VR 90FPS - my personal Solution

[vinnefreerider]

After almost 2 months testing possibilities to put my Assetto Corsa running at VR 90fps over online sessions (such as Nordschleife Tourist with 32 connected players) I think I finally found the solution.

Recap, my setup is:
i7 7700 + 16gb DDR4 Hyperx Fury 2133mhz + RTX 2070 + 1tb HDD + Lenovo Explorer + Win10

Andrew_WOT recommend me to use the AC Shaders Patch tweaks and it did a GREAT improvement performance but not solved 100%. (thank you Andrew for the tips)
It has improved the FPS from AVG 62 to 64, it is 3,23%

So I started looking for the reason I'm having CPU Bound and I discovered a simple and basic fact: RAM memory has the responsibility to provide the communication with CPU and GPU. If you are using a slow memory kit, it will affect dramatically your FPS performance on the games.

Also I noticed there is a lot of people running good memory kit but without BIOS tweaks such XMP profiles to enjoy full speed of the equipment.

My SOLUTION:

1.
I did a memory overclock, from 2133mhz to 3066mhz.
The bandwidth speed was improved in +-40% and the latency was reduced in +-11% from 57.4ns to 50.2ns (latency also influences a lot in games) HUGE IMPROVEMENT HERE.
(It is stable with more than 5 hours of stress tests with Aida64 and MemTest86)
check the pics

*** I removed the part what I was saying about to turn Hyperthread OFF to improve the performance. After few more tests I noticed I was wrong, thanx for the people here who has give the same feedback.***

This moment your AC fps have improved a lot.
Assetto Corsa is running with 90fps locked 90% of the time over online sessions even on Nordschleife Tourist with 32 connected drivers!

A in game benchmark on post No. 18 bellow this

 

[Kek700]

Tested in game benchmark Single Display mode all settings maximum.

DDR4 2133mhz
AVG: 59 fps
MIN: 46 fps
MAX: 65 fps

DDR4 3066mhz
AVG: 64 fps (8.47% faster)
MIN: 52 fps (13.04% faster)
MAX: 69 fps (6.15% faster)

................stumped !
lost for words,

will give this another go, but take a bit more care in getting my results.

 

[vinnefreerider]

I've tested HT on/off on my i7 2600k @ 4.5 GHz (RAM 2133 MHz) with Assetto Corsa: performance was worse with HT off.

Same here. Performance was worse with HT (in BIOS) turned off. I'm running a I9-9900K overclocked to 5.1 Ghz and a GTX 1080ti with Oculus Rift .. with HT off performance dropped signifficantly.

You guys are correct.
I was wrong about my impressions with HT on and off.
Today I did an test with HT on and OFF and the result is: OFF is worst = AvgFPS came from 64 to 61 = 4.69% (is a lot)

I'll edit my first post with this report.
Sorry for my misperception.

 

[RasmusP]

You guys are correct.
I was wrong about my impressions with HT on and off.
Today I did an test with HT on and OFF and the result is: OFF is worst = AvgFPS came from 64 to 61 = 4.69% (is a lot)

I'll edit my first post with this report.
Sorry for my misperception.

Happens to the best! I had the same impression once but out of 10 benchmark runs only 2 were better with HT off. Thought your "on the fly HT off" might make a difference so I kept it quiet

 

[Deleted member 197115]

Most likely different CPUs respond differently to HT settings.
I have seen some marginal boost with HT off on i7-5930K (6 cores), guess if you are low on physical cores, HT off will make things worse if you have number of active threads in games exceeding the number of cores.
There is video incidentally testing the same CPU.

 

[RasmusP]

Most likely different CPUs respond differently to HT settings.
I have seen some marginal boost with HT off on i7-5930K (6 cores), guess if you are low on physical cores, HT off will make things worse if you have number of active threads in games exceeding the number of cores.
There is video incidentally testing the same CPU.

Not wanna call bs on that video but with HT on the cpu is 100 MHz lower clocked and the system uses over a GB less of ram...
Also a cpu can't be "low of physical cores" as long as it's not used to its full extend (close to 100% overall cpu load) as a not fully loaded cpu means that there are not enough threads to get it filled.
I would be very curious though to find out the reason behind that "special" cpu showing that behavior. Seems for me like there's more involved than just HT on/off.
BTW, cs:go reacts very strange to a lot of things. The engine is not made for high fps although everybody wants them (me too!). You change anything and the fps will change a lot no matter whether the cpu load changes or not.
And far cry uses frostbite afaik which had massive stuttering on my system with HT on in Bf4 and bf1. 3 and 5 are fine.
No idea about pcars though and the rest of the video...

 

[Deleted member 197115]

Also a cpu can't be "low of physical cores" as long as it's not used to its full extend (close to 100% overall cpu load) as a not fully loaded cpu means that there are not enough threads to get it filled.

Only multi core processing it truly concurrent, on a single core it's scheduler juggling threads, the more you have sharing the same core, the less CPU time each gets.

 

[RasmusP]

Only multi core processing it truly concurrent, on a single core it's scheduler juggling threads, the more you have sharing the same core, the less CPU time each gets.

Wouldn't the single cpu core be loaded to 99% then though? I can watch exactly this happening by deactivating cores and seeing the overall cpu load go up (each core gets more load ofc). With 2 cores assetto corsa goes to 100% overall cpu load but even with 3 cores (no HT) it no longer clips the 99% and dips down every now and then.
I think I'll do some performance graph and run AC on 1-4 cores + HT on/off to prove this for myself

 

[Deleted member 197115]

Because it is highly improbable for a typical execution thread to consume all CPU cycles, they need to wait for resources, block on IO, cache misses, wait for results of previously issued instructions, etc.
Pretty decent read explaining all that here.

 

[Kek700]

i5 6600K......

DDR4 = 3200Mhz :-

C VERSION: 1.16.3 (x64)
POINTS: 35391
FPS: AVG=241 MIN=89 MAX=294 VARIANCE=13 CPU=87%
LOADING TIME: 15s
GPU: NVIDIA GeForce GTX 1080 (3440x1440)
OS-Version: 6.2.9200 () 0x100-0x1
CPU CORES: 4
FULLSCREEN: ON
AA:4X AF:16X SHDW:1024 BLUR:0
WORLD DETAIL: 3 SMOKE:1
PP: QLT:3 HDR:0 FXAA:1 GLR:3 DOF:3 RAYS:1 HEAT:1

DDR4 = 2666Mhz :-

C VERSION: 1.16.3 (x64)
POINTS: 32648
FPS: AVG=222 MIN=47 MAX=273 VARIANCE=14 CPU=88%
LOADING TIME: 15s
GPU: NVIDIA GeForce GTX 1080 (3440x1440)
OS-Version: 6.2.9200 () 0x100-0x1
CPU CORES: 4
FULLSCREEN: ON
AA:4X AF:16X SHDW:1024 BLUR:0
WORLD DETAIL: 3 SMOKE:1
PP: QLT:3 HDR:0 FXAA:1 GLR:3 DOF:3 RAYS:1 HEAT:1


Ac bench mark, I have always found a bit confusing, cannot argue with results though.

 

[vinnefreerider]

i5 6600K......

DDR4 = 3200Mhz :-

C VERSION: 1.16.3 (x64)
POINTS: 35391
FPS: AVG=241 MIN=89 MAX=294 VARIANCE=13 CPU=87%
LOADING TIME: 15s
GPU: NVIDIA GeForce GTX 1080 (3440x1440)
OS-Version: 6.2.9200 () 0x100-0x1
CPU CORES: 4
FULLSCREEN: ON
AA:4X AF:16X SHDW:1024 BLUR:0
WORLD DETAIL: 3 SMOKE:1
PP: QLT:3 HDR:0 FXAA:1 GLR:3 DOF:3 RAYS:1 HEAT:1

DDR4 = 2666Mhz :-

C VERSION: 1.16.3 (x64)
POINTS: 32648
FPS: AVG=222 MIN=47 MAX=273 VARIANCE=14 CPU=88%
LOADING TIME: 15s
GPU: NVIDIA GeForce GTX 1080 (3440x1440)
OS-Version: 6.2.9200 () 0x100-0x1
CPU CORES: 4
FULLSCREEN: ON
AA:4X AF:16X SHDW:1024 BLUR:0
WORLD DETAIL: 3 SMOKE:1
PP: QLT:3 HDR:0 FXAA:1 GLR:3 DOF:3 RAYS:1 HEAT:1


Ac bench mark, I have always found a bit confusing, cannot argue with results though.

Nice report!
You got an improvement in
AVG fps: 8.56%
MIN fps: 89.36%
MAX fps: 7.69%

 

[Peter Boese]

Nice report!
You got an improvement in
AVG fps: 8.56%
MIN fps: 89.36%
MAX fps: 7.69%

I wouldn't take the minimum value in account. There could be 1 little event, like stutter, which completely change this value.

 

[Fernando Deutsch]

The biggest improvement aspect related to higher memory speeds are usually the min fps. Nice results.

 

[Kek700]

It has made a difference, a bit shocked to be honest. I ran Brands GP, 23 Ai , all settings as high as
makes no difference ( started at back of grid ). I get a constant 120 fps including start with two frames dropped occasionally though out a lap, depending on race circumstance. ( 120hz monitor )
For a few years now I have completely discounted memory speed.

Forgive me for bashing on

 

[Deleted member 197115]

Why would anyone deliberately run DRAM at the lower speed anyway?
Overclock to the highest and enjoy all the free benefits.
I can only run mine at 4000MHz, can't even push to rated 4133, the problem with Intel memory controllers, they guarantee only 2666, everything else is overclock, which may or may not work, regardless of what is stamped on memory module.

 

[Kek700]

I would clock higher, but I am expecting diminishing returns ! also much more concerned about reliability.
my dram is rated at 2666. I needed to push voltage to 1.3v to get 3200. that's really as far as I want to go.
it has been quite straight forward to get 3200, I don't think I am prepared for any associated hassle with going any further.

 

[RasmusP]

Thing about memory speeds is that:
- at some point the additional price becomes higher than the additional performance
- the difference to buy the next better cpu is bigger

So imo you grab the cpu you want/have the budget for and then buy the ram you can afford or if you want the best for your money you take the memory right before the price goes steep.

 

[vinnefreerider]

Why would anyone deliberately run DRAM at the lower speed anyway?
Overclock to the highest and enjoy all the free benefits.
I can only run mine at 4000MHz, can't even push to rated 4133, the problem with Intel memory controllers, they guarantee only 2133, everything else is overclock, which may or may not work, regardless of what is stamped on memory module.

After the overclock, I recommend you test the memory with AIDA64 stress test for an 50minutes or more and after, test with MemTest86 Passmark software... which is a more complex test... if in both are stable, your setup is done! hehe

 

[RasmusP]

I would clock higher, but I am expecting diminishing returns ! also much more concerned about reliability.
my dram is rated at 2666. I needed to push voltage to 1.3v to get 3200. that's really as far as I want to go.
it has been quite straight forward to get 3200, I don't think I am prepared for any associated hassle with going any further.

Der8auer did an experiment to prove his statement about memory being able to take a lot more voltage than the specs suggest. He ran ddr3 memory at 1.75V (Intel tells you to run 1.5V) and let prime95 run for 30 days non stop. All went fine!
So no guarantee ofc but from my knowledge you can go a lot higher than Intel tells you!

 

[scottiedawg66]

^ holy crap!!! thats insane....

 

[RasmusP]

Addition: he doesn't state the same for other voltages regarding the cpu! He simply says that he never ever seen a cpu failing due to a faulty memory controller. If a cpu stopped working it was because of too much core voltage or just being extremely old.