What is the fastest GPU possible without bottlenecking this CPU?

  • Thread starter The_Absolute
  • Start date
  • Tags
    cpu Gpu
In summary, the conversation discusses the possibility of a GPU bottleneck with an AMD Phenom II X4 965BE and the necessity of a Core i7 clocked at 4.0 GHz to properly run the latest graphics cards. It is argued that a bottleneck is not a major concern and that having enough video memory is more important. The conversation also mentions the potential performance increase with a new system and a Radeon HD 5870 graphics card. It is noted that the HD 5870 is almost three times faster than the HD 4850 and that a properly built game should exhaust the arithmetic performance of the GPU regardless of bandwidth constraints. The conversation concludes by discussing the relative speeds of the two graphics cards in terms of arithmetic performance and memory
  • #1
The_Absolute
174
0
What is the fastest GPU setup you can have without encountering any major bottlenecks with an AMD Phenom II X4 965BE? Is it absolutely imperative to have a Core i7 clocked @ 4.0 GHz to properly run the latest graphics cards, especially in SLI/CrossfireX?

My monitor only goes up to a 720p resolution. I heard that a bottleneck only appears at a certain resolution.

My current system is a C2Q Q9400 2.66 GHz @ 2.66 GHz, 6GB 800Mhz DDR2 RAM, and a Radeon HD 4850. (512 MB) I was considering building a Phenom II X4 965 desktop, with 6-8GB of DDR3 1333 MHz RAM, an ASUS Crosshair Formula III motherboard, and a Radeon HD 5870 graphics card.

Even if I have a bottleneck with my new PC, should I still see a huge performance increase across every graphical setting, compared to my old PC and HD 4850?

The HD 5870 is almost Three times faster than my HD 4850. Most games are GPU bound, rather than CPU bound. A bottleneck with a Phenom II X4 965 @ stock speeds won't pose a crippling hit to performance with an HD 5870?
 
Computer science news on Phys.org
  • #2
The GPU bottlenecking the CPU or vice-versa phenomenon is a myth. Although it's possible in practice, it is an atypical situation regardless of virtually any reasonable system specs.

A sufficiently slow CPU will limit the framerates you can obtain in a game, but this is not a GPU bottleneck, it is merely a bottleneck in the CPU-part of code execution. It's the CPU bottlenecking itself, the GPU can still run as fast as it pleases regardless of CPU insufficiency.

A slow CPU is evident at lower resolutions rather than higher. At higher resolutions you increase the GPU part of the work load. Only when the GPU is relieved of work does the performance difference between CPUs become more evident.

There is a very real limit on the bandwidth of the PCIe bus, and there is a limit to how much data a CPU can pipe through a bus. You can run into circumstances where bandwidth across the bus and/or the processor's ability to pipe information is the limiting factor in an application, rather than arithmetic performance or local latency/bandwidth on the graphics card, however this is a very rare issue in a well developed game. A properly built game should exhaust the arithmetic performance of the GPU regardless of bandwidth constraints, so even on an extremely slow CPU and bus, the GPU does not need to be constrained. Your limiting factor at high resolutions will be video memory. If your video memory is insufficient, the CPU will need to do more work to perform constant swapping, however in this case a faster CPU will merely bandage the source of the problem. Latency to data will be dramatically increased and no CPU of any magnitude will help. The golden rule is simply to have enough video memory. If you have that, your CPU can go on vacation as far as the GPU is concerned.

This may all seem theoretical but it is generally true in practice in modern games. In fact in the vast majority of cases you can go from a middle of the road Core 2 Duo to a top of the line Core i7 Extreme and you won't witness any meaningful performance improvement, if any improvement at all, regardless of your graphics devices. Core i7s have a habit of occasionally performing worse than their predecessors at high gaming resolutions.

There is no realistic limit to how fast a GPU/GPUs you can put in there. Two 5970s would work without concerns. I would also strongly recommend against the Phenom. The only justifiable upgrade to your system as it is right now is to give it as many GPUs as you can afford. You would without question see far bigger gains in performance pushing the CPU budget into the GPU to almost no end.

I intend to keep my Core 2 Duo on 4GB RAM and PCIe 1.0 until at least the 6800/GT500 generation of GPUs, and probably some years after. CPU upgrades have shown to make remarkably little sense in recent years.
 
Last edited:
  • #3
So what you're saying is that a GPU/CPU bottleneck is exaggerated in most cases, and that any of the latest processors should be able to handle any GPU setup without any noticeable lag?
 
  • #4
You figured it out :smile:
 
  • #5
Negatron said:
You figured it out :smile:

So a Phenom II X4 965 shouldn't have any issues running a single Radeon HD 5870, and I should still notice a HUGE performance increase in all of my games compared to my HD 4850? With our without a bottleneck.

How much faster is a Radeon HD 5870 as opposed to an HD 4850?
 
  • #6
Your limiting factor at high resolutions will be video memory. If your video memory is insufficient, the CPU will need to do more work to perform constant swapping, however in this case a faster CPU will merely bandage the source of the problem.

I don't think it'll do even that. If you don't have enough video memory, the PCI-X bus becomes the bottleneck (Radeon 4850 internal video memory: 64 GB/s; Radeon 5870 internal video memory: 150 GB/s; PCI-X 2.0 x16: 8 GB/s.)

A properly built game should exhaust the arithmetic performance of the GPU regardless of bandwidth constraints, so even on an extremely slow CPU and bus, the GPU does not need to be constrained.

That's debatable. Depends on resolution and settings. Every modern video game you can think of will be constrained by GPU if you run it in 2560x1600 with full antialiasing. The lower you go, the more the CPU and the bus matter. Older video games are more easily constrained by the CPU than newer video games. Something like Doom 3 or Call of Duty 2 in 1280x1024 resolution will reach the same 200 fps on the 4850 and the 5870.

Radeon HD 5870 is about 2.5 times faster than HD 4850, both in terms of arithmetic performance and memory bandwidth. HD 4850 has 800 cores @ 625 MHz, HD 5870 has 1600 cores @ 850 MHz.
 
  • #7
IMO, you can have the top notch GPU in crossfire/SLI without any problems.
 
  • #8
dE_logics said:
IMO, you can have the top notch GPU in crossfire/SLI without any problems.

Can an HD 5870 run STALKER Clear Sky on it's maximum graphical settings (720p resolution) and still get playable framerates? Or do I have to go with an HD 5970 if I want to do that?
 
  • #9
That should be somewhat playable on the 4850 already. On the 5870 you should be getting 60-70 fps.
 
  • #10
hamster143 said:
That should be somewhat playable on the 4850 already. On the 5870 you should be getting 60-70 fps.

I can play STALKER CS on DX 10.1 on all "high" quality settings on 1024x768 and get >50fps without sun shadows activated. With sun shadows turned on, I get <15 fps.

It doesn't really matter, whatever game I play my PC powers itself off for no apparent reason after about 30-40 minutes. With the exception of a few, older, less demanding DX9 games like Battlefield 2, and Call of Duty 4.
 

1. What is a GPU?

A GPU, or graphics processing unit, is a specialized electronic circuit designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer intended for output to a display device.

2. What is a bottleneck?

A bottleneck occurs when one component in a computer system is unable to keep up with the speed of another component, causing a decrease in overall performance. In the context of a GPU and CPU, a bottleneck can occur when the GPU is too slow to keep up with the processing speed of the CPU, limiting the overall performance of the system.

3. How do you determine the fastest GPU possible without bottlenecking a CPU?

To determine the fastest GPU possible without bottlenecking a CPU, you need to consider the specifications of both the GPU and CPU. Look at the clock speeds, number of cores, and memory bandwidth of both components. Ideally, the GPU should have similar or higher specifications to the CPU in order to avoid a bottleneck.

4. Can a bottleneck be avoided entirely?

In most cases, it is difficult to completely avoid a bottleneck between a GPU and CPU. However, by choosing components with similar specifications and ensuring that both are properly cooled and not overclocked, you can minimize the potential for a bottleneck.

5. What happens if a bottleneck occurs between a GPU and CPU?

If a bottleneck occurs between a GPU and CPU, the overall performance of the system will be limited. This can result in slower loading times, lower frame rates, and overall reduced performance. Upgrading either the GPU or CPU to better match the specifications of the other component can help alleviate the bottleneck and improve performance.

Similar threads

  • Computing and Technology
Replies
2
Views
4K
  • Computing and Technology
Replies
1
Views
2K
  • Computing and Technology
Replies
2
Views
3K
  • Computing and Technology
Replies
11
Views
6K
  • Computing and Technology
Replies
5
Views
3K
Replies
4
Views
5K
  • Computing and Technology
2
Replies
43
Views
3K
Replies
5
Views
7K
  • Computing and Technology
Replies
7
Views
3K
  • Computing and Technology
Replies
11
Views
3K
Back
Top