After I picked up some new PC parts late this week, I decided to compare how much a GPU like HD7950 will be bottlenecked by a weaker CPU as well as PCIe X4 bandwidth VS full PCIe X16.

I decided to use 3D Mark Vantage as it strikes a good balance between CPU and GPU and clearly shows how much a slower CPU bottlenecks GPU performance.

Rig 1
CPU - FX 8350 Stock
RAM - Kingston HyperX 2133
GPU - Radeon HD7950 Stock

Rig 2
CPU - 3870K APU - OC to 3.6 Ghz
RAM - Kingston HyperX 2133
GPU - Radeon HD7950 Stock

I will let pictures do the talking.

Benhcmark 1 - FX 8350 + 7950 PCIe X16

Benchmark 2 - 3870k + 7950 PCIe x16

Benchmark 3 - 3870k + 7950 PCIe X4

So the shocker here is a that there is absolutely no noticable performance loss between PCIe x16 and PCIe x4.

Also FX 8350 which has 2X more cores and is probably twice as powerful as 3870K apu when those cores are utilized only improves the GPU performance by roughly 10%.

My conclusion looking at these numbers is that x8-x8 crossfire/SLI setup is perfectly fine for current gen video cards in terms of bandwidth. Even x16-x4 crossfire setup should be fine actually.

Also a much slower CPU does not cause GPU performance loss nearly as much as it does on the CPU side. If the game or application you want to use isn't CPU intensive, you can get away using a much cheaper CPU and invest in a better GPU.

==================================================================================================

Update

Benchmark 4 - FX8350 with 4 cores disabled + 7950

So it appears that more than 4 cores does not really really get utilized on 3D Mark Vantage GPU tests, which I find is the closest benchmark to actual gaming.

The GPU score is slightly lower probably because of the select few moments when the CPU does limit the frame rate but most of the time, the GPU would still be the bottleneck.

It will however be interesting to see how the numbers look with HD7950 in crossfire with 4 and 8 cores. Will attempt that shortly.

Another thing that stood out were the temperatures of the FX 8350 while running on 4 cores. 28 degrees only while running these benchmarks is remarkable. I can understand why disabling 4 cores can push this chip passed 5GHz.

Yeah, that's why I keep telling people that spending a ton on a CPU is not worth it unless you are using applications that utilize it, which are not games in general but more on the side of media related work or running mass amounts of VM. GPU is much much more important for gaming, console makers know this too, and that's how the Wii U and PS4 resulted in their current state with weaker CPUs.

With APU set ups the GPU is going to become even more important since it will open the GPU up for processes original done for CPU work. It might be young for APUs still and they look weak by current computer understanding of how separate CPUs and GPUs work, but the design is going to be quite the advancement in processing. I don't see why not our high end CPUs will become incorporated into APU designs in the future.

You should have compared CPUs of the same architecture and if possible core count.

The A8-3870K is Llano based which has higher per-core performance than the Piledriver-based FX-8350. The FX has four more cores, but they really aren't used in a 3Dmark type benchmark because the CPU threads for game graphics handling are one or two threads at most.

Your test doesn't show what you want it to show because the results are not comparable.

"An i5" is stronger than either of these CPUs for gaming. Both CPUs he tested could be considered weak.

The 3DMark Vantage uses more than two for sure.

But your point is valid and true... so different to make any conclusion... in fact the high score for CPU is because the 8 cores and not the performance per core itself.

@disolitude

Can you disable 4-core of the FX 8350 to make a better comparison? Some time the BIOS have that type of option.

Thanks for the graph ethomaz. I would be interested to see if the scaling continues past four. Also, 3Dmark is an artificial test and I wouldn't be surprised if games scale worse than it in this regard (since his point is to apply to real games).

Yes, but the thread's hypothesis is about gaming right?

Some tests I found in web.

8-core does make difference in score.

The FX 8350 Stock and 3870K APU - OC to 3.6 Ghz are more on pair in game performance per core in @disolitude tests.

I have to add that too with the PS4/Nextbox comming the games will start to be more optimized to 8-cores than they are today... devs will make the games run better in 8-cores CPUs (except the PC exclusives).

So PC will have a grest advantage in games in two or three years in terms of CPU with 8-cores.