Collection of Conroe Data. (Core 2 Duo and Core 2 Extreme!)

[lbax]

Deleted by author

 

[306maxi]

Updated the Hexus data, (it was better for Conroe than I orginally had), and added a link to the original article.

Thanks to mxyztplk.

You mean thanks to me 😛

I posted it on Thursday and he posted it on Sunday 😛

 

[iterations]

Ah, my bad, thanks to both of you. He sent me a PM.

 

[iterations]

For folks that don't want to deal with percentage crap, here are charts for the real world apps on the Hexus list that are CPU limited:

 

[306maxi]

Ah, my bad, thanks to both of you. He sent me a PM.

No problems Just wanted to point that out

 

[DavidC1]

EE version - It there's no 3.3 Ghz processor w/ 1333 FSB.
What are you paying so much for? The extreme added something extra besides speed. - more cache last time. To add a little speed and call it the extreme version. Doesn't make sense, and from a market ploy,
it makes it a joke.

The ORIGINAL rumors were that the Core Extreme 3.33GHz/1333FSB was $1169 and Core 2 Duo at 3.2GHz/1066FSB was $969.

Cache doesn't seem to matter on Pentium 4's. It mattered back in Northwood vs. Willamette, but as architecture got old, it reached a limit where cache doesn't do much.

However, unlike Pentium 4, Core 2 SCALES WELL with cache, and FSB. So if people were planning to get the Core 2 Duo 3.2/1066 for $969, its better for them to get Core Extreme 3.33/1333 for $1169.

 

[DavidC1]

What is even more phenomenal in the gaming benches is from this page:
http://www.pconline.com.cn/diy/evalue/evalue/cpu/0605/791941_13.html I am not sure if I believe it though, it seems too good to be true.

Ok, I don't know what's there not to believe. They are using their own CPUs they managed to get hands on, which makes sense as Japan and China gets exotic samples faster.

I believe it. Here's why.

If 2.66GHz version is faster than 2.8GHz FX-62 by 40%, can 2.13GHz version be slower somehow?? If it scales linearly, which never happens, 2.13GHz would be 12% faster. However CPUs doesn't scale linearly with clock.

If the numbers show in sites like Anandtech or Tomshardware, which for some reason people think its more "true" than the chinese site, they are gonna be shocked.

 

[ltcommander_data]

However, unlike Pentium 4, Core 2 SCALES WELL with cache, and FSB. So if people were planning to get the Core 2 Duo 3.2/1066 for $969, its better for them to get Core Extreme 3.33/1333 for $1169.

Hmm, I'm generally of the opinion that going from a 1066MHz FSB to a 1333MHz FSB on Core 2 wouldn't make much difference at the same clock speed. Seeing the types of numbers that Merom turns in it doesn't seem like Core 2 is that bandwidth limited, probably thanks to the large 4MB L2 cache. A 800MHz FSB would probably be adequate, a 1066MHz FSB preferred, and a 1333MHz FSB probably has room to spare. The one benefit of a 1333MHz FSB would be reduced latency, but I don't think the difference would be drastic and again the large L2 cache really buffers everything. I believe I've already said this earlier in this thread, but I'd love to see a comparision once everything's released of a 2.33Mhz Merom on a 667MHz FSB, a 2.4GHz Conroe on a 800MHz FSB, a 2.4GHz Conroe on a 1066MHz FSB, and a 2.33MHz Woodcrest on a 1333MHz FSB all with a 4MB L2 cache. I guess a Conroe Extreme Edition could also be used to clock at those different FSB speeds. That would really settle whether Core 2 is really limited by using an FSB.

 

[DavidC1]

Post deleted as it was moved.

 

[DavidC1]

Seeing the types of numbers that Merom turns in it doesn't seem like Core 2 is that bandwidth limited, probably thanks to the large 4MB L2 cache.

Cache and Bandwidth is related to each other. Looking at the Chinese site, PCOnline, it shows that 2.13GHz 4MB L2 is 11-19% faster than 1.86GHz 2MB L2. It shows that Conroe loves cache, unlike the other CPUs(this is why saying Athlon 64's should have same cache as Conroe before being compared argument is false, since Conroe benefits more).

Think of it this way:
Computerbase.de showed(http://www.computerbase.de/news/hardware/prozessoren/intel/2006/maerz/cebit06_benchmarks_intel_merom/) that Merom at equal clock is 20% faster than Yonah with same platform, that is equal FSB, memory, and chipset, in Quake 4 benchmark, even though it was using a crappy video card, a mobility X1400.

Anandtech showed that X2 with 1MB L2 per core is 4.8% faster than Yonah at same clock: http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2648&p=13

We can also roughly extrapolate that an X2 that clocks equally faster compared to Yonah as FX-60 2.8GHz is to 2.66GHz Conroe, will be around 7-8% faster than Yonah from the same benchmark.

Yet, Anandtech's IDF Conroe preview showed that Conroe is over 22% faster than FX-60 at 2.8GHz, which is basically the same CPU as X2 but higher clock.

So, there is roughly, 1.22*1.07=31% advantage of Conroe vs. Yonah, which puts 1066MHz FSB vs. 667MHz FSB about 8% difference.

However, if you read Anand's second preview carefully, you'll notice there are TWO Quake 4 values, one with Intel's demo, and one with Anand's demo. One with Anand's demo, which is the same thing as one used on their Yonah benchmark, Conroe is actually 28% faster, not 22%.

Then the difference between 1066MHz FSB and 667MHz stretches to 12-13%.

Very rough, I know, but it shows that Conroe also needs a lot of memory bandwidth. And it makes sense, as more powerful CPUs don't show full potential unless they have enough bandwidth.

 

[DavidC1]

You know what guys?? Especially regarding two posts I put with response to iterations and JumpingJack, please don't take it personally. I wanted to express my opinions and why I think of it that way. I describe it very detailed since there will be some others that respond to me.

It seems to good to be true. I guess it makes sense as Yonah performed so well already, and even I didn't look that as being much. I thought Conroe couldn't be much faster than Yonah in games, maybe in other apps it will have big advantage, just like Pentium 4's did.

I estimated back when first IDF preview came that only 2.13GHz will be needed to equal FX-60. I also thought it will get better as time passes by, as the following showed that advantages became greater as Anand used he's own test programs, not supplied by Intel. It seems to become all true, even better than what I originally thought.

 

[gOJDO]

What about Conroe 2.4GHz (9x267MHz) FSB1066. I wonder if it can be stable overclocked at 3.2GHz (9x400) with air cooling, FSB 1600MHz(16GB/s) with DDR2 1GHz(+50% FSB overclock means +50% DDR2 overclock, 667MHz * 1.5 = 1GHz). I wonder if there will be chipsets that will support more than 50% overclock on the FSB/RAM.

 

[iterations]

Shouldn't be a problem gOJDO, the folks who have the 2.4GHz chips seem to be hitting 3.6-3.7 consistantly on air. Of course that is with a mobo and RAM that can handle a 1600MHz FSB.

The RAM has to run at 400MHz, for a 1600MHZ FSB, not 1GHz.

 

[Guest]

Its funny I did the same calculation in a boring class...The 2.4 will probably what I will be getting although a lower clocked part with 4 megs my do even better if the OC headroom is so big.

I've been looking at september/october for a upgrade and it seams like a perfect time, especially with the lauch being Juin 23rd, that'll give it some times for platforms to mature and to see what memory works best. Wich bring me to my next point, many people stated it, all of the benchies are on a 975x chipset, I just cant wait to see whats the potential of this chip in nf5/ATI/newer Intel chipset/etc. I think it'll be intersting and really fun too follow from an enthusiast perspective, can't wai too build my next Pc 😀

Also on a side note, I will be attending Intel's ICC connference in montreal on friday. I dont really know what too expect since its my first attendance to this kind of conference, but I'll be sure to bring a camera and to report anything interesting. Also they talk about big prizes, to me it means some Conroe's /keep dreaming

 

[iterations]

New May 29, 2006: Complete analysis of Conroe.Merom SuperPi performance:

Conroe's SuperPi performance scales perfectly linearly with clock speed and is showing no signs or bottlenecking. Nifty.

 

[DavidC1]

Makes you wonder if the advantage of an IMC was not realized at K8 introduction, and the bottle neck now is actually in the core and IPC does it not?

K8, was SIGNIFICANTLY faster than AthlonXP. Majority of the performance is IMC alone. AMD estimates 20% from it.

 

[S7A88Y]

Not to heckle you on your little fanboy war/fantasy preformance fourm but what about the 64bit preformance. In a year a vast majority of 64bit computers will be running Vista. I asked you because you seem to have all the answers.

 

[iterations]

Not to heckle you on your little fanboy war/fantasy preformance fourm but what about the 64bit preformance. In a year a vast majority of 64bit computers will be running Vista. I asked you because you seem to have all the answers.

No need to heckle or be immature. If you want info and I can find it, I'm happy to provide it.

It is interesting you say 64-bit is so important, as I've had a hard time finding a pure 64-bit example (As in 64-bit capable chip, OS, drivers and application) for desktop systems.

Nevertheless, I did find one example here:

Past Pentium systems would actually suffer from degraded 64-bit performance, but Conroe gains performance when moving to 64-bit as it should be.

CINEBENCH 2003 32-bit on Conroe/Win2003: 642/1190
CINEBENCH 2003 64-bit on Conroe/Win64: 759/1401

These are at the same clockspeed and FSB/memory timings.

32-bit:

64-bit:

 

[tehue]

quick as this:
1) is the first core 2 cpu going to be a "conroe" or an "allendale" core?
2) how come noone is mentioning "millville"? i'd surely like to see some benchies on this cpu
that's all for now

 

[DavidC1]

quick as this:
1) is the first core 2 cpu going to be a "conroe" or an "allendale" core?
2) how come noone is mentioning "millville"? i'd surely like to see some benchies on this cpu
that's all for now

1) Conroe for desktop, Merom for laptop, Woodcrest for Xeon
2) Because it doesn't show that code name in roadmaps. Besides it doesn't make sense Intel has a code name for single core CPU and lower cache CPUs when they didn't have for Prescott, 2MB L2 cache Prescott was called Prescott 2M.

 

[iterations]

Allendale and Millville will come later, they are precursors to the Kentsfield Quad core chip. Allendale is scheduled for Q4 06, I'm not positive on Millville off hand.

 

[iterations]

New May 30, 2006: Core 2 Extreme has officially broken 5GHz and a new SuperPi_1M record has been set (10.281sec):

 

[ltcommander_data]

It's unclear whether those 2MB Conroes are really Allendales, but they are probably just reject 4MB Conroes. In any case, Intel really should market the 2MB Conroes as part of the E6xxx series since it's confusing and missrepresentative.

Regarding Millville, I already posted an article about it in this thread.

http://www.dailytech.com/article.aspx?newsid=2532

They are now calling it the Conroe-L and it's scheduled for Q2 2007 as a Celeron D replacement. The likely specs are 2MB of L2 cache and a 667MHz FSB to match the DDR2 667 memory that is now becoming more popular since that's what the AM2 Semprons support. I'm more inclined to believe a 2MB L2 cache since previous Celeron M products base on the Pentium M and Core architectures have maintained half the cache of the full versions rather than 1 quarter the catch as in the case of Netburst based Celeron Ds. There really wouldn't be a need for a single core value processor to have a 800MHz FSB so a 667MHz one should be more than sufficient since it's already more than Dothan.

In regards to Kentsfield being related to Allendale, I'm hoping that that is not the case. It's wishful thinking for Intel to double the cores yet reduce both the L2 cache and the FSB and expect decent performance gains. I can see it being weakly justificable in a server environment which would more likely take advantage of the extra cores inspite of the bandwidth limitations, but for the desktop world most things are barely optimized for 2 cores much less 4. Kentsfield will hardly be a hit for it's gamer targeted market if its single and dual core performance is limited by FSB and cache versus the current Conroe Extreme Edition and a possible future 1333MHz one.

Luckily, recent Intel documents have said that Cloverton will be using 4MB of L2 cache per die which indicates that Woodcrest or Conroe cores are now being used rather than 2MB Allendale dies. The larger cache will definitely assist in memory bandwidth. Now if only they can get the 1333Mhz 3 electrical load FSB figured out, although those 1600MHz+ FSBs the Conroe samples are achieving gives hope that the signal strength is there. Seeing that Cloverton is using 4MB Woodcrest cores, it's likely that Kentsfield will also be using 4MB Conroe cores. Using Allendales would have made Kentsfield and Cloverton an extreme ripeoff too. Seeing that the cheapest Conroe 2MB is only going to be $183, Allendale would have been simiarly priced. Putting 2 of them together would have meant that Intel would have charged $999 for less than $400 worth of dies.

 

[AlexanderTheGreat]

New May 30, 2006:
SuperPi_1M record has been set (10.281sec):

Core 2 EX @ 5GHz+

Why ...Why am I not surprised?

 

[306maxi]

New May 30, 2006:
SuperPi_1M record has been set (10.281sec):

Core 2 EX @ 5GHz+

Why ...Why am I not surprised?

Because you're not easily surprised :?: