Where do you think the gaming framerate bottleneck should be?

[chaosmassive]

its easy
GPU, because maxed out GPU = only 3d performance is limited
but CPU = ENTIRE SYSTEM WILL SLOW ( background process, system process, and so on.. those processes will fight each other for process time in CPU)

and finally PCI-E interface has backward compatibilty and it will works with any GPU vendor from any OEM, you could plug in RTX 2080 Ti on top of G31 mainboard and it work just fine.
CPU? you want to upgrade to i11 9999X? oh too bad, you need new mainboard X999 with DDR9 !

 

[Bill_Bright]

The question is kinda wrong.
Optimally, you want your GPU to always be maxed out, so you could potentially always have the highest graphics performance.

I agree, the question is wrong.

The question isn't wrong. As I said right at the start, if there was no bottleneck, then the computer would run infinitely fast.

And that is why the question is wrong. The answer is, there should be no bottlenecks.

The answer is clear, the computer should ALWAYS be waiting on the user. ALWAYS!

 

[Vya Domus]

You're always bound by something, CPU, GPU, memory even storage.

Why would you sleep better at night knowing that it is because of your GPU you're dropping frames and not the CPU, beats me. You're still dropping frames regardless.

 

[EarthDog]

You're always bound by something, CPU, GPU, memory even storage.

Why would you sleep better at night knowing that it is because of your GPU you're dropping frames and not the CPU, beats me. You're still dropping frames regardless.

But hey, let's start a thread discussing where we THINK it SHOULD be......... that will get the collective REALLY going, no? Even better, attach a poll to it as that makes it official, the opinion of others.

Smart ass aside, I still don't understand what the OP is actually trying to get from this thread... some clarity would be great.

 

[argon]

I used to play for a lot of month with a 2011 years old phenom x6 1090t before I got upgraded from a HD6950 to my R9290X back in 2014 and definitely worth every cent to not upgrade the cpu before buying a better faster gpu, If you play games, always buy a powerful GPU over upgrading CPU, still in lots of game I see no changes

 

[eidairaman1]

I feel the dam cracking... but not a thing has changed in 6 months since 'leaving' so........yeah... doubtful it will be like before, but will post occasionally outside of my FS thread.

EDIT: I just noticed I lost all my 'knowledge' tags......... I guess I'm a moron and nobody should listen to me since I don't have them anymore.

Just not active as much is all.

 

[EarthDog]

I know exactly why they are gone.

They are nothing more than post count and frequency of a section...just muttering about its relative usefulness. This and the plus/minus system can both be binned.

...but I digress. This is a bit OT. Apologies. Will wait for qubits to see what I'm missing here.

 

[Voluman]

Erm, im not sure, i am understand the question, but there should be NO bottleneck at all

 

[Divide Overflow]

Going with Other - modern gaming is becoming more multi threaded and expectations are for much more complex AI in games. Ideally that should be taxing your CPU to it's limit. The graphics quality and special effects should also be maximized, which should be pushing your GPU to it's limit. So both.

 

[AsRock]

So, in a gaming system, where should that bottleneck be, CPU or GPU? Please vote in the poll.

Bottleneck ( no were P ), if there is going to be one would depend on the games you play which would be the deciding if it's a problem or not.

Typically for me it's GPU related for the most part, again though totally depends on the games youu play as some are heavy on one or the other or even both.

CPU bottleneck i find worse than GPU bottleneck.

 

[Bill_Bright]

Ideally that should be taxing your CPU to it's limit.

I am not buying that. No component should be taxed to its limits. IMO, if any resource is maxed out, it is now a bottleneck - or at least a lid on the bottle. What if tomorrow, you need just a little bit more?

I say ideally, there should always be a little headroom left. The only exception to that rule might be when benchmarking - when the goal is to see just where that taxed limit is.

 

[EarthDog]

Is it really a bottleneck if it is the fastest out? This is why I don't subscribe to the nothing should be a bottlneck... something HAS TO BE the bottleneck!

Reasonably speaking, with EXISTING technology, a stock (or overclocked) Intel CPU and a 2080Ti is the 'fastest' thing out there gaming wise. If one wants to be that technical about it, the GPU is always going to be your bottleneck in cases like 2080Ti and 4K. There is literally nothing that EXISTS that could improve it. So while it is nice to say in theory NOTHING should be a bottleneck, something always has to be. Be it CPU at low res or GPU at high res and every other option between.

A GPU is intended to be run at 100% levels. To do anything less many would be considered a rip off to consumers. Leaving meat on the bone in hardware comes in the form of overclocking, not in running a GPU at less than what it is intended to run at (an no, GPUs are intended to run out of the box what they come as due to binning. We have to have SOME faith says card partners and nvidia aren't intentionally running things slower than allows for good yields/profits/availability...

If you need a little bit more, you overclock it. If its a car, you buy go fast parts. But there shouldn't be a switch to simply enable performance while still fitting within the same parameters (yields/profits/availability). Its a fine line and one consumers really have to think is a reasonable limit already. There is evidence of that with the latest GPUs and CPUs honestly. Look at AMD CPUs. Almost literally ZERO overclocking capability past its all core boost speeds. 100-200 MHz. Intel OTOH similar specs, it can get past its all core boost by a few hundred MHz at least, but both aren't what they used to be a couple of generations ago. New GPUs, barely 100-200 MHz overclock on the cores (memory is a different story). So board/card partners, I feel ARE giving us what is realistically the best they can already without dipping into yields and profits (among other things)..and there is still a bit of headroom, just not nearly as much as back in the day.

 

[Papahyooie]

The central premise that "if there is no bottleneck, the computer will run infinitely fast" is simply incorrect.

It's not likely to actually happen, but one could easily imagine a CPU that is JUST fast enough to feed the GPU as fast as the GPU is capable of spitting out frames, while still maintaining OS operation. In this case, you'd have no bottleneck. That doesn't mean that somehow the GPU or the CPU gets to go faster, resulting in infinite performance... it simply means that the two are perfectly paired.

Now obviously this only applies to a SINGLE theoretical application, as different applications have different CPU/GPU requirements... Some require more CPU power than others, some require more GPU power than others. So it's not like this is going to result in any real discussion, because the situation will change per application, and even at different stages of the same application.

In this fantasy world, you'd want to match them as closely as possible. In the REAL world, where CPUs and GPUs cannot be perfectly matched:

If money is no object, then you'd definitely want a CPU that is more than capable of feeding your GPU and doing all the other stuff it needs to do, as that will result in the highest possible minimum framerates. In reality, the interesting question comes about when budgets are involved. A CPU that isn't quite capable of maxing out your chosen graphics card will be money better utilized, than buying a super high end CPU and sacrificing GPU power to fit into the budget. Sure, if you go high end CPU it will mean your bottleneck is on the GPU side, but you'd be sacrificing more performance than if you went with a CPU that can't max out your high end graphics card.

All in all, the OP is a flawed question.

 

[R0H1T]

If I had to pick one it'd be storage (A chain is only as strong as its weakest link) btw as some have said the poll seems kinda

 

[Bill_Bright]

Is it really a bottleneck if it is the fastest out?

Yes. Of course! Just because something is the fastest currently on the market, or what Man is currenly able to create, that does not mean something faster would not be better.

This is why I don't subscribe to the nothing should be a bottlneck... something HAS TO BE the bottleneck!

Why? By definition, a bottleneck implies something is being backed-up and delayed because it cannot get past unhindered. It is causing a slow down. That's not good - even if that slow down is not "perceived" at the final destination.

It's not likely to actually happen, but one could easily imagine a CPU that is JUST fast enough to feed the GPU as fast as the GPU is capable of spitting out frames, while still maintaining OS operation.

True but you are talking about just two links in the chain.

The data has to come from somewhere to feed the CPU in the first place. Where did it come from? A slow, bottlenecked drive? A slow bottlenecked network card? A slow bottlenecked network? And how did the data get to the drive?

How did the data get from the CPU to the GPU? Through a slow bottlenecked motherboard bus? How did the data get from the GPU to the monitor and then on to the screen?

And then there is the system RAM and graphics RAM, other various I/Os, and more.

I guess when the process from the source to final destination though every potential bottleneck moves with zero resistance at the speed of light, then there will be no bottlenecks - except that the speed of light "ONLY" travels at 186,000 miles per second - then that will be the bottleneck!

 

[eidairaman1]

That L4 cache is overrated. There is a reason neither Intel nor AMD chooses to include such a thing, the cost far outweighs the performance improvement. High frequency DDR4 made the on chip eDRAM redundant and obsolete.

Besides L1 is Static Ram, L2 is slowerthan L1, L3 is slower than L2. No need for L4-infinite...

 

[jboydgolfer]

not a thing has changed in 6 months since 'leaving'

sadly your correct.

 

[Divide Overflow]

I am not buying that. No component should be taxed to its limits. IMO, if any resource is maxed out, it is now a bottleneck - or at least a lid on the bottle. What if tomorrow, you need just a little bit more?

I say ideally, there should always be a little headroom left.

Extra headroom is underutilized, wasted resources. Ideally your CPU is feeding your GPU and allowing it to be at max, while still fulfilling gaming logic to it's maximum. You can always worry about future proofing, but that's future speculation and not the question at hand as I understand it.

 

[spectatorx]

I would approach that problem from different perspective. You should be picking parts as balanced as possible: mid-range gpu with mid-range cpu, high-end cpu with high-end gpu and so on. But if we are talking about pc config as a whole then i would say if you have to create bottleneck and this should be only because of tight budget then ram speed and main disk drive. Ram speed differences in some usecases make huge difference but still in fairly many it makes almost no difference. About disk i would prefer to have more powerful gpu and cpu instead of loading few seconds faster, really, i am patient person and i can wait these few more seconds for a game to load.

PSU is also interesting topic. You need to calculate how much wattage you really need, consider if you gonna overclock anything or not and pick proper 80+ certification. PSU should never be bottlenecking system and safely better be a bit overpowered instead. Similar with motherboards. If you are going for overclocking then pick higher rate motherboard, otherwise anything will be good for you, even cheap ones.

With this post i'm a bit contradicting myself as i just performed an upgrade, long time awaited one, though. This includes asrock x470 taichi ultimate, 32GB ddr4 3000MHz ram (2x16GB) and ryzen 2700x. I'm not going to overclock this cpu as after reading some articles and watching some videos i came to conclusion that oc makes no sense as this cpu usually clocks up by oc to 4,2GHz while it boosts up to 4,3GHz. If i am wrong on this please someone correct me, i am open for opinions on that matter. Probably, if ram will handle it, i will oc it to 3200 or 3333MHz. This means i could have go with cheaper motherboard and would be totally fine but i wanted to have one of top motherboards just "because i can". My psu is ocz zs550 which is 550W "just" 80+ bronze certified, gpu is radeon r9 380 4GB. These are at the moment the weakest parts of my build and i decided to be so as i am going with progressive upgrade plan. Next is going to be psu (650 platinum), then hard drives and somewhere later in time gpu. All this is an upgrade from fx-6300 and 16GB 1333MHz ddr3. With gpu i am fairly satisfied and gonna wait for something really worthy, maybe radeon 590 or vega 56/64 or one of navi gpus when they will be launched. That's what i meant by saying "a bit contradicting myself" as at the moment this build is a bit bottlenecked by gpu and a bit by psu but they are things in progress.
As a side note about this my upgrade: i'm doing performance comparison in games, before/after upgrade and gonna post it as video on youtube.

 

[Vayra86]

My god this topic is really going nowhere at all.

Better off examining the bottleneck of a nice beer IMO

 

[Bill_Bright]

Ideally your CPU is feeding your GPU and allowing it to be at max, while still fulfilling gaming logic to it's maximum.

No! Sorry, but that is incorrect!

Ideally, your CPU is feeding your GPU as much as your GPU calls for (and will ever call for!) as fast as the GPU can take it - while still fulfilling all the demands of the operating system, all other running programs (like security), and all other system interrupts like storage and networking access, HID I/Os (human interface device - keyboard and mouse inputs) and whatever other multitasking demands are going on at that moment in time (like processing and handing off audio to the sound device).

Extra headroom is underutilized, wasted resources.

Wasted? No. Not unless you are talking about extreme amounts of extra headroom, and we're not. Having some resources in "reserve" is simply smart.

If your computer drew 250 watts, would you buy a 250W power supply? No, you would buy a PSU with extra headroom, like a nice 400W supply. Using a 1000W supply would be wasting resources - as in your budget resources.

If your CPU was capable of running without issues up to 80°C would you configure your cooling to maintain your CPU temp at 80°C. No, you would configure cooling to maintain the CPU well within its designed operating range.

Each step along the entire path must always be able to handle everything thrown at it without any delays (or overheating). And each step must be able to deliver to the next step everything the next step can take, as fast as it can accept it. That's where the balance comes in. It is not between two points, but between all points.

i would say if you have to create bottleneck and this should be only because of tight budget then ram speed and main disk drive.

I disagree with RAM "speed" unless you already have "more RAM than you will ever need!" (whatever that means). That is, as a general rule (which means there are exceptions, of course), more RAM trumps faster RAM every day. I would much rather have 8GB of slow RAM than 4GB of fast RAM.

I do agree with you about disk drives. Sadly, many folks think SSDs only matter for boot and wake times. Not true. Since operating systems are very disk intensive, when the OS can finish tasks faster with a SSD, that frees up resources more quickly that can then be utilized elsewhere. So I would much rather take a slow SSD over the fastest hard drive any day of the week. And when it comes to "virtual memory" (RAM + Page File on the disk), more system RAM can mean fewer calls to the slow PF.

PSU is also interesting topic. You need to calculate how much wattage you really need, consider if you gonna overclock anything or not and pick proper 80+ certification.

Ummm, while I am all for picking an efficient PSU, 80+ certification has nothing to do with power demand. If your computer (CPU, graphics, RAM, drives, etc.) need 300W, they will pull from the PSU 300W regardless if the PSU is a 500W 70% efficient PSU or 750W 85% efficient PSU. All the 80+ Cert means is how much power from the wall is wasted in the form of heat. The PSU is still going to pull from the wall what the computer needs, regardless the size or efficiency of the PSU (assuming the PSU is big enough). The difference is how much extra power will be pulled that will be wasted. Whatever the inefficiency amounts to has nothing to do with bottlenecks.

As far as the poll question, when it comes to gaming framerate bottlenecks, does everything boil down to the relationship between the CPU and GPU? No.

 

[Divide Overflow]

Wasted? No. Not unless you are talking about extreme amounts of extra headroom, and we're not. Having some resources in "reserve" is simply smart.

If your computer drew 250 watts, would you buy a 250W power supply? No, you would buy a PSU with extra headroom, like a nice 400W supply. Using a 1000W supply would be wasting resources - as in your budget resources.

If your CPU was capable of running without issues up to 80°C would you configure your cooling to maintain your CPU temp at 80°C. No, you would configure cooling to maintain the CPU well within its designed operating range.

If optimized gaming code is not maximizing your available resources, the software is the bottleneck. You're bringing in other factors outside the scope of the original question, but so am I when I point to the software.

 

[Bill_Bright]

If optimized gaming code is not maximizing your available resources, the software is the bottleneck.

Nah! Sorry, but totally wrong. Software should be using resources in the most efficient manner, not maximizing it.

What is a major concern for browsers and security software, operating systems etc.? If they are big resource hogs or not.

When it comes to gaming, what matters is the speed in which backgrounds and moving objects are drawn. And that depends on the design and flow of the game as programmed by the game developer. If the programming calls for instant imaging, it is up to the hardware to deliver.

Why is there a constant drive to create more powerful and faster hardware? To crunch the numbers (process the data) faster.

 

[Papahyooie]

True but you are talking about just two links in the chain.
The data has to come from somewhere to feed the CPU in the first place. Where did it come from? A slow, bottlenecked drive? A slow bottlenecked network card? A slow bottlenecked network? And how did the data get to the drive?
How did the data get from the CPU to the GPU? Through a slow bottlenecked motherboard bus? How did the data get from the GPU to the monitor and then on to the screen?
And then there is the system RAM and graphics RAM, other various I/Os, and more.
I guess when the process from the source to final destination though every potential bottleneck moves with zero resistance at the speed of light, then there will be no bottlenecks - except that the speed of light "ONLY" travels at 186,000 miles per second - then that will be the bottleneck!

Yes, I'm only referring to two links in the chain, because those were the two links mentioned in the OP. In the thought experiment that I derived, obviously it's assuming *either* instantaneous function of those other components, or only slightly more realistically, all those other components are also perfectly matched to provide the CPU and GPU with the data they need in order not to bottleneck each other.

My point stands that it's a flawed assumption that a non-bottlenecked system runs infinitely fast, and also that it's a superfluous exercise to consider them without the context of cost.

 

[Bill_Bright]

My point stands that it's a flawed assumption that a non-bottlenecked system runs infinitely fast

Totally agree with this. The laws of physics (and the speed of light) limits (bottlenecks) electron flow. Thus infinitely fast will never be possible - at least not until quantum superposition becomes an everyday reality.

and also that it's a superfluous exercise to consider them without the context of cost.

Not really. If total budget for both CPU and graphics solutions are a factor, then that budget should be stated up front. If no budget, then current technologies and product availability, and/or theory is what matters - especially when the costs of such items are currently within reach of many members of this site. We are talking a couple $1000, not $millions.

It is not superfluous because striving to reach theoretical limits is a very valid goal and necessary exercise to consider. And if the costs are not totally prohibitive, seeing that that goal is within reach can be a real motivator to get there.

That said, I stick with my comment above. "There should be no bottlenecks. The computer should ALWAYS be waiting on the user. ALWAYS!"