C2D L2 cache performance?

[rogue1979]

I think I must be missing something?

I have an E2180, E6400, and E6600 with 1MB, 2MB and 4MB of L2 cache respectively.

I clocked them all to 3GHz and used the memory divider to keep memory speeds between 750-800MHz, since they all have different multipliers.

Nothing scientific, just some 3DMark06 and running fraps in the following games:

Crysis, COD4, GRAW2, GRID, COH Opposing Fronts.

Using various midrange video cards, all overclocked:

9600 GSO, 8800GT 512MB, 8800GTS 320MB, 9600GT, HD 3850 512MB.

All of the above mentioned cards score 11 - 12K in 3DMark06. Almost no difference with 1MB, 2MB or 4MB of L2 cache on each of the video cards. Casually looking at fraps I can't detect any significant difference in gaming either.

According to some of the benchmark reviews and cpu charts, should't I notice some performance difference? Or am I just gpu limited? Most of my gaming is done at 1152 x 864 16AF 4XAA except Crysis, which seems to run decent at 1280 x 1024 8AF 2XAA.

 

[Tempered81]

So you had an E2180 + 9600GSO score the same as E6600 + 8800GT ?

 

[rogue1979]

Originally posted by: jaredpace
So you had an E2180 + 9600GSO score the same as E6600 + 8800GT ?

11.1K for the 9600GSO on the 2180, 12.2K for the 8800GT on the E6600.

The 8800GT scores 400 pts less on the 2180, not much difference.

In Crysis, the 8800GT is about 10-15 fps ahead the 9600GSO, averaging low 40's
at 1280 x 1024 with 2X AA. The 9600GSO has to play at 0X AA to get the same framerates.

The E6600 or E2180 runs either card about the same in any games.

 

[Denithor]

3Dmark06 exaggerates the processor impact on games. Repeat the tests with 05 instead, that one focuses more on GPU performance, the cpu really only factors in majorly if it limits the GPU.

And that still sounds about right. 1k difference is about a 9% difference going from 1MB to 4MB cache and I've heard that you typically gain 3-5% per double cache increase (1MB -> 2MB -> 4MB each step gets 3-5% faster).

EDIT: So you will probably see bigger differences in 3d05 but it will be due to GPU strength not CPU cache levels.

 

[AstroGuardian]

Denithor is absolutely right.

 

[rogue1979]

Another insight:

Put more volts into the E6600 (1.46v) and got it to 3350MHz on a Biostar G31 M7 TE.

Prior to this I had a personal safe limit of 1.4v max on any of my C2D's.

With a 350MHz increase in cpu speed 3Dmark06 only went up to 12,500 on the 8800GT.

Only 300 points?

This leads me to believe any C2D at 3GHz or higher is more than enough cpu power to push a midrange video card to the max, regardless of the L2 cache size.

I bet if I repeated the tests with a HD 4870 or GTX 260 the extra cache would make a significant difference because those much faster gpu's would both be cpu limited.

I'll load in 3DMark05 and see what happens.

And that still sounds about right. 1k difference is about a 9% difference going from 1MB to 4MB cache and I've heard that you typically gain 3-5% per double cache increase (1MB -> 2MB -> 4MB each step gets 3-5% faster).

That 1.1K difference was not just the L2 cache, that increase in score included switching video cards from a 9600GSO to an 8800GT.

 

[Tempered81]

Originally posted by: rogue1979
This leads me to believe any C2D at 3GHz or higher is more than enough cpu power to push a midrange video card to the max, regardless of the L2 cache size.

I bet if I repeated the tests with a HD 4870 or GTX 260 the extra cache would make a significant difference because those much faster gpu's would both be cpu limited.

Sounds about right. When you bog down the gpu with a game like crysis, or some other game at high-res with AA and good IQ, cpu impact is marginal.

Edit:
- However, consider PCGH's benchmarks. This review was done at 1920 x 1200 MAX IQ Settings 4xAA 16xAF, using a single GTX280 video card in COD4. (All cpus are at 2000mhz):
http://www.pcgameshardware.com..._CPUs_reviewed/?page=4

- And here is that review again with the E8xxx processor scaling from 2000mhz --> 4000mhz (this time the resolution is 1650 x 1080):
http://www.pcgameshardware.com...647744&image_id=839047

Are these the benchmarks you are referencing in the first post?