Mobile CPU Wars: Core 2 Duo vs. Core Duo
by Anand Lal Shimpi on August 3, 2006 9:25 AM EST- Posted in
- CPUs
Core Duo vs. Core 2 Duo
We've spent a lot of time comparing Intel's Core architecture to NetBurst and AMD's K8; however, we've stayed away from quite possibly one of the most confusing comparisons: Core 2 Duo vs. Core Duo.
Unlike its desktop predecessor, Core 2 Duo comes from the same genealogy as the Core Duo. Despite the similarities in name and in architecture, there are some fairly major differences between the two CPUs, some of which won't become apparent until next year. The table below should help summarize the differences:
| Core Duo (Yonah) | Core 2 Duo (Merom) | |
Manufacturing Process |
65nm |
65nm |
Die Size |
90.3 mm^2 |
144.9 mm^2 |
Transistors |
151M |
291M |
Clock Speeds |
1.20GHz - 2.33GHz |
1.06GHz - 2.4GHz+ |
FSB Frequency |
533MHz - 667MHz |
533MHz - 800MHz |
L1 Cache Size |
32KB + 32KB |
32KB + 32KB |
L2 Cache Size |
2MB Shared |
2MB - 4MB Shared |
Pipeline Stages |
12 |
14 |
Decoders |
1 complex + 2 simple |
1 complex + 3 simple |
Maximum Decode Rate |
3 |
4+1 |
Reorder Buffer |
80 |
96 |
Issue Ports |
5 |
6 |
Scheduling |
Unified Reservation Station |
Unified Reservation Station |
Scheduler (# of Entries) |
24 |
32 |
FP Units |
FMUL/FADD: 1 |
FMUL: 1 |
SSE Units |
1 |
3 |
Integer Units |
ALU: 2 AGU: 2 |
ALU: 3 AGU: 2 |
Load/Store Units |
Load: 1 Store: 1 |
Load: 1 Store: 1 |
Socket Interface |
Socket-M (PGA/BGA) |
Socket-M (PGA/BGA) & Socket-P (PGA/BGA) |
Compared to the desktop Core 2 Duo (Conroe), the mobile version is architecturally no different. Obviously clock speeds (both CPU and FSB) are lower because these things will be going in notebooks where power consumption is more of a concern, but other than that the architectures are identical.
Compared to Yonah, Merom has some very clear advantages; on the surface the larger L2 cache is responsible for the 140M increase in transistor count, but architecturally the improvements extend far beyond that. You can get the details from the table above or from our previous articles on Intel's Core 2 processors, but simply put Merom is wider and slightly deeper than Yonah. The slightly deeper pipeline helps increase clock speeds on Merom (which will bump performance a bit), but the added decode and execution width will increase overall performance.
Not listed in the table above are the improvements to the cache subsystem and memory accesses on Core 2 Duo. Merom features more aggressive prefetchers than Yonah, as well as Intel's Memory Disambiguation technology that allows for out of order loads. In other words, not only is Merom able to operate on more data at once, at a faster speed, but it can also get access to that data quicker.
New CPU today, new platform in 2007
The first versions of Core 2 Duo are completely backwards compatible with the Napa platform that Core Duo currently uses, and thus they share the same Socket-M interface. Unfortunately for Merom, Napa only supports a maximum of a 667MHz FSB and thus has almost 40% less bandwidth to the CPU than the desktop version, and is identical to what the fastest Yonah CPUs use. The problem with FSB limiting Merom like this is that Merom is a hungrier core (as we've seen by the table above) than Yonah, so it needs a faster FSB in order to truly stretch its legs. The tradeoff is that a faster FSB consumes more power, thus reducing battery life, not to mention that you'll need a "new" chipset to support the faster FSB.
That new chipset is part of Intel's Santa Rosa platform, to be introduced in early 2007. Santa Rosa is composed of Intel's upcoming Mobile 965 chipset, ICH8M and a new wireless solution with 802.11n support. The new chipset will add official 800MHz FSB support, and thus Core 2 Duo processors released next year will be able to use the faster FSB. The Santa Rosa platform also introduces a new pin-out, Socket-P, for Core 2 Duo processors. Unfortunately that means you won't be able to use current Core 2 Duo and Core Duo processors in Santa Rosa based motherboards and notebooks.
Facebook
Twitter
RSS
46 Comments
View All Comments
Spacecomber - Thursday, August 3, 2006 - link
I know that we are perhaps past the time for this, but I'd be curious how the Pentium M stacked up against its replacement, the Core Solo. It might shed some light on the roots for the Core line of processors.ksherman - Thursday, August 3, 2006 - link
I for one hope Apple pops these babies in the MacBook as well as the MacBook Pro. I have been reading a lot of rumors suggesting that Apple will only put Merom in the Pro model at first... Seems kinda goofy, since they purchase processors in *relatively* low quantity. Ive got my eye on the MacBook, so any performance increase with no price premium is always a plus, and I do plan on doing a lot of video editing/renderingmlittl3 - Thursday, August 3, 2006 - link
I'm waiting for the third revision of the Macbook Pro in the middle of next year. By this revision most of the problems with the new designs should be ironed out and these notebooks will probably be based on the Santa Rosa platform (800 MHz FSB). Right now I have a 1.5 GHz G4 Powerbook and it does what I need. I will upgrade to the Macbook Pro 2.4 GHz Core 2 Duo (Merom) on the Santa Rosa platform. If you like your notebook right now, I would wait until then. This would give you the most stability and bump in performance in the near future. Expect the third revision sometime next summer.mlittl3 - Thursday, August 3, 2006 - link
To be more clear:First Revision - Macbook Pro 2.16 GHz Core Duo 667 MHz FSB 2 MB (1Q 2006)
Second Revision - Macbook Pro 2.33 GHz Core 2 Duo 667 MHz FSB 4 MB (3Q 2006?)
Third Revision - Macbook Pro 2.4 GHz Core 2 Duo 800 MHz FST 4 MB (2Q 2007?)
AndrewChang - Sunday, August 6, 2006 - link
Santa Rosa... At the earliest? I mean, I supose with the Core 2 Duo being 'crippled' by a slow(er) FSB, it might be worth the wait. But what do you think Anand means by, at the earliest?Whats next after Santa Rosa? Does he know something we don't know? Well, thats a given, but now I'm sketched out about all this... Should we expect some early adopter problems with the introduction of this newfangled Robson technology? God, for a hardware enthusiast, who would've thought that making a new hardware purchase could be so tough. All I want is the fastest performing Merom/Leopard based Macbook Pro available. Am I really going to have to wait until at least Santa Rosa next year? It's going to be a long wait...
Olaf van der Spek - Thursday, August 3, 2006 - link
What three Compaqs?
Anand Lal Shimpi - Thursday, August 3, 2006 - link
Didn't you see the three compaqs in the review? ;)Take care,
Anand
yacoub - Thursday, August 3, 2006 - link
now this is a test i can totally appreciate: everything is identical except the CPU, so you get to see what the REAL WORLD benefit of changing the CPU is in your REAL WORLD system that people might actually buy/own. ie, instead of maxing everything else out with parts 99% of people don't buy / can't afford.of course the result is that you see that the real world difference is only noticeable in some situations and with some programs. but hey, that's the reality of it and actually it's easy to see that since the pricing is comparable and all else the same, it's a decent upgrade and certainly a level of future-proofing as well.
jones377 - Thursday, August 3, 2006 - link
It was interesting to compare the numbers in this review with the previous Core 2 Duo desktop review where 2MB vs 4MB L2 cache was examined (although at 1,83GHz/1066)http://www.anandtech.com/cpuchipsets/showdoc.aspx?...">http://www.anandtech.com/cpuchipsets/showdoc.aspx?...
Its not a perfect comparison but from what I can gather, there are significant improvements performance wise coming from the core, even in the non-FP/SSE related benchmarks. A favorite argument among some people is that the extra cache makes all the difference, I hope this will shut them up! (tho I really doubt it)
iollmann - Tuesday, September 26, 2006 - link
In SSE code, I see close to a factor of 2 performance increase from Yonah to Merom much of the time. These benchmarks are depressing. The improvement should be better than what we see. Does no one vectorize?