GPU vs CPU Bottlenecks

[ccubed]

What are the bottlenecks for systems with differing CPUs vs GPUs?

Is it just clock speed? So a Q6600 at 3.2 ghz can handle the same video card as an I5 or I7 at the same ghz?

So has anyone tested the thresholds of performances for the various CPUs and GPUs?

I'm outdated and have an older CPU. I'm wondering if a GPU uprade will be worth it or whether my CPU will hold it back. And I'm looking for more general advice than specifics. Although I'll list mine later.

 

[BrightCandle]

It all depends on the game/program. Some games have frame rates that depend enormously on the CPU (Grand Theft Auto 4) and the GPU is much less important. However there are games where the converse is true, where they could run on a P4 class CPU but with modern graphics run similar to those with i7's.

The thing is about clock speed is its only one important aspect of performance. These days the architecture of the chip is making a big difference to the amount of instructions actually executed at any given clock speed. At the same clock speed the modern i7's (Sandy Bridge) are considerably faster than a core 2 quad and they achieve higher overclocks as well.

It just depends. In general if you want the widest array of games to work then you want balanced equipment in terms of age. Gamers can get away with moderate CPU power and higher end graphics in a lot of cases but once you start doing high end SLI and Crossfire you start to push even the highest end CPU into decent overclocks to get back the most benefit.

List your specs and then someone can comment about whether a GPU upgrade would be of benefit or not.

 

[Denithor]

Without knowing specifics we can only talk about general trends.

-Higher clock speed within a cpu family is better but you cannot really compare between families based solely on speed (C2D @ 2GHz would kick the crap out of a dual Pentium 4 @ 3GHz)
-CPU has more impact on fps at lower resolutions, GPU more impact at higher resolutions and/or higher levels of detail (more eye candy enabled)

For your specific case, even with an older CPU you may see improvements from upgrading the GPU if you're running high resolution or you may be able to turn on more eye candy (AA, shadows, etc) without hurting your fps.

 

[Seero]

Although Q6600 has 4 cores, it can't be utilized efficiently like I7 because it is based on an older architecture, Front Sided Bus(FSB). In short, the newer design QuickPath Interconnect(QPI) allows better communication between devices, as well as the integrated memory controller, which allow faster memory access. This 2 things allow better utilization on CPU and GPUs. Unfortunately, that means you also need Mobo, CPU and RAM upgrades.

Higher clock doesn't help much as the problem is CPU spend most of its time waiting instead of working. You can have your Q6600 running at 4ghz and games will still be shuttering compare to I7 920 stock.

The difference between I7 and SB (2nd gen) is simply the dynamic OC. The down side of SB is the missing 16xPCIe which are connected to its iGPU. The up side is USB3 and PCIe 3.0(some older SB mobo does only support PCIe 2.1).

As to mobo, there are H67/Z68 for average user. Both of them has 16 PCIe lanes. Z68 featured iGPU OC and SSD caching, which IMO useless except some Z68 mobo supports PCIe 3.0. If you are using 1-2 video cards, it is good enough. If however you are going for 3-4 video cards, there are not enough lanes (bandwidth) to unleash the full potential of those GPUs, you will need X79 which features 40 PCIe 3.0 lanes. In quadfire or 4way SLI configuration, each card gets 8 lanes.

Aside from that, the difference between PCIe 2.0 and 3.0 is that 3.0 doubles the bandwidth of 2.0, from 500MB/sec to 1Gb/sec per lane. However, only PCIe 3.0 video card will take advantage of those extra bandwidth, and current generation video cards are barely bottlenecked by the bandwidth of PCIe 2.0. That is, 16x/16x is just a bit faster than 8x/8x. By a bit I mean under 5%, games are programmed assuming user don't have monster setups, the communication between CPU and GPU is minimized, and therefore the bandwidth is not a problem. However, in CF or SLI configuration, loads are not splitted, the GPUs render frames alternatively, so the more video cards, the more load on PCIe. 8xPCIe lanes is still okay, but that is for current gen video cards. Once 3.0 PCIe really hits the market, games will increase the utilization on the PCIe, and PCIe 2.0 will become a bottleneck. Don't worry, I am talking in several years.