2009 i5 750 OC 3.6Ghz out performing AMD FX 8300 Black OC to 4.2Ghz

DblGonzo

Prominent
Mar 27, 2017
2
0
510
I thought I had a bad MB or CPU. My i5 rig bought in 2009 was shutting down sometimes before POST and other times after it had been on for a while. Swapped out everything PS, Mem, Video Card.
I was not ready to spend a bunch of money so I went with an AMD 8300 Black and MSI 970A MB. Added an SSD and reused my Graphics cards (ASUS GTX 960 reference) in SLI. After I got everything installed and booted up. I started getting the same shut downs. Random times no blue screen.
Turned out my power switch on my Antec case was bad. So I swapped the reset button with the power button on the MB. Works great, let it run for 24hrs.

So I hooked up all my old hardware put a Cooler Master T4 on the i5 and off I went into the OC world. I got it to boot and run at 4Ghz but was not happy with the temps running almost 80c under load. So I set it to 3.6Ghz and temps are great.

Now the crappy part. The 3D mark scores are better than my AMD rig.

3d Mark firestrike with AMD 8856 using 8 gigs of ram and GTX 960 2Gdr5 in SLI. With my intel i5 the score was 9987 using 4gigs of ram and the GTX 960's in SLI. Both running the latest drivers from Nvidia. The time spy bench mark was also 1000 higher with the intel i5.

Did I just waste money? Or is there something going on here that I am missing. I admit I am a noob at overclocking. And may not have my MSI 970A set up correctly for a good overclock.

Or is the intel i5 750 just a better chip. I know the i5 is only a couple years older than the FX in technology. But I did not expect it to out perform the 8300 with half the cores.
 
Solution
Even though the FX 8300 has a higher clock rate after the overclock that your i5 750, the i5 750 performing faster is not unexpected. Clock rates are not all that matters, clock rates only show the number of cycles a processor can do in every second.
This then requires a look into the microarchitectures of the CPUs. It is known that Piledriver FX CPUs require more clock cycles to perform the some instructions then the Nehalem Core CPUs. I will explain vaguely below.
Firstly the FX CPUs use a Clustered Multithreading design (CMT). This means that for every two cores in each cluster there is only one FPU. This results in conflict between cores and delays when two cores share the FPU in floating point heavy calculations.
Secondly, each...

Faike

Notable
Mar 27, 2017
256
0
860
I tested a 750 ti, 760, 750 ti super clock single card set ups with 8350 and realized my card performed a lot better when I ran it with another card. Not to sli, but for Physx, in this case I used a 660 TI. But by itself it did not perform as well as it could've. Now when I stream compared to my old i7 I do get better performance, but I don't know how you're getting such high temps. I was easily able to OC my CPU to 4.4 ghz without going above 40 C using just the stock cooler.

Intel tends to perform better with Nvidia in general, but when I switched to AMD my streaming performance was outstanding. So like the guy above me said, unless you're using all your cores (in my case, gaming and streaming), intel probably would run better for the 750.

I did build a PC with a 1050 ti and a 8350 fx and she can run the games on ultra with 90 FPS (on overwatch anyway), no overclocking necessary, plus a smooth stream.

I did check and saw that your 960 shouldn't bottle neck. Have you tried not SLIing the cards? Just using one card might fix the issue. What is your PSU as well?
 

Faike

Notable
Mar 27, 2017
256
0
860


Yep, if you're streaming or something while playing I found that my 8350 processor outperformed my friend's i5's pretty drastically. Mostly because the game used all 4 cores and the other 4 cores available allowed me to stream with high settings without any issue in graphics.

 

kgt1182

Reputable
Jun 8, 2016
420
0
5,160
Even though the FX 8300 has a higher clock rate after the overclock that your i5 750, the i5 750 performing faster is not unexpected. Clock rates are not all that matters, clock rates only show the number of cycles a processor can do in every second.
This then requires a look into the microarchitectures of the CPUs. It is known that Piledriver FX CPUs require more clock cycles to perform the some instructions then the Nehalem Core CPUs. I will explain vaguely below.
Firstly the FX CPUs use a Clustered Multithreading design (CMT). This means that for every two cores in each cluster there is only one FPU. This results in conflict between cores and delays when two cores share the FPU in floating point heavy calculations.
Secondly, each cluster only has one decoder. This means that when two cores of the same cluster are used, the decoder works on an effective clock rate of half the core clock relative per core. This bottleneck causes some decode-intensive tasks to slow down massively.
Lastly, the FX CPU cores have high rates of branch misprediction due to the cut-down branch predictor for executing chains of interconnected boolean operators. This was done to raise core clocks which unfortunately results in high branch misprediction penalty when prediction fails proportional to pipeline length, even though FX CPUs have equal pipeline lengths to Intel CPUs, since this happens commonly, many cycle are wasted. Also, cache misses are common due to weak memory controller and L3 cache, leading to slow L3 cache access, and misses lead to even longer RAM latencies. This compounded with the weak IMC, where DDR3 1600+ is rarely supported, again incurring large clock penalties. All because L3 cache is connected to core clock.

In a nutshell with a running analogy, Nehalem CPUs take fewer steps each second, but slightly longer steps, while Piledriver CPUs take many steps a second, but shorter steps and fall down more frequently.
 
Solution