Is L2 Cache or Clock Frequency better?

R

Ruskin

n00b
Joined
Apr 3, 2005
Messages
43
Is it better to go with the AMD X2 4400+ dual core processor that has a 2.2GHz clock with a 1Mb L2 cache (per core) or the X2 4600+ dual core that has a higher clock frequency of 2.4GHz, but a lower L2 cache (of 512Kb per core)?

Not sure if the extra cache provides more performance, or if a lower cache, with higher clock speed would be an advantage?

If I do comparisons, based on prices then;
4400+ is NZ$1100
4600+ is NZ$1500
4800+ is NZ$1880

Thus, the difference between 2.2GHz 1Mb L2 and 2.4GHz 512Kb L2 is NZ$400. Whereas the difference between 2.4GHz 512Kb L2 and 2.4GHz 1Mb L2 is NZ$380. The prices would imply, that the step of .2GHz clock is similar to a difference of 512Kb/1Mb L2 cache.
 
Cache helps when you're gaming at lower resolutions. But clockspeed is supremely better by a longshot.
 
You can make up for clock speed with overclocking. Theres nothing you can do to replace L2 cache. So I vote for the chip with more L2 and simply overclock it to 2.4GHz or however high your willing to go.
 
Sir-Fragalot said:
You can make up for clock speed with overclocking. Theres nothing you can do to replace L2 cache. So I vote for the chip with more L2 and simply overclock it to 2.4GHz or however high your willing to go.
Click to expand...

Yeah, but the 4400+ might not OC as high as the 4200+, and vice versa. It's a crapshoot either way when you bring OC'ing into the equation.
 
Thanks for the answers, but I'm not going to be overclocking.

Just wanted to see if L2 cache or clock frequency was more important.

cheers...
 
Cache is be more important. If you have a high clock speed but a less L2 cache, then each cache miss will take more clock cycles to load the value from main memory, which is valuable time the processor could be using to graduate (commit) other instructions (assuming the pipeline needs that data soon). L2 cache misses are painfull for computers. Some types of programs can cover up L2 cache misses if the program has really parallel instructions, but for really squential programs you want L2 cache (espically if they are dealing with lots of memory).
Also, someone said that the 4600+ (which has 2x 512k) beats the 4400+ (which has 2x 1mb). I have not seen the any of thoes benchmarks, but I would be intersted in seeing what type of applications they are running on them (perhaps you could post a link or something). Cache and multiprocessors becomes a complicated issue, with things such as cache coherency, false sharing, and a bunch of other junk. If it were me buying the processor, I would choose the cache over the clockspeed. and If it were a single processor, you sould take cache anyday, but for the multiprocessing it really depends on the type of stuff you want to do.
 
zappa86 said:
Cache is be more important. If you have a high clock speed but a less L2 cache, then each cache miss will take more clock cycles to load the value from main memory, which is valuable time the processor could be using to graduate (commit) other instructions (assuming the pipeline needs that data soon). L2 cache misses are painfull for computers. Some types of programs can cover up L2 cache misses if the program has really parallel instructions, but for really squential programs you want L2 cache (espically if they are dealing with lots of memory).
Also, someone said that the 4600+ (which has 2x 512k) beats the 4400+ (which has 2x 1mb). I have not seen the any of thoes benchmarks, but I would be intersted in seeing what type of applications they are running on them (perhaps you could post a link or something). Cache and multiprocessors becomes a complicated issue, with things such as cache coherency, false sharing, and a bunch of other junk. If it were me buying the processor, I would choose the cache over the clockspeed. and If it were a single processor, you sould take cache anyday, but for the multiprocessing it really depends on the type of stuff you want to do.
Click to expand...
Every benchmark says otherwise.
 
zappa86 said:
Cache is be more important. If you have a high clock speed but a less L2 cache, then each cache miss will take more clock cycles to load the value from main memory, which is valuable time the processor could be using to graduate (commit) other instructions (assuming the pipeline needs that data soon). L2 cache misses are painfull for computers.
Click to expand...
Yes, cache is important in this aspect, but there is such a thing as "having enough cache." You can never have too much clock frequency, however, heh.

Also, someone said that the 4600+ (which has 2x 512k) beats the 4400+ (which has 2x 1mb). I have not seen the any of thoes benchmarks, but I would be intersted in seeing what type of applications they are running on them (perhaps you could post a link or something).
Click to expand...

I and others have posted this all over this forum regarding this issue, but here it is AGAIN.
http://www.gamepc.com/labs/print_content.asp?id=a64x2

10 FPS more on HL2 at a low resolution with having more cache per core. This is probably do the the extra space on the core for calculating physics. What would be nice is if they benched it at a resolution that someone using an X2 would play at, i.e. one that a screen larger than a GameBoy's would display natively.
It's also worth noting that nothing other than HL2 gets a noticable benefit.
My opinion is that the cache is something you'll never see unless you benchmark and only benchmark, whereas the $100+ you save will definately be noticed.
 
Supasso said:
Every benchmark says otherwise.
Click to expand...
I just went to the link below your post and there are some benchmarks where the 4400 beats the 4600, such as the airboat chase. Now the conclusion says the 4800+ did not benifit from more cache. I bet that AMD would not have designed the chip with more cache if they did not think it could be used. (Happy Moon Landing Day, JULY 20th)
 
robberbaron said:
Cache helps when you're gaming at lower resolutions. But clockspeed is supremely better by a longshot.
Click to expand...

Cache helps when your gaming at any resolution, it just depends on how CPU limited the game is as to how much it will help at higher resolutions. At higher resolutions clock speed will be less of a factor as well in a GPU limited game.

robberbaron said:
The 4600+ will beat the 4400+ in every application. EVERY application.
Click to expand...

I think its pretty much impossible for you to say EVERY application considering only a handful of applications are even benchmarked in reviews. There are thousands of applications out there and four or five of the most popular being benchmarked doesn't equate to overall performance in ALL of them. Some applications will benefit more from extra L2 cache. Up until more recently applications havn't been coded to really take advantage of extra L2 cache because the majority of CPU's only had 512kb but with Intel and AMD both adding more L2 cache i think applications will start taking more advantage of it in the future because of the fact L2 is much faster then RAM.

http://www.xtremesystems.org/forums/attachment.php?attachmentid=32498

The X2 4200+ and 4400+ are going to overclock to about the same speed. You might get ither one to scale to 2.7-2.8 if your lucky but likely they'll top out at 2.5-2.6. If you can find a 4200+ for cheap then it would probably be the best buy but if your paying nearly the same price as the 4400+ you might as well pay a little extra for the extra L2 cache. I personally would never buy an OEM version of one of these X2's because they are just too expensive and the money saved isn't worth what you'll lose if something happens to it and you have no warranty.
 
The A64 has a relatively large L1 cache. It also has much shorter pipes than the P4, which means clearing the pipes from a cache miss isn't nearly as hard on performance. A very efficient, low latency memory controller also helps in recovering from cache misses that require data to be retrieved from main memory. As such the A64 does not benefit from larger caches the way the P4 does. While the larger L2 cache may allow the slower clocked chip to win in some cases, benchmarks are going to show that the chip with a smaller L2 cache and higher clock speeds will win far, far more than it will lose.
 
Wow, if this doesn't show the absolute stupidity of the new rating conventions, I don't know what does. Why can't AMD/Intel just stick with a static cache size for a while on a processor, and just change it when they change cores?

Seems to me that cutting cache is just a way to sell slightly defective CPUS, i.e. if a full 1mb of L2 cache doesn't work, shut off half and ship it as a different part number.
 
skyeandangus said:
Wow, if this doesn't show the absolute stupidity of the new rating conventions, I don't know what does. Why can't AMD/Intel just stick with a static cache size for a while on a processor, and just change it when they change cores?

Seems to me that cutting cache is just a way to sell slightly defective CPUS, i.e. if a full 1mb of L2 cache doesn't work, shut off half and ship it as a different part number.
Click to expand...
of course it is, it improves the efficiency of the production by reducing the number of parts that need to be tossed. If you look at most recent processor dies, you will see that the cache is a very large area compared to the actual execution core. While I believe that producing a large, defect free cache is easier than producing a large, defect free execution core, probability requires that defects in the cache are more likely than in the exec. core. For that reason, slightly faulty chips can be sold with parts of the cache disabled.

I don't see anything wrong with that, since it allows me to get a cheaper chip in the end, than if AMD only sold 1MB L2 cache chips.
 
skyeandangus said:
Seems to me that cutting cache is just a way to sell slightly defective CPUS, i.e. if a full 1mb of L2 cache doesn't work, shut off half and ship it as a different part number.
Click to expand...

I totally agree with you. I think they should just throw away the dies, since putting them to good use is stupid. What the hell are those jerks thinking when they sell value chips?
 
If the X2's act anything like San Diego vs. Venice, then the extra cache will only increase speed by 1-5% while running clock for clock. Not worth the extra money in my opinion.
 
The 4600+ will beat the 4400+ in every application. EVERY application
Click to expand...

Unless you're cache-limited. Most hardware sites don't bother to test real applications (compilers, databases, etc), so you never see benchmarks that show the advantages of having a large cache.
 
pduan87 said:
If the X2's act anything like San Diego vs. Venice, then the extra cache will only increase speed by 1-5% while running clock for clock. Not worth the extra money in my opinion.
Click to expand...

See and that's what counts. It does help, but I think its worth having more cache because you can't add that. I can overclock though. Of course there are no guarantees with OC'ing, but such is the way of things.
 
rhashem said:
Unless you're cache-limited. Most hardware sites don't bother to test real applications (compilers, databases, etc), so you never see benchmarks that show the advantages of having a large cache.
Click to expand...

If you're going to be doing such things (compiling and other professional applications) then you probably won't need to post a thread about it asking, "WATS DA CACHE DO?"
 
Having more cache in your processor is like having more torque....cache = torque, clockspeed = horsepower....you're better off with both, but real men love the torque. *hugs his claw* :D

But seriously, if you're going to stay stock, I always recommend higher clockspeeds (a 3800+ over a 3700+, for example) but if you plan on overclock I say get more cache....as you can't add more cache but you can add a few cycles (no guarentees tho).
 
You must log in or register to reply here.
Back
Top