Graphics card VS CPU 'bottlenecking' in games

[Kalm_Traveler]

Wanted to get a discussion going to figure out the best ways to determine where the bottleneck lies on a given setup. 

 

This came to mind because last year when I upgraded from 1 1080 to 2 1080's, then again to a pair of 1080 Ti's I had checked Witcher 3 (the only modern game I've played recently, since WoW only sort of counts) FPS with all the same maxed out settings, and everything else in my setup remaining the same more or less.

 

My screen all this time has been a 3440 x 1400 100Hz Gsync Asus display, the CPU has been the same 6900k @ 4.5 GHz, but RAM was replaced from 8 x 8gb 2666 CL16  on the 1080 and 1080 Ti testing to 4 x 16gb 3200 CL14 with the Titan V testing now, so that could play a role as well.

 

Basically, as I recall with the SLI 1080's, Witcher 3 would be around 40-60 fps, with 1 1080 Ti it was about the same, with SLI 1080 Ti's it was around 67-80 FPS. I finally got around to playing it again last night with the Titan V (technically both are installed and the 2nd one is being used for PhysX automatically by the Nvidia Configuration panel) and now Witcher 3 is running between about 95-110 FPS.

 

-------------

 

The reason this is strange to me is that when I looked at CPU and GPU utilization last year with the 1080 Ti's, It appeared that 2 or 3 CPU cores were pegged on 99% utilization when running W3, while both graphics cards were around 80% utilization.  This led me to believe that my CPU was then the bottleneck, but all I've done is install a bit faster RAM, and better graphics cards (not even in SLI) and seen a very huge performance increase.

 

I didn't check CPU util last night but just seeing the FPS increase seems pretty obvious that in my previous testing with the 1080 Ti's, the 6900k was not actually the bottlenecking part, right?

[Overl0rd]

Perhaps the Titan V being a completely different architecture with different drivers could have surely been a factor.

[Princess Luna]

11 minutes ago, Kalm_Traveler1 said:

but all I've done is install a bit faster RAM,

Memory frequency takes much bigger role these days, the gap of performance between 2133mhz to 3200mhz is huge on the latest more modern titles... also yes 100% usage is just the easier way to spot a CPU bottleneck not the only way... I think single core wise the i7 6900K indeed can lag a dual 1080 Ti setup somewhat... Also a single card configuration will always work better than a dual setup.

 

Scaling of performance will simply always better on single cards configurations.

[Samfisher]

Not many have done testing with RAM speeds on W3, but for Overwatch going from 1600MHz to 3000MHz RAM can net you over 100+ FPS, everything else in the system being the same.

[Kalm_Traveler]

5 minutes ago, Princess Cadence said:

Memory frequency takes much bigger role these days, the gap of performance between 2133mhz to 3200mhz is huge on the latest more modern titles... also yes 100% usage is just the easier way to spot a CPU bottleneck not the only way... I think single core wise the i7 6900K indeed can lag a dual 1080 Ti setup somewhat... Also a single card configuration will always work better than a dual setup.

 

Scaling of performance will simply always better on single cards configurations.

 

I understand that RAM speed will have an impact, but 2666 to 3200 is only half the increase of the speeds you mentioned. 

 

From the older testing, it is obvious that all else being equal, Witcher 3 'works better' with two of the same card in SLI than one, as the FPS increased a bit with both SLI 1080's over a single 1080, and SLI 1080 Ti's over a single 1080 Ti.

 

The reason I'm confused is that it looked like I was hitting a CPU bottleneck since the SLI 1080 Ti's were only around 80% utilization, but I was seeing several cores on the CPU pegged at ~ 100%. 

 

At that point, several forums had suggested that I needed to get a (at the time) 7700k if I wanted to see 100 fps stable in WC3 with everything maxed at 3440x1440 but obviously that has proven to not be the case.

 

Is there a more accurate way to figure out where a performance bottleneck is?

I'm going to switch to SLI Titan Xp's probably (unless the new 1180/2080/whatever it is bests the Titan Xp) since the handful of games I play all benefit from SLI and I'm pretty sure two Titan Xp's on water will outdo what I'm seeing from a single Titan V in those games, but at the same time this is making me want to really make sure this build is as balanced as possible  - IE there's not one glaring part that is way behind the rest of the system in overall performance.

[Kalm_Traveler]

15 minutes ago, Overl0rd said:

Perhaps the Titan V being a completely different architecture with different drivers could have surely been a factor.

for sure, it's a completely different part - though to be fair the Titan V runs on the Geforce driver stack just like all the Geforce-branded cards.