I5-750 vs i3-530 Huge memory performance differences

[AstroGuardian]

Hello all,

I got this two rigs:
Gigabyte P55-UD3R
2 x 2Gb Kingston DDR3-1333
Sapphire Radeon 5570

One of them has i3-530 and the other has i5-750.
i5-750 - memory read: 14000 MB/S on Everest 5.50 and 58ns latency (write and copy are similar)
i3-530 - memory read: 8800 MB/S on Everest 5.50 and 88ns latency (write and copy are similar)

They both have BIOS version F6. CPU clocks are default (2.66 for 750 and 2.93 for 530)

i5-750 - CPU Queen = 22156
i3-530 - CPU Queen = 18767

Why so huge memory performance difference? I know i am missing something in the specs so would appreciate if someone tells me what
Aren't they supposed to have on-die memory controller which will lower latency?

Thx all.
Astro

 

[NumericalMethods]

From my limited knowledge of integrated memory controllers - The i3-530 has an on package memory controller, not on die. The memory controller is actually bundled with the integrated graphics chip on the 45nm die next to the CPU. The i5-750 (not like the other i5s) is the nehalem based architecture that we are so used to that does include the IMC on die.

The latencies associated with the separate die transactions of the i3 memory controller are far greater than those of an on die controller. There is little difference to this approach than previous generation s775 cpu - northbridge architecture, which is why you see similar bandwidth results as DDR3 775 systems.

This is something that Anand commented on when he introduced the Clarkdale processors. Good job the Intel cache architecture allows the reduced memory bandwith to not really effect the performance of the chip.

 

[AstroGuardian]

This is something that Anand commented on when he introduced the Clarkdale processors. Good job the Intel cache architecture allows the reduced memory bandwith to not really effect the performance of the chip.

Thanks a lot mate! So are you suggesting that i should not be concerned about the same memory latency and bandwidth i had with my previous Gigabyte G41 with E5300 and DDR2-800? I used to have nearly identical memory performance NUMBERS as with my new i3 rig.
So what you suggest is that despite having identical memory performance NUMBERS in Everest 5.50 i have better real life performance with the i3? Do i get this right?

 

[NumericalMethods]

No probs buddy, you shouldn't be concerned if you had similar bandwidth performance with your previous system at all.

The point I was making about the cache situation was that the modern trend of having large on die cache mitigates most memory bandwidth bottlenecking sufficiently. You are not crippling you CPU performance by having a lower memory bandwidth. Of course there are certain situations where memory bandwidth is very important but not for the everyday user. You wouldn't be using an i3 chip for those kinds of applications anyway.

You will see slightly better performance clock for clock than your E5300 due to improved architecture design. There might be some situations where the E5300 beats the 530 but I'm speculating. But most importantly you don't get all the cool features that the i3 has over the E5300 like HT, improved performance per watt, integrated graphics etc.

 

[AstroGuardian]

I just read that the i3 has two dies on the chip. One is the CPU and the other is the GMCH which means that the CPU and memory controller are on a separate die connected with QPI.
That explains a lot.
I didn't have the time to read about these new CPUs but it's a must. I shouldn't ask stupid questions before i read the "what's new" @ Anandtech. I learned that lesson alright...

 

[NumericalMethods]

meh, don't worry about it. There is so much happening out there in Intel space alone x58, p55, h55. i3, i5, i7. Then all the rules change about features on the chips when you move to laptops. It's enough to make a man turn AMD

 

[AstroGuardian]

It's enough to make a man turn AMD

HA! Best quote ever!!!

 

[IntelUser2000]

I don't think there's ANY chance that a E5300 can beat i3 530. The 3.2GHz i5 650 beats 3.33GHz E8600 in almost every app(http://anandtech.com/bench/Product/54?vs=144), and even the 3.06GHz i3 540 almost does that.

northbridge architecture, which is why you see similar bandwidth results as DDR3 775 systems.

Bandwidth is higher, but latency is sometimes little higher also. http://techreport.com/articles.x/18216/7

Everest results show 8GB/s in average for Clarkdale and 5GB/s for Core 2, while on STREAM its 14GB/s and 7GB/s respectively.

So the on-package memory controller did help, but only on bandwidth. I also have a feeling they might have also sacrificed a bit of latency for more bandwidth to help the GPU.