Quad Cores: Intel Kentsfield Preview
by Anand Lal Shimpi on September 28, 2006 7:00 PM EST- Posted in
- CPUs
At this year's Spring IDF, before the launch of Intel's new Core 2 processors, in an unprecedented move (for Intel) we were allowed to benchmark the new Core 2 processor alongside an AMD system and publish benchmarks. While we've previewed Intel processors in the past before their launch, this was the first time we were able to test something this far in advance of its launch, with Intel's support, and also test a competing AMD part at Intel's own show. Previously, Intel would hardly even recognize that it had a competitor, but in the past year alone Intel's attitude towards competition has changed dramatically. The change in heart was also quite evident at this year's Fall IDF, where featuring competing AMD solutions on stage for performance comparisons was far from taboo.
Generally speaking, Intel had a lot of success with its early previews of the Core 2 processors. Readers could get a good idea of performance well before its launch and they could plan upgrades accordingly, and as it turned out, the final shipping performance of Core 2 CPUs was nothing short of impressive. With the November launch of its quad-core Kentsfield and Clovertown based processors quickly approaching, Intel offered members of the press a similar chance to take an early look at quad-core performance at IDF.
This time around there were only two systems, both of which were Intel based. There's no need for a reference AMD system since we've already determined that Intel's Core 2 Extreme processor is faster than anything AMD currently has to offer; what we're now interested in is how much of a performance improvement you can get from going to quad-core. Both systems were identically configured with 2GB of DDR2-800 running at 4-4-4-12 timings on an Intel BadAxe2 motherboard based on the 975X chipset.
Unfortunately, we weren't very pleased with the benchmarking circumstances this time around. When we tested Core 2 at last IDF, we were able to run some of our own benchmarks and the benchmarks represented a reasonable subset of the performance tests we were already running in house. With Intel's quad-core setup, we were not able to run any of our own benchmarks and the majority of the tests we were able to run were of benchmarks we wouldn't normally use to begin with. While we could adjust settings and use some of our own demos/test configurations last time, this time we couldn't do anything other than click go and see the results generated in real time. There's no reason to believe that Intel did anything fishy, and the performance we saw agreed with internal performance testing of Kentsfield, but overall the time benchmarking at IDF was not all that useful.
Keep in mind that the benchmarks Intel selected to show off were specifically chosen to maximize the impact of four cores, and thus we're looking at a lot of significantly multi-threaded applications like 3dsmax. Realistically, the majority of tests in our normal CPU benchmark suite will not see a benefit from quad-core and thus the real purchase decision between dual and quad core is far more difficult to make than you'd think based on these numbers alone. At the same time, when your options at $999 are a 2.93GHz Core 2 Extreme X6800 (dual-core) or a 2.66GHz Core 2 Extreme QX6700, we'd generally recommend going for the latter if you're multitasking at all or plan on running any applications that are well threaded. Even games like Alan Wake and Unreal Tournament 2007 will be able to take advantage of the extra cores, and you can always overclock the processor to make up for the clock speed difference, but you can't enable more cores than you have.
With the stage set, let's take a look at the actual performance Kentsfield brings to the table.
Generally speaking, Intel had a lot of success with its early previews of the Core 2 processors. Readers could get a good idea of performance well before its launch and they could plan upgrades accordingly, and as it turned out, the final shipping performance of Core 2 CPUs was nothing short of impressive. With the November launch of its quad-core Kentsfield and Clovertown based processors quickly approaching, Intel offered members of the press a similar chance to take an early look at quad-core performance at IDF.
This time around there were only two systems, both of which were Intel based. There's no need for a reference AMD system since we've already determined that Intel's Core 2 Extreme processor is faster than anything AMD currently has to offer; what we're now interested in is how much of a performance improvement you can get from going to quad-core. Both systems were identically configured with 2GB of DDR2-800 running at 4-4-4-12 timings on an Intel BadAxe2 motherboard based on the 975X chipset.
 |
 |
| Click to enlarge | |
Unfortunately, we weren't very pleased with the benchmarking circumstances this time around. When we tested Core 2 at last IDF, we were able to run some of our own benchmarks and the benchmarks represented a reasonable subset of the performance tests we were already running in house. With Intel's quad-core setup, we were not able to run any of our own benchmarks and the majority of the tests we were able to run were of benchmarks we wouldn't normally use to begin with. While we could adjust settings and use some of our own demos/test configurations last time, this time we couldn't do anything other than click go and see the results generated in real time. There's no reason to believe that Intel did anything fishy, and the performance we saw agreed with internal performance testing of Kentsfield, but overall the time benchmarking at IDF was not all that useful.
Keep in mind that the benchmarks Intel selected to show off were specifically chosen to maximize the impact of four cores, and thus we're looking at a lot of significantly multi-threaded applications like 3dsmax. Realistically, the majority of tests in our normal CPU benchmark suite will not see a benefit from quad-core and thus the real purchase decision between dual and quad core is far more difficult to make than you'd think based on these numbers alone. At the same time, when your options at $999 are a 2.93GHz Core 2 Extreme X6800 (dual-core) or a 2.66GHz Core 2 Extreme QX6700, we'd generally recommend going for the latter if you're multitasking at all or plan on running any applications that are well threaded. Even games like Alan Wake and Unreal Tournament 2007 will be able to take advantage of the extra cores, and you can always overclock the processor to make up for the clock speed difference, but you can't enable more cores than you have.
With the stage set, let's take a look at the actual performance Kentsfield brings to the table.
Facebook
Twitter
RSS
34 Comments
View All Comments
Mclendo06 - Monday, October 2, 2006 - link
Will this chip have an unlocked multiplier like other "Extreme Edition" processors, or will it be extreme in core count only?Ealdric - Saturday, September 30, 2006 - link
These 4-core chips seem to be out ridiculously soon after the C-2-D. They could (should) have just gone straight to quad. Seems like the vendors will have a hard time keeping up.feraltoad - Saturday, September 30, 2006 - link
You guys see any pricing for the "lower cost" Kentsfields?I thought in the past quad-core? Wut game even uses dual-core? but with Crysis saying it will use multiple cores if they are there. I can't imagine a better way to drive amazing games with crazy physics/AI/enviroments than by using available cores esp. since Aegia is flagging, and the 360 and PS3 are themselves multicored. Plus the video encoding would rawk.
JNo - Friday, September 29, 2006 - link
I know time based percentages can be a little confusing at first but you guys need to sort out your maths...[given and identical task...] "As an example, it is not accurate to state that a score of 40 seconds vs. 80 seconds is twice as fast, but rather that the 40 second score takes half as long or the 80 second score is twice as slow."
Half as long IS twice as fast!! The 40 sec cpu can do the same task twice in the time that it takes the other cpu to do the task i.e. it can work twice as hard ergo it is twice as fast. Twice as fast is 100% faster. Now to be twice as slow it would have to do the task in 160 secs obviously (not 120 secs).
If it did the same task in 20 secs (vs 80 secs), where are we? Well it is 4x as fast! It can do the same task 4 times when the other cpu can only do it once. It is 300% faster (NOT 400%).
Speed difference (as a multiple) is Old time/New time but percentage difference is (Old time/New time -1)
This works the other way round too of course i.e. 80 sec vs 40 sec is 100% slower and 80 vs 20 is 300% slower.
JarredWalton - Friday, September 29, 2006 - link
I updated the text after one reader pointed out the error. Technically, it *is* correct now. 80 seconds is twice as long as 40 seconds; 40 seconds is half as long as 80 seconds. Some like to say "twice as fast" but that is slightly wrong. In situations where higher scores are worse, you have to change the syntax to remain grammatically correct.Time is a duration measurement, not a speed measurement. Would you disagree that 80 seconds is twice as long as 40 seconds? Or that 20 seconds is one fourth as long as 80 seconds? "Fast" is the wrong term to use for such a comparison, other than to say that 40 seconds is faster than 80 seconds. You could talk about rate of travel and say one guy is moving at 40 MPH and that's twice as fast as 80 MPH. Call it a symantic difference of opinion, but I don't like "fast" as a way of describing time.
JNo - Wednesday, October 4, 2006 - link
I understand that time is a duration measurement, not a speed measurement. That's why you have to think of it in terms of the amount of time taken to complete *a given task*, which is a defined amount of work, which will then allow you to make the speed comparison. If a car travels the same distance in half the time another car does (eg 40 secs instead of 80 secs), it is twice as fast. PERIOD."Some like to say 'twice as fast' but that is slightly wrong." - No it isn't! It is undisputably correct!
[quote from following comment] "What is 10% faster than 100 seconds? I suppose you can day it's 90.91 seconds if you want." - I don't just want to - it is! You can not say otherwise without being incorrect!
If you want to make this easier, use the INVERSE of time to define speed (which is scientifically correct - think of the equation distance = speed x time). Now, speed = distance (or workload - eg calculating pi to 2million dp or encoding a video file) divided by the time taken to do said task. Just make the workload = 1 for the sake of argument (as the answer will still be proportionally correct) and you have speed equalling the inverse of time.
This way, 1/40 (faster cpu) is twices as fast as 1/80 (slower cpu) and a cpu that does the task in 20 secs is 4 times as fast as the 80 sec cpu (1/20 is 4x 1/80). Use a calculator if you don't believe me. What if the faster one does task in 73 secs? It's speed is 1/73 compared to 1/80, which is 9.59% faster (1/73 divided by 1/80).
It is not a matter of semantics. It is fact. I am certainly not saying this just to be annoying cos most of the stuff you guys understand (cpu architecture etc) and evaluate goes way over my head. But this is quite straight forward - I just want to try to help you to get it right in future.
vhx500 - Friday, September 29, 2006 - link
.. it's "semantic", not symantic.*Ducks
yacoub - Friday, September 29, 2006 - link
I think his issue is regarding % speed improvement.If something scores 100 points on something, what's a 10% improvement over that? 110 points. What's 100% improvement over 100 points? 200 points.
Thus something that scores 200 points is not 200% faster than something that scores 100 points, it is 100% faster.
There was never a problem with 40s vs 80s being twice as fast. That's the same thing as "taking half as long".
JarredWalton - Friday, September 29, 2006 - link
But that works only for instances where higher is better, without becoming confusing. What's 10% faster than 100 seconds? I suppose you can say it's 90.91 seconds if you want. I prefer to stick with your math and simply state that 110 seconds is 10% slower. :)Kougar - Friday, September 29, 2006 - link
Great idea to create that Kentsfield compatible motherboard list! I've heard that a board needs an EPS12v connector to be truly compatible though... can you confirm this? Already seeing good results with the Abit AW9D-MAX being able to overclock these!Since "Core" processors are the best thing out there for gaming, I can't see why anyone would turn down "Core" based Kentsfield, just simply overclock it. Worst case you get the same performance of a overclocked C2D and have some extra cores... ;)
Again, thanks for the mainboard list!