Monday, January 08, 2007

AMD tells Anand to go back to school

Learn how to measure cache latency. Brisbane cache latency is 14 cycles, not 20.

LostCircuits people do have college degeres, and they got the cache latency numbers right.

18 Comments:

Anonymous Anonymous said...

At least Anandtech has the courage to stand up and admit when they're wrong, unlike you, who, when confronted with how wrong you are, use distraction and made up numbers.

The only people with an IQ lower than yours are the people who actually believe what you post and say.

2:24 PM, January 08, 2007  
Anonymous Anonymous said...

You would think that Anand is all wrong by the title of the post. But surprise, surprise, every web review I've visited (Xbit, Tom's, TR, and Anand) have all reviewed the cache latency issue. Even Anand's article had a revision based on AMD representatives' feedback (that would be on page 3) vis a vis the cache latency issue.

The good doctor would like you to believe that Anand has done something underhanded, when in fact his journalistic integrity is well-demonstrated in his corrections/updates. Most importantly, he did not arrive at any different conclusions that any other major web-review site. Those are, in no particular order:

1) A slight decrease in performance in certain applications due to the cache latency.

2) Poor overclocking headroom.

3) Lower power consumption, a plus.

4) Ability to move to future processors that feature a larger cachev - the reason for the cache latency issue in the first place.

What I don't know is....can/will AMD really do that with K8, when K8L should be just around the corner? Or is this just AMD-generated smoke?

Please, will someone post a response to the above question? Preferably someone with some credibility? Speculation is encouraged.

3:25 PM, January 08, 2007  
Blogger Sharikou, Ph. D said...

The good doctor would like you to believe that Anand has done something underhanded, when in fact his journalistic integrity is well-demonstrated in his corrections/updates.

I was questioning his education and IQ.

3:45 PM, January 08, 2007  
Anonymous Anonymous said...

Now Intel has Anandtech on their website too :

http://www.intel.com/core2duo/c2e.htm?cid=cim:ggl|competition_us_amd|k7B33|s

4:27 PM, January 08, 2007  
Blogger Sharikou, Ph. D said...

"The Intel® Core™2 Extreme X6800 didn't lose a single benchmark in our comparison, not a single one."

July 14, 2006 | AnandTech.com


That's only because Anand picked the benchmarks.

4:57 PM, January 08, 2007  
Anonymous Anonymous said...

Anandtech have been exposed for the incompetent idiots they are!Intel are reduced to quoting them to boost Core2Dud!I rest my case!

5:19 PM, January 08, 2007  
Anonymous enumae said...

Sharikou said...

"That's only because Anand picked the benchmarks."

Well does that make the statement false?

Also in the link you provided to Lost Circuits, the people with college degrees, what benchmark aside from memory bandwidth (which on the desktop, would seem irrelevant to performance) did Intel loose?

5:36 PM, January 08, 2007  
Anonymous Anonymous said...

Is this silly Pretnender for real

"That's only because Anand picked the benchmarks."

Count on Sharikou to do the bean counting in 2008 when he claims INTEL is BK..

5:38 PM, January 08, 2007  
Anonymous Anonymous said...

"I was questioning his education and IQ."

Ohhh the irony of this statement. LOL!!!

5:58 PM, January 08, 2007  
Anonymous Anonymous said...

So, did the latency STILL GO UP by more than 15%? Was the performance still degraded over 90nm? Was Anand's performance benchmark data wrong?

What was your latency measurement Mr. PhD?

Specifically why did it go from 12-14? Longer circuit path? Increased resistance? Incomplete dielectric scaling?

Please enlighten us about the 15% delta.

You are a technical joke...the same joke that claimed a multistage explosion was due to the 2 cores in an Intel chip (2 explosions = 2 cores right? You know the Si vaporizing, right?)

The only thing you are capable of is pulling out random links in a vain attempt to suppport a pre-formed conclusion (like every good scientist does, right?). Every time you've tried to do an analysis you show your lack of education - perhaps you will school us on how AMD makes 0mm EE wafers with no street for dicing again?

8:01 PM, January 08, 2007  
Anonymous Anonymous said...

Tell me, Sharikou, why do you spend so much time trying to tear down Anandtech? I mean, you've claimed before that almost every major hardware website and publication in the world is being bribed by Intel, yet you spend a lot of time going specifically after Anandtech. Why are they any worse than anyone else?

1:27 AM, January 09, 2007  
Anonymous Anonymous said...

K8L is 5 months away. 2.8GHz is faster than 2.8 X2 core to core by 70% thanks to L3 cache, HT3 and other changes. There will be no answer for that from Intel side. Expect Conroe clock increase by 20%, 45nm shrink, more cores for less money. But who wants quad-core when dual-core will be faster or single core faster than dual-core?

2:47 AM, January 09, 2007  
Anonymous Anonymous said...

You would think that Anand is all wrong by the title of the post. But surprise, surprise, every web review I've visited (Xbit, Tom's, TR, and Anand) have all reviewed the cache latency issue. Even Anand's article had a revision based on AMD representatives' feedback (that would be on page 3) vis a vis the cache latency issue.

The problem is that without Sharikou bringing this to my attention, I would still believe that there was a massive problem with AMD's 65nm chips.

The first impression is always a lasting one, I applaud sharikou for keeping us up to date.

How often do you return to an article that you have already read?

3:03 AM, January 09, 2007  
Anonymous Anonymous said...

Sharikou said...

"I was questioning his education and IQ. "

Like you have the right to question *anyone's* education when you so blatantly LIE about your own?

6:52 PM, January 09, 2007  
Anonymous Anonymous said...

"Anonymous said...
K8L is 5 months away. 2.8GHz is faster than 2.8 X2 core to core by 70% thanks to L3 cache, HT3 and other changes. There will be no answer for that from Intel side. Expect Conroe clock increase by 20%, 45nm shrink, more cores for less money. But who wants quad-core when dual-core will be faster or single core faster than dual-core?

2:47 AM, January 09, 2007"

Please quit posting under an anonymous ID, Sharidouce - everyone can tell by the broken english that this is you...you only make yourself look worse, if that's possible...

6:55 PM, January 09, 2007  
Anonymous q9ray said...

"Also in the link you provided to Lost Circuits, the people with college degrees, what benchmark aside from memory bandwidth (which on the desktop, would seem irrelevant to performance) did Intel loose?"

You missed the point. LostCircuits reported that the tools used to measure latencies were
"inconsistent & non-reproducible".
LostCircuits found "two cycles
increase latency" of Brisbane's L2.
So that's 12 to 14 cycles but not 12 to 20 as reported by other reviewers.

10:43 AM, January 10, 2007  
Anonymous enumae said...

q9ray said...

"You missed the point...So that's 12 to 14 cycles but not 12 to 20 as reported by other reviewers."

I understand why he posted a link to Lost Circuits.

The context in which I made my comment is refering to Sharikou's claim about Anand's bencmark results for Core 2 Duo and Anand's IQ, it is not related to AMD's latencies differences.

12:23 PM, January 10, 2007  
Blogger N4CR said...

Maybe they will be enlarging the cache on the k8l platform...
interesting. I wonder if K8L is more cache dependent design than current and previous amd designs.

8:33 PM, January 11, 2007  

Post a Comment

Links to this post:

Create a Link

<< Home