AMD's Feldman says ARM processors have some advantages over x86 chips

[sontin]

I had a little time and read the summary of an AMD rep who talked about ARM:
http://semiaccurate.com/2013/05/15/amds-andrew-feldman-talks-about-arm/

It's a little bit shocking that AMD thinks that a standard 64bit ARM core is much better than Bulldozer to take market share from Intel...

 

[MisterMac]

Well that's some serious flirting if there ever was some.

I do wonder how they're going to convince huge unit customers to re-purpose all their tech\codepaths to ARM and get them using it - at a cost advantage to x86.

...while still also making money themselves.

But it'll be interesting nonetheless.

 

[inf64]

Please provide exact quote of what was being said. Performance wise what you stated is of course not true. But Feldman said:

For server workloads Mr. Feldman believes that ARM processors have an advantage over x86 chips in terms of power consumption, cost, and die space. AMD’s value proposition in the market is two-fold; first off it’s the only ARM vendor with experience in traditional servers and AMD’s off-the-shelf IP blocks will give them an advantage over the competition which may not be able to match IP portfolio that AMD has built for itself over the years. There are of course trade-offs in choosing one instruction set over the other, but AMD really sees the additive IP blocks that they can offer on their server chips as their biggest advantage over the other ARM licensees. To quote Mr. Feldman directly, “This isn’t religion, this is instruction sets.”

So the advantage is power draw, cost and die area. Performance is not an advantage and probably won't be in near future. AMD is covering the HPC segment with SR/EX cores in next few years. Beyond that it's possible we might even see some mix of x86/ARM/GPU all on the same die.

 

[R0H1T]

This is actually quite believable because AMD has dug themselves a grave with their rather inefficient x86 chips, as compared to Intel, so they have a second chance with ARM. Just looking at the title does however tell me that AMD is referring to efficiency here & not pure & absolute performance !

 

[zebrax2]

I've read the article but nowhere does it say that the ARM chips is better than their current x86 proc., more like the needs of servers are shifting and ARM is a better fit for them to adapt to that shift simply because it is cheaper to develop (specially with their current financial status)

 

[Hitman928]

You're title is misleading as it isn't an actual quote from AMD. Also, the point of the article is that in a segment of the server space, it makes more sense to make semi-custom chips specifically for the needs of the server and that the server market is shifting this direction. AMD feels that using ARM in this space makes a lot of sense as you can "take an order" and produce a semi-custom product tailored to the customer's needs a lot faster and cheaper with ARM. Obviously we'll have to see if they're right.

 

[SlowSpyder]

Misleading anti-AMD OP by the sontin... I'm shocked, totally shocked I tell you.

 

[RoarTiger]

Where does he say "ARM is better than our own x86 processors". I only found a few quotes in the whole article and none of them came close to saying that. Is this just a shit troll thread OP?

Summary of the Article:
AMD believe in ARM SERVER market and have invested in it heavily for various reasons explained the the article.

The only quotes in the article are these:
“This isn’t religion, this is instruction sets.”
“This [ARM based server chips] is the final stake in the heart of the one-size-fits-all world of just using different bins of the same chip.”
“Intel lost every handset on Earth because ARM was better than Atom.”

What has happened to this place when a SemiAccurateLOL news article has less crap and more truth in it than a thread on this forum.

 

[mrmt]

It's a little bit shocking that AMD thinks that a standard 64bit ARM core is much better than Bulldozer to take market share from Intel...

Shocking? Bulldozer wiped out AMD market share on servers. They become a blip on the market, not even competitive in web servers and other low-profile markets. If anyone had a doubt about Bulldozer being a failure and that CMT/modular being the wrong design trade offs, here comes the answer from AMD's mouth.

Did you get the feeling that they are divesting from x86? With AMD shrinking engineering resources, it is rather obvious that they needed to find a way out of x86 market, and the way out is manufacturing SoCs. Looks like Charlie was right, and AMD big core line is toast.

Two things caught my attention in the presentation:

- The same weak points in Intel line up they pointed out in their presentation are even worse in AMD's own x86 line up. If Atom, including the 22nm, is crappy and ARM is simply better, what's left for Kabini and Temash? Why did they bother to develop Jaguar in the first place?

- The always obtuse nature of AMD marketing message: Calxeda is actually selling ARM servers and is developing a custom core, Samsung pooched an entire CPU team from AMD, and yet they still think they can be ahead of the pack with a vanilla ARM core. And ARM chips have nowhere near the level of server features that big x86 chips have, let alone the performance levels, and yet they directly compare the costs of the two.

 

[blackened23]

Misleading anti-AMD OP by the sontin... I'm shocked, totally shocked I tell you.

I've posted my unfavorable views of AMD CPUs, but the OP is clearly baiting and trolling with the thread title. It isn't uncharted territory for him, he does it continually in subtle ways.

 

[Cerb]

I had a little time and read the summary of an AMD rep who talked about ARM:
http://semiaccurate.com/2013/05/15/amds-andrew-feldman-talks-about-arm/

It's a little bit shocking that AMD thinks that a standard 64bit ARM core is much better than Bulldozer to take market share from Intel...

I don't see why not. They can hardly give away BDs. Personally, I'm pessimistic, because I think they missed the boat my a mile by not having Bobcat-based server chips. At that time, there would have been no competition for them, having multicore x86-64 parts, and could have gotten a foothold that 1/2-1/4-speed ARM CPUs, stuck in 32-bit land, had no chance of.

In 2014, they will have to fight off other SoC vendors for their sales.

Mr. Feldman was confident that AMD would lead the ARM server market because of AMD’s Freedom fabric interconnect and its unique IP blocks.

Nobody else has ever implemented proprietary high-speed networks that look like COTS networks to the OS. Not anybody. Ever.

Oh, and, "per say"? SemiProofread articles can be found on SemiAccurate .

 

[inf64]

Misleading anti-AMD OP by the sontin... I'm shocked, totally shocked I tell you.

Yep, this sums the OP perfectly.

 

[Gikaseixas]

Fail thread. Title has nothing to do with what was actually said. Only one other intel fanboy jumped in and tried to further stir the pot.

 

[sontin]

Please provide exact quote of what was being said. Performance wise what you stated is of course not true.

There are a lot of hints:

Right out of the gate Mr. Feldman laid down a few reasons why he believes that ARM server chips will take market share from traditional x86 offerings.

Apparently the server market is no longer about having the highest single threaded performance, it’s now about using lots of little power efficient cores to solve simple problems, like search queries. Servers are changing to meet this new workload by adopting ARM chips, which are better suited for this role than traditional x86 servers.

For server workloads Mr. Feldman believes that ARM processors have an advantage over x86 chips in terms of power consumption, cost, and die space

...

But Feldman said:
So the advantage is power draw, cost and die area. Performance is not an advantage and probably won't be in near future. AMD is covering the HPC segment with SR/EX cores in next few years. Beyond that it's possible we might even see some mix of x86/ARM/GPU all on the same die.

Performance comes with better perf/watt. See how people react to Jaguar. And don't forget that they are selling the server cpu in their APU since Trinity.

 

[inf64]

The fact is that he never said what you wrote in the title of the thread, so it's misinterpretation at best or misinformation at worst. You pick which one.

 

[Gikaseixas]

nm

 

[Imouto]

- The same weak points in Intel line up they pointed out in their presentation are even worse in AMD's own x86 line up. If Atom, including the 22nm, is crappy and ARM is simply better, what's left for Kabini and Temash? Why did they bother to develop Jaguar in the first place?

I guess AMD wants the whole gaming market for itself.

- The always obtuse nature of AMD marketing message: Calxeda is actually selling ARM servers and is developing a custom core, Samsung pooched an entire CPU team from AMD, and yet they still think they can be ahead of the pack with a vanilla ARM core. And ARM chips have nowhere near the level of server features that big x86 chips have, let alone the performance levels, and yet they directly compare the costs of the two.

General purpose chips for servers will die. You can build a custom SoC based on your needs instead of blowing millions of dollars in overkill server hardware. Google and Facebook are heading this way and any smart player will replace their servers for customs from now on.

 

[MisterMac]

I guess AMD wants the whole gaming market for themselves.



General purpose chips for servers will die. You can build a custom SoC based on your needs instead of blowing millions of dollars in overkill server hardware. Google and Facebook are heading this way and any smart player will replace their servers for customs from now on.

Custom SoCs like in embedded systems with specific purposed design takes time - probably not as muc as AMD\Intel are going for x86 but still time.

Alot of time.


And then that time has to be counted as investment - and then even if using ARM they will most likely use some proprietary block of IP.

Which makes it less useable for bog standard ARM cores and dillutes the point of using ARM cause of ease\cost.

I think it's clear intel is destroying it's a "middle-class" and going for lower class and extreme high end sales.

READ: Mobile + Server - desktop is to be a blimp sadly.

 

[Idontcare]

General purpose chips for servers will die. You can build a custom SoC based on your needs instead of blowing millions of dollars in overkill server hardware. Google and Facebook are heading this way and any smart player will replace their servers for customs from now on.

You are a good level-headed poster, so I am baffled by this position you have taken here.

You are trying to generalize what might be true for say the top 100 companies in the world and apply that "benefits of scale" to the other 99% of the market volume that is created by companies that very much can only afford (or will only ever invest) to buy general purpose COTS servers.

How many Googles and Facebooks are there? Versus how many other businesses that are going to purchase a rack-server every year from DELL or HP?

Sure proprietary servers will exist and serve a market space. But to suggest they will utterly displace the pedestrian general purpose server is akin to saying the rise of the aerospace industry will result in the auto-industry disappearing. It just doesn't work like that.

 

[Imouto]

@IDC You only need a custom SoC for searching engines, other for web hosting, another one for cloud storage and so on like SQL and other light stuff. There are not that much different types of workloads. I know you will still need x86 for things like virtualization or specialized computing but common stuff will be ARM SoC realm. If a bunch of randoms made their own ASICs for Bitcoin mining I don't know why Qualcomm, Calexa, HP or other SoC designer can't do the same. Even the chinese government is making their own CPUs for their Google clone servers and AMD just started a custom design department for example.

 

[Abwx]

Lol , i was banned for two weeks because i said that the OP
was doing viral marketing , wich is blatantly the case in this
thread that has a fraudulous quote as title...

 

[mrmt]

@IDC You only need a custom SoC for searching engines, other for web hosting, another one for cloud storage and so on like SQL and other light stuff. There are not that much different types of workloads.

And what makes you think that the added costs for the custom blocks will be able to offset the smaller scale of those developments? And what about release cycle, do you think you can sustain a two-year release cycle with exponentially higher R&D costs with custom SoCs? Not even Facebook is developing custom ARM cores for them, they are developing basically a custom interconnect, they are using vanilla processors.

The console example as you touted as example illustrated just that. You keep manufacturing the same chip in different nodes for 5, 6, 7 years and there isn't changes in performance there. The nature of the console market (and of any embedded business) allows just that. But for almost everything, especially servers, a 7 year old x86 chip is as good as junk now, because the performance and performance/watt bar is so high when compared to the past that a 7 years old server is essentially uneconomic to operate.

This is yet another AMD "I'm out of the bleeding edge" moment, not something for companies of the league of Intel, Qualcomm or Samsung for that matter.

 

[Imouto]

@mrmt Consoles ain't the same as servers. A console player would feel ripped having to buy a new system every 2 years.

Your own "6-7 years x86 servers are junk now" proves you wrong. x86 is still there as ARM v7 code would. Take mobile SoC makers for instance, they sell their SoCs for barely 1 year before they become obsolete as walkmans. Qualcomm for example made an integrated LTE modem for their SoCs giving an edge over other designers and put a custom ARM core on top. Servers are the same, custom is the way to go.

 

[mrmt]

Your own "6-7 years x86 servers are junk now" proves you wrong. x86 is still there as ARM v7 code would. Take mobile SoC makers for instance, they sell their SoCs for barely 1 year before they become obsolete as walkmans. Qualcomm for example made an integrated LTE modem for their SoCs giving an edge over other designers and put a custom ARM core on top. Servers are the same, custom is the way to go.

Who, except Nvidia, is marketing SoCs for one year only? Life of a common SoC is 3 years in the bleeding edge, more in embedded.

 

[sontin]

Qualcomm is putting out new SoCs nearly every 6 months right now:
Krait, Krait Quadcore, S400, S600 and S800.