Wednesday, November 30, 2005

INTEL has efficiency problem in the Yonah core

Anandtech has an exlcusive benchmark on INTEL's Yonah dual core notebook chip against AMD's Athlon 64 X2 3800+. The Yonah is 2GHZ with 2MB cache made on a 65nm process, the Athlon 64 x2 3800+ is 2GHZ with 1MB cache (2 x 512KB). As you can see, despite having twice the cache, the 65nm Yonah CPU is slower than Athlon X2 3800+ at the same frequency in most of the tests, sometimes significantly so. For example, Yonah 2GHZ is 20% slower than Athlon 64 X2 3800+ in video encoding tests. In all, the score is 16 to 6, the x2 3800+ won 16 tests, and the Yonah won 6 tests.

This is not surprising, in a comparison of INTEL Dothan 760 (2GHZ, 2MB cache) and AMD Turion ML 37 (2GHZ, 1MB cache) done by laptoplogic, the Turion won 15 tests, the Dothan won 7 tests.

From this, one can conclude INTEL's Dothan and Yonah CPU cores are lower than Athlon 64 core on 32 bit IPC (Instructions Per Clock).

Furthermore, Yonah is a 32 bit chip based on the old Pentium III. In June 2005, Mooly Eden, an INTEL VP claimed that enterprise demand for 64 bit is low. INTEL "made a conscious decision" to not add 64 bit to Yonah to achieve reasonable battery life. Eden claimed that "It may take many years for enterprises to demand it (64 bit)". As a result, INTEL's Sossaman CPU, which is adopted from Yonah for servers, will also be a 32 bit chip. The Athlon 64 x2 is a 64 bit chip by design and can run future operating systems such as Windows Vista, which is a 64 bit OS to be released in 2006. The Athlon 64 X2 gets a 30% performance boost under a 64 bit OS.

AnandTech also compared the power consumption of two systems under max load. The Athlon 64 X2 3800+ desktop consumes 144 watts, the Yonah consumes 108 watts. The Athlon 64 X2 consumes 36 watts more. According to AMD, the Athlon 64 x2 3800+ is a 89 watt chip, therefore, the Yonah is at least 89- 36 = 53 watts*, which is basically the power consumption of two Dothan cores. This indicates that INTEL gets no power reduction by going from 90nm to 65nm. A very bad sign, because INTEL's speed will be again limited by heat. In comparison, the popular AMD Opteron 870HE, a 8 way 64 bit server processor at 2GHZ is 55 watts.

From AMD's persentation, with DSLSSOI, AMD's 65nm is much lower on power.

* Total_power = CPU_power + MB_power + Component_power

1) 144 watts = CPU_power_x2_3800 + MB_power_ASUS_A8N_SLI + Component_power

2) 108 watts = CPU_power_yonah_2GHZ + MB_power_yonah_unknown + Component_power

Noting that the same components were used in the X2 3800+ test system and Yonah test system, and CPU_power_x2_3800 is 89 watts, substract 1) from 2), we have

CPU_power_yonah_2GHZ = 89 + 108 - 144 + MB_power_ASUS_A8N_SLI - MB_power_yonah_unknown

CPU_power_yonah_2GHZ = 53 + MB_power_ASUS_A8N_SLI -MB_power_yonah_unknown

We can safely assume that the ASUS A8N SLI desktop MB consumes more power than the Yonah Notebook motherboard.

Proof: Yonah is at least 53 watts

2 Comments:

Blogger Sharikou, Ph. D said...

In the Dempsey vs Opteron test, AnandTech chose to run
Opteron in slow mode and run Dempsey in fast mode. see
comments on AnandTech's response:

http://sharikou.blogspot.com/2005/12/why-did-anandtech-handicap-opteron.html


In the original Yonah vs X2 3800+ test, AnandTech drew
conclusions in favour of Yonah despite the fact the X2
won the tests with a score of 16:6 with only half the
cache:

http://sharikou.blogspot.com/2005/11/intel-has-efficiency-problem-in-yonah.html

Now, AnandTech redo the benchmark, again the X2 won by
a score of 11: 7

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2648&p=14


What puzzles me is why AnandTech keeps saying stuff
like:
"We continue to see that the Core Duo can offer, clock
for clock, overall performance identical to that of
AMD’s Athlon 64 X2 - without the use of an on-die
memory controller. The only remaining exception at
this point appears to be 3D games, where the Athlon 64
X2 continues to do quite well, most likely due to its
on-die memory controller.
"

AMD's DCA is an archictectural advantage, the use of
on die memory controller is an industry trend, all
modern CPUs like Alpha EV7, AMD64, UltraSparc T1 and
Power5 have on die mem controller, only the outdated
ones don't. Using its huge R&D budget, INTEL may have
squeezed out some additional performance out of the
Pentium III core without an on die controller, but
that's poorman's kludge and it won't last. Especially,
considering Sossaman32 is a server version of Yonah,
without a on die mem controller, its system bandwidth
is severely limited.

It seems that AnandTech like to remove the on die
controller of the X2 and then do a "fair" comparison"
with the Yonah.

If AnandTech worried so much about X2's on die mem
controller giving X2 an advantage, why didn't
AnandTech clock down the Yonah's memory frequency by
100MHZ so a better comparision of the cores can be
done?

Why didn't AnandTech think about X2's biggest
advantage: test X2 in 64 bit mode which will give X2 a
30% boost and then compare it with the Yonah, and then
wish Yonah had 64 bits?

Yonah is a 32 bit chip and won't be able to run 64 bit
apps. The result is this: for AMD64 users, they can
download any Windows app, 32 bit or 64 bit, and run
them just fine. I am doing this now with my Windows Xp
x64. For Yonah and Sossaman users, what they will
encounter is this: after hours of download, when they
click the program to run, it pops a message saying:
"Sorry, this is not a valid Windows program". Our poor
fellow is going to call support, put on hold for 30
minutes, then get an explanation if they are lucky:
"The app you downloaded was 64 bit and your Yonah32
can't run it". If they are unlucky, the support guy
will have no clue and will ask the poor fellas to
reformat hard drives and reinstall Windows Xp 32.

Shouldn't this be a concern for AnandTech readers?

4:38 PM, December 19, 2005  
Blogger Sharikou, Ph. D said...

Ok, look at AnandTech.com's Yonah review benchmark pages, you see huge INTEL ads. Again, money talks, everyone has a price. I wonder how much INTEL paid in this ad deal.

9:06 PM, December 19, 2005  

Post a Comment

Links to this post:

Create a Link

<< Home