A question about CPU bottleneck when upgrading GPU

[hsjj3]

There is one scene in Hitman where my CPU usage is maxed out, and my GPU usage drops below 99%. It can be easily reproduced by me.

To me that's a textbook definition of a CPU bottleneck, right?

So in this scene, I get about 52fps.

Now, were I to play it with a GTX 1060 (on the way, currently have a 960), at the exact same settings and PC specs otherwise, will I still be stuck at 52fps in this particular scene that's bottlenecked by the CPU?

I'm not fully clear as to what to expect and how CPU bottleneck works.

Thanks!

 

[Headfoot]

What CPU do you have

 

[poofyhairguy]

To quote the funniest single post on this forum in months:

There is a one Russian flipping tables somewhere right about now because you could make the same decision back when you purchased your 960.

 

[Pottuvoi]

If by lowering resolution you do not get any performance, you most likely have a CPU bottleneck.

 

[Yuriman]

Sounds like a classic CPU bottleneck. However, NV drivers tend to be lighter on the CPU (in DX11 games), so if you're moving from and AMD card to an NV card, you might still gain some frames in CPU bottlenecked circumstances.

 

[Flapdrol1337]

If gpu usage drops you're cpu limited, cpu doesn't even have to have all cores on max.

Unless you have vsync or some other limiter.

 

[Dresdenboy]

Maybe interesting in this regard: A review of GTX 1060 and RX 480 in old computers:
http://www.hardwareunboxed.com/gtx-1060-vs-rx-480-in-6-year-old-amd-and-intel-computers/

 

[Ketchup]

What CPU do you have

I am still waiting for the answer to this question. And unless your screen is a 120, I would be VERY surprised if you noticed visual issues at 52.

But I guarantee you would notice a difference if you turned off the monitoring apps and enjoyed your game.

 

[hsjj3]

It's a i5 3330. And sure it's nothing super apparent, but still got me thinking.

Now, if 52fps is the limit in this one particular scene, and let's assume that it is indeed caused by a CPU bottleneck...all I want to know is, keeping everything else constant (system specs and game settings), will a GPU upgrade improve performance at all, or is the 52fps the limit no matter how good a GPU I get?

 

[Flapdrol1337]

This scene is in hitman? Could run it under dx12, should improve it. Or are you already?

 

[ZGR]

There are tons of games that need as much single threaded performance as they can get, and the i5 3330 is very poor in that regard.

The only thing that would help (if you kept your CPU) would be to reduce driver overhead. In the past, Nvidia's DX11 driver overhead caused weak CPU's to get higher FPS than using a comparable AMD card.

Things have changed. AMD's Polaris has lower overhead as well, possibly lower than Maxwell. We need more data regarding Polaris vs Pascal though. In DX12 and Vulcan, Polaris driver overhead is miniscule.

What I would suggest you do is upgrade to a used i7-3770k and overclock it to a conservative 4.4 GHz. This lets you use a cheap $20 cooler and you get much faster single threaded performance.

Your current i5 will bottleneck a lot of GPU's in certain areas on a lot of games. Upgrading to a 480/1060/1070 will still net you a solid performance boost.

If this is the only game you play that you notice you are CPU limited, then I wouldn't worry about it.

One cool thing you can try is downloading Throttlestop. This lets you change the CPU multiplier. So if you cut your CPU clockspeed in half, and your framerate scales linearly with CPU clockspeed, then it is very easy to know which games need more clockspeed.

 

[Yuriman]

It's a i5 3330. And sure it's nothing super apparent, but still got me thinking.

Now, if 52fps is the limit in this one particular scene, and let's assume that it is indeed caused by a CPU bottleneck...all I want to know is, keeping everything else constant (system specs and game settings), will a GPU upgrade improve performance at all, or is the 52fps the limit no matter how good a GPU I get?

All else being equal, changing your GPU would not improve framerates.

However, as others have mentioned, there are plenty of other factors involved. Driver overhead is different between manufacturers, and different in DX11 vs DX12.

 

[Ketchup]

It's a i5 3330. And sure it's nothing super apparent, but still got me thinking.

Now, if 52fps is the limit in this one particular scene, and let's assume that it is indeed caused by a CPU bottleneck...all I want to know is, keeping everything else constant (system specs and game settings), will a GPU upgrade improve performance at all, or is the 52fps the limit no matter how good a GPU I get?

When I look at the benchmarks that have been done for this game, I am seeing a nice jump in performance based on video card. This usually means that, while a more powerful CPU will get you a couple more FPS, it won't be a huge jump.

you can see the difference between your 960 and a 980 (which should match your 1060) here:

In short, don't buy this CPU just to make this one game play better.

 

[MajinCry]

There are tons of games that need as much single threaded performance as they can get, and the i5 3330 is very poor in that regard.

What I would suggest you do is upgrade to a used i7-3770k and overclock it to a conservative 4.4 GHz. This lets you use a cheap $20 cooler and you get much faster single threaded performance.

Your current i5 will bottleneck a lot of GPU's in certain areas on a lot of games. Upgrading to a 480/1060/1070 will still net you a solid performance boost.

Surely he'd be best off getting an i5 3570k and super clocking that? Save some pennies and get a wee bit better power usage, seeing as how it's single threaded perf he's looking for.

 

[ZGR]

I consider it a waste to spend money on another i5. I consider i5's too anemic in modern games. Hyperthreading makes a big difference in mulithreaded titles. OP probably doesn't have a Z motherboard anyways, so a measly BCLK overclock is all they can do.

Better off not doing anything until a new build is necessary.

 

[Silverforce11]

The solution here is to run Hitman in DX12 mode. It's good for weaker CPUs.

 

[hsjj3]

The solution here is to run Hitman in DX12 mode. It's good for weaker CPUs.

I will be honest, the game runs worse on DX12. When I check the graph, yes there is slightly less CPU bottleneck, but it still exists.

 

[hsjj3]

Surely he'd be best off getting an i5 3570k and super clocking that? Save some pennies and get a wee bit better power usage, seeing as how it's single threaded perf he's looking for.

No a K series i5/i7 is out of the question. I live in a hot place.

Like I said in the OP above, I really haven't felt any major slowdown from my CPU. It works, so I'm not upgrading the CPU till Kaby Lake comes out. Then I'll get either the mid or highest non K i5 CPU.

 

[hsjj3]

All else being equal, changing your GPU would not improve framerates.

However, as others have mentioned, there are plenty of other factors involved. Driver overhead is different between manufacturers, and different in DX11 vs DX12.

Hmm, two posters above say it will improve regardless, but you say it will stay exactly the same.

This is what I actually wanted answered in the first place, couldn't find a definite answer on the Internet. [Again, like I said, assuming that the bottleneck does fully origin from the CPU]

 

[Yuriman]

When CPU bottlenecked, upgrading your GPU (and keeping all else equal), your framerate will not increase. However, you're probably not CPU bottlenecked 100% of the time, and when you aren't, your framerate will go up, which means you'll probably still see higher averages. Additionally, NV GPUs are generally lighter on the CPU in DX11, so if you move from an AMD to an NV GPU in a DX11 CPU-bound game, you'll also see an improvement.

As others have said, DX12 reduces CPU load a lot, so if a game offers a DX12 rendering path, you can also get much higher framerates with your existing CPU.

 

[hsjj3]

Guys, to share, this is the graphs I get on MSI AB when running the Hitman benchmark in DX 11 and DX 12. The GPU usage is much smoother in DX12 but the true fact is that I see lower average framerates with DX12 for whatever reason.

I am running the game on Medium settings + High textures. VRAM and RAM are not a bottleneck, I have 4/8GB respectively.

DX11: http://imgur.com/a/0Zz54

DX12: http://imgur.com/a/juR83

If somebody can have a quick look at those graphs and tell me some useful info, I'd appreciate it. Especially in the first graph, I want to know why there is a long period somewhere in the middle of the benchmark where the GPU usage is well under 99%.

 

[hsjj3]

I would also like to ask an additional question. Do in-game settings in general have zero impact on CPU usage, or do they? Stuff like AA, AO, shadows, draw distance, texture filtering etc. Were I to run a game a random AAA game at Ultra and then at Medium, should the CPU performance be consistent between the two, or will it improve at Medium settings?

And if these settings do make a difference, what are the ones that make the most difference to counter a a CPU bottleneck?

Simple hypothetical situation. Imagine at EPIC settings with a GTX 1080 I get ~100 fps in Overwatch. I am always limited by CPU. Were I to drop down to the MEDIUM preset, can I expect however little at least some improvement in my framerate due to less CPU bottleneck?

 

[AtenRa]

There is one scene in Hitman where my CPU usage is maxed out, and my GPU usage drops below 99%. It can be easily reproduced by me.

To me that's a textbook definition of a CPU bottleneck, right?

So in this scene, I get about 52fps.

Now, were I to play it with a GTX 1060 (on the way, currently have a 960), at the exact same settings and PC specs otherwise, will I still be stuck at 52fps in this particular scene that's bottlenecked by the CPU?

I'm not fully clear as to what to expect and how CPU bottleneck works.

Thanks!

No, you will get higher fps

 

[AtenRa]

All else being equal, changing your GPU would not improve framerates.

yes it would

 

[AtenRa]

I would also like to ask an additional question. Do in-game settings in general have zero impact on CPU usage, or do they? Stuff like AA, AO, shadows, draw distance, texture filtering etc. Were I to run a game a random AAA game at Ultra and then at Medium, should the CPU performance be consistent between the two, or will it improve at Medium settings?

And if these settings do make a difference, what are the ones that make the most difference to counter a a CPU bottleneck?

Simple hypothetical situation. Imagine at EPIC settings with a GTX 1080 I get ~100 fps in Overwatch. I am always limited by CPU. Were I to drop down to the MEDIUM preset, can I expect however little at least some improvement in my framerate due to less CPU bottleneck?

Yes a lot of in-game settings affect CPU performance. If you lower the settings you become more CPU limited if you have a fast card, but it will also increase fps vs the Ultra settings no matter if you are still CPU limited.

Example,

CPU A + 1080 Ultra settings = 100fps
CPU B + 1080 Ultra settings = 100fps

CPU A + 1080 Medium settings = 140fps
CPU B + 1080 Medium settings = 160fps

with CPU A you increase fps going lower settings to Medium but you are still increase fps vs Ultra settings no matter if you are CPU limited vs CPU B.