Multitasking, physical core vs virtual cores

[Scali]

It is a 20-40% gain in specific scenarios that might not be important to most people. I am not so sure that the 5% die increase couldn't be used in a better way for a more general performance increase, e.g. I'd easily take a 3% increase in everything over 30% in encoding. Maybe it really can't and HT is well worth, but I've not been convinced so far, and for sure I don't see it as a game changer.

I think it's pretty obvious that AMD has consistently failed to improve their core performance since the K8, and SMT is probably the 'low hanging fruit' at this point. 20-40% gain is a lot more than AMD has managed in the past 5-6 years of development.

And there was some discussion here with people arguing that large HT gains might be due to apps not being well optimized. I can see how SMT can be used to e.g. hide memory latencies, but scheduling more CPU-bound threads than physical cores on a SMT enabled CPU will probably do more damage than good, even if you will see a gain compared to running the same number of threads with SMT off.

Not if you understand how a modern x86 works.
For example, the Core2 and later have 4 ALUs per core. However, because of the x86 instructionset and the decoding complexity, it is physically impossible to execute more than 3 instructions per cycle on average. In practice, most of the time, you should be happy with 2 instructions per cycle.
So that means 1 or 2 ALUs are pretty much always idle. And then I'm not even getting into the fact that there are a lot more units than just ALUs in the CPU (FPU/SSE/LSU etc).

It's physically impossible to feed x86 code that uses ALL your execution resources. Most of the time, more than half of them are sitting idle.
In fact, with a bit of clever application design, you can get almost perfect results from HT... for example, if you run one thread that only uses the ALU, and another that only uses SSE, they can pretty much coexist on a single physical core without any compromise.

x86 is just the most horrible instructionset ever. It's like trying to fly a space shuttle with a set of tweezers.

 

[Nemesis 1]

Denial in its maximum expression. Give it a rest man, in things like Virtualization, stuff that uses FPU, Encryption, AMD platform is quite competitive, specially in multi CPU platforms, HyperTransport scales better than QPI.

Wrong quote

 

[Nemesis 1]

Then sir, you are a troll, as viral as you can get. We need competition, get that in mind if there's any left. AMD is quite competitive in the server market and midrange/low end market. I don't think you can buy a brand new Intel Quad Core processor under $99, Athlon II X4 620, the 635 model offers the 70% of performance of it higher end sibling, the Phenom II X4 955. AMD isn't a holy company, specially with Hector "Satan" Ruiz. I'm certainly sure that if the roles reversed and AMD was the one who had the crown, the same thing would happen, like in the Athlon 64 era. Paying almost $300 for an Athlon 64 3200. . .

PS. My gf is enjoying her Athlon II X4 635 with a decent Biostar motherboard and 4GB of RAM for less than $235, I gave her a HD 3870 and can max her games at 1680x1050, currently playing Lost Planet in DX10 with such resolution and 4x FSAA, min: 28 Max:50 Ave: 38, not groundbreaking but enough for her.

Someone is trolling here . But Who.

Yes it is . So when ya going to stop denial. Price is what you pay for . Intel >AMD by alot in almost all things but price. I not sure when enthusiast sites became The dollar store but it sure as hell has occurred.

Their are millions of cars in America Most can run on our highways. But to run at daytona a high performance car is required or you will cause hugh problems for other high performance cars if your not up to speed . The only way lowperformance cars can run daytona is with other low performance machines or they cause bottlenicks and crashes.

AMD doesn't belong on the same track as Intel and thats simple fact. Go ahead buy what you want. But stop with the AMD is as good as Intel because that is a Flat out LIE. .

 

[jaqie]

....wow. seriously. people are comparing the consumer x4 and x6 to the subserver and server quality intels and saying they are better... uhm, no duh? Has everyone forgotten about magny-cours?

 

[Scali]

....wow. seriously. people are comparing the consumer x4 and x6 to the subserver and server quality intels and saying they are better... uhm, no duh?

The only thing keeping Intel from releasing 6-cores in the same consumer market is AMD.
Intel doesn't need anything more than their 4-cores to compete with AMD's 6-cores, and that says enough already.

Has everyone forgotten about magny-cours?

What about it? It's just 12 Athlon64s lumped together. Why would any enthusiast get excited about it? It's about as ridiculous as AMD's 'quadfather' platform a few years ago.

 

[Nemesis 1]

None forgot about MC. But that only a 4s system . Intels new systems are 8S that equals 48 real cores for AMD and 48 real cores for Intel . Its an asskicking . Performance for performance. Intel has covered all the bases except the one were AMD starts giving away Cpus in ceral boxes. Intel does have 8 core cpus ya know.

 

[Voo]

There we have the one-dimensional thinkers again.
So you're saying that there are only ultra-cheap cars and Ferrari's? No middle ground? I hope you realize how stupid that sounds.

No I only paraphrased what you said, because money is very much an important decision factor, even for people who work and not just poor students, that's exactly why the "middle ground" is so popular - ahm good "performance" for a reasonable price.

Most people only have certain demands on their hardware and if you know those it's rather reasonable to get the best price/performance CPU that fulfills those, which in most cases probably is not the fastest CPU out there.
Intel surely has the technological lead but they demand a nice premium for it - though Amd better finds another way to compete with them then just more cores, because most consumers atm just don't need 12 cores or even 6 at the moment..

 

[JFAMD]

Oh god... don't tell me you don't even know that Intel has special instructions for encryption that AMD doesn't have...
Luckily Anand covered it: http://www.anandtech.com/show/2901/5
"...or at least AES-NI will probably keep AMD out of the running for consideration."
Or how about a synthetic test to show the REAL power of those AES instructions?
http://www.anandtech.com/Show/Index...gny-cours-opteron-6174-vs-intel-s-6-core-xeon
AMD is NOWHERE in that test.
"Once the Xeon X5670's AES instructions can do their work, encryption is lightening fast. Here the new Xeon is 19 times faster than its older brother and 9 times faster than the best Opteron."

As for the other stuff: yea right, dream on. I recall a blog post of... I believe it was Ryan Smith, talking about potential performance problems with virtualization on Nehalem because of a disadvantage in the TLB caching on paper. But once they benchmarked actual hardware, Nehalem showed no weaknesses.
FPU? Yea right... Not that anyone cares anyway, that's what GPUs are for.
HyperTransport scales better than QPI? In what alternative reality is that?

Just more of the same rubbish as you had been spouting about HyperThreading earlier.

The problem with the AES-NI istructions are that they show up well in benchmarks but actual applications need to be recompiled to support them. Security vendors are generally slow to recompile applications because each new version requires significantly more review and debug than normal software because of the potential for introducing vulnerabilities.

There aren't that many applications that are supporting the instructions just yet, and fewer people taking advantage of them because they have changed their software out (again, enterprise customers take security VERY seriously; secure is far more important than fast.)

Over the next year this situation will change, but don't expect that there was a wholesale jump to AES-NI upon the westmere launch.

Wouldn't it be better to compare Int and Floating Point if this was a pure processor to processor discussion? Magny Cours beats Westmere in both of those areas. And does it at a lower price.

 

[Scali]

The problem with the AES-NI istructions are that they show up well in benchmarks but actual applications need to be recompiled to support them.

The problem with you is that both your nickname and your signature made me stop reading after this line

 

[JFAMD]

None forgot about MC. But that only a 4s system . Intels new systems are 8S that equals 48 real cores for AMD and 48 real cores for Intel . Its an asskicking . Performance for performance. Intel has covered all the bases except the one were AMD starts giving away Cpus in ceral boxes. Intel does have 8 core cpus ya know.

8P x86 equals .12% of the market in 2009 and shrunk 10% YoY from 2008. That is about as much of a niche as you can find.

48 cores for AMD has a top end of ~$5.4K. 8 top bin 75xx processors will set you back more than $29.5K.

No matter how you slice it, it becomes hard to justify the cost of an x86 8P when there is such a huge delta to the cost of a 4P.

8P is a tiny part of the market and the delta in price is not going to help intel grow a market that is tiny and in decline. Most of the apps in that space are going to be competing head to head with RISC and there is a lot more lacking from any x86 platform relative to RISC. The easy stuff moved to x86 years ago when Linux made that transition possible. Anything remaining in RISC at that level probably isn't going anywhere because of a ton of reasons, none of which could be impacted by an x86 value proposition.

So the 7500 sits in a bad position. Too expensive to make a stand in x86, too underfeatured to take on Itanium or RISC. And fighting hard at a shrinking market that represents 1/8 of one percentage point of the x86 market. Good luck with that.

 

[Scali]

No I only paraphrased what you said

No, you made it one-dimensional.
I never said that you should go for the Extreme Edition CPUs. I have never bought such a CPU in my life, and probably never will.
I buy the "middle ground" (say around i860-i870 range currently), and AMD can barely cling on there with their 6-core CPU, which I would never buy, because I value better single-threaded performance and lower power consumption.

 

[taltamir]

No matter how you slice it, it becomes hard to justify the cost of an x86 8P when there is such a huge delta to the cost of a 4P.

8P is such a niche market BECAUSE there is such a huge price delta.
I really don't see why though.

 

[JFAMD]

8P is such a niche market BECAUSE there is such a huge price delta.
I really don't see why though.

Either do we. That is why, with our latest products, we removed the 4P tax. 2P and 4P processors cost the same now. No price premium because it is the same silicon. I have yet to meet a customer that thinks this is a bad idea. Nobody wants to pay more money than they need to, everyone has budget constraints.

 

[taltamir]

Either do we. That is why, with our latest products, we removed the 4P tax. 2P and 4P processors cost the same now. No price premium because it is the same silicon. I have yet to meet a customer that thinks this is a bad idea. Nobody wants to pay more money than they need to, everyone has budget constraints.

Yes. plus you could save power and space by using 4P...
But what about 8P? doesn't this still have an 8P tax?

 

[Scali]

Either do we. That is why, with our latest products, we removed the 4P tax. 2P and 4P processors cost the same now. No price premium because it is the same silicon. I have yet to meet a customer that thinks this is a bad idea. Nobody wants to pay more money than they need to, everyone has budget constraints.

Everyone knows it's the same silicon... it's just historically been a way for companies to get a quicker return on investment.
It sounds like AMD admits that they really do need to drastically lower prices, because their technology doesn't have anything else going for it.
Big question is: where will the profit be coming from?

 

[evolucion8]

Someone is trolling here . But Who.

Yes it is . So when ya going to stop denial. Price is what you pay for . Intel >AMD by alot in almost all things but price. I not sure when enthusiast sites became The dollar store but it sure as hell has occurred.

Their are millions of cars in America Most can run on our highways. But to run at daytona a high performance car is required or you will cause hugh problems for other high performance cars if your not up to speed . The only way lowperformance cars can run daytona is with other low performance machines or they cause bottlenicks and crashes.

AMD doesn't belong on the same track as Intel and thats simple fact. Go ahead buy what you want. But stop with the AMD is as good as Intel because that is a Flat out LIE. .

You go away man, you are the worst of all and I'm not gonna waste my time arguing with someone so biased for Intel and that speaks an extraterrestrial english that's very hard to understand. I was having a nice conversation with someone who knows what's talking about, you don't know a thing. So I will just ignore you.

Wouldn't it be better to compare Int and Floating Point if this was a pure processor to processor discussion? Magny Cours beats Westmere in both of those areas. And does it at a lower price.

You are right, we may say as well that AMD processors are faster in optimized SSE4A software, something that Intel doesn't have, but it would be heavily one sided and plain stupid. And this topic is about general multi core performance.

The problem with you is that both your nickname and your signature made me stop reading after this line

Wow, every time that you attack someone, it makes you look smarter, keep going...

 

[Scali]

Wow, every time that you attack someone, it makes you look smarter, keep going...

So you think we should actually take AMD employees who 'magically' appear in this thread to 'defend' AMD's products seriously?
(Or whatever other company for that matter, but somehow I've only ever seen AMD people do it... not counting nVidia 'focus group' members, they're not really nVidia employees).

 

[taltamir]

So you think we should actually take AMD employees who 'magically' appear in this thread to 'defend' AMD's products seriously?

As long as he makes legitimate claims and points, yes you can.
Its not like he is a shill, pretending to be unaffiliated with a company while secretly being paid to mislead people in favor of the company paying him.
He is an AMD employee, he admits to be one, and he has inside knowledge of what is going on. I have no problem talking to him cordially and accepting any valid claim he makes just because it was he who made it.

PS. the way you say it sounds so dehumanizing. S/He might work for AMD but S/He is just a human being; a person.

BTW, is it He or She?

 

[Scali]

As long as he makes legitimate claims and points, yes you can.

I could, in theory...
But his first response to me already showed his lack of knowledge (not understanding the difference between 'more efficient' and 'faster'), or worse, deliberately trying to mislead people, so I'm not going to bother to entertain him. He's not worth my time.

 

[evolucion8]

I could, in theory...
But his first response to me already showed his lack of knowledge (not understanding the difference between 'more efficient' and 'faster'), or worse, deliberately trying to mislead people, so I'm not going to bother to entertain him. He's not worth my time.

And why then you don't work for AMD? Or Intel if you know more than him/her? Its quite easy to accuse someone that has lack of knowledge, but you had done anything to prove him/her wrong. You are just pointing fingers and running backwards without evidence of your claims...

So you think we should actually take AMD employees who 'magically' appear in this thread to 'defend' AMD's products seriously?
(Or whatever other company for that matter, but somehow I've only ever seen AMD people do it... not counting nVidia 'focus group' members, they're not really nVidia employees).

We could say the same for nVidia's focus group members. And yet they can provide some nice tips for nVidia users, after all, all are humans, not robots engineered to say whatever it was programmed to say.

 

[Scali]

Its quite easy to accuse someone that has lack of knowledge

I am not accusing. I established that fact, given his response to my post, and his lack of understanding of the material. He was actually accusing me of being wrong first (which I wasn't, obviously... you have to be REALLY good to catch me on a mistake. And I hate nothing more than people who think they're smarter than me, try to 'correct' me, while it's actually them who are the clueless ones).
You could say it was his first impression that spoilt his chances.

 

[Voo]

No, you made it one-dimensional.
I never said that you should go for the Extreme Edition CPUs. I have never bought such a CPU in my life, and probably never will.
I buy the "middle ground" (say around i860-i870 range currently), and AMD can barely cling on there with their 6-core CPU, which I would never buy, because I value better single-threaded performance and lower power consumption.

Ah, so saying "money is only a concern for poor students, etc." is now the same as "no I only buy middle ground because it has the best price/performance and fulfills my needs"? Interesting interpretation, suddenly money is a concern, well hard to argue otherwise.

 

[taltamir]

I could, in theory...
But his first response to me already showed his lack of knowledge (not understanding the difference between 'more efficient' and 'faster')

just to play the devils advocate, this could be a case of miscommunication rather then lack of knowledge. He could have misunderstood you.
I don't recall the specific instance you are referring to, so I can't tell how obvious what you meant was.

 

[Scali]

Ah, so saying "money is only a concern for poor students, etc." is now the same as "no I only buy middle ground because it has the best price/performance and fulfills my needs"? Interesting interpretation, suddenly money is a concern, well hard to argue otherwise.

Well, since this is AMD vs Intel, you have to put it in that context.
AMD's most expensive processor is a LOT cheaper than Intel's.
So if you are arguing in favour of AMD, and giving price as an argument, you are probably saying you should buy somewhere in the region of $50-$200 (assuming you're not arguing about AMD's most expensive CPU alone).
That still leaves a HUGE gap with Intel's most expensive CPUs, which are over $1000.
So we are talking about an $800 gap that you are just ignoring?

No, obviously I said something like "Well, $200... $300, $400, not such a big deal to most people", not necessarily stretching it out to the maximum of $1000+ CPUs, as this makes no sense whatsoever, except to AMD fanboys in search of fallacies.

I personally *could* afford the high-end stuff, but like most people, I can spend money only once. I don't like to blow it all on the absolute fastest stuff, I think the challenge is to spend the money wisely. Then I save money which I can spend on other, more important stuff.
I generally buy in the $400-$600 range. Very close to the high-end in performance (well, most of the time, currently the 980X is exceptionally fast), while only costing about half. I'll close the gap with a bit of overclocking and that's that. As I said, I'd probably go for an i860 or i870 currently. Seems to be the 'sweet spot' for me.

 

[Scali]

just to play the devils advocate, this could be a case of miscommunication rather then lack of knowledge. He could have misunderstood you.
I don't recall the specific instance you are referring to, so I can't tell how obvious what you meant was.

Here is the exact post:
http://forums.anandtech.com/showpost.php?p=30072262&postcount=21

The very first word he has ever uttered to me is 'wrong'. Obviously that doesn't go down well.
The rest of his post just demonstrates how he misunderstood what I said, and how he thinks he knows better, while he is actually coming to the same conclusion as I did (or well, Amdahl did, I'm just the messenger here): "It all comes down to how much of the application can be done in parallel versus in serial."
If you take that one step further: "In the case of a perfectly parallellizable application, you get the same efficiency from N cores at M GHz as you do from one core at (N*M) GHz. With any less than perfect parallelism, the serial version is the more efficient one". Which is exactly what I said to begin with.