AMD CPU speculation... and expert conjecture

[truegenius]

^ i see what you are trying there

 

[gamerk316]

Are edits working? Because I tried to edit and it wouldn't take. Don't think the thread is stable yet...

I'm assuming we're running into a post count limit?

 

[truegenius]

edits are not working (atleast for me)
so now again , we can post, we can delete but we can't edit

 

[gamerk316]

Fair enough.

Here's a must read for those who want a better understanding of how "things" work internally:

http://danluu.com/new-cpu-features/

 

[juanrga]

K10 arch was at its limits and, lacking any better idea, K11 engineers decided to use the old MCMT architecture invented in 1999 by Andy Glew.

which makes me more angry.
why did they wasted precious money and time just because of the fact that they didn't had any new idea. they could have used this wasted time and money to do more experiments to modular design to make is beat sandy clock for clock for real. and could have released real 32nm k10 with those improvements discussed above.
engineers at amd they must be using intel pc that is why they didn't spotted the uncore angle which would have got 5-10% better performance ( even a mere +5% is bettar than -20% of bd) for general softwares and even better in games.

This must give some clarification,

Bulldozer was on the drawing board (people were even working on it) even back when I was there. All I can say is that by the time you see silicon for sale, it will be a lot less impressive, both in its own terms and when compared to what Intel will be offering. (Because I have no faith AMD knows how to actually design chips anymore). I don’t really want to reveal what I know about Bulldozer from my time at AMD.

BTW, you ask how AMD could have competed? Well, for one thing, the could have leveraged K8 and the K8 team’s success and design techniques instead of wasting years of time on a project that eventually got cancelled using people that had never achieved any success. It took Intel years to come out with Nehalem, and AMD could have been so far ahead by that point that they’d have enough money in the bank that they wouldn’t have to accept a low-ball settlement offer in the antitrust suit and they wouldn’t have to sell off their fabs.

The question is did management learn from the lessons of the pass? Seeing how the company is being run in late years and the decisions are being taken about roadmap and strategy I doubt.

 

[jdwii]

Something weird is going on people continue to talk about previous products from Amd and claiming that's what will make them stay around. The point is a company could be successful 15 years ago but suck today. For example Sony was doing quite a bit better back in the 90's and 80's then now. Not that they suck but its an example.

The reason why Amd won back in the day in terms of making a worthwhile design is over them being able to actually fund their research department. Its just not going to happen without the money required to pay for engineers. Sure they have Jim but he has to deal with the constants of time and money as well.

Does this mean they are dead or does that mean they need to find themselves and make a product they can sell in mass quantities. I'm not going to pretend to know something about their future products and the most we can do is pretend to know something.

Anyways they are quiet, people keep leaving a CEO dropped out of thin air. If it wasn't for the consoles i'm not sure they would be around.

 

[Reepca]

Personally I'm still on the lookout for the Berlin Server APUs. In servers it makes a lot more sense (to me at least) to use an APU for GPGPU computing, due to having to process a large amount of data in a relatively short amount of time, and neither of those requirements sound like PCI-E. Also because it's my excuse for dreaming big ("sure, true-to-geometry server-side collision-detection/physics processing sounds like a lot, but it's a massively parallel HSA server!").

What ever happened to those? The latest I can find information about those seems to be early last year. Did they go the way of the fx-7600p?

 

[palladin9479]

Fair enough.

Here's a must read for those who want a better understanding of how "things" work internally:

http://danluu.com/new-cpu-features/

Amazing link, I believe everyone here should read this. For those of us who already know this, it's a good refresher and reminds us that not everyone is familiar with these concepts, for those who don't know these things it will enable them to better understand the "why" about processor design's and performance metrics.

 

[gamerk316]

Yeah, as mainly a software guy, it's nice to have an easy to understand refresher handy.

 

[juanrga]

Something weird is going on people continue to talk about previous products from Amd and claiming that's what will make them stay around. The point is a company could be successful 15 years ago but suck today. For example Sony was doing quite a bit better back in the 90's and 80's then now. Not that they suck but its an example.

The reason why Amd won back in the day in terms of making a worthwhile design is over them being able to actually fund their research department. Its just not going to happen without the money required to pay for engineers. Sure they have Jim but he has to deal with the constants of time and money as well.

Does this mean they are dead or does that mean they need to find themselves and make a product they can sell in mass quantities. I'm not going to pretend to know something about their future products and the most we can do is pretend to know something.

Anyways they are quiet, people keep leaving a CEO dropped out of thin air. If it wasn't for the consoles i'm not sure they would be around.

Excellent analysis. This is a good summary:

http://seekingalpha.com/article/2386725-one-simple-reason-why-amd-cant-compete

• ■ AMD is very good at creating hype, but at the same time it's been slashing its R&D spending.
  ■ Intel vastly outspends AMD in R&D, and toward the end of last year Nvidia surpassed AMD as well.
  ■ In the long-term, there is no way that AMD can remain competitive given these R&D spending gaps.

Sometimes seeing is believing

Not shown above but in 2006--2007, AMD did spend about the half than Intel.

AMD Zen could be competitive in one or two metrics, if AMD plays ok its cards, but nobody can really believe that AMD will do a comeback and that Zen performance will superior to Skylake. The Zen core will be less powerful (I think will be a kind of Puma core on steroids) and the only possibility to outperform is if we compare a quad-core Skylake to a Zen chip with moar cores (there are rumors about 20-core FX CPUs where each core is "very small").

 

[juanrga]

Personally I'm still on the lookout for the Berlin Server APUs. In servers it makes a lot more sense (to me at least) to use an APU for GPGPU computing, due to having to process a large amount of data in a relatively short amount of time, and neither of those requirements sound like PCI-E. Also because it's my excuse for dreaming big ("sure, true-to-geometry server-side collision-detection/physics processing sounds like a lot, but it's a massively parallel HSA server!").

What ever happened to those? The latest I can find information about those seems to be early last year. Did they go the way of the fx-7600p?

Berlin APUs vanished in the air the same like the Seattle SoCs. Either got canceled or it was only a paperlaunch to hype. In any case the server head manager abandoned the boat before Rory did.

 

[-Fran-]

Fair enough.

Here's a must read for those who want a better understanding of how "things" work internally:

http://danluu.com/new-cpu-features/

Amazing link, I believe everyone here should read this. For those of us who already know this, it's a good refresher and reminds us that not everyone is familiar with these concepts, for those who don't know these things it will enable them to better understand the "why" about processor design's and performance metrics.

It indeed refreshed my memory as well.

Yeah, as mainly a software guy, it's nice to have an easy to understand refresher handy.

Thanks again for the link, gamerk. It saved me from reading about memory management, haha.

Cheers!

 

[8350rocks]

YAY!

BTW: Juan, please stop this crap...it is bad enough you troll me in this thread...I do not need you trolling my PM inbox:

To: 8350rocks

From: juanrga
Sent on: January 8, 2015 9:26 PM
I am sending you a private message because the thread is broken again and cannot post.

My question is when will you share with us that Excavator introduces no change in FP unit and continues being a weak 8 FLOPS/core (confirming my prediction and disproving what you said us before), and that the integer IPC gains will be compensated by further reduction in base frequencies (probably sub 3 GHz)?

This is ridiculous, even the Denver core for tablets achieves 16 FLOPS/core. The dual core K1 produce about the same FP throughput than a quad-core excavator at same clocks...
From: 8350rocks
Sent on: January 11, 2015 11:17 PM
I will not say something that flies in the face of what AMD are telling me...as I understand it, clocks should be higher, IPC higher, and an improved FPU.

I thought you had an "inside source" for this stuff...you should know what is coming then...?
From: juanrga
Sent on: January 12, 2015 10:43 AM
I was giving you the possibility to retract in public from your comments and misleading information, but I see you continue the same.

I hope you are enjoining how your beloved Byrne has been fired from AMD :-D
From: 8350rocks
Sent on: January 13, 2015 5:10 AM
Good thing he is not the only one I know there then...huh? He was just the highest ranking person I personally knew...
From: juanrga
Sent on: January 13, 2015 8:03 AM
Yeah but I suppose that he cannot do all that miraculous things that you said us he was going to do at AMD :-D

I did just read your ad hominen. It is so hilarious that doesn't even deserve a reply.
From: 8350rocks
Sent on: January 13, 2015 8:09 AM
Juan, you have no technical qualifications at all...you are literally a forum warrior...
From: juanrga
Sent on: January 13, 2015 9:26 AM
And with all your 'qualifications' still I am right 90% of time whereas you only 10%... :-D
From: 8350rocks
Sent on: January 14, 2015 5:13 AM
Juan...you are wrong more often than you admit.

Look at your blustering about Carrizo, when I said all along it would be mobile only.

There are many other examples.
From: juanrga
Sent on: January 14, 2015 10:13 AM
You never admit you are wrong. We all know that.

You said that Carrizo was coming only to mobile and I corrected you because I knew was coming to desktops as well. Now AMD has announced that will use Carrizo on certain desktops, but you can continue trying to hide your error.

And of course all your claims about Zen beating Skylake will be soon shown to be wrong.

I wonder whom you think you are fooling. Nobody in the thread trust you anymore!
From: 8350rocks
Sent on: January 14, 2015 10:59 PM
I said specifically..."Carrizo is ULP only" (basically mobile aside from AM1 type setups for very low demands)

Close, but no cigar Juan. Go back and quote me where I said Carrizo was ULP only please...so you can show you are wrong.

Sure, you might have some desktop parts, but they will be bargain bin garbage for email and other stuff...you cannot possibly play any PC games on them aside from maybe something like angry birds.
From: juanrga
Sent on: January 15, 2015 3:57 AM
You continue beating a dead horse. Everyone knows what you said and how you were wrong (as usual) and that is why I received congrats in public by my prediction. I wonder whom you believe you are fooling by sending me this garbage.

People knows you have been spreading misinformation and hype during years. Nobody trust you anymore. Stop posting nonsense about AMD leaving Nvidia in the dust for ten years. Stop pretending that Zen will be faster than Skylake. Just stop.

 

[gamerk316]

Welcome back everyone!

 

[jdwii]

^^^^ It's over them thinking they can make a X86 CPU for android. Still waiting for the day Intel gives up and just makes Arm CPU's for android. I mean they have the money and the engineers but unfortunately they also have a bias.

I assure you its not over their laptop CPU sales

 

[juanrga]

This is my explanation of AMD recent strategy for Carrizo.

1) AMD is stuck with 28nm node.
2) To compete with 14nm Broadwell on efficiency they have to reduce clocks. Whereas performance increases with frequency is about linear, power increases as cube frequency: P ~ f^3
3) Architectural changes (probably at cache level) will increase the IPC and will about compensate the clocks reduction.
4) Now Excavator doesn't scale well above 3GHz and anything above 35W is canceled because a 65W Carrizo @3GHz would be slower than 95W Kaveri @3.7GHz.

This is more or less the history. Integer cores, FPU, all is about the same than Steamroller.

Apparently all the efforts are put in the iGPU side of Carrrizo. It seems will be more advanced than Tonga or any other discrete GPU.

 

[juanrga]

intel already makes cpu's running android ? But it is not just tablets, laptops and every other ulv cpu has been bleeding money. Not to hard to receive design wins when you just about give away your product for free. forget about android it will stay in mobile until Microsoft releases a arm version of windows, if ever. But i see no need going forward both intel and amd have driven tdp down to a acceptable point to offer full windows comparability on mobile devices. Given a choice most users will take windows over android.

The loses are from the incentives to tablet OEMs. The goal of the incentive program was to ships 50M x86 tablets or something like that.

 

[noob2222]

I can see nothing has changed. one person sais something, another person takes that and make the assumption of comparing the top end to the lowest available and tries to defend his rantings.

this forum thread is still ran by one ego that thrives on misinformation and twisting the facts.

just fyi, a CMT runs at 80% efficiency when both cores of a single module are loaded. so at 80% efficiency, even a quad to "quad" comparison, which one is truly faster when they both score the same?

who knows, maybe ill check back in a couple of months and see if AMD has anything worth talking about. in the meantime, enjoy the hatred one person has towards AMD and everything they can't do because they don't spend as much money as Intel does on fab R&D. oh, that's right, AMD doesn't even have a fab budget, while Intel is spending 90% of their budget trying to shrink their way into the mobile market.

I guess juan was right about "ARM will win" since it never launched.

@intel's mobile market loss, that's what happens when your cpus are given away for free, aka marketing, aka repeat of Intel's good monopolistic strategy to stop AMD from getting design wins with bema/mullins.

peace out.

 

[juanrga]

just fyi, a CMT runs at 80% efficiency when both cores of a single module are loaded. so at 80% efficiency, even a quad to "quad" comparison, which one is truly faster when they both score the same?

Precisely Steamroller eliminated the 20% module penalty by duplicating the decoder. However, the penalty is of 50% in anything from Bulldozer to Excavator when running the FPU in fused mode, because when one core is accessing the fused FPU, then the other core in the same module cannot access to the FPU.

 

[noob2222]

^^ stop the lies

Piledriver’s single-threaded performance in CB11.5 is actually a bit better than Bulldozer’s clock-for-clock, as we’ll see, but the chip still takes a significant multi-threading penalty when we run all four threads across just two modules. Microsoft’s scheduler improvements don’t do much to help here.

http://www.extremetech.com/computing/138394-amds-fx-8350-analyzed-does-piledriver-deliver-where-bulldozer-fell-short/2

cya.

 

[juanrga]

^^^^ It's over them thinking they can make a X86 CPU for android. Still waiting for the day Intel gives up and just makes Arm CPU's for android. I mean they have the money and the engineers but unfortunately they also have a bias.

Some analysts want Intel to give up x86 and do ARM CPUs, instead trying to get x86 in the mobile market.

 

[jdwii]

intel already makes cpu's running android ? But it is not just tablets, laptops and every other ulv cpu has been bleeding money. Not to hard to receive design wins when you just about give away your product for free. forget about android it will stay in mobile until Microsoft releases a arm version of windows, if ever. But i see no need going forward both intel and amd have driven tdp down to a acceptable point to offer full windows comparability on mobile devices. Given a choice most users will take windows over android.

No body wants a windows tablet today. X86 just doesn't matter in those markets. Maybe it will change but i doubt it based on the evidence of apple and samsung and other major companies being able to control how the architecture is being made instead of being forced down to what Intel or Amd does.

 

[juanrga]

I guess juan was right about "ARM will win" since it never launched.

The full phrase was "ARM will win in the long-run". It doesn't have anything to do with AMD's Seattle.

Seattle was AMD's attempt to cover its mistake of coming late to the server party. K12 will be ready by 2016--2017 and AMD decided to launch Seattle with A57 cores in 2014 for taking part of the market. The problem is that other companies did release custom cores and now nobody want Seattle when there are better alternatives. That is probably why we don't hear anything about Seattle anymore.

Moreover, I doubt that AMD will play a leading role on ARM. Several server makers are using APM hardware. Cray joined Cavium to build future ARM-based supercomputers. Broadcomm has the fastest known ARM CPU (more powerful than Haswell Xeons) and Qualcomm will probably dominate the microserver market. What is left to AMD?

 

[Cazalan]

2) To compete with 14nm Broadwell on efficiency they have to reduce clocks. Whereas performance increases with frequency is about linear, power increases as cube frequency: P ~ f^3

Where do you come up with this pseudo math?

P = C*F*V^2

http://en.wikipedia.org/wiki/Frequency_scaling

Power remains linear with frequency. The reason the power starts going crazy is they have to bump the core voltage to remain stable.

 

[jdwii]

^^^ Yes and as stated its over them caring about tablets and smartphones. I'm not sure what your point is