L2 Cache in gaming - 6mb vs 12 mb

[arh2o]

http://www.insidehw.com/Review...de-By-Side/Page-2.html

They basically took a QX9650 and brought it down to Q9300 speeds. So both processors were running at 2.5 ghz. After looking at that chart, all I can say is WTF? Is there an error or something?

The average FPS they got in FEAR with a Q9300 is 299; for a QX9650 400 FPS. Similarly, in World in Conflict they got 166 FPS with a 9300 and 199 fps with the 9650. Furthermore the max fps they got in in FEAR is almost 200 more, and 100 more for WIC.

People have said that cache is not that important in real-life applications and it won't be noticeable to the average user. However if in gaming, I can get 200 MORE FPS just with double the cache, than that is pretty damn important.

Is that test flawed? I've been considering either getting the Q9400 or the Q9450. They're identical except for the cache size, and i'm thinking the extra ~$70 in getting the 9450 might be worth it for 100~200 more fps

 

[BTRY B 529th FA BN]

Are you sure those comparisons are between the same GHz e.g. 3-3 or did they clock down the QX9650 to 2.5Ghz?

 

[Tempered81]

nah, he's right. Cache makes a huge difference. a 4ghz e8400 is about twice as fast as a 4ghz e5200 at the same low-res CPU bench of a video game because of triple L2$. Similarly the test between a 2.5ghz q9300 and a 2.5ghz qx9650 will put the qx far ahead solely because of L2$.

 

[amenx]

Depends on the application or game. It can make a big difference in some cases while in others it may not:

http://www.tomshardware.com/re...ize-matter,1709-5.html

 

[noriseghir]

For C2Q the cache size is even more important due to the MCM nature of these CPUs.

 

[v8envy]

I can definitely see that. A game like WiC is one of the handful of titles trying to take advantage of as many cores as possible. It's not like each cpu has 1.5 megs of dedicated cache, there could be premature cache evictions caused by one CPU of data needed immediately by another CPU, while a larger cache thrashes far less.

You're looking at a degenerate case there, I don't believe many current titles will exhibit this behavior.

 

[Idontcare]

Something is not right with those benches. Look at the Everst numbers for the Q9300 vs. 2.5GHz QX9650 vs. stock QX9650.

The QX9650 at 7.5x333 has a latency of 84ns while the QX9300 at tthe same 7.5x333 has a mere 73ns latency. This is reflected in the poor read and copy bandwidths as well.

 

[soccerballtux]

That caught my eye as well.

 

[imported_Scoop]

Those framerates are rather interesting. Something tells me they didn't run at 1900x1200.

 

[Idontcare]

I know it an older article and relates to dual-cores with smaller caches but this toms link is pretty thorough in my opinion:

http://www.tomshardware.com/re...-size-matter,1709.html

 

[VirtualLarry]

Originally posted by: jaredpace
nah, he's right. Cache makes a huge difference. a 4ghz e8400 is about twice as fast as a 4ghz e5200 at the same low-res CPU bench of a video game because of triple L2$. Similarly the test between a 2.5ghz q9300 and a 2.5ghz qx9650 will put the qx far ahead solely because of L2$.

That's interesting, now that you put it that way, it actually makes the E8400 seem like the better value, because the E5200 is not half the price of the E8400.

 

[apoppin]

Well Cache made a big difference in most games when i upgraded from e4300@3.33Ghz to e8600@3.33Ghz

- of course i didn't stop there

 

[dguy6789]

Not directly the same, but I went from a 2.8Ghz 2mb cache dual core to a 3.4Ghz 12mb cache quad and my UT3 fps (min, avg, and max) went up by an extremely noticeable degree. Running at 1680x1050.

 

[harpoon84]

Originally posted by: jaredpace
nah, he's right. Cache makes a huge difference. a 4ghz e8400 is about twice as fast as a 4ghz e5200 at the same low-res CPU bench of a video game because of triple L2$. Similarly the test between a 2.5ghz q9300 and a 2.5ghz qx9650 will put the qx far ahead solely because of L2$.

Please don't exaggerate, an E8400 @ 4GHz is not nearly twice as fast as an E5200 @ 4GHz in gaming, even in an extremely cache bound game like GRID benched at a CPU bound res (1024 x 768):
http://www.pcgameshardware.com...rs_overclocked/?page=3
E8200 (6MB L2) @ 4GHz = 103fps
E5200 (2MB L2) @ 4GHz = 79fps

Even comparing an E8x00 to E2xx0, the difference is nowhere near 100% improvement:
E8500 (6MB L2) @ 3.17GHz = 80fps
E2180 (1MB L2) @ 3.2GHz = 65fps

Here's a summary of the differences between 3MB L2 vs 6MB L2 in C2Ds : http://www.xbitlabs.com/articl...duo-e7200_4.html#sect0

The difference is games (at 1024 x 768) is about 10% per clock between an E7x00 and E8x00 at the same clockspeed. This diminishes once we take GPU bottlenecking into the equation at higher resolutions, something inevitable for most people without SLI/CF setups.

Originally posted by: VirtualLarry
That's interesting, now that you put it that way, it actually makes the E8400 seem like the better value, because the E5200 is not half the price of the E8400.

See above.

 

[Tempered81]

ahh, okay 31% faster.


sorry.

 

[Idontcare]

Please AMD and Intel, if you are reading this, please send us Deneb and Nehalem sooner instead of later. We enthousiasts desparately need something to beat each other up over. Its gotten desperate here in the CPU forums.

I fear the worst is coming for us this long 31-day October month, next we will be debating whether 7200rpm harddrives actually spin at exactly 7200 rpm or maybe some spin at 7199 while others at 7201 rpms. Please help us.

 

[Tempered81]

Originally posted by: Idontcare
Please AMD and Intel, if you are reading this, please send us Deneb and Nehalem sooner instead of later. We enthousiasts desparately need something to beat each other up over. Its gotten desperate here in the CPU forums.

I fear the worst is coming for us this long 31-day October month, next we will be debating whether 7200rpm harddrives actually spin at exactly 7200 rpm or maybe some spin at 7199 while others at 7201 rpms. Please help us.

haha well just saw a 4ghz nehalem do superPI at 12 seconds per loop on the 16MB test.

I tried that test with my 65nm Merom at 9 x 200 (1800mhz) and was getting 31 seconds per loop on the 16MB test. If i clocked to 4ghz, that would be about 14-15 seconds per loop. 10% IPC improvement to 45nm part would put me around 13 seconds. I can't wait to see some CPU charts with these things!

Anyone out there with a 4Ghz cpu want to try it out and compare to a 4Ghz i7 965 EE?

Download superPI here (make sure to do the 16mb test):
http://www.xtremesystems.com/pi/

Screenshot of Nehalem Bench (4 Minutes 46 Seconds):
http://img216.imageshack.us/my.php?image=qpi210nf9.png

Edit: hopefully this brings back some excitement - I'm going to make a thread for this!