|
The importance of L2 cache and system bus speed
#295372 - 09/09/12 07:13 AM
|
|
|
Given the following two Core2Duo CPUs:
E7300 - 3MB L2 Cache - 1066 MHz System Bus
E5800 - 2MB L2 Cache - 800 MHz System Bus
Assuming that they both overclock equally to 3.6GHz, should I expect any appreciable performance gains in MAME on the second CPU due to the larger L2 cache and faster 1066 MHz system bus?
|
GroovyMAME support forum on BYOAC
|
|
| B2K24 |
|
MAME @ 15 kHz Sony Trinitron CRT user
|
|
|
|
|
Reged: 10/25/10
|
|
Posts: 2663
|
|
|
|
|
|
Send PM
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: krick]
#295373 - 09/09/12 08:36 AM
|
|
|
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: B2K24]
#295374 - 09/09/12 09:00 AM
|
|
|
I'm going to overclock them to the same final speed. I'll end up dropping the multiplier on both and bumping up the FSB.
So the difference comes down to the extra L2 cache and the difference in system bus speed interacting with the memory.
From what I've been reading, the faster system bus speed can increase memory bandwidth, but the system usually needs to introduce a higher memory latency to make everything play nice together, so the net result may not actually be better overall performance, depending on the actual application you're running.
I wish it was easier to swap CPUs for benchmarking.
|
GroovyMAME support forum on BYOAC
|
|
| B2K24 |
|
MAME @ 15 kHz Sony Trinitron CRT user
|
|
|
|
|
|
Reged: 10/25/10
|
|
Posts: 2663
|
|
|
|
|
|
Send PM
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: krick]
#295378 - 09/09/12 11:11 AM
|
|
|
The thing is your asking about which CPU will give you the best overall performance, but your leaving out important details like exactly which MOBO and RAM you will be using assuming these are already in your possession.
And why have you already decided on a final speed before getting your hands on them? C2D chips can run 24/7 at 80C and beyond, but personally, I don't like exceeding 70C at 100% load.
overclocking on LGA775 it's important to understand the different chipset straps and the effects of high FSB on your NB
http://www.xtremesystems.org/forums/showthread.php?128913-FSB-Strap-for-noobs.
http://www.thetechrepository.com/showthread.php?t=30
|
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: B2K24]
#295411 - 09/09/12 09:30 PM
|
|
|
I already own both CPUs and I've already overclocked them both in different systems to the same final speed, so I know that they will both do 3.4-3.6 GHz with no problem.
The E5800 is in my primary desktop PC (the one I'm typing on now), the other CPU is sitting in a box.
The PC in my MAME cabinet is using a E5200 (2MB cache, 800 MHz system bus) currently overclocked to 3.5GHz.
All three chips are Wolfdale core so architecturally, they're the same. The only variables are the cache and the system bus.
The CPU currently in my MAME PC is essentially a slower version of the E5800. The cache and system bus are the same speed and they're both wolfdale core.
What I was thinking about doing was dropping the E7300 into my MAME PC.
The motherboard is a ASUS P5KPL-CM, so it's a tricky board with limited overclocking options, but once you know how it works, it's not a problem.
I'm currently using G.SKILL 4GB (2 x 2GB) DDR2 1066 (PC2 8500) memory.
The current CPU in the MAME PC is overclocked using the following settings...
FSB: 280
FSB to Memory divider: 3:5
CPU multiplier: 12.5
Effective memory speed: 467 MHz (DDR2 933 MHz)
Effective CPU speed: 3.5GHz
When overclocking the E7300 I'd probably use these settings...
FSB: 333
FSB to Memory divider: 2:3
CPU multiplier: 10.5
Effective memory speed: 500 MHz (DDR2 1000 MHz)
Effective CPU speed: 3.5GHz
|
GroovyMAME support forum on BYOAC
|
|
| B2K24 |
|
MAME @ 15 kHz Sony Trinitron CRT user
|
|
|
|
|
|
Reged: 10/25/10
|
|
Posts: 2663
|
|
|
|
|
|
Send PM
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: krick]
#295425 - 09/09/12 10:52 PM
|
|
|
Since all the chips are very similar and will be running at the same clockspeed, I'm thinking a swap is not worth the trouble. If you saw any performance increase from swapping I'm sure it would be extremely minimal.
If you do decide to swap you should consider running some benchmarks with MAME and some other tools before and after swap, which would provide proof of answering what your curious about.
http://www.xtremesystems.org/forums/show...erPi-and-PiFast
http://www.tweakers.fr/timings.html
|
|
|
| R. Belmont |
|
Cuckoo for IGAvania
|
|
|
|
|
|
Reged: 09/21/03
|
|
Posts: 9716
|
|
Loc: ECV-197 The Orville
|
|
|
|
Send PM
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: B2K24]
#295441 - 09/10/12 02:32 AM
|
|
|
> If you do decide to swap you should consider running some benchmarks with MAME and
> some other tools before and after swap, which would provide proof of answering what
> your curious about.
Yeah, I'd love to see MAME benches between those two configs. That's an area where we assume some things but don't necessarily have actual numbers.
|
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: R. Belmont]
#295444 - 09/10/12 05:49 AM
|
|
|
It's been a while since I've run any MAME benchmarks.
I know there's now a -bench command line parameter, but does anyone have a batch file that runs through a list of specific games and saves the results to a file?
Speaking of games, what specific games should be benchmarked?
Should I be using the stock MAME 0.146 64-bit binary, or should I be compiling my own optimized for Core2Duo?
If I should be building my own, should I be using the new GCC 4.6.3 toolchain, or the older GCC 4.4.7 toolchain? (Assuming the 0.146 code will even build with the new toolchain)
As a side note, it appears that the stock MAME makefile doesn't specify any -march setting, from what I can find online, the default if not specified is -march=i386 or possibly -march=native, I'm not sure which. Is this intentional?
Edited by krick (09/10/12 09:09 AM)
|
GroovyMAME support forum on BYOAC
|
|
| Quantum Leaper |
|
OCRer and Monkey Typist for Galaxy Game
|
|
|
|
|
|
Reged: 03/08/04
|
|
Posts: 198
|
|
Loc: Orion spiral arm of the Milky Way
|
|
|
|
Send PM
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: R. Belmont]
#295630 - 09/12/12 06:48 PM
|
|
|
> > If you do decide to swap you should consider running some benchmarks with MAME and
> > some other tools before and after swap, which would provide proof of answering what
> > your curious about.
>
> Yeah, I'd love to see MAME benches between those two configs. That's an area where we
> assume some things but don't necessarily have actual numbers.
I have even seen a difference in speed of MAME, with the only thing different was the Motherboard. When I first got my P4 3GHz, I had a Gigabyte mother board (~$150), I had power problems, and it killed my motherboard. I then got a some cheap motherboard, cost me about $60. MAME and the rest of my games ran slower. I used to have some benchmarks, though not with MAME, showing the CPU 10% to 20% slower and the video just a little slower. The hardware was an Audigy 1 soundcard, 120G HD (Maxtor) which I still use and nVidia 5200.
|
|
|
|
Re: The importance of L2 cache and system bus speed
[Re: Quantum Leaper]
#297430 - 10/11/12 10:44 PM
|
|
|
Back when I was upgrading my cab, I had a really old PIII in it, so I bought a cheap processor in the C2D familly with 512KB L2 cache. Concerned about the cache level, I tested it against my desktop computer's C2D with 2MB cache on the same motherboard, memory etc., both overclocked to the same speed. The result was few fps (as in ~4) in favor of the C2D with 2MB but I was never convinced it was due to cache, and attributed it to the fact that to get to the same speed, I had to overclock the 512KB processor more, since it had a lower clock.
I sincerely doubt very much you will get anything noticeable, due to the difference in L2 cache.
I also played with memory size and clock and dual vs. single channel and it was all completely irrelevant to MAME. My conclusion was clock speed and CPU architecture are the only things that really matter for MAME performance.
Edited by "The Manuel" (10/11/12 10:45 PM)
|
|
|
Previous 
Index 
Next 
Threaded 
The importance of L2 cache and system bus speed
Reply
[Re: 