• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

GPU-CPU Bottlenecking Ever End?

C

Caliche

Limp Gawd
Joined
Jun 1, 2004
Messages
356
Simple Question. Will the bottlenecking that is seen between the GPU-CPU end in the upcoming future? If I am asking the question right? With the incorporation of two more channels into the DDR3 interface and storage performance slowly but progressively gaining ground, I thought by now, at least, this sector would have seen an end to bottlenecking. I'm a little confused though, cause when I look back 5 years till present, we've seen very significant increase in CPU performance and GPU performance. Not to mention multi-core support on both hardware, plus the add support in the software also. Yet the more progression in this market with newer hardware, there still is abundant bottlenecking present. To me it seems like it comes back to the Motherboard and design?

So either the motherboards are going to get larger in size to accomadate more channels for Ram or the Ram itself will have to be inbedded on the motherboard itself? Only last option is some other advanced interface that will replace Ram. Not to mention updates in the the North and South bridges as well. Hhhmmm will bottlenecking ever end....

So what do you all think?
 
There are two precious resources to game developers.. CPU cycles and GPU cycles. As each piece of hardware advances, the code (also see "code slop") expands to fill their capacities, in terms of calculation complexity and effects.

See the joke quote "Any given program will expand to fill all the available memory."

The reason for this is that they'll want ot make use of what they have, not just create something then be done with it, despite the fact that they have a bunch of extra CPU and GPU power left unused... they want to add in more effects and use everything they can to differentiate their games and make them look and play as good as they can.

So no, this phenomenon will not go away unless there is some discontinuity where CPUs instantly get many orders of magnitude faster while GPUs do not.
 
it really isn't possible for "bottlenecking" to end as long as their is a CPU and a GPU. All bottlenecking is saying is that a program (game in this case) will only run as fast as the slowest part since one realies on the other.
 
here is how you can stop the bottlenecking. Remove all but 256mb of RAM. I am certain that neither the CPU nor GPU will be a bottleneck.
 
I'm not sure what the OP is asking. Something will always be the bottleneck unless all games start frame-limiting themselves to 30fps.
 
IMO, doubtful. New GPU architectures are coming out and evolving way faster than their CPU counterparts. Core 2 has been out since June 06' if I recall correctly and we're just now getting something new. K8 has been out for even longer, 03'?, and it's still around and in modern processors. During just the time since Core 2 was released how many GPU architectures have come out?
 
Kuschelweich said:
IMO, doubtful. New GPU architectures are coming out and evolving way faster than their CPU counterparts. Core 2 has been out since June 06' if I recall correctly and we're just now getting something new. K8 has been out for even longer, 03'?, and it's still around and in modern processors. During just the time since Core 2 was released how many GPU architectures have come out?
Click to expand...

Not really. G80 was a major change over the 7xxx series, but G92 and GT200 are both revolutionary changes, with the same core architecture. You could argue the change from Core 2 to i7 was much larger than the change of G92 to GTX 2xx
 
Well, I may be dreaming of the impossible, but if someday both general purpose processing and graphic rendering can be executed in one unified processor with massive parallel execution units, then perhaps we can balance out the workload dynamically the way we did with vertex/pixel shading in GPU these days with unified shaders. That way, we can remove such bottlenecking as the processing resources can be utilized depending on situation

In a CPU intensive game, more resources can be use for the general purpose stuff, and in a graphic intensive game, more resources is allocated to render graphics

Just a guess, I may be out of my mind
 
Well, when you look at how much we ALL are spending in terms of cold hard $$$, would you want to see vastly increased performances gains. That's how it was in the beginning of this decade. Every year the industry was showing increased gains with technology. Now a days it seems like every 6 months or so, a new GPU or CPU comes out and the gains suck for the most part. Here is my point: Remember when we were all playing back in the day, and we couldn't wait for more of this and that in our hardware and now it is finally here and our games dont look that much greater in some respects. That is why if the industry would reduce this bottlenecking, than Yes, as some of pointed out, the games would looked crazy epic awesome, that is what I think.

Atech said:
Actually it will in the future:
http://techresearch.intel.com/articles/Tera-Scale/1421.htm

Like Tim Sweeny said, the GPU is a temp fix for something that acutally is a general computing problem:
http://arstechnica.com/articles/paedia/gpu-sweeney-interview.ars

Ever wonder why NVIDIA is so desperate to expand the use of their GPU's outside the 3D world? ;)
Click to expand...

Atech, is right, the GPU is a temp. fix for right now. I totally forgot about this. I can definitely see this happening in the next 5-10 years. This is one solution that can or also most eliminate bottlenecking.

rally9x said:
it really isn't possible for "bottlenecking" to end as long as their is a CPU and a GPU. All bottlenecking is saying is that a program (game in this case) will only run as fast as the slowest part since one realies on the other.
Click to expand...

Good Point!

Kuschelweich said:
IMO, doubtful. New GPU architectures are coming out and evolving way faster than their CPU counterparts. Core 2 has been out since June 06' if I recall correctly and we're just now getting something new. K8 has been out for even longer, 03'?, and it's still around and in modern processors. During just the time since Core 2 was released how many GPU architectures have come out?
Click to expand...

Puts some rendition on what I was writing above. All these years later and look where the industry still is now.

Hornet said:
Well, I may be dreaming of the impossible, but if someday both general purpose processing and graphic rendering can be executed in one unified processor with massive parallel execution units, then perhaps we can balance out the workload dynamically the way we did with vertex/pixel shading in GPU these days with unified shaders. That way, we can remove such bottlenecking as the processing resources can be utilized depending on situation

In a CPU intensive game, more resources can be use for the general purpose stuff, and in a graphic intensive game, more resources is allocated to render graphics

Just a guess, I may be out of my mind
Click to expand...

This goes back to the fact that is seems, the industry will have to intergrate the CPU and GPU.

My next question really pertaining to bottlenecking is: Storage Performance. Will we, as consumers, need more RAM? How do you all think that RAM should be intergrated into our computer hardware to combat the effects of bottlenecking?
 
I think your all wrong. The reason there is CPU or GPU bottlenecking is that computer programs (i.e. games) continue to move forward trying to do more things. Unless programming became stagnant, and made no forward progress (think "Half Lift 2 engine is the pinacle of computer games and can not be improved upon, only imitated") there will always be a bottleneck.
 
Comte said:
I think your all wrong. The reason there is CPU or GPU bottlenecking is that computer programs (i.e. games) continue to move forward trying to do more things. Unless programming became stagnant, and made no forward progress (think "Half Lift 2 engine is the pinacle of computer games and can not be improved upon, only imitated") there will always be a bottleneck.
Click to expand...

And what exactly does that have to do with the relationship between CPU and GPU?
 
meatfestival said:
And what exactly does that have to do with the relationship between CPU and GPU?
Click to expand...

Nothing...I look forward to have an x core CPU that does it all.
It will be good in all games, no matter if they are CPU or GPU limitied by today standards.
 
meatfestival said:
No, it implies optimum performance, not infinite. If you make one chip perform the functions of 2, where's the bottleneck?
Click to expand...

A chip isn't a monolithic processor running in lock step. Even within a single ASIC there are various data paths, processors, caches, external links etc. In a single GPU there are varying bottlenecks in the texture units, shaders, memory controller etc. Making one chip instead of two won't make a difference - it would just load balance better. Same way some CPU loads are bandwidth bound, FPU bound, etc.....

The only way to remove a bottleneck is for every single stage in the process to run at the exact same speed and nothing ever waits on anything else. That's not going to happen - something must be the slowest.
 
trinibwoy said:
A chip isn't a monolithic processor running in lock step. Even within a single ASIC there are various data paths, processors, caches, external links etc. In a single GPU there are varying bottlenecks in the texture units, shaders, memory controller etc. Making one chip instead of two won't make a difference - it would just load balance better. Same way some CPU loads are bandwidth bound, FPU bound, etc.....

The only way to remove a bottleneck is for every single stage in the process to run at the exact same speed and nothing ever waits on anything else. That's not going to happen - something must be the slowest.
Click to expand...

That's certainly true for today's processors, but when the CPU and GPU eventually become one, I feel that the architecture will be so different that a lot of the above will no longer be relevant.
 
Perhaps but some things will forever remain - not the least of which is off chip data storage. As long as there's external memory that feeds the chip there will be a bottleneck somewhere. Maybe some day far in the future texture LOD calculation, addressing, filtering, AA, blending et al will all run on the shader core. But the data still has to come from somewhere.
 
trinibwoy said:
Perhaps but some things will forever remain - not the least of which is off chip data storage. As long as there's external memory that feeds the chip there will be a bottleneck somewhere. Maybe some day far in the future texture LOD calculation, addressing, filtering, AA, blending et al will all run on the shader core. But the data still has to come from somewhere.
Click to expand...

But it will no longer be a question about CPU/GPU bottlenecking...as the OP asked about.
 
Rockjay420 said:
i thought that was AMD's whole thing, integrating gpu's on the cpu die.
Click to expand...

not sure about AMD's concept there, but to remove the GPU-CPU bottleneck, we need a processor that is capable of executing both general purpose instructions and graphic rendering, not because you have the GPU and CPU on one single die, but because the processor's execution unit can execute either type of instructions

putting a GPU and a CPU on one same die will not solve the problem.It has to be a unified processor
 
CPU's have been built for specific tasks, just like GPU's, SSL accelerators, IPS accelerators etc. They are architecturally different for a reason. Now I'm not saying that integrating a GPU and a CPU into one chip isn't going to help, but in doing so, you also raise the cost of the chip.

The majority of people couldn't give a crap about having kickass video processing built into the CPU, let alone pay more for something they won't ever utilize. That would be such a drastic change to computer architecture, I just can't imagine it happenning.
 
i think the cpu bottleneck is pretty close to going away permanently if some form of hardware physics takes off.
i think at that point, any quad core cpu would be pretty much equal for gaming, and the real differences would show in other apps. when you think about it, outside of physics, not alot of really complex things in a game for the cpu to do that the current architectures would struggle with. im thinking of multiplayer fps games specifically.
 
You must log in or register to reply here.
Back
Top