CPU core count for a complete sim rig?

[Deleted member 1066209]

Hello there!

What is the minimum amount of CPU cores that you would need for an optimal experience to run a complete sim rig with the following: game, direct drive wheel, VR, tactile transducers (SimHub, SimVibe), 6DOF motion (FlyPT or SRS), G-seat (Sim Commander) etc.

I would guess that 8 cores would be enough for the multitude of programs to be run simultaneously, but maybe I'm wrong. I'm sure I'm also missing some essential programs for a sim racing PC, and maybe the addition of those would warrant even a 12-core CPU?

[edit]: forgot to include the G-seat.

 

[RasmusP]

My benchmarks in ACC showed that you don't gain any fps with more than 5 cores.
I deactivated cores and switched Hyperthreading on/off.
With 4 cores, HT improves fps but with 5 cores it doesn't anymore.

Currently no other sim uses more cpu threads than acc.
Apart from maybe project cars 2 and therefore ams2.
I don't have either so can't test.

The better fps with an i9 vs i5 come from the bigger cache and the higher possible clockspeeds due to binning.

The costs per fps go through the roof though, so upgrading your 6 core cpu every second generation often makes more sense.

About additional programs: they barely increase cpu load and overall, even at 6 cores, you will have enough spare performance.

More cores are only needed if you stream your gameplay with the encoder running on the cpu.
Or if you're playing triple AAA games like Assassin's Creed that can use a lot more cores since they don't have to follow realistic physics calculations.

 

[Deleted member 1066209]

Wow, thank you very much for the informative explanation.

 

[RasmusP]

Wow, thank you very much for the informative explanation.

You're welcome! I'll upload my benchmark overview tonight when I'm at the pc.
Generally as an explanation:
Although Taskmanager will show only little load for each cpu core/thread, it only shows an average!
At 1 second refresh rate and a 5 GHz cpu , you'll have 5.000.000.000 cpu cycles averaged in that statistics.

And CPUs can only do something or do nothing.
So 20% load means the cpu core worked for 0.2s and did nothing for 0.8s.

Then there are the application threads.
For simracing you can't really split these easily. Small developer teams etc lead to most simracing titles only using 2-4 threads.

Now if you'd only have 1 thread, task manager would still show some load for all cores (on amd you sometimes see one core at full load though).

This is because Windows moves the threads around. Probably to let the next core start caching or whatever.

So for 1 second, 4 cores, 1 application thread you might see 25% load on all cores.
This simply means that this 1 application thread was being worked on for 0.25 seconds on each cpu core.

And you will only see 25% "cpu load" and think "hey I got 75% performance left. Why are my fps not higher?!".

That's because your cpu simply had nothing to do for the other 0.75 seconds because the application couldn't feed it.


I hope that makes sense?
In the end I call that the "hidden single thread performance limit".
You can't see it, but it's there and although you might only see 50% load on cpu and gpu, your fps are stuck.

More cores won't help. It would just be more cores doing nothing..

Running the tiny extra programs for simracing equipment will use the cpu core, when it's done with the game work.

So no fps loss, mostly.

You will lose some fps if Windows screws up the priorities and because of the scheduling overhead for the management.

What does help though is a bigger cache and higher single core performance.

So for simracing:
Get a cpu with as high single core performance as possible and with enough cores to not get 100% overall cpu load.
Currently that's 5 cores minimum and I don't see that changing to be honest within the next 5 years..

I'll post my acc overview later, which will make things very clear

 

[RasmusP]

So here are the Screenshots:

For the process explorer (little program you can download for free) Screenshots:
I have 12 CPU thread (6 cores + HT), Intel i5 10600k.
So one CPU thread has a maximum of 100% divided by 12 = 8.33%.
Anything above 8% means that thread is maxed out. The other 0.33% are because you can't display the 4.900.000 CPU cycles accurately.

I ran ACC with minimum resolution and minimum resolution scale with medium-high settings and a 24 AI grid. Graphics card had only 40-60% load so my i5 was the limiting factor.

ACC:
2 big threads hitting the limit, 2 smaller threads with quite some load.
Makes for 4 cores being used for these.
Then there are a few smaller threads, so 5 cores give higher fps than 4 cores.
More than 5 cores won't result in higher fps though.

For comparison, here's AC:
2 big threads like ACC, but only one other medium small thread and a few very little ones.
4 cores are enough, but sometimes with Discord, SimHUB etc, you can max them out so 5 cores is better again.

And now here's my big ACC benchmark overview:

I did 3 runs for each setting. Then for comparison disabled HT to get a higher clockspeed.
Averaged the 3 runs and then for some reason decided to average the avg/min/max too.
The runs weren't that long, about 40 seconds iirc so the 1% low and 0.1% low can be ignored.

I then took the best performing setting and tested the impact of memory.
Affinity means that I went into TaskManager while ACC was running and stopped ACC from using both threads of one core so it would use one core for every big application thread.
In theory Windows should do this anyway but the minimum fps showed that it's not perfect.

BTW my replies seem a bit long and overkill for your initial question but this questions pops up every few months and your title is very good so hopefully it will help more people over time to not spend too much money on unneeded cores

 

[Deleted member 1066209]

Unbelievable. That is such a thorough investigation of this topic. Thank you very much for sharing your data; it must have been tedious to collect it all, but I'm glad you did because I am now able to easily understand your recommendations.

 

[RasmusP]

Unbelievable. That is such a thorough investigation of this topic. Thank you very much for sharing your data; it must have been tedious to collect it all, but I'm glad you did because I am now able to easily understand your recommendations.

Back in 2019 with my old i7 2600k I just couldn't figure out why my fps dropped below 60 during race starts in AC.
No core showed more than 50% load, overall load wasn't higher either and my graphics card load was around 70%.

So I researched and thanks to other very helpful and knowledgeable people I found out about the detailed threads view in process explorer and how it all works.
I only scratched the surface and still have no idea how things really work inside windows or the CPU, but it's good enough to explain what we experience!

I then had enough of all the talks about ACC needing 8 cores, mega fast memory and that you should overclock the cpu cache too on Intels.
I couldn't find any good benchmarks so decided to spend half a day to run all these and record the data via msi afterburner.

BTW I once got my hands on some 4100 mhz memory and could get it to run 3800 cl19.
But the fps didn't improve at all.

So 6 cores, 3200 MHz ram and highest clock speeds you can get are what's needed for simracing.
More is better ofc, but if the i5 costs 280€ vs the i7 which costs 409€, that's 46% more money for 1-6 fps.

If you need these fps to be able to run VR without reprojection, it might be worth it.
But if you're using a monitor and no vsync, a few fps more or less don't really matter.
It's better to sell and buy the next gen i5 if there are improvements.

With AMD, you can even put the new cpu in your existing mobo (the current CPUs are the last gen for the existing mobos though.., so I'm waiting for the next gen to upgrade).

 

[2stains]

So here are the Screenshots:

For the process explorer (little program you can download for free) Screenshots:
I have 12 CPU thread (6 cores + HT), Intel i5 10600k.
So one CPU thread has a maximum of 100% divided by 12 = 8.33%.
Anything above 8% means that thread is maxed out. The other 0.33% are because you can't display the 4.900.000 CPU cycles accurately.

I ran ACC with minimum resolution and minimum resolution scale with medium-high settings and a 24 AI grid. Graphics card had only 40-60% load so my i5 was the limiting factor.

ACC:
2 big threads hitting the limit, 2 smaller threads with quite some load.
Makes for 4 cores being used for these.
Then there are a few smaller threads, so 5 cores give higher fps than 4 cores.
More than 5 cores won't result in higher fps though.
View attachment 548338

For comparison, here's AC:
2 big threads like ACC, but only one other medium small thread and a few very little ones.
4 cores are enough, but sometimes with Discord, SimHUB etc, you can max them out so 5 cores is better again.
View attachment 548342

And now here's my big ACC benchmark overview:

I did 3 runs for each setting. Then for comparison disabled HT to get a higher clockspeed.
Averaged the 3 runs and then for some reason decided to average the avg/min/max too.
The runs weren't that long, about 40 seconds iirc so the 1% low and 0.1% low can be ignored.

I then took the best performing setting and tested the impact of memory.
Affinity means that I went into TaskManager while ACC was running and stopped ACC from using both threads of one core so it would use one core for every big application thread.
In theory Windows should do this anyway but the minimum fps showed that it's not perfect.
View attachment 548343


BTW my replies seem a bit long and overkill for your initial question but this questions pops up every few months and your title is very good so hopefully it will help more people over time to not spend too much money on unneeded cores

Sorry for jumping from thread to thread but this is a more appropriate thread to continue on.
What do i do now you reckon? They have 2 XMP profiles, one 3600 c18-22-22-46 and this which is active and stable.

 

[RasmusP]

Sorry for jumping from thread to thread but this is a more appropriate thread to continue on.
What do i do now you reckon? They have 2 XMP profiles, one 3600 c18-22-22-46 and this which is active and stable.
View attachment 548627View attachment 548626

There are 2 things for memory:
Bandwidth/throughput and latency.
Throughput is higher with higher frequency, while latency is mix of frequency and timings.
3600 CL18 = 10 nanoseconds
4400 CL19 = 8.6 nanoseconds
Just Google for ram latency calculator.

So 4400 CL19 has both.
Higher throughput and a lower latency!
Awesome that it's stable!

 

[2stains]

There are 2 things for memory:
Bandwidth/throughput and latency.
Throughput is higher with higher frequency, while latency is mix of frequency and timings.
3600 CL18 = 10 nanoseconds
4400 CL19 = 8.6 nanoseconds
Just Google for ram latency calculator.

So 4400 CL19 has both.
Higher throughput and a lower latency!
Awesome that it's stable!

It crashed, dropped to 4200 dropping to c18 and running. I am wondering about the other timings at 26-26-46-771 2T So high from my previous 15-15-15-35 kit. I may try one by one to lower them.

 

[2stains]

There are 2 things for memory:
Bandwidth/throughput and latency.
Throughput is higher with higher frequency, while latency is mix of frequency and timings.
3600 CL18 = 10 nanoseconds
4400 CL19 = 8.6 nanoseconds
Just Google for ram latency calculator.

So 4400 CL19 has both.
Higher throughput and a lower latency!
Awesome that it's stable!

Crazy, even though everything would run and seem okay 4200 c18 just wasn't right (choppy game play even though fps stayed normal) so i left the XMP 4400 c19 enabled but reduced to 3800 c16-22-22-38 so its still 1.45v and now in ACC i have over 100 fps where before i wouldn't get over 85 like EVER at race start and stayed between 95-108 the whole race! All my worry i needed a new CPU is gone Best $130 ive spent on my PC. Here is my ACC settings and screenshot of race start with CPUz and HWInfo. I think my PC is done for a few years now

 

[RasmusP]

Crazy, even though everything would run and seem okay 4200 c18 just wasn't right (choppy game play even though fps stayed normal) so i left the XMP 4400 c19 enabled but reduced to 3800 c16-22-22-38 so its still 1.45v and now in ACC i have over 100 fps where before i wouldn't get over 85 like EVER at race start and stayed between 95-108 the whole race! All my worry i needed a new CPU is gone Best $130 ive spent on my PC. Here is my ACC settings and screenshot of race start with CPUz and HWInfo. I think my PC is done for a few years now
View attachment 548910
View attachment 548911
View attachment 548912
View attachment 548913View attachment 548914
View attachment 548915

Awesome!
Are you sure that your ram was running 3200 mhz before though and not the default 2600?

 

[2stains]

Awesome!
Are you sure that your ram was running 3200 mhz before though and not the default 2600?

Yeah for sure, I played with that set till the cows came home trying to get lower timings, managed to get 3200 but still at XMP timings. XMK was 3000 15-15-15-35 and everything showed it as 3200. It must have been terrible ram even though it was supposedly good Trident Z.
I always knew something just wasn't right though and now I feel sorted...finally.

 

[2stains]

Hello there!

What is the minimum amount of CPU cores that you would need for an optimal experience to run a complete sim rig with the following: game, direct drive wheel, VR, tactile transducers (SimHub, SimVibe), 6DOF motion (FlyPT or SRS), G-seat (Sim Commander) etc.

I would guess that 8 cores would be enough for the multitude of programs to be run simultaneously, but maybe I'm wrong. I'm sure I'm also missing some essential programs for a sim racing PC, and maybe the addition of those would warrant even a 12-core CPU?

[edit]: forgot to include the G-seat.

Just to be more on topic, I always run SimHub with a buttkicker and phone dash, also have MSI Afterburner, HWINFO64 and Equaliser APO running. The 6 core seems golden for this job, could have sparkles if boosted higher but golden is good enough for me.

 

[RasmusP]

Sounds great! I wonder why the 3800 memory didn't give me a single fps more..
Maybe I'm gonna order the same ram you now have and test again with your settings!

 

[2stains]

Sounds great! I wonder why the 3800 memory didn't give me a single fps more..
Maybe I'm gonna order the same ram you now have and test again with your settings!

3800 didn't t give you anything over 3200?

The ram moduals themselves are sort of low end, when holding them in hand with the Trident Z I pulled out they felt really cheap, but in my PC I don't see anything, little window, no lighting, it is already pretty mixed in colours so I'm not worried as long as they don't overheat and stay together.
If you've got a spare $100 I say grab them and play, I will be interested to hear what you manage so I can copy your timings, lol

This is the sticks and the shop I git them from

msy.com.au | Australia's No.1 Computer Store

msy.com.au is Australia's No.1 Computer Store, with the best prices on Computer Parts, Desktops, Laptops, Tablets, Gaming, Drones, Printers and more!

www.msy.com.au

 

[RasmusP]

Thanks for the link and info! Not available in Germany but there are other kits for around 100€ with slightly better timings. No idea if they would run better or worse than yours. I might try such a kit if I feel like experimenting!

3800 didn't t give you anything over 3200?

Not a single fps... It was really unexpected and I just figured that beyond 3200 mhz there was little to gain with a 10600k.

Apparently that's not the case for you at all. Very strange!

But I'm happy for your haha

 

[2stains]

Thanks for the link and info! Not available in Germany but there are other kits for around 100€ with slightly better timings. No idea if they would run better or worse than yours. I might try such a kit if I feel like experimenting!


Not a single fps... It was really unexpected and I just figured that beyond 3200 mhz there was little to gain with a 10600k.

Apparently that's not the case for you at all. Very strange!

But I'm happy for your haha

From what I've learnt in the last few days makes think my 3000 kit at 3200 was throwing errors, it wasn't able to budge on timings so maybe that 3200 wasn't a good oc. Maybe I would have been better off with them at 3000??
I have no idea but games are definitely smoother, I thought AMS2 was smooth, but now it really is amazing.
I look forward to hearing what you do with this RAM.