The rise and fall of AMD

[Exophase]

I'd like to know where this idea that Vishera does much better on Linux comes from. This is taking into account the scheduling patches made available for Windows 7 and incorporated into Windows 8. It's not taking into consideration what compiler was used - note that you can also get ICC for Linux and far from everything on Windows uses it.

When I look at the Phoronix scores I see something largely similar to what I see on Windows benches - wins on a few heavily threaded benches, loses hard on weakly threaded benches. Which is what you'd expect taking everything about the design into consideration. What Phoronix has is not just a different OS but several different sorts of benchmarks, many of which won't matter for many people. It's up to the reader to determine what has the most relevance to them and go forward with that.

Kind of weird to see complaints about Cinebench when it's one of the better showings for Vishera.. and AMD themselves have been using it to promote Jaguar vs IB..

 

[galego]

That's true -- I don't need to exaggerate, because it really is that low:

Based in NetMarket estimates, which disagree with other estimates of clients. In any case look the close than linux and W8 are.

Why do you dismiss one but not the other?

 

[frozentundra123456]

Conspiracy theory? No! Just facts. It is well-known that AMD chips tend to behave better under linux than under windows.

You explicitly mentioned "tests in anands bench" and I simply add the fact that those are only windows bench. Linux benchs say otherwise.

I wonder why Anand tests under windows?

Could it be because the vast majority of uses operate under that system? Naw, it must be because he hates AMD and wants to show them in the worst possible light.

 

[galego]

I find your attitude is highly unprofessional and strikes of out and out hate for AMD. Now if you will really get satisfaction if AMD stops fielding competitive products to Intel just to spite anonymous people on the Internet, that speaks for itself.

+1

 

[Charles Kozierok]

Nine posts in the span of an hour. Like I said, if he's not being paid, he should be.

Don't really see any point in continuing this as long as it's being spammed by puerile nonsense from someone obviously lacking even a speck of objectivity.

 

[AnandThenMan]

Not just process advantage. R&D advantage. Capital advantage. Fab advantage. Management advantage. Server market position advantage (a very big issue not discussed often enough, and IMO, one of the real reasons AMD has gotten in so much financial trouble.)

And given all of that, somehow AMD is still able to keep themselves within spitting distance of Intel's performance. But I would imagine your completely objective and factual opinion is this happens because Intel is purposely not trying very hard.

If you thought about it for even a few seconds, you'd quickly realize that AMD is orders of magnitude more efficient than Intel, and squeezes every last drop out of their available R&D. If you doubt this, remember the computer you are using right now is executing AMD64 instructions. All of Intel's resources and superiority, and they were unable to bring x86 into the 64 bit age, not to mention they lagged for years using ancient interconnects, all the while telling people we don't even need 64 bit computing on the desktop.

I don't agree. I actually think the degree to which AMD is being led has been masked both by the general stagnation in improvement rates over the last few generations, and by Intel's change of focus to the mobile space and power efficiency. If anything, I think Intel has now fishtailed too much away from the P4 era and is not putting out enough parts that emphasize performance over "performance per watt", and that is part of why their lead over AMD isn't as evident.

Another way of saying, if Intel wanted to they could have a much bigger performance lead. That's your opinion, plus it doesn't matter what matters is products we can buy.

I express no vitriol to people who honestly present benchmarks. I have a big problem with people who use intellectually dishonest tactics, like ignoring a 2013 market share report and constantly repeating a link to a 2012 report. Or constantly trying to claim that I should care about how well AMD does in benchmarks that have nothing to do with real world use.

Benchmarking is an extremely volatile and variable subject. There are no cold hard facts, it is a field full of landmines, caveats, exceptions, misrepresentations, cheating, optimizations, hidden vender ID checks etc. etc. it never ends. But for some reason you seem to think that you are the only one that knows exactly what is considered honest benchmarking.

People like that are not honestly discussing a subject with a different viewpoint. They are pushing an agenda.

That's your opinion. Some reading your posts could be forgiven in thinking you have an unhealthy hatred for AMD and anyone that defends them.

Nine posts in the span of an hour. Like I said, if he's not being paid, he should be.

Are you accusing him of something here? Either come out and say it, or don't go there.

Don't really see any point in continuing this as long as it's being spammed by puerile nonsense from someone obviously lacking even a speck of objectivity.

And again that's your opinion, which I'm sure you want others to respect, so show the same respect back.

 

[nehalem256]

And given all of that, somehow AMD is still able to keep themselves within spitting distance of Intel's performance. But I would imagine your completely objective and factual opinion is this happens because Intel is purposely not trying very hard.

If you thought about it for even a few seconds, you'd quickly realize that AMD is orders of magnitude more efficient than Intel, and squeezes every last drop out of their available R&D. If you doubt this, remember the computer you are using right now is executing AMD64 instructions. All of Intel's resources and superiority, and they were unable to bring x86 into the 64 bit age, not to mention they lagged for years using ancient interconnects, all the while telling people we don't even need 64 bit computing on the desktop.

Its not that they couldn't, it that they didn't want to. They wanted to push the Itanium architecture.

 

[Phynaz]

Then why are reviewers running lots and lots of benchmarks on Windows 8 which has only about a 3% of worldwide share?

Please stop derailing my thread and start your own.

Thank you.

 

[AnandThenMan]

Its not that they couldn't, it that they didn't want to. They wanted to push the Itanium architecture.

Agreed, which IMO was an extremely ill conceived plan (I should have said unwilling more than unable, although that is debatable IMO). The only reason Intel got away with it is because they are able to hold a gun to the head of anyone licensing x86 from them. Anything AMD comes up with, Intel can use and copy without the slightest downside, this is quite unbelievable actually. AMD no longer pays Intel a licensing fee, but for the longest time they did.

 

[CHADBOGA]

Agreed, which IMO was an extremely ill conceived plan (I should have said unwilling more than unable, although that is debatable IMO).

Not by anyone who understands anything about CPU's.

The only reason Intel got away with it is because they are able to hold a gun to the head of anyone licensing x86 from them. Anything AMD comes up with, Intel can use and copy without the slightest downside, this is quite unbelievable actually.

Anything Intel comes up with, AMD can use and copy without the slightest downside, is this quite unbelievable??

 

[Abwx]

I'd like to know where this idea that Vishera does much better on Linux comes from.

I wonder why Anand tests under windows?

Could it be because the vast majority of uses operate under that system? Naw, it must be because he hates AMD and wants to show them in the worst possible light.

Actualy he used Linux and compared it to windows with
a Magny Cours CPU and the results did show Intel in such
bad light that he didnt renew the experience with BD and PD
despite saying that he will publish the numbers later when
he reviewed those CPUs...

http://www.anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon/7

 

[NTMBK]

Actualy he used Linux and compared it to windows with
a Magny Cours CPU and the results did show Intel in such
bad light that he didnt renew the experience with BD and PD
despite saying that he will publish the numbers later when
he reviewed those CPUs...

Considering that that was an alpha (i.e. totally unfinished) version of the software, I'm not surprised that it had big performance deltas across platforms...

Also, it makes sense to do Linux in server chip reviews, as a very large number of servers run Linux. Very few desktops do.

 

[galego]

As an well-informed poster suggested before, the fastest supercomputers in the world (Jaguar and Titan) use cheap AMD Opterons and a free OS as linux because are in a "budget".

Only really really serious people use Intel and Windows 7.

Considering that that was an alpha (i.e. totally unfinished) version of the software, I'm not surprised that it had big performance deltas across platforms...

Except final version shows the same.

http://www.reddit.com/r/hardware/comments/184kcj/amd_fx8350_or_i7_3770k/

Also, it makes sense to do Linux in server chip reviews, as a very large number of servers run Linux. Very few desktops do.

By your own logic why are using W8 in desktop chips reviews? Very few desktops do?

 

[ShintaiDK]

As an well-informed poster suggested before, the fastest supercomputers in the world (Jaguar and Titan) use cheap AMD Opterons and a free OS as linux because are in a "budget".

Only really really serious people use Intel and Windows 7.

Yet Cray got tired of their 3 year locked contract with AMD when Dulldozer failed. So as soon as it expired this happend:

http://investors.cray.com/phoenix.zhtml?c=98390&p=irol-newsArticle&ID=1789449
http://www.engadget.com/2012/11/08/cray-launches-100-petaflop-xc30-supercomputer/

 

[sontin]

As an well-informed poster suggested before, the fastest supercomputers in the world (Jaguar and Titan) use cheap AMD Opterons and a free OS as linux because are in a "budget".

Only really really serious people use Intel and Windows 7.

Titan uses AMD because Cray had a contact with them. The next gen system comes with Intel CPUs and support for Kepler and Phi.

Cray kicked AMD out of their portfolio.

 

[frozentundra123456]

As an well-informed poster suggested before, the fastest supercomputers in the world (Jaguar and Titan) use cheap AMD Opterons and a free OS as linux because are in a "budget".

Only really really serious people use Intel and Windows 7.



Except final version shows the same.

http://www.reddit.com/r/hardware/comments/184kcj/amd_fx8350_or_i7_3770k/



By your own logic why are using W8 in desktop chips reviews? Very few desktops do?

Because win 8 is the logical next successor to win 7, and is also in the vast majority of computers being sold today.

 

[NTMBK]

By your own logic why are using W8 in desktop chips reviews? Very few desktops do?

The vast majority of new desktops use it. And that is where reviews of new chips matters.

 

[mrmt]

And given all of that, somehow AMD is still able to keep themselves within spitting distance of Intel's performance. But I would imagine your completely objective and factual opinion is this happens because Intel is purposely not trying very hard.

What do you call spitting distance? Just because they sell two processors for a similar price with comparable performance you can say they are at spitting distance? Well, let me tell you, they aren't. They need more power, more die area to get to Intel's performance levels. This inefficiency is reflected in their margins. While Intel can withstand close to 60% margins, AMD is below 40%.

When we remove the power and die size constraints, when we are able to see the full power of both architectures, AMD simply cannot reach Intel's performance levels, or LP levels. Just look at IVB 13W TDP chips, or a SNB-EP chip.

AMD no longer pays Intel a licensing fee, but for the longest time they did.

That's fair, isn't it? AMD got access to Intel IP, could copy Intel fancy instructions just by replicating the code in clean room without having to pay someone to think on developing them, just in replicating them, and on top of that got protection from competitors because Intel would protect the market with their IP lawyers.

 

[Charles Kozierok]

When we remove the power and die size constraints, when we are able to see the full power of both architectures, AMD simply cannot reach Intel's performance levels, or LP levels. Just look at IVB 13W TDP chips, or a SNB-EP chip.

Or, as I've said, look at the server market, because that's the place where the real performance crunch is -- speed matters, efficiency matters, cost matters. And AMD is getting obliterated in the server market. They peaked at over 20% in the mid-2000s, and are now below 5%.

The server/workstation market is Intel's cash cow -- it's a constant stream of revenue and income that powers the entire company. It is not the desktop that is the real problem for AMD right now -- it is servers (in addition to mobile).

That's fair, isn't it? AMD got access to Intel IP, could copy Intel fancy instructions just by replicating the code in clean room without having to pay someone to think on developing them, just in replicating them, and on top of that got protection from competitors because Intel would protect the market with

Another good point that gets lost in all the "David versus Goliath" cheerleading around here.

I think competition is important, and I don't want to see AMD disappear. But at the same time, innovation is also important, and Intel has been responsible for nearly all the innovation in the x86 world over the last 30 years. AMD gave us x86-64, which was important, and they deserve credit for that. (ETA: AMD gets credit for the first integrated memory controllers, too.) But that was ten years ago -- literally, the first 64-bit Opteron was released April 2003 -- and what else has AMD really done that's been innovative since then?

The companies that do the R&D, that invent new technologies, that lead the way to greater efficiency -- they are more deserving of success than copycats, who arguably are only still in the running at all because of the Sherman Act.

 

[dastral]

and what else has AMD really done that's been innovative since then?

Fusion and we'll see the first real use with PS4, A10 using DDR3 is probably a preview to the future.

 

[Abwx]

and Intel has been responsible for nearly all the innovation in the x86 world over the last 30 years.

Which are thoses "innovations"?.And please dont come with ISAs...

Their whole history was to prevent any competitor from
innovating in the X86 market for fear that they would
no more hold a tight grip over it...

 

[Enigmoid]

Fusion and we'll see the first real use with PS4, A10 using DDR3 is probably a preview to the future.

I like fusion but intel has beat amd to market with both of those things. Arrandale had an igp on chip (separate die) that was perfectly fine for office work and 90% of the stuff people do on a laptop. It was released in january 2010 a year before llano. Even sandy bridge was out before llano. Likewise haswell will support a unified memory address, releasing before kaveri. Further thing to consider is that intel had NO gpu IP to leverage to build these chips. Of course they were handily outperformed by llano but you can hardly call their efforts no innovation.

 

[Charles Kozierok]

Fusion and we'll see the first real use with PS4, A10 using DDR3 is probably a preview to the future.

What was the big innovation in Fusion? They certainly weren't the first IGPs.

Which are thoses "innovations"?.And please dont come with ISAs...

Let's see, off the top of my head:
- Invented the microprocessor.
- First commercial DRAM IC.
- Created the entire x86 ISA infrastructure.
- Superscalar processing.
- Clock multiplying.
- Integrated L2 cache.
- L2 cache running at CPU speed.
- OOO execution.
- multicore chips.
- turbo mode.
- integrated graphics.
- temperature sensors.
- numerous RAS features.
- SIMD instructions.
- Numerous other ISA improvements (whether you like it or not).
- hyperthreading.
- instruction fusion.
- countless branch prediction improvements.


How about outside CPUs?

- Northbridge/southbridge chipset architecture?
- High-speed nb/sb links?
- The entire line of USB standards?
- PCI?
- The ATX form factor?
- First mainstream SSDs?

I could find dozens more if I wanted to spend the time.

And what has AMD done that's really been innovative? Not much. They mostly just copy Intel.

Their whole history was to prevent any competitor from
innovating in the X86 market for fear that they would
no more hold a tight grip over it...

This is a joke. How did they "prevent competitors from innovating"? By being better than they were?

 

[Exophase]

To the people who responded about Linux vs Windows performance: showing that a particular program performs much better in Linux than Windows is not evidence that the OS itself is responsible for the performance delta. Let me reiterate:

1) You can use GCC on Windows, many programs do
2) You can use ICC on Linux, no idea if anything does
3) You can run the Phoronix test-suite on Windows and use the same compiler versions and flags
4) A patch was made for Windows 7 to improve the scheduling for BD devices (this was a long time ago, back in late 2011), Windows 8 never had the problem
5) Linux does penalize some code on BD (up to 3% penalty) because address space randomization aggravates icache evictions due to VIPT aliasing (64KB / 4KB > 2 ways). The "fix" was to make it not randomize bits 12:15 or maybe 13:15. Linus was very opposed to this so I doubt he let this into mainline. I don't know if Windows randomizes aggressively enough to have this problem. You can turn off ASLR on both. Intel CPUs don't have this problem because 32 / 4 = 8 ways.

If you can run a large set of the Phoronix test suite using the same source, same compiler, same flags, and still show Linux has a tangible average performance gain then that'll mean something.

 

[galego]

My dear friends I am really fond of your enthusiasm.

Supercomputer makers such as Cray chose AMD over Intel because AMD chips were ahead on performance during the last ten years. Offering at least 40% more computing performance than Xeons. Jaguar and now Titan (#1 supercomputer) use AMD chips... Cray "loves" AMD's interlagos architecture.

Intel has moved piece and did buy part of Cray by $140 Million, including 74 Cray engineers... 100000 Xeon chips will be used in the new XC, but AMD Opterons will continue to be used in the XE models. It seems that the Titan will be upgraded again but now to the new AMD 6300 chips.

About W8. Are you aware that most people only buy W8 because is pre-installed and that next they downgrade (I would say upgrade to W7? Are you aware of people who update their computers and don't get W8 true? Are you aware of people that build a new rig and buys a W7 license true?

Still waiting for a reliable explanation...