Overclocking 2070 doesn't appear to make any difference in fps

[mgh24]

Using MSI afterburner.

NOT adjusting core voltage.
Increasing power limit to max
Core clock +115
Memory clock +500

Just going by in-game displayed fps, I am gaining nothing in AC or ACC.
Seems strange to me that it does not appear to be helping anything.

Using latest custom shader in AC, and enabling AMD FX, I do get some gain in fps.

Am I expecting too much from the over clock, or doing something in correctly?

Thanks!

 

[Alex Ball]

Might be a CPU bottleneck then?

 

[Neilski]

Might be a CPU bottleneck then?

Yeah, but the other thing to do is check what Afterburner says is limiting the GPU. If it's voltage, there's your answer. If it was power (on the default settings), then raising the power limit should have helped. If it's the CPU that is the bottleneck of course, then the GPU may not be limited by either voltage or power.
Caveat: I'm a noob at GPU-overclocking

 

[RasmusP]

Using MSI afterburner.

NOT adjusting core voltage.
Increasing power limit to max
Core clock +115
Memory clock +500

Just going by in-game displayed fps, I am gaining nothing in AC or ACC.
Seems strange to me that it does not appear to be helping anything.

Using latest custom shader in AC, and enabling AMD FX, I do get some gain in fps.

Am I expecting too much from the over clock, or doing something in correctly?

Thanks!

Check the following things:

1. Is your GPU load above 95%? If it's below, you're CPU bottlenecked and the 2070 OC won't do anything.
2. If your GPU load is above 95%, you need to make sure that you actually gain real core clock from the OC:
   1. You need to detach the graph at the bottom of afterburner (I'm using the old skin, not sure how it looks on the latest skin with the full re-design...).
   2. Check what LIMIT is triggered while playing AC.
      1. Power Limit: you need to either raise power limit or undervolt to get clock headroom with the same power consumption.
      2. Temp Limit: cooling is insufficient or you need to raise the limit (90°c is totally fine for GPUs).
      3. Voltage limit: this should be ok to hit, as long as you're hitting the desired clock speeds.
      4. Usage limit: that's the limit if no other limit is triggered, so that's fine to be "1".
   3. Have a look at the core clock graph. Is it showing more than without the oc WHILE you're playing AC? If one of the limits is triggered, your GPU will downclock.
3. To have more control over the actual clock I would suggest to OC by using the ctrl+F curve:
   1. Open the curve via ctrl+f
   2. hold shift and drag any point down until the highest point at the right is BELOW your stable core clock (I would suggest to drag it all down until the highest point is at 1800 MHz for 10xx, 20xx and 30xx)
   3. Look up what the maximum stable clock at which voltage was during reviews. (Or use the lighter coloured default curve in the background to orientate)
   4. Drag the point where you want to lock your core clock to the correct position
   5. You will now have a FLAT line from the desired voltage until the far right
   6. Now find the sweetspot be either raising the point of the default voltage higher until you see artefacts or get a driver crash
      OR
      raise points further at the left to the height of the flat line. This will undervolt the card (important when hitting the power limit).
4. Overclocking the memory can give you a bigger performance boost than the core clock. This was the case with my GTX 1070.
5. 've found these values during a very quick search:
   1. 1740 MHz @ 850 mV, Memory +0
   2. 1950 MHz @ 950 mV, Memory +300
   3. 1980 MHz @ 936 mV, Memory +1000
   4. 2139 MHz @ 1038 mV, memory +1000
6. Important note:
   1. Always put power limit to the maximum Afterburner allows!
   2. When hitting the power limit, the bios will downclock the card quite roughly, causing the famous power spikes the early 30xx cards had when going back up, making PSUs shutting off!
   3. Your PC will thank you for not reaching the power limit and instead adjusting the undervolting/overclock.

Here's my undervolting curve of the 3080:
1830 MHz @ 818 mV (+205 MHz for that voltage point)

Default clocks or maximum stable clock would be somewhere slightly above 2000 MHz but as soon as you put load to this 3080, the power will go straight to 330W+, hitting the power limit and throttling down to as low MHz as 1700 MHz.
With this undervolting, the card never reaches the power limit, keeps the clock completely stable to have consistent FPS and depending on the game I either lose up to 5% fps or gain up to 5% fps.
Average power consumption is lowered to 180-250W instead of constantly being over 300W!

 

[RasmusP]

@Neilski wtf is going on with the paragraphs in lists? Tried to edit it a few times but it's just getting worse...
Even in BB-Code the paraphs are cut out when changing back to normal mode.
Some paragraphs are transferred though but sadly, the wrong ones

I'd like to have an empty line after each point

 

[Neilski]

@Neilski wtf is going on with the paragraphs in lists? Tried to edit it a few times but it's just getting worse...
Even in BB-Code the paraphs are cut out when changing back to normal mode.
Some paragraphs are transferred though but sadly, the wrong ones

I'd like to have an empty line after each point

I have noooo idea I'm afraid It may be that the CSS has been... adjusted in some way by someone

 

[Sascha Reynders]

@RasmusP you spend waaayyy too much time with your GPU

 

[R8pilot]

This may not be helpful, but I don't think overclocking a 2070, or almost any recent video card, will change the performance of a racing game to a noticeable degree in actual game play. The only meaningful improvement you would likely find is in benchmarks - as you have seen.

If "overclocking" would make a meaningful difference, then graphics cards would be sold with higher clock speeds out of the box. The reasons that there is any headroom left to overclock is to account for variability in chip and board performance, and to help prevent instability and premature failures. That is, leaving a bit on the table allows more chips to pass quality control at that speed, and it help lower returns due to potential instability, etc.

These days, in my opinion and experience, there just isn't much left to gain by overclocking, and not enough to make a significant or even noticeable difference in actual game play.

 

[mgh24]

I'll have to dig into Afterburner more, I was not aware of the graphs it could display.

However, driving home today, I thought I am probably chasing something that is not possible to do. My GPU is already running at 99% (CPU around 25%), so if it is already at 99%, there is nothing more to gain, correct?

 

[R8pilot]

My GPU is already running at 99% (CPU around 25%), so if it is already at 99%, there is nothing more to gain, correct?

Well, that's exactly the condition that could let overclocking do something. But see my cross-posted post above - it won't matter anyway unless you spend your days staring at a FPS counter instead of driving.

 

[RasmusP]

This may not be helpful, but I don't think overclocking a 2070, or almost any recent video card, will change the performance of a racing game to a noticeable degree in actual game play. The only meaningful improvement you would likely find is in benchmarks - as you have seen.

While I agree in general, I don't think this is always the case. Testing of the chips seems to be lacking, like the PSU issues with early 3080 drivers showed and marketing probably has some role in the MHz too.
I gain a few % in ACC, while saving 100W. I find this to be pretty insane.. Forums and reviews all show the same MHz @ same mV that I'm using so it doesn't seem likely that I have a "super chip".

With my GTX 1070 I could gain up to 10%, which isn't just a tiny bit.

But as I said overall I agree. Modern products need to perform stable even when being choked in a dusty case with only the CPU fan in a climate where you have 35°c in the rooms.
That's no comparison to my high airflow build with 25°c max during summer.

I'll have to dig into Afterburner more, I was not aware of the graphs it could display.

However, driving home today, I thought I am probably chasing something that is not possible to do. My GPU is already running at 99% (CPU around 25%), so if it is already at 99%, there is nothing more to gain, correct?

If you read my post again (edit: sorry! I had a word flipped there! Fixed it!), then, as R8pilot said:
Having the graphics card load at 99% is exactly the usecase where overclocking the graphics card will result in directly higher FPS!

If your graphics card is below 95-99% (depends on the application and exact GPU), then running higher MHz will only result in more headroom, since you don't need the maximum performance anyway.

The load value is always in relation to the maximum possible load at the current MHz.
This results in some funny stats like my 3080 showing 50% load while watching a youtube video.
It's running at 510 MHz only though

 

[RasmusP]

@RasmusP you spend waaayyy too much time with your GPU

Hehe well I started overclocking my components in 2013 so I got some years to learn the features and figure out how things work
Studying a mix of mechanical and electrical engineering is a good base to have fun spending time with GPUs.
*insert picture of me cuddling with my 3080*

 

[mgh24]

What I was hoping for with my fps gazing, was a way to measure performance, to see if I could get enough boost that AI in ACC did not look so cartoon-ish.

At present, with triple 2K monitors, I am at 80 fps with video settings to "mid" level. When adding AI to the mix, the cars appear to float above the surface of the road, and do not look realistic at all. Not a big deal, just looks bad. Also, I really don't know what to expect, how well AI can, or usually is, represented in ACC

 

[Novopaine]

I'll have to dig into Afterburner more, I was not aware of the graphs it could display.

However, driving home today, I thought I am probably chasing something that is not possible to do. My GPU is already running at 99% (CPU around 25%), so if it is already at 99%, there is nothing more to gain, correct?

When monitoring the CPU usage, it's best to report all cores/threads rather than the average CPU usage. Most games can only use a few CPU threads, and the rest goes unused no matter what the game conditions are. For example, if your CPU has 16 threads but the game can only use 4 threads, 25% average usage could mean that the 4 available threads are at 100%. So in my example you'd actually be CPU limited.

Just some food for thought.

 

[RasmusP]

When monitoring the CPU usage, it's best to report all cores/threads rather than the average CPU usage. Most games can only use a few CPU threads, and the rest goes unused no matter what the game conditions are. For example, if your CPU has 16 threads but the game can only use 4 threads, 25% average usage could mean that the 4 available threads are at 100%. So in my example you'd actually be CPU limited.

Just some food for thought.

Yes and no.
With amd Ryzen CPUs, Windows tends to keep a single application thread on a single cpu thread.
But on Intels you will see a single thread application showing load on 3 to all threads with varying loads. It's more efficient, somehow.

So basically apart from the overall cpu load maxing out or not, you can't see a cpu limit at all.

So simply look at the gpu load, if it's not 95% or higher, your cpu is the bottleneck.

 

[Novopaine]

I'm not sure I concur on the more efficient part, but there are a ton of factors that can affect performance beyond just hardware.

Here's an example of what I was talking about regarding the CPU threads:

 

[RasmusP]

I'm not sure I concur on the more efficient part, but there are a ton of factors that can affect performance beyond just hardware.

Here's an example of what I was talking about regarding the CPU threads:

I know what you meant with cpu threads. Put that rtss overlay over a single thread application and on a Ryzen cpu you will probably see one cpu thread going close to 100%.
I can tell you that with my old I7 2600k and with my current 10600k, I might see one cpu thread going going to something around 40-60% and the others changing and bouncing between 0-30%.
In some cases you could also see all cpu threads at a very low load like 5-20%.

If you want to see this, download cinebench (any version) and msi kombustor.
Both can fill the cpu with different amounts of threads so you can simulate a 1 thread app or a 40 thread app and see what's happening.

I did this with quite a few different CPUs out of curiosity.

 

[Novopaine]

To be clear I was only pointing out that when looking for bottlenecks, the overall CPU usage can be misleading. If a GPU is running at 50%, and the CPU is at 30%, one might assume that the PC has some kind of other problem like memory bandwidth, or maybe a driver issue. I've even read about people swapping out power supplies out of desperation. When you dig deeper, you may find that one or two cores are hitting a brick wall and limiting framerates.

I saw this issue frequently in the digital audio world. Users would complain that they are getting audio pops and crackling but not understand why. The CPU usage was only running at 20%, so it can't be the CPU they would say. After more digging, they realized one core was completely maxed out, and they had multi-core/multi-threading disabled in their DAW. Whoops.

 

[RasmusP]

To be clear I was only pointing out that when looking for bottlenecks, the overall CPU usage can be misleading. If a GPU is running at 50%, and the CPU is at 30%, one might assume that the PC has some kind of other problem like memory bandwidth, or maybe a driver issue. I've even read about people swapping out power supplies out of desperation. When you dig deeper, you may find that one or two cores are hitting a brick wall and limiting framerates.

I saw this issue frequently in the digital audio world. Users would complain that they are getting audio pops and crackling but not understand why. The CPU usage was only running at 20%, so it can't be the CPU they would say. After more digging, they realized one core was completely maxed out, and they had multi-core/multi-threading disabled in their DAW. Whoops.

Yep, 100% agree!
My point was just that the all the tools (Taskmanager, perfmon, afterburner etc) often don't show that "One or two cores are hitting a brickwall".
Here's how Cinebench R15 single thread looks like in Taskmanager and in Process Explorer:

Taskmanager not indicating any bottleneck, because we only see an average and not what's happening within a single CPU cycle.

Important here: 100% divided by 12 CPU threads = 8.33% maximum load for one thread. CB 15 is hitting 8.05%, fluctuating up to 8.3%. Clearly showing that the maximum single thread performance is reached.


What's interesting though is that due to Windows shuffling the single thread load around and using 6 CPU threads within one Taskmanager refresh tick, the efficiency is improved and the overall CPU load goes to 13%, instead of the theoretically possible 8.33%!