Budget CPU Shootout: Clash of the 'rons
by Derek Wilson on December 4, 2003 10:55 AM EST- Posted in
- CPUs
Final Words
It is very obvious from these tests which line of budget processors is worth the money. When we can find a 1.6GHz Duron for just over half the price of a 2.6GHz Celeron and get better performance consistently in almost every test we ran, the choice is clear.It's obvious that the long pipeline of the Pentium 4 just can't handle the crippled cache of the Celeron. With more cache misses and pipeline stalls, the processor isn't getting as much useful work done as it is trying constantly to refill the pipeline. We are seeing these results for the same reason we saw the performance gains from the P4 Extreme Edition with its 2MB L3 cache: the pipeline needs to stay full for the P4 to really shine.
The Pentium III based Celerons offered, at one time, acceptable performance. However, it is clear that in the value segment today, Intel has nothing to offer but a high clock speed. AnandTech readers will know to stay away from the Celeron at all costs; however, what is troublesome are the number of retail customers who are faced with the decision between a higher priced 2.6GHz Celeron system and an Athlon XP 2200+. We would highly encourage system vendors like Compaq and eMachines to shift their low-end focus to AMD if their customers are of any importance at all. As we've seen through our extensive benchmarking, the Celeron's performance is truly dismal; so while Intel is quite competitive in the mid-range and high-end segments, their value processors are inexcusably slow compared to AMD.
This review really isn't complete without taking a look at overclocking performance. For enthusiasts who want a lot of performance for a small amount of cash (cache?), pushing a cheap processor beyond its limits is the way to go. Every overclocker remembers the original Celeron processor and its amazing ability to run incredibly fast because of its lack of cache. At this point (however unlikely), such a feature would be the only saving grace of the Celeron line. Of course, even if the Celeron is a good overclocker, it will be very interesting to see how high the Duron can be pushed with its cut cache as well.
The conclusion we can make from all this is that the Duron processor is a solid purchase. If you have the extra 40 to 50 dollars to spend, a Barton processor would be a nice addition to any system for that added dimension of performance to a tightly budgeted system. Hopefully, system builders will take note and start offering better performing systems for an even lower price based on the Duron processor rather than the Celeron. For those who want the cheapest possible system, AMD will give you the best performance every time.
Facebook
Twitter
RSS
97 Comments
View All Comments
arejerjejjerjre - Saturday, December 6, 2003 - link
MoronBasher buy a celeron machine and compare the results to anandtechs review you'll see the difference yourself! (By the way i didnt say that a celeron could ever beat anything else then amds low end!)I wonder why they used so high latencies in the tests? Of course because they noticed a celeron would perform poorly with that kind of settings!
arejerjejjerjre - Saturday, December 6, 2003 - link
Now I know why the internet has been so slow lately!!!!!!! Its because Intel based servers have been replaced with amds crappy systems!!!!!!!MoronBasher - Saturday, December 6, 2003 - link
People, keep returning your "dead" procs back to the store, cuz guys like me get them for free. LOL!!!What i found amzing was the fact that anandtech used ddr400 and they also used pretty high latency timings
novice - Saturday, December 6, 2003 - link
Wow, while the socket 370 Celerons also trailed the Durons, at least they were close. The current crop is really sad, compared to the AMD products. Definitely proving once again that clockspeed doesn't really mean much and AMD's "Performance Rating System" is not just a marketing tool.MoronBasher - Saturday, December 6, 2003 - link
arejerjejjerjre, you are a moron. Do you honestly believe amd chips just die? I am a computer maintenance technician and from the returns we get from stupid customers, it's not the proc that's the problem, it was the mobo... i could guarantee 99% of the time, the proc is not to blame. i have a 2500+ and a 2.8C and i have no problems with either. Errors? usually comes from drivers. or a moron like you who doesn't know how to set a computer up properly.DerekWilson - Friday, December 5, 2003 - link
from #51:"This latency is seen in some of the tests where the Barton performs worse than lesser clocked A-XPs despite a larger L2 cache."
The first XP chip clocked lower than the Barton is the 2200+ (1.800 GHz) which doesn't ever come close to touching the 1.833 GHz 2500+ Barton. AFAIK, all Athlon XP L2 caches (including Duron) have the same latency. I don't know this for a fact (though I highly suspect it), but it would be an easy test (just need to plug in sisoft sandra and look at cache latency).
Honestly, the fact that there are only a very few benchmarks where the HIGHER clocked 2400+ (2 GHz) can touch the Barton shows how important large cache size is in increasing overall system performance. Even the high latency L3 cache (which is still much lower latency than main memory) on the P4 EE helps to push performance much higher than on similarly clocked P4 CPUs.
The whole point of any ondie cache is to reduce latency between main memory and the processor. Having a large L1 L2 and even L3 cache doesn't increase latency, it decreases it overall. Without a cache, every single memory access takes a large number of cpu cycles to get to the processor, and much time is spent waiting for data.
Things get complicated when looking between Intel and AMD. AMD has a larger low latency L1 cache, but Intel's L2 cache is lower latency than AMD's L2 cache. But I think I'll save that analysis for another day :-)
Really the only advantage (aside from the cost savings) of cutting out a large ammount of cache is that you can more easily clock the chip higher. But that only really gives you a performance advantage if you can increase the clock enough to overcome the performance loss due to the lack of cache. The original Celerons could actually exceed this performance and that's why they were so sought after. The difference here is that P4 architecture is so much more sensitive to memory latency that we really can't hope for these kinds of performance gains.
I don't think that even overclocking a P4Celeron to 3.6GHz would help enough to matter. But hopefully we'll find out in an upcoming article ;-)
Pumpkinierre - Friday, December 5, 2003 - link
What people forget about the original celeron 300 (no L2 cache) was that it was lousy in benchmarks but gamers loved it because of its low latency and overclockability both of which are hindered by cache addition. The same applied to K6-2(no L2 cache) vs K6-3(256Mb L2). This is why these celerons are still out there as the demo benchmarks dont reflect the true gaming experience ie spontaneous response by the user. The other requirement is raw grunt(floating point calculation), province of the K7, which explains the Duron's longevity despite low clock speed. This CPU also has a very small 64K exclusive L2 cache, ideal for low system latency (and 64Kb is the memory unit of the X-86 based systems). This latency is seen in some of the tests where the Barton performs worse than lesser clocked A-XPs despite a larger L2 cache. And as I stated above real world gaming accentuates this quality further.Its true that the vast gain in cpu speed cf. memory speed has required a middle man ie cache. But if that middle man doesnt have the goods then re ordering takes time. My favorite theoretical gaming CPU would be a K8 without L2 cache (perhaps with 256Kb L1 cache) which in conjunction with the on die memory controller and optimised fpu would have very low system latency.
sonyboy851 - Friday, December 5, 2003 - link
all of you AMD haters need to get a clue. 48: what nforce 2 board did you use? And AMD systems are just as stable as Intel ones. Why have IBM and Sun chosen AMD Optorons for their servers? I doubt its because they are unstable. Maybe its because Intel cant offer such a great product.Anyways, this is a budget review, so I shouldnt even mention that. So you think Celerons are good? Is that what your saying?
CRAMITPAL - Friday, December 5, 2003 - link
Same old shit from the criminally insane Intel fanboys. Must really bruise your ego to have your face punched in every day by online reviews showing how pathetic Intel products are! Get use to it as things are only going to get worse for Intel and it's fanboys.Xbit Labs is reporting Intel can't even produce EE's and that the Flame Throwing Prescotts are a disaster. Intel has so many production and design problems that they can't resolve, they may need to release Tejas as an even bigger FLAME THROWER than Prescott. And by all accounts the Xeon even with L3 is dead.
I'd suggest anyone with a clue, buy stock in water-cooling companies and liquid nitrogen producers.
arejerjejjerjre - Friday, December 5, 2003 - link
And for those amd folk out there! You will never have the stability and flexibility what intel based machine can offer! I've seen too many amds aka too many problems,errors,etc!Theres allways something that doesnt work on an amd system!!! :)
By the way if anyone hasnt noticed NFORCE 2 is the worse I've ever seen! Nothing works as it should! :) LOL!