AC GT3 @ PORTIMAO, Sunday 2nd May 2021

[jackdyson31]

New event is set up:
Sign up here: https://beta.simracing.gp/events/TY43xveg0VK94My4N7MDR#cars

And please read the 1th and 3rd post of this thread: https://www.racedepartment.com/threads

Thanks Han

 

[jackdyson31]

You should try to run either MSI Afterburner (recommended to see the different GPU "limits" and to undervolt or OC your card, which is very much needed with a 3090, later more to that!)
or run "OpenHardwareMonitor", which is very lightweight and has a 24h history graph (you need to enable the "plot" to get the graph and zoom in/out is done by hovering the mouse over the axis you wanna rescale and then using the mouse wheel or click+drag).

Looks like this for me, when compressed to fit into RD picture uploads

View attachment 471322

If your GPU load is not 95% or more, your CPU is the bottleneck (or an fps limiter ofc).
So you can easily find out what's happening and what you should adjust.

Mod tracks can be heavy on either side of things, depending on how optimized and how they're built in general.


About undervolting:
The 3080 and 3090 really suffer from running into their power limit and creating massive frequency jumps and power jolts.
There's an article from Igor's Lab:

GeForce RTX 3080 and RTX 3090 Undervolting - When Reason and Experimentation Meet NVIDIA Ampere | igor´sLAB

Our community member Gurdi was able to get his hands on a GeForce RTX 3080 just in time for the launch and was kind enough to invest some time to provide you with new insights around this card.

www.igorslab.de

I read into this after being unhappy with my 3080, now absolutely amazed by it

The issue is basically that nvidia built some automatic clock adjustments and they have a few limits the algorithm leans at:
- Temprature
- Power
- Voltage
- (Usage)

- Temp is easy: hit a certain temp and the clocks go down.
- Voltage is only hit when you overclock the card without having big loads on it so the voltage will restrict the card from clocking higher.
- Usage isn't a real limit.. It's just when the card is completely loaded and no other limit is reached

- Power:
This is causing some issues... When the card hits the power limit, it will be clocked down quite massively. A moment later though it will try to go back up.
During this clock step upwards, it will create quite a power jolt for the PSU.
With the early drivers, this caused even some 800W+ PSUs to shut off...!

So anyway, you can even gain fps by undervolting. It's a bit tricky though, since you can't just pull down a voltage slider...
You need to limit the maximum frequency while putting the max frequency earlier on the voltage curve.

The card will always try to reach the maximum frequency point, but if it's already reached at a lower voltage, it will stay there...
Yeah.. I needed a moment or two to understand this behavior..

Anyway, you can mostly undervolt all 3080 and 3090 so that they'll reach their normal max clock with a lower voltage.
And then they'll reach their "usage" limit before hitting the power limit.

My 3080 puts out 2% more fps now with 80-120W lower power consumption!!!
Coil wine is gone completely and the room temperature and air quality are a lot better


So two things I wanna tell you:

1. Check with OpenHardwareMonitor if your GPU is the bottleneck or if it's your 5800x, although being unlikely, but check it before you adjust anything!

2. If your GPU is fully loaded, install msi afterburner, check if your Power limit is constantly reached and if yes, drop me a PM

My "OC curve" looks like that:
Sweetspot is mostly 750-850 mV and around 1800 MHz.
View attachment 471323

Stock Clock/Voltage curve:
View attachment 471324



Effects:
Playing an AC replay with full grid, it was 260W, 43% load with stock settings ; 188W, 47% load with undervolting.
The load is calculated by taking in the clock speeds. At ~45% load the stock settings will clock a bit higher.
If you load the cards completely, they will both reach about 350W for my 3080. But the clockspeeds will be more stable and the fps will slightly increase, while the 1% lows will be a lot higher, since there are no clockspeed drops.


Well.. Should have created a new thread lol.
Sorry to everyone!

Ask a silly question and like Razza writes out the scientific equivalent of "War and Peace"

Fantastic detail in the answer, much appreciated

 

[RasmusP]

Fantastic - its definitely a code issue then, even Dmitry was unhappy with the rendering so ... its probably one of those things like "its not designed for it"

Can't really confirm that.
With the Oculus Rift the performance was exactly as expected:
CPU load being about 2% higher with VR so about 2% lower CPU related fps.
My 10600k is hovering between 80-90 fps for race starts, so with the Rift it was hovering between 78-88 fps.

My graphics settings were a good compromise between "still looking okay" and 60-80 fps.

Now my monitor has 3440x1440 pixels = 4.953.600
The Rift has 2x 1080x1200 = 2.592.000

Which is roughly x1.9 between them.

So I plugged the Rift in and the fps where all over the place due to the jumping between 45 and 90 fps.
I went into single player and had steady 90 fps without issues.

Now racing was impossible with native 90 fps so I forced ASW (45 fps) and my CPU could keep the frametimes steady and I could put some super sampling on top. No stuttering at all and it was looking as good as the pixel party of the Rift can get.

It's all scaling as you would expect. It's just that people don't expect their high-end CPUs to struggle with 80 or 90 native VR fps in ACC. But they do, sadly.
On the other hand most high end CPUs are just hovering around the 100 fps mark for AC race starts so the headroom is very small...

Small headroom + 5 years newer game = Problems.

The other issue is that ACC suffers from input lag and at 45 fps, you will feel that lag...

 

[jackdyson31]

Can't really confirm that.
With the Oculus Rift the performance was exactly as expected:
CPU load being about 2% higher with VR so about 2% lower CPU related fps.
My 10600k is hovering between 80-90 fps for race starts, so with the Rift it was hovering between 78-88 fps.

My graphics settings were a good compromise between "still looking okay" and 60-80 fps.

Now my monitor has 3440x1440 pixels = 4.953.600
The Rift has 2x 1080x1200 = 2.592.000

Which is roughly x1.9 between them.

So I plugged the Rift in and the fps where all over the place due to the jumping between 45 and 90 fps.
I went into single player and had steady 90 fps without issues.

Now racing was impossible with native 90 fps so I forced ASW (45 fps) and my CPU could keep the frametimes steady and I could put some super sampling on top. No stuttering at all and it was looking as good as the pixel party of the Rift can get.

It's all scaling as you would expect. It's just that people don't expect their high-end CPUs to struggle with 80 or 90 native VR fps in ACC. But they do, sadly.
On the other hand most high end CPUs are just hovering around the 100 fps mark for AC race starts so the headroom is very small...

Small headroom + 5 years newer game = Problems.

The other issue is that ACC suffers from input lag and at 45 fps, you will feel that lag...

I knew I should have kept my mouth shut ...

Thanks again - yeah you need to lock the framerate to monitor refresh rate to stabilise the system, I have no experience with VR so can't comment on that unfortunately. AC is underusing CPU and GPU released in 2018 and above generally and it is also an idea to keep the resolution down to 2K for this sim on monitors. Conjecturing, on VR probably a lot higher due to the pixelation issue ...

 

[RasmusP]

I knew I should have kept my mouth shut ...

Thanks again - yeah you need to lock the framerate to monitor refresh rate to stabilise the system, I have no experience with VR so can't comment on that unfortunately. AC is underusing CPU and GPU released in 2018 and above generally and it is also an idea to keep the resolution down to 2K for this sim on monitors. Conjecturing, on VR probably a lot higher due to the pixelation issue ...

Hehe yeah my comment wasn't about that the ratio between visuals and performance would be great for ACC.. AC is a lot better in this regard!
My statement was only that ACC sucks no matter if you're on a monitor or in VR
It's just that it's less noticeable on a monitor and that you have more GPU headroom on a monitor, generalized.

Also most people either don't see the stuttering when having unsynced fps/monitor as long as their monitors have 100 Hz or more and that most modern monitors now can use either freesync or gsync-compatible, so some fluctuations aren't too bad.

Meanwhile if you get tearing or stuttering in VR... Your brain melts

 

[jackdyson31]

wowza

My statement was only that ACC sucks no matter if you're on a monitor

This is pretty weird - runs at 200 fps smoothly at 2K res on mine ... rf2 runs at near 500 fps lol ...

Yeah my laptop monitor is gsync and its no difference or not much - but the bigger one needs the locking as you said because it can't.

 

[RasmusP]

wowza

This is pretty weird - runs at 200 fps smoothly at 2K res on mine ... rf2 runs at near 500 fps lol ...

Yeah my laptop monitor is gsync and its no difference or not much - but the bigger one needs the locking as you said because it can't.

Put in 20 AI's (15 visible) and it looks like this, when the rolling start begins:

This is a clear CPU limit, GPU only at 83% load...
However it's with settings maxed out from my benchmarking session against the 3070, with a "good compromise", things will improve a bit.
No FPS limit active.

This is AC with my CSP config during a Road Atlanta race start (but with 30 AI). Didn't have the RTSS overlay active, sadly. But you can see: very similar CPU bottleneck performance.
GPU load is roughly around 50% for race starts on normal tracks.

 

[jackdyson31]

Put in 20 AI's (15 visible) and it looks like this, when the rolling start begins:

View attachment 471346

This is a clear CPU limit, GPU only at 83% load...
However it's with settings maxed out from my benchmarking session against the 3070, with a "good compromise", things will improve a bit.
No FPS limit active.

This is AC with my CSP config during a Road Atlanta race start (but with 30 AI). Didn't have the RTSS overlay active, sadly. But you can see: very similar CPU bottleneck performance.
GPU load is roughly around 50% for race starts on normal tracks.
View attachment 471347

Crikey - for ACC I get 180 fps with 24 cars and rf2 ... 500 ish fps while racing lol ...

I thinketh with all of my heart that you must be at 4K to be getting these numbers - and if not then the Nvidia driver is crap.

That is a very powerful GPU you have there and the CPU is nice too !

 

[RasmusP]

Crikey - for ACC I get 180 fps with 24 cars and rf2 ... 500 ish fps while racing lol ...

I thinketh with all of my heart that you must be at 4K to be getting these numbers - and if not then the Nvidia driver is crap.

That is a very powerful GPU you have there and the CPU is nice too !

Mhh.. Are you sure that you don't have an "x2" or "x3" in your fps read out?
Just joking ofc but something is weird then...

There's a CPU benchmark thread in the AC forums and my performance makes perfect sense comparing it, your fps don't make any sense though

The ACC CPU Benchmark Thread

This is a spin-off to "The ACC Benchmark Thread" & "The ACC GPU Benchmark Thread" :...

www.assettocorsa.net

 

[jackdyson31]

Mhh.. Are you sure that you don't have an "x2" or "x3" in your fps read out?
Just joking ofc but something is weird then...

There's a CPU benchmark thread in the AC forums and my performance makes perfect sense comparing it, your fps don't make any sense though

The ACC CPU Benchmark Thread

This is a spin-off to "The ACC Benchmark Thread" & "The ACC GPU Benchmark Thread" :...

www.assettocorsa.net

lol ... I swear to god its 150 - 220 fps at 2K ACC lol Don't ask me how but remember I am on an ASUS RoG so those are custom built and integrated - it could be that ? Plus mine is set to overdrive.

PS those numbers are for naked ACC and AC (no CSP etc)

 

[demetri]

Thanks for confirming an impression Dmitry, I moved over for you and then I was totally shocked at your lines - I was off note myself with toothache, headaches and whatnot - and I thought "man Dmitry's not having any better a day than I am". I thought it was just one of my high speed delusions but reading your report confirms what I felt as we drove the Portimau backwaters together for a lap. Just wanted to ask - what mod of Porti did you drive, and, I am sorry you had an unforgiving race by your standards.

It is "Algarve by AC TrackReboot and Zwiss". I don't think I can post a link to it right here, but you can just google that or if you still can't find let me know and I can PM you the link

Here it is in all its glory:

You can compare that to the onboard video I posted on page 2 and see that it is much closer to the real thing.

 

[demetri]

@HF2000
I have no useful advice, just observations from my driving seat looking at your 488.
it looked quite skittish, you would in random places get a lot of sudden oversteer.
Now I have no doubt on certain circuits this would not occur. I just think this circuit brought the worsts out, handling wise, from your 488 setup.
I tend to alter my Nissan setup quite dramatically race to race, it is a case that on some circuits I can get away with different setups.
This circuit is one in particular, curbs here and I am off..an oversteer Nissan here is not good news here.
A long winded and probably pointless way of saying if you could find a way of giving the car a more predictable handling setup, ok, maybe slower, but more predictable for the race.
My qualifying setup was not the same as my race setup, it was just unfortunate that I could not get the best from my qualifying setup.
My race setup was very stable, and I used that to good advantage here for my race.
Undoubtedly could have been a bit quicker with the qualifying setup, but one off and that-would negate any small advantage I got from that setup.
Sorry for the waffle.

I concur. Having a setup that you can reliably drive with is the key for a relatively long race. Some people can drive for a long time balancing right there on the edge, but I can't, hence why I'm not driving 488 or 650. Both of them are kind of a car that is looking to kill you any suitable moment. You can make them safe with a setup, but that just guts them and they aren't competitive anymore. Both don't have the top speed, so you have to lean on their grip in the corners and use an oversteery setup to get good laptimes.

Oh, and thank you, @HF2000 for organizing the event, I forgot to do that in my post after the race, silly me.

 

[demetri]

Thanks for confirming an impression Dmitry, I moved over for you and then I was totally shocked at your lines - I was off note myself with toothache, headaches and whatnot - and I thought "man Dmitry's not having any better a day than I am". I thought it was just one of my high speed delusions but reading your report confirms what I felt as we drove the Portimau backwaters together for a lap

T1 comparison.

RD version

"Other" version

Both clips are taken from about the same spot and the same scale (based on the length of the Audi). The other track is significantly narrower which calls for a later entry at a lower speed. Pit exit is also much closer to T1

 

[640er]

Thanks @HF2000 for organising and congrats to @Chris Down . Sorry you had such a boring race up front

Was there any difference in track/temp than last time? I couldn't get close to my PB. Best was +0.5. Anyways, had a good quali and start of the race. Settled in 2nd and stayed there until Patrick and Dmetri with their "only refuel" strategy overtook me in the pits. But the fresher softs gave me enough advantage to finish P2. First time I managed brake bias during every lap. Before entering the curvy middle/last sector, I reduced it by 2% which helped me keeping the rear in place. Maybe that's something I'll try more often.

 

[HF2000]

Thanks @HF2000 for organising and congrats to @Chris Down . Sorry you had such a boring race up front

Was there any difference in track/temp than last time? I couldn't get close to my PB. Best was +0.5. Anyways, had a good quali and start of the race. Settled in 2nd and stayed there until Patrick and Dmetri with their "only refuel" strategy overtook me in the pits. But the fresher softs gave me enough advantage to finish P2. First time I managed brake bias during every lap. Before entering the curvy middle/last sector, I reduced it by 2% which helped me keeping the rear in place. Maybe that's something I'll try more often.

Ait temperature 20C, track 27C. I think Chris used 26C track. I used 27 because server manager said that realistic with air 20C. But I can't imagine that cost you 0.5s. Maybe it's because you're older now ?

 

[demetri]

Track temp almost doesn't matter in AC. +1c is not even noticable, unlike the air temp

 

[SOLO59]

I was pretty sure that I had P7 finish in my pocket after that, but apparently the fuel estimate I was given was wrong and I was 1L short. I only realized that at the end of the penultimate lap and tried everything from coasting to shortshifting, but that was not enough and I ran out of gas in the final turn. I still had enough momentum to coast uphill to the finish line but got passed by two cars when doing that.

oh snap! I figured thats what happened. I simply got used to crew chief saying I got 10 minutes of fuel with only 2 minutes to go lol

Didn't get to practice too much, but I did gather a couple hours of practice leading up to qualifying.
I made too many errors in qualifying, and couldn't even match my pb from official practice session. Started in P11, I was too cautious on the first lap and lost a couple spots early, but kept digging in. Stayed pretty consistent at my own pace, and after pitsops, found myself in P2 until I couldn't handle the softs any longer lol. Stayed out for a lap or 2 I think, then pit. Came out I think just outside the top ten. Managed to get down to P9 until this happened with Brian ahead of me lol

Then Demetri ran out of fuel at the end. Just wanted to round up the top 10, but through those unfortunate situations, brought the R8 home in one peace P7

Thanks Han for hosting. Congrats to Chris for the win and pudiums!
I joined the Simracing604 Sunday league beginning this Sunday. Its 6 weeks, won't be able to do these races again until then. But these club races I hope has prepared me for my second attempt at a league. First was RD Mini Challenge season 8 and aside from one race, the rest was not too good. That was then and this is now lol

I did record the race, but won't have time to put it together till tomorrow

 

[Interslice]

oh snap! I figured thats what happened. I simply got used to crew chief saying I got 10 minutes of fuel with only 2 minutes to go lol

Didn't get to practice too much, but I did gather a couple hours of practice leading up to qualifying.
I made too many errors in qualifying, and couldn't even match my pb from official practice session. Started in P11, I was too cautious on the first lap and lost a couple spots early, but kept digging in. Stayed pretty consistent at my own pace, and after pitsops, found myself in P2 until I couldn't handle the softs any longer lol. Stayed out for a lap or 2 I think, then pit. Came out I think just outside the top ten. Managed to get down to P9 until this happened with Brian ahead of me lol

Then Demetri ran out of fuel at the end. Just wanted to round up the top 10, but through those unfortunate situations, brought the R8 home in one peace P7

Thanks Han for hosting. Congrats to Chris for the win and pudiums!
I joined the Simracing604 Sunday league beginning this Sunday. Its 6 weeks, won't be able to do these races again until then. But these club races I hope has prepared me for my second attempt at a league. First was RD Mini Challenge season 8 and aside from one race, the rest was not too good. That was then and this is now lol

I did record the race, but won't have time to put it together till tomorrow

You seemed to be going alot better with the change of tires! Pity the internet said no, would have been a nice battle there with the 3 v10s. Your car literally ended up inside my cockpit and scared the bejaysus out if me. Good luck with the league Dean

 

[UHF79]

Having just recently purchased a Reverb G2 for racing and spent a lot of time reading and testing for myself, I believe the issue is as follows.

The per eye resolution of Reverb G2 is 2160x2160 so thats 4,665,600 pixels per eye which is 9.3mp, this is obviously higher than 4k which is 8.3mp so a full 1mp more pixels to run natively.

But here is the crack, because of the distortion the lenses add, the picture has to be sent to the headset in a higher resolution in order to for there to be enough resolution to make it fit. There seems to be some argument online about how much higher but if we work on 25% then thats 2700x2700 which is then 7.2mp per eye so 14.5mp overall which is significantly higher than 4K.

Add to this that VR requires the scene to be rendered twice, once from each eye's perspective, so its not 4320 x 2160, its 2 lots of 2160 x 2160 and everything that goes with it, lighting, shadows, object occlusion etc.

So if you run your Reverb G2 at native resolution with a % super sampling to account for lens distortion, then you're up around double the GPU requirements of 4k, so 7 year old game aside, its still a LOT of work!

Add in some overhead for the WMR interface too (which BTW if you click the pause button inside it, it results in significantly lower overhead)

Personally, I run the render resolution close to native so i'm not getting the full experience and with most settings in AC disabled i can get near to 90fps with 1-2% reprojection (filling in missing frames).

If you can stand 60hz (which I cant), then you can enable that and up the render resolution.

I run a 8086K with all cores on 5Ghz on water and a cheap version dual fan 2080ti.

As for ACC, I dont think i'll touch it in VR until 4080Ti because I don't think a 3090 will cut it at native resolution.

This helped too (all small gains): https://docs.microsoft.com/en-us/wi...rosoft-edge#monitor-and-input-handling-issues

 

[pattikins]

...
So if you run your Reverb G2 at native resolution with a % super sampling to account for lens distortion, then you're up around double the GPU requirements of 4k, so 7 year old game aside, its still a LOT of work!

Add in some overhead for the WMR interface too (which BTW if you click the pause button inside it, it results in significantly lower overhead)

Personally, I run the render resolution close to native so i'm not getting the full experience and with most settings in AC disabled i can get near to 90fps with 1-2% reprojection (filling in missing frames).

If you can stand 60hz (which I cant), then you can enable that and up the render resolution.

I run a 8086K with all cores on 5Ghz on water and a cheap version dual fan 2080ti.

As for ACC, I dont think i'll touch it in VR until 4080Ti because I don't think a 3090 will cut it at native resolution.

This helped too (all small gains): https://docs.microsoft.com/en-us/wi...rosoft-edge#monitor-and-input-handling-issues

I think AC (early access)came out in 2013, along with rfactor 2 & Raceroom (a vintage year!) & I remember playing them all on a triple monitor setup (6.7-7.2 mega-pixels) with an i7 3770 + AMD R9 290 with a frame-rate in the low 40s for AC. With each gpu upgrade, it was clear that the previous gen video card was a bottleneck for AC performance. When I switched to VR in 2018 (Vega56/1080ti Oculus Rift cv1 - 2.7 mega-pixels), the roles were reversed with the cpu being the more obvious bottleneck. This continued to be the case when I upgraded my cpu to an AMD 3600X, and less so now with a 5600X & a Samsung Odyssey+ (4.9 million pixels). With the vanilla version of AC (no post-processing, csp or SOL...), the gpu workload is less than 50% on a 1080ti, so I imagine my ageing 1080ti could cope with the graphical demands of the HP Reverb G2 VR headset, let a lone a 2080ti*.

taken from Sunday night's race @ Portimao.

*that said, with Project Cars 1&2, AMS2, Raceroom & even rFactor2, it's a different story!