Sunday, September 03, 2006

Lack of IQ is a problem in the hardware review profession

You can't write code, you can't design circuit boards, you can't write good prose..., you can't do anything mildly creative, you don't have any in depth understanding of anything, such as an operating system, but you can install and run benchmark programs, what do you do? You become a PC hardware reviewer.

Hexus.net wrote this piece of report, they got really excited seeing an AMD dual core machine streaming two videos.

Retards!

My socket 754 Sempron can stream 100 videos simultaneously, and CPU usage is below 20%. Why? Because Direct Connect Arhictecture offers plenty of bandwidth for I/O and memory access. As long as my network card and hard drive are fast enough. Sending off data over network is a piece of cake.

A dual core AMD Live box is as powerful as a 2P single core Opteron 252 server in CPU cycles, and has 8GB/s I/O bandwidth and 12.8GB/s memory bandwidth. Intel has only a fraction of that.

27 Comments:

Anonymous enumae said...

Sharikou said...

"...what do you do? You become a PC hardware reviewer."

Or your create a blog and bash companies or review sites you don't like...

Just kidding, couldn't resist :)

3:02 PM, September 03, 2006  
Blogger Kalle said...

Too bad they didn't say what streams they used.

Streaming tens to hundreds of videos targteted to 56k is a piece of cake.

Now take couple of 1080p@30FPS and you have a healthy 200MB/s per stream for only filling the framebuffer. Add decompression to that and you can easily get to several times bigger bandwidth requirements.

Of course with H264 you will hit CPU limitation way sooner than memory bandwidth limit :)

3:10 PM, September 03, 2006  
Blogger Kalle said...

Forgot to mention, your Sempron has about 3.2GB/s theoretical memory bandwidth. In real-world you probably hit about 3GB/s peak.

Top-end HDD can deliver average of ~50MB/s of constant reading bandwidth over long time. GLan can give about as much. It takes several HDD's and nic's to hit memory bandwidth limits with streaming data :)

3:13 PM, September 03, 2006  
Blogger Sharikou, Ph. D said...

Streaming tens to hundreds of videos targteted to 56k is a piece of cake.

A Sempron with GbE (1000Mbs) can sustain at leat 100 streams of 6Mbs. Video decoding is done at the client.

3:14 PM, September 03, 2006  
Blogger Sharikou, Ph. D said...

Or your create a blog and bash companies or review sites you don't like...

I start the blog, because I found there are so many retards running big companies and even more retards writing about them and their products.

Look, I make the most accurate predictions with only publicly available data. Why? I have a higher than high school education and enough IQ to apply my knowledge in a scientific and logically sound analysis.

3:17 PM, September 03, 2006  
Blogger Scientia from AMDZone said...

I guess I'm overqualified to be a reviewer then. I can write code, have built systems, setup networks, and administered both Windows NT and Unix systems.

However, I don't think the real problem is the IQ of the reviewers; I think it is Intel's $8 Billion/year marketing budget. I think review sites follow the free hardware and ad money as I've mentioned in my own blog.

3:26 PM, September 03, 2006  
Anonymous Anonymous said...

What about being humble?

3:58 PM, September 03, 2006  
Blogger Kalle said...

Yet another thing I forgot to mention :)

"A dual core AMD Live box is as powerful as a 2P single core Opteron 252 server in CPU cycles, and has 8GB/s I/O bandwidth and 12.8GB/s memory bandwidth"

No. AM2 has rather low memory controller efficiency compared to s939. Sure, dualchannel DDR2 800 can theoretically provide up to 12.8GB/s bandwidth but because of a bit worse efficiency it can provide about 7-8GB/s in theoretical benchmarks. In real-world it is around 6-7 or less.

Of cource that is still more than Intel's ~5GB/s with 1066MHz FSB :)


"A Sempron with GbE (1000Mbs) can sustain at leat 100 streams of 6Mbs"

If I may ask then what will be sitting on the other end of the ethernet cable?

5:26 PM, September 03, 2006  
Anonymous Anonymous said...

Look, I make the most accurate predictions with only publicly available data. Why? I have a higher than high school education and enough IQ to apply my knowledge in a scientific and logically sound analysis.

Since YOU brought it up, no, you do not analyse things in a logically sound way.

You are in the business of having an opinion, then selectively surrounding yourself with data to support your opinion.

I see no hard facts here. You are someone who leeches information from around the net to either pump AMD or discredit Intel.

Unlike other reviewers who generally set aside their biases when they review a product, you do not even 1) review anything, 2) run your own benchmarks, 3) come to the table with a mindset that disregards business tactics when looking at technology.

These people (reviewers and benchmarkers) at least are one step up on the food chain above you. And yet YOU criticise them?

If you wanted anyone to take you seriously, go out and do your own benchmarks. That way at least you know you're not a paid Intel pumper and the information in front of your own face is real...instead of calling reviewers everywhere "paid by Intel" when their reviews come out SCIENTIFICALLY in favor of Intel.

Guess what Sharikou? Sometimes AMD just loses. Grow up, be a man, and admit when your little favorite company is only in second place.

6:04 PM, September 03, 2006  
Blogger Sharikou, Ph. D said...

Unlike other reviewers who generally set aside their biases when they review a product, you do not even 1) review anything, 2) run your own benchmarks, 3) come to the table with a mindset that disregards business tactics when looking at technology.

It's the other way around, dude. Look at those reviewers, they run some benches, and they jump to conclusions. However, if you read the benches, you would come to a different conclusion. For instance, AMD X2 3800+ frags Pentum XE 965 in most tests, you never saw Tom or Anand cheer. However, when Conroe XE 6800+ leads FX62 by 10-15%, all these people, Anand, Toms, Hexus dance like crazy. Why? They smell money from Intel.

These dudes are paid pumpers. They can never be trusted.

You can trust this blog, because it's not tainted with money. I am most interested in getting the right analysis to point to the right direction of IT. I am proud of the accuracy of my predictions.

7:34 PM, September 03, 2006  
Blogger Ajay S. said...

Rudiger at PPCNUX finds Athlon64 faster than core2 when loaded with two different process.

http://www.ppcnux.com/modules.php?name=News&file=article&sid=6552

I think that is a very valid usage scenario and is what differentiates dual core from single core,

Scientia from AMDZone has blogged about this draw back in core2 earlier

http://scientiasblog.blogspot.com/2006/08/dishonesty-of-overclocking.html

7:56 PM, September 03, 2006  
Anonymous Anonymous said...

As somebody who is trained in business, marketing, and accounting...

you are way off about Intel heading towards bankruptcy. There is no way that happens in the forseeable future, sorry.

Even if what you said is true (ceteris paribus, Intel will be bankrupt in 2008), they would be aware of this and make the necessary corrections. They have far, far, far, far more information than you have.

8:04 PM, September 03, 2006  
Anonymous Edward said...

"You can't write code, you can't design circuit boards, you can't write good prose...,"

Haha... if they could have done those, they'd have got a much better and more productive job at some true companies of the industry. ;-)

I agree, that most of the online reviews are paid craps, whether they're Intel-favoring or AMD-favoring.

10:59 PM, September 03, 2006  
Anonymous Anonymous said...

You can trust this blog, because it's not tainted with money. I am most interested in getting the right analysis to point to the right direction of IT. I am proud of the accuracy of my predictions.

How can we trust that you will not corrupt your own analysis of a situation with your obvious bias against Intel then?

Understand it from an outsider's opinion then. You obviously hate Intel and want AMD to succeed because they are the option vs. Intel. And not just a LITTLE bit biased, you outright hate them.

Now, you may actually be right from time to time....but...

How does anyone who visits here and wants a truthful analysis know your bias isn't influencing your whole analysis? The only safe bet is to not listen to you. We'd be better off listening to a paid pumper who has at least a small vested interest in appearing fair.

I'd like it if you responded to this post. It seems like a fair request, no?

6:52 AM, September 04, 2006  
Anonymous Green Nose said...

There is a lack of IQ in EVERY profession, not just hardware review.

You could go on to say that there is a lack of IQ in the human race.

But then you'd get a bunch of PC comments... the kind of PC no one wants.

You'd be telling the truth. But no one wants to hear the truth out in copper top land.

And there is no greater mind slave copper top than someone who is PC... can you hear me out there, you god forsaken PC morons?

7:34 AM, September 04, 2006  
Blogger S said...

Sharikou, if you believe Hardware reviewers are retards, why do you quote them so very often ? Especially when the review contains comments favourable to AMD ? How come the reviewers are retards when what they say doesn't match up to your conclusions ?

You start with a bias and look around for data which supports your biased views. There isn't much intelligence in you either.

8:14 AM, September 04, 2006  
Blogger Sharikou, Ph. D said...

You start with a bias and look around for data which supports your biased views.

No. I just didn't turn the bad stuff and ugly stuff of Intel into good things.

Look, Intel is laying off hundreds of Israelies, people who saved Intel, at least for a short while. Intel is a ruthless and cruel company, it inherited a lot of stuff from the Nazi and Soviets. Intel's quota system in workforce elimination is real. Expect the Israelis to retaliate.

As for technology, Intel is anti-progress. It has become an AMD64 clone maker and a bad one of it.

Why do we need Intel?

The BK of Intel will benefit the whole IT industry.

9:46 AM, September 04, 2006  
Blogger Sharikou, Ph. D said...

Sharikou, if you believe Hardware reviewers are retards, why do you quote them so very often ? Especially when the review contains comments favourable to AMD ?

I don't even read reviewers' conclusions. Most of them are incapable of correctly interpret benchmark data--reveiwers are just software installer and runners, they know little. I just read their data and draw my conclusions.

I have pointed out long time ago that Conroe has a 10% IPC lead in integer and lags behind K8 in floating point, using the scarce but reliable data I had. ScienceMark results from independent individual test and Cinebench result from Intel's VP. All those Anand, Hexus, Toms results were mostly crap, especially the way they make their analysis.

9:56 AM, September 04, 2006  
Anonymous Anonymous said...

You start with a bias and look around for data which supports your biased views.

No. I just didn't turn the bad stuff and ugly stuff of Intel into good things.

Look, Intel is laying off hundreds of Israelies, people who saved Intel, at least for a short while. Intel is a ruthless and cruel company, it inherited a lot of stuff from the Nazi and Soviets. Intel's quota system in workforce elimination is real. Expect the Israelis to retaliate.

As for technology, Intel is anti-progress. It has become an AMD64 clone maker and a bad one of it.

Why do we need Intel?

The BK of Intel will benefit the whole IT industry.


That's just the thing...you base, say, your interpretation of Conroe on the business practices of Intel.

How is that intelligent? The world may be a better place without Intel, but you CAN'T deny that Conroe performs well.

Your opinions of a product offering are thereby tainted with your opinion of a corporation, regardless the quality of the product.

I see no point in accepting any of your opinions, since they may or may not be tarnished by your bias. However, other people's opinions in this blog (for the most part) are not obviously tainted by bias and deserve to be heard.

If you can intelligently debate my viewpoint, please do so. However, you've already established your motives and I don't believe that you could capably argue with me on this point.

1:34 PM, September 04, 2006  
Blogger Sharikou, Ph. D said...

you CAN'T deny that Conroe performs well.

Conroe is OK for a desktop chip, but its lack of bandwidth is the limitation of an aging architecture. Conroe has no future. It will ship for a few million units, then it will be totally obsolete as K8L roll out.

8:39 PM, September 04, 2006  
Anonymous Anonymous said...

The BK of Intel will benefit the whole IT industry.

How exactly will a monopoly controlled by AMD (or anyone else for that matter) "benefit the whole IT industry" in any way?

3:17 AM, September 05, 2006  
Blogger S said...

Conroe is OK for a desktop chip, but its lack of bandwidth is the limitation of an aging architecture. Conroe has no future. It will ship for a few million units, then it will be totally obsolete as K8L roll out.

Conroe lacks bandwidth compared even to AMDs current products. Isn't that correct ? But it more than makes up for it in other aspects of design, thus outperforming all of AMDs current products (at least in 90% of industry opinion). What will K8L bring to the table other than extra bandwidth ? How useful will the extra bandwith be when interfaced with other devices in a PC ? Having a high bandwidth itself may not be of much use unless there are designs to take advantage of the bandwidth. How can you be sure the next gen Intel products will not have features which will offset AMDs bandwidth advantage ?

Please backup your statements with some facts.

Remember Intel will be one generation ahead in manufacturing process for atleast another 6-7 quarters. AMD is also much behind Intel in platformization and is trying hard to catch up

7:04 AM, September 05, 2006  
Anonymous Anonymous said...

Intel is a ruthless and cruel company, it inherited a lot of stuff from the Nazi and Soviets.

Oh mighty scientific and logical analyzer- how about a source for this statement? Seems like more of your cherished inflammatory rhetoric. I could see someone making a very weak argument that AMD, with the bulk of its manufacturing in the former communist East Germany, is benefiting from the Nazis and Soviets, but how so Intel? Because Grove was a refugee from the Nazis and Soviets?

Enlighten me!

10:04 AM, September 05, 2006  
Anonymous Edward said...

"Conroe lacks bandwidth compared even to AMDs current products. Isn't that correct ? But it more than makes up for it in other aspects of design, thus outperforming all of AMDs current products (at least in 90% of industry opinion)"

That's why he said Conroe is an okay desktop chip, because desktop applications require less bandwidth than server ones.

When your application is bandwidth bottlenecked, nothing can make that up. The more high-end and heavy-loading you go, the more bandwidth your applications will take.

Those CPU benchmarks, however, do not measure bandwidth purposely. They measure how well CPU process data assuming bandwidth was not an issue. That's why they always include a test of memory bandwidth as well - because they left this part out in other tests.

12:24 PM, September 05, 2006  
Anonymous Edward said...

"GLan can give about as much. It takes several HDD's and nic's to hit memory bandwidth limits with streaming data :) "

You'll need more than Gb/s bandwidth if you want to do anything useful on the GbE data.

The CPU reads in the data, process them, then write them out back to memory. One simple pass takes twice the bandwidth of the raw data. SATA disk bandwidth is much more than 50MB/s for sequential reads; that's what SATA's 1.5Gbps and 3.0Gbps are meant for.

Things multiply if there are more than one GbE NIC and RAID stiping - typical for streaming servers.

We know that K8 is a 3-issue core; lets conservatively assume only 1 is useful, and that only processes two 32-bit numbers. That means 8 bytes per clock cycle, and dual-core makes that 16 bytes per cycle; at a relatively slow 2.0GHz it's 32GB/s processing power. As we said above, each byte of processing requires two bytes of memory bandwidth, which now is 64GB/s.

Note that we conservatively assumed a one issue per cycle, 2.0GHz dual-core processor. Our conservative estimate has reached Intel's theoretically maximum bandwidth.

If the data to process are 64-bit, the bandwidth doubles. If they are 128-bit SSE, the bandwidth doubles again. One cannot but wonder how is Intel's fast SSE implementation useful (other than for benchmarking) if the processor's memory bandwidth can't support 1/4 of the data required?

The fact is, Core 2 Duo relies on large cache to make its 4-issue and fast SSE slightly more useful, but it is not a good processor for video streaming, network processing, and network server.

12:58 PM, September 05, 2006  
Anonymous george said...

want to join rubyworks formz you need to join. it is a fourm founded by madmodmike and you just nead to advertize it on your blog and join.

9:45 AM, September 09, 2006  
Anonymous george said...

so for supercomputing amd has two advantages
A: hypertransport
B: improved floating point

9:50 AM, September 09, 2006  

Post a Comment

Links to this post:

Create a Link

<< Home