Multi-Core CPUs

Dual core processors have suddenly exploded upon us. We

were nicely chugging along, totally content with our 64FX processors when

suddenly poof! AMD hit us with their X2s and Inte with their Pentium Ds. The

battle for dual core supremacy has continued on and now Intel has launched Yonah,

or Core Duo/Solo as it chooses to call it.

We have long wondered what each processor would do in real

world tests and what kind of real, tangible performance gain we can expect from

them. When we got our hands on a couple of Yonah powered notebooks, we felt this

was the right time to compare the top notch performances of these processors and

figure out which one is better and from past experience, where Intel has reached

in terms of performance Vis-à-vis AMD.

The processors used were AMD's Athlon X2 4800+, 4200+,

Intel's Pentium D 840 and the W5F, a Core Duo T2500 based notebook from ASUS.

What's Yonah



To re-iterate, Yonah is Intel's dual core offering for the mobile

platform. Yonah is really the codename. The marketing name is Core Duo or Core

Solo depending on whether it is dual core or single. It is based on a whole new

fabrication process, it is thinner than the earlier processors, faster than the

Pentium M and uses far lesser power to go about its business. If this was not

enough, its light as air and if you were to hold one in your hand, you would

think it was a silicon dummy.

The promises that Intel has made with Yonah are quite

serious. The most important claim being that it can run two cores at the same

power consumption levels. So you get twice the performance while your battery

life at least stays the same if it is not indeed enhanced.

The Yonah core that forms Intel's dual core strategy. This is also called the CoreDuo

The processor still is a 479 pin variety however, this in

no way means that you can take a 479 pin board from your old Pentium 4s and

stick the Yonah in it. No way. Even though the processor uses 479 pins, Intel

has changed the pin configuration so no backward compatibility. If you just

invested in a new notebook with the Pentium M, hard luck.

The main reason for the unbelievable size of the processor

is that unlike the Pentium D (which is rather huge) it is based on the 65nm fab

process and each core accesses a shared cache. In Pentium D, the 2MB cache is

basically divided between the two processors so that each has access to a full

MB.

With the Yonah, Intel has once again changed the

nomenclature of its processors. The three digit numbers are now gone and they

have been replaced by an alphabet followed by 4 digits depicting the expected

performance level of the processor. There is no necessity however that a

processor with the tag of T2500 is running at 2.5 GHz. The names are perhaps

indicative of the expected performance levels or classification of the

processors. If this were so Intel would've done something rather ironic. It

would've copied AMD's nomenclature concept. Funny thing being that AMD's

names were based on Intel in the first place!

Anyway, the T followed by a 2xxx means that the processor

is dual core. A T followed 1xxx implies a singe core or Core Solo processor. All

the Txxxx processors consume only 31W of power and have all the goodies such as

EIST and Execute Disable.

The fact that they use such a small amount of power in no

way means there are no ULVs. The ULV dual core range currently consists of the

L2400 and L2300, both of which utilize a mere 15W of power and run at 1.66 and

1.5GHz respectively.

The key specs of the Athlon X2. Virtualization technology will make its way into the CPU in the coming year

Yonah is the result of some rather splendid engineering by

Intel guys over in Israel. They have at once, managed to retain some of the best

features and learnings from processors gone by (both desktop and mobile). They

have also managed to iron out some of the issues plaguing the current generation

of processors. All this has led to rather incredible performance gains both in

speed as well as power maximization.

The Instruction Pipeline



The first thing that changed quite dramatically is the instruction pipeline.

It is down to around 12 from the mega 30+ in the Pentium 4s. Think of a pipeline

as an assembly line.

To give you a small example, imagine there are 10 people

assembling a PC on the assembly line. Each guy is doing one part of the job.

There is one guy attaching the hard drive, another attaching the RAM, third,

screwing the cabinet together and so on. Obviously, since each person is doing

just one job, he will need much lesser time and the PC will move from one stage

to the next pretty fast. However, you will need a lot of people to finish the

PC, which will make your assembly line rather long. This is akin to a long

instruction pipeline, which results in a higher clock speed.

On the other hand, if you make each person do multiple

jobs, it will reduce the overall assembly line's length but each stage will

obviously take longer. This is equivalent to a shorter pipeline, which results

in a lower clock speed.

The fact is that the 65nm fabrication coupled with a

shorter pipeline ensures lesser power consumption and overall heat dissipation

which does give Intel a bit of a head-room to crank up the clock speeds of the

processor. We think that even though right now, the Core Duo processors start at

a mere 1.66 GHz and go up to only 2.16 GHz, we would not be surprised if there

are more Core Duos launched with speeds up to 2.4 GHz /core.

Smart Cache



This is another indication of why Intel's Pentium D has been languishing

behind AMD's X2 for so long. Intel didn't realize that just putting in two

cores into a processor will not help performance all that much. What is equally

important is the way the two cores interact. So far, the way Pentium D was doing

this was rather silly.

In AMD's X2, there is a system, which ensures that the

cores keep accessing the common cache and are able to interact with each other

at core speeds. This ensures that precious clock cycles are not wasted while the

system waits for the painfully slow bus to do the needful.

In Pentium D, if core 1 wants to talk to core 2, the

message is taken out of the core using a painfully slow bus, and then finally

enters the other core where it is processed and sent back. All this while, core

1 is obviously waiting for the message to come back because it has obviously

kept things on hold (out of order execution capabilities not withstanding).

In Yonah, Intel has fixed this problem and has done a

brilliant job at that. Yonah does away with the entire concept of different

caches for each processor. It just pools in all the cache and connects both

processors to it with a bus. The bus then works somewhat like the on-die memory

controller found on the AMDs and ensures that the cores can talk to each other

at much faster speeds than before.

Obviously, this still does not mean that the Yonah cores

can communicate at core speeds and to be fair, AMD's on die controller is

still a master solution to this problem as it does away with the need for a BUS

for all practical purposes.

Now you might ask why the two cores would want to

communicate to each other in the first place? The reason is quite simple. Just

because there are two cores doesn't mean they are doing things independently

and are processing completely different streams all the time. There will be

times when they will need to work on the same data. This means that each core

needs to know the most current data storage location/processing status so it can

work on the correct data set.

Another interesting thing is that the cache allocation to

each core is dynamic. In fact, if there is very little requirement for cache,

the memory controller will flush out everything stored in the cache to the main

memory and put the cache in the off stage. This again helps in power saving but

ensures that the cache latencies are higher, indeed, in Yonah, the L2 cache

latency is as high as 40% more than what we have seen on Dothan (Pentium M).

Intel's Digital Media Boost



Intel was the pioneer of bringing multimedia home when it started with the

MMX and multimedia technologies back in the days of Pentium and Pentium II etc.

Even though multimedia capabilities of the processors have increased many folds

on the desktop, their mobile counterparts have suffered in this area. The prime

reason has been that such applications require enhanced SIMD (single instruction

multiple data) operations as well as multi thread processing at the instruction

level. This obviously sacrifices power savings and has not been so far

incorporated into a processor.

This has however changed with Intel giving Yonah completely

media oriented capabilities. All of Yonah's decoders will support SSE3

instruction decoding, where SSE3 is the 3rd generation of Single SIMD

Extensions. The first was obviously MMX and the second being SSE2 which was seen

on the Pentium IIIs.

All this will ensure that if you were to use your notebook

for heavy multimedia use like graphics editing and perhaps even 3D modeling (by

extension gaming? We hope so!), it would give you much better performance than

you would otherwise get on a notebook. Again, all this while your notebook's

battery life stays the same!

Napa V/S Sonoma



The earlier mobile platform was codenamed Sonoma. This was the platform that

was used in the Pentium Ms (Dothan) that we all use today. Napa is the newest

mobile platform from Intel. It is based on the Yonah processor and the 945

mobile chipset. It promises to upgrade the overall speed as well as power

consumption optimization of the entire platform.

The reason why we need to understand the platform

specifications is simply because Intel itself has moved away from providing

processor centric solutions to providing entire platforms that make everyone's

lives a little easier. In Napa, for the first time, the FSB is clocked higher

from the 533 MHz of Sonoma to a cool 667 MHz. We suspect that if Intel really

wanted, it could've given us a much higher clocked FSB but that obviously

would've translated into more power consumption, something Intel is not too

keen on right now.

The Test Setup



The notebook ASUS sent us consisted of 1 GB of RAM, a 40 GB hard disk drive

and of course the Core Duo T2500 processor. This is the second fastest processor

in the Core Duo stable. To compete with it, we lined up the X2 4800+, the X2

4200+ and the Pentium D 840.

The test bed used was an ASUS A8N-SLI, 1 GB DDR400 RAM,

7200rpm SATA HDD and NVIDIA 7800GT graphics.

We benchmarked the processors on a variety of tests, from

office productivity to raw power and rest assured the results were nothing short

of stunning!

How they fared



We started with an overall system benchmark using SysMark 2004. As expected,

the X2 4800+ zipped ahead of everyone with a score of 222 but guess what the

T2500 scored, a cool 206! Just 16 points short of a 4800+ is perhaps a statement

in itself. When you think about the fact that this score was achieved on a

notebook with a slower HDD, it becomes even more impressive.

In almost all the other tests, the T2500 was nowhere near

the Athlon X2, which was the fastest processor in every test except two (more on

this later). What we did find very intriguing was the fact that Yonah

outperformed the Pentium D 840 quite considerably in most tests. This in itself

is testament to the Smart Cache and Media Boost technology implementations that

Yonah enjoys.

We also did some hardware independent purely processing

tests like Super Pi and WinRAR compression. In Super-Pi, we were simply stunned

to see that the T2500 actually outperformed the X2 4800+ by a good 5 seconds! We

were so surprised by this that we ran this test twice just to be sure!

In WinRAR, the X2 with its higher core clock did manage to

outdo the T2500, which came in second, leaving the X2 4200+ and the Pentium D to

bring up the rear. To ensure that we are using a practical dataset, our 526 MB

of data consisted of ppts, TIFFs, word files, excel files, outlook express

folders and so on.

We were unable to run the gaming benchmarks on this

notebook because sadly, even now, Intel's IGP is practically good for nothing.

We are just waiting for an Xpress600 or an NVIDIA C51 based board to come along

for Yonah so we can have workable onboard graphics. However, we suspect that

Yonah, without an on-die memory controller would've certainly lost out to the

AMD processors.

Bottom line: To sum up, if you read our previous article on

Intel's nomenclature and took our advice to hold on before buying a notebook,

you did well by listening to us. Yonah, or Core Duo, is one of the best mobile

processors we have ever come across. The mere fact that it can give X2s a run

for their money while giving you atleast the same battery life as a Pentium M is

truly commendable. Take a bow Intel, excellent job!

Varun Dubey