Friday, June 02, 2006

Global Crown Capital Equity Research's David Wu needs to have some IQ

According to INQ, David Wu at Global Crown Capital Equity Research (GCRE) wrote that

We wereperplexed by AMD's performance-per-watt leadership claim as 1) AMD deliberately compared its Opteron to Paxville (soon-to-be 2 generations old) and avoided Intel's extremely competitive Woodcrest processor that will be launched in a few weeks; 2) AMD again decreased Opteron's total system power consumption by ignoring North Bridge components and increased Woodcrest's by doubling the number of memory slots (16 FB-DIMMs instead of a typical 8 FB-DIMMs for DP). According to trusted 3rd party reviews, Woodcrest DP system consumes roughly 20% less power and outperforms at least 20% at the same time.

Such low IQ analyst is extremely annoying. First, David Wu said AMD avoided Woodcrest, then David Wu said AMD doubled Woodcrest memory slots. David Wu's brain coherence can't be sustained within one sentence. A stupid analyst is harmful even to the con-analyst profession.

Marty Seyer's presentation slides are here. The webcast video is here. On slide 7, he compared the currently shipping processors, the Xeon Paxville and Opteron. Then, he said since Intel is hyping Woodcrest, he would take the oppurtunity to set the record straight. On slide 8, Seyer compared the upcoming chips, the Intel Woodcrest and AMD Santa Rosa Opteron (Socket 1207).

Seyer's estimate of Woodcrest is based on the following

1) The Blackford north bridge chipset in Woodcrest consumes a minium of 22 watts. In AMD system, the north bridge is inside the CPU. Both AMD and Intel have similar south bridge chips for I/O.

2) Each 1GB FB-DIMM (for Woodcrest) consumes 11 watts. Each 1GB DDR2 DIMM (for Opteron) consumes 4.4 watts. Seyer used 16 GB memory for the 2P server for the estimate. If you look at typical Bensley systems, most of them have 16 FB-DIMM slots, they can hold up to 64GB. During the Q&A, Seyer was asked about the average memory size. Seyer said if you only fill half the slots (8), the Opteron would still consume 15% less power (Woodcrest: 144+22+ 8*11 = 254, Opteron: 190+ 8* 4.4 = 225, then add 28% for conversion loss). It obvious that for Woodcrest to consume less power, the system must have less than 4GB of memory.

3) Seyer further pointed out AMD's TDP spec is for max power. Intel's spec is for typical power.

4) David Wu talked about independent 3rd party tests on Woodcrest. So far, all Woodcrest tests are done in Intel's lab with Intel configured system under Intel's watch. We exposed Intel's fraud previously.

I hope this set the record straight.

Some analyst actions:

Sumit Dhanda of B of A, rates INTC at $25, dump below $20, buying AMD like crazy.

Joe Osha of Merrill Lynch, rates INTC at "buy", Joe Osha says "we are buyers". Merrill Lynch dumped large portion of its INTC holdings at huge loss. Joe Osha rates AMD at "sell", Merill Lynch buying AMD like crazy.

30 Comments:

Blogger "Mad Mod" Mike said...

It get real annoying hearing Intel Fanboys saying "Intel announcing typical TDP's is going away with Conroe" and I think "Did Paulie tell you this?" I really wonder why people think NGMA is a revolutionary change for Intel and some how they are now this great, honest and truly caring company....it's a little disturbing if you think about it.

Woodcrest 80w and Conroe 65w are still typical, because if it was MAX, you know without a shadow of a doubt that Intel would be touting this out loud, and they'yre not; so i doubt it's true. AMD will maintain Performance Per Watt forever in the Server market, that is a fact that Intel fanboys should realise.

On the mobile front, the Turion consumes far less at Idle and typical environments than the Pentium M, so unless a mobile user in an office is playing Doom 3, than AMD also has the PPW crown for mobile.

For Desktops, AMD has 89w MAX for a 2.8GHz Dual Core....that is freakin' insane...and it solidifies AMD's Leadership in All-Platform PPW king.

6:28 PM, June 02, 2006  
Blogger "Mad Mod" Mike said...

If you really look at TDP's, AMD has had the lead and continues forever for 1 huge reason: They have to account for the Memory Controller. I would bet their MC consumes 20% of their TDP, so if you really look at CPU vs. CPU, Intel is 40% or more higher than AMD on TDP for just CPU.

7:03 PM, June 02, 2006  
Anonymous Anonymous said...

APM, heart of AMD, and no other companies can copy it.

10:51 PM, June 02, 2006  
Anonymous Anonymous said...

"We have the best x86 execution engine today, and we will have the best one next year," said Dirk Meyer, AMD’s chief operating officer. "We have a lot of great engineers and they haven’t been sleeping," he added.

This quotation explains it all.

11:00 PM, June 02, 2006  
Blogger Sharikou, Ph. D. said...

have to account for the Memory Controller. I would bet their MC consumes 20% of their TDP

Yes. The Woodcrest's MCH, the Blackford chipset is a quite complex piece of crap just to do 2P. Two independent 1333MHZ buses, and I bet there is a whole bunch of cache inside it to buffer traffic. Also, keep in mind, this chipset is still being made on ancient 130nm technology. So it must be very power hungry.

12:35 AM, June 03, 2006  
Anonymous Anonymous said...

AMD's presentation of the wattage comparison between the Opteron and the Woodcrest appears promising -BUT-

There was nothing said about a performance comparison.

The Woodcrests performance is significantly better than the Opteron and will more than make up for the wattage difference.

6:52 AM, June 03, 2006  
Blogger Sharikou, Ph. D. said...

The Woodcrests performance is significantly better than the Opteron and will more than make up for the wattage difference.


Most benchmarks indicate that Woodcrest is 10% slower than Opteron at the same clock speed (scroll down this blog and read). Now, Woodcrest't top model will be 3GHZ, the fastest Opteron officially launched is 2.6GHZ. The 3GHZ Woodcrest has 5% lead over 2.6GHZ Opteron. However, if you google "Opteron 890", you find it has been out there for a long time. The 890 is a 2.8GHZ chip. Furthermore, AMD is launching the Rev F Socket 1207 Opteron this month, before Woodcrest.

You also have to understand that Woodcrest architecture is just an imitation of the anicent Athlon MP. There is NO WAY for Woodcrest to do MP. Woodcrest is limited to 2P ONLY. Opteron can go glueless 8P. So there is a capability difference there.

8:46 AM, June 03, 2006  
Anonymous Anonymous said...

"Opteron can go glueless 8P"

LoL. That's right, "glueless 8P". Thanks god noone was stupid enough to try 8P with current generation opterons.

10:36 AM, June 03, 2006  
Anonymous Anonymous said...

Anandtech's view
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2768&p=1

10:37 AM, June 03, 2006  
Anonymous Anonymous said...

The IT industry do their home work,
these Intel bumps compared to Amd's
concerted design are nothing more than a meaningless nonsence in a an effort to stave of the collapse of Intel.

maybe we will be talking about Nvidia buying Intel..No their not stupid.!

1:48 PM, June 03, 2006  
Anonymous Anonymous said...

hmm you're a bit talking, near the question here, it was will woodcrest outperform the opteron, i think it will on some points (depending on application) if you believe some synthetic benches even xeon was better then opteron, but they won't have the performance/power crown they hope. this is very important in future... not only there tdp rating is a joke but the use of fbdimm and nb will add lots off additional power usage look at the blades from HP for example... 5 out of 6 types are amd. i hope socket F will bring some power decrease and additional performance. i think it will since most opteron benches are done on cas3 ecc, so the move to ddr2 is less cas dependent like the desktop k8

if you look at performance difference of conroe and k8 (where conroe is better....i wonder what went wrong with woodcrest performance "hasty job??"

Intel is still strong because marketing behind is strong and people are still in the mind of the days that the only decent proc seller was intel "pentium" lol

2:23 PM, June 03, 2006  
Anonymous Anonymous said...

Get over it dudes.. AMD can promise anything they want.. But by end of june, they are underdogged and overrated until they prove otherwise..

Now, now.. Be a sport

2:47 PM, June 03, 2006  
Anonymous Anonymous said...

"You also have to understand that Woodcrest architecture is just an imitation of the anicent Athlon MP. There is NO WAY for Woodcrest to do MP. Woodcrest is limited to 2P ONLY. Opteron can go glueless 8P. So there is a capability difference there."

And don't forget K8L (or RevG), which can do 32P gluelessly.

4:43 PM, June 03, 2006  
Anonymous Anonymous said...

Check this out
I would reserve judgement till I see the 65nm AMD parts with updated compiler.

6:02 PM, June 03, 2006  
Anonymous Anonymous said...

"Such comprehension-impaired analyst is extremely annoying. First, David Wu said AMD avoided Woodcrest, then David Wu said AMD doubled Woodcrest memory slots. I prefer to read Joe Osha's con-man type analyst report than this kind of stupid crap. A con-analyst is at least useful to their bosses, a stupid analyst is harmful to the con-analyst profession."

And you Shakira (can't be bothered to spell your name correctly) have proven yourself to be the perfect example of integrity? That makes me laugh hard. The pot calling the kettle black. Dude you suck. When it turns out that Conroe chips are way faster than AMDs best desktop chips will you eat crow and issue mea culpas? I won't hold my breath waithing for that.

6:06 PM, June 03, 2006  
Anonymous Anonymous said...

The Techreport benchmarks show that Woodcrest is faster than the Opteron by 15 - 20 percent. Not sure where you're getting 5 percent.

6:24 PM, June 03, 2006  
Anonymous Anonymous said...

Xeon MP is a joke vs AMD64 Direct Connect architecture!

7:08 PM, June 03, 2006  
Anonymous Anonymous said...

The question is what part of the market is 4p? if Woodcrests is ahead on single or 2p where most of the volume is than Intel will be ahead

8:48 PM, June 03, 2006  
Blogger Sharikou, Ph. D. said...

That's right, "glueless 8P". Thanks god noone was stupid enough to try 8P with current generation opterons.

Google "Galaxy X4600", you find SUN's 8P 16 core Opteron. IWill, Fabric 7, Tyan also offer 8P 16 core Opteron servers. No chipset needed, just connect the wires of the ccHT links, you are done. Dude, that's innovation, that's leading edge.

9:36 PM, June 03, 2006  
Blogger Sharikou, Ph. D. said...

When it turns out that Conroe chips are way faster than AMDs best desktop chips will you eat crow and issue mea culpas?

If Conroe was indeed good, why does Intel have to resort to massive global fraud to hype it. Why just give Anand a chip for a drill? Why did Intel cheat when comparing benchmarks. All verifiable benchmarks show CORE2 slower than K8 clock/clock.

9:39 PM, June 03, 2006  
Anonymous Anonymous said...

What these analysts are saying is that they belive in Intel and all these biased benchmarks. The market is prepared for huge performance gain and the question is what is going to happen - Conroe benchmarks are invalid - all of them, so we don't know what is the truth. The only question is what the customers are going to say, are they going to flee from Opterons?
Even if Woodcrest IS 10% faster, it doesn't mean that Santa Rosa is going to loose anything. AMD should speed up ramping 65nm and launch Operons first not Athlons. On the desktop side it is enough to lower the prices.

11:11 PM, June 03, 2006  
Anonymous Anonymous said...

AMD 65nm production in July/Aug, shipping in Oct/Nov
According to one analyst anyway. If true, then final version of engineer samples should be out soon.

12:19 AM, June 04, 2006  
Anonymous Anonymous said...

All verifiable benchmarks show CORE2 slower than K8 clock/clock.

please explain the use of clock/clock comparisson, when the processor designs differ that much? do you also compare ibm power with amd on a clock/clock base?

the clock rate of a processor doesn't matter for users (amd tells us that for several years now). what's important ist the performance of the chip, and the efficiency of energy use.

12:54 AM, June 04, 2006  
Anonymous Anonymous said...

fortunately the industry listens to the likes of Cray, Sun, Microsoft, Vmware...the list goes on, AMD has the platform for all of them to grow the x86-64 business, anyone who buys intel shares based upon those lie's without doing their home work deserves them.

1:13 AM, June 04, 2006  
Blogger Sharikou, Ph. D. said...

please explain the use of clock/clock comparisson, when the processor designs differ that much?

Clock for clock comparison is comparing IPC. With IPC determined, we can the determine performance by clock speed. 3GHZ will be Woodcrest's top speed.

3:16 AM, June 04, 2006  
Anonymous Anonymous said...

"Clock for clock comparison is comparing IPC. With IPC determined, we can the determine performance by clock speed."

Gosh, you must be a huge Itanium fan if you care about IPC so much.

Honestly, why do you corner yourself in this IPC discussion when all that matters today is performance/watt?

9:20 AM, June 04, 2006  
Blogger Sharikou, Ph. D. said...

Gosh, you must be a huge Itanium fan if you care about IPC so much.


Dude, you have no clue. I will let others to educate you more. Put it briefly, IPC, clockspeed, bandwidth, power consumption are the four key parameters that determines the characteristics of an architecture.

1) Itanium: High IPC, low clockspeed, high wattage, low bandwidth, high cost--low performance hot expensive crap.

2) Pentium 4: low IPC, high clock, low bandwidth,high wattage -- performance better than Itanium(that's why Itanium didn't fly even before Opteron emerged), but too hot and run into clockspeed limit.

3)K8: good IPC, good clocks (3GHZ), high bandwidth, low wattage, low cost(no north bridge needed).

4) CORE2: IPC at 10% lower than K8, clocks close to K8 (maybe slightly higher due to longer pipeline), low bandwidth, low wattage (we are so told).

9:50 AM, June 04, 2006  
Anonymous Anonymous said...

"4) CORE2: IPC at 10% lower than K8, clocks close to K8 (maybe slightly higher due to longer pipeline)"

Sharikou... I'd differ. Conroe wouldn't have a lower IPC if it's a 4-issue machine, versus K8's 3-issue. I actually think Conroe's IPC is 15% higher than K8.

As for clockrate, I don't think these two are much different. Intel has process advantage, but its use of non-SOI makes its chips less stable @higher clockrates. So if one is lucky, he probably can overclock Conroe more than K8; but eventually K8 might reach a higher stable clockrate than Conroe. Just my guess, could be wrong though.

12:01 AM, June 05, 2006  
Anonymous Anonymous said...

Hey! What happened to my 40%?!?

http://www.tomshardware.com/2006/06/05/first_benchmarks_conroe_vs_fx-62/

2:07 PM, June 05, 2006  
Blogger "Mad Mod" Mike said...

IIRC, Intel uses a form of SoI on their CPU's, but I'm not sure to what extent. The real test will come at 45nm for Intel, because electron leaking at 45nm will increase significantly and the stability of their CPU's at high clock rates will be diminished, especially as they start adding cores and more cache.

3:19 PM, June 05, 2006  

Post a Comment

<< Home