I know about the Infinitely Fabric a little bit yes.

And yes the ram is @ 3000mhz.

It says 1500mhz>x on that program because you have to double what you see there to get numbers found in bios.

My bios is set to 3000mhz on a quad dimm 64GB kit
-
Coming from 2666mhz on 32gb of Corsair LPX I used on X99 (4 of 8dimm 64 kit) I've seen what I believe to be up to a 10fps increase when using 12 or more cars.

I don't think I was running around 45fps averages with 15 cars on west coast before that. Pretty confident it was around 36fps average.

All this is while streaming the game through discord as well... It feels much better with even this high latency 3000mhz from what I can see.

Also under load mine cpu runs at 4.2-4.3ghz on Air, 4.1ghz if all cores are loaded.

@ 2666mhz it was 4.35ghz under low-medium load (this includes BeamNG running the CPU 50% across most of not all cores in practice btw)

4.2Ghz in Cinebench pegged @2666mhz
4.1Ghz @3000mhz

I run a Noctua D15-S with one fan. It doesn't peak past 80-81C during heavy/full load.

Usually around the 60-70c mark I sadly don't recall CPU temps in beam. But I do know 7 Days to Die will hit 77-80c and runs the cpu much differently and hotter then BeamNG oddly enough, and that's just the game no server or nothing lol.

I wanted to run AIR which seems fine so far over the past month because I run my systems for long term power on stability.

I'd hate to troubleshoot with a loop or have to replace a pump. >.>

But it is interesting to see what I looked like a massive improvement from such a slight increase in ram. Now I know the Infinity fabric is parallel in ratio to the Ram up to 3600mhz. And some have even tinkered with ratios to get crazy performance @ 3800mhz ram.

3600mhz might be worth it next year yet if I can manage to get it running on my machine.

On a side note I have a 750W 80+Gold EVGA but want to move to an 850W Seasonic Titanium in the future, or something like that.

I've been nothing but impressed with the system nonetheless. Coming from a 5820K X99 platform I built in 2016.
--- Post updated ---
--

Also my Motherboard is an X570 Asus Crosshair VII Hero (Wifi)

I've heard the MSI boards are better memory over clockers on X570. I bought the Asus for it's good phase and I/O.
--- Post updated ---
--

Also my Motherboard is an X570 Asus Crosshair VII Hero (Wifi)

I know the MSI boards are better memory over clockers on X570. I bought the Asus for it's good phase and I/O

I am also fairly aware that memory is best on four slot boards ran with two dimms in duel channel for lower latency and better stability.

Yeah the Ryzen options are simply a better buy with often more performance for less, especially for non-gaming compute based applications with the 3000 series closing the gap so much that Ryzen over Intel is kind of the only thing that makes sense unless you really want the 9900K for the slight edge only in games sometimes.

That's my two cents... We'll see another cpu war e I reckon in 3 years.. I refuse to fanboy for anyone.

Overall the 2700x is a great CPU.

I think I've seen some people with threadripper around here. I'd be interested to see any of the 3000 series Threadrippers tested around here.


-
--- Post updated ---

Okay so this is 3000mhz with my loose as heck timings.. I'm gonna do some tinkering to see what I can observe.
--- Post updated ---
Hmmm...

^ @ 2666mhz (same loose timings)

"Loose Timings" = 22 21 21 - 50
--- Post updated ---

Ram @ 3266mhz ~ same awful, unaltered timings. ~~3400mhz prooved unstable, unsurprizingly.. yes voltage was accounted for~~
--- Post updated ---

Ram @ 3000mhz XMP ~ 16 17 17 35 Tighter timings

--- Post updated ---

Ram OC'D W/ XMP settiings ~ @ 3266Mhz ~ 16 17 17 35

btw I run all these on a clean boot with nothing and remove as many programs as possible and run 2 passes to make sure the results are relatively, somewhat accurate.
--- Post updated ---

OC'd Manually with Ryzen mem calculator FAST
3266Mhz (CL14) ~ 14 17 17 30

Final Verdict.

3266Mhz CL14 - 17 - 17 - 30 | 4 x 16gb = @ 64gb Duel Channel ~ Still doesn't make up very much ground here in BeamNG.


On a side note.. it would seem to me that Banana Bench can provide inconsistent results test to test due to how it handles each pass when it simulates physics, it drops the load and replaces it causing the CPU to dive to base or otherwise fluctuate wildy which I don't see in practice within BeamNG during real world testing/playing the game.

While playing BeamNG I usually observe between 4.25-4.30Ghz pretty consistent across loaded cores. This can cause even low numbers of cars to suddenly have one pass that randomly nose dives and becomes an outlier.
--- Post updated ---
Can confirm all this didn't make a huge difference at the end within BeamNG in-game, about the same performance.
CPU stays around the 4.25-4.275 mark under 70C consitantly with 16 random cars spawned and running in traffic on Utah.

Although this'll probably help elsewhere in other games and applications across the computer now.

 

20 cars and still 163% above real time performance fuck yeah man, and this is with "just" a 3950X and Waterfox open:


3950X, C8DH, 64GB of B-die running at 3200 CL14 XMP (because overclocking 64GB's of B-die is absolute hell even on Zen 2)

I should try a rather populated KissMP server next!

 

thats pretty cool... meanwhile im still stuck with my 9900k:/