Wednesday, October 24th 2012

AMD Could Place x86 and ARM Cores on the Same Silicon: Report

Following last week's 15% workforce cut that came with the company's Q3 2012 results, AMD plans to consolidate its workforce and eliminate "bureaucracy." According to a VR-Zone report, the company plans to announce its future strategy next week, on October 29. This strategy could ride heavily on recent moves the company made in the area of "heterogeneous computing," and a working alliance it built with the likes of ARM, MediaTek, Qualcomm, Samsung and TI, under the heterogeneous system alliance (HSA).

According to the report, AMD's future looks to be heading towards building highly versatile SoCs for devices in the post-PC era. It's not far-fetched to imagine, for example, that AMD could integrate ARM IP onto APUs with x86 cores, with the ARM cores based on recent micro-architectures such as Cortex-M5, performing specific kinds of tasks (such as data-encryption). Down the line, AMD could also integrate baseband components into the APU, beating Intel to the punch on similar efforts.
Source: VR-Zone
Add your own comment

30 Comments on AMD Could Place x86 and ARM Cores on the Same Silicon: Report

#1
btarunr
Editor & Senior Moderator
Maybe they should start with a RISC core dedicated to calculating Pi. AMD's market-share will shoot up in no time. :rolleyes:
Posted on Reply
#2
Steevo
btarunrMaybe they should start with a RISC core dedicated to calculating Pi. AMD's market-share will shoot up in no time. :rolleyes:
I f they just created a app for trans coding applications into OpenCL on all those beautiful unused cores.........
Posted on Reply
#3
cadaveca
My name is Dave
btarunrMaybe they should start with a RISC core dedicated to calculating Pi. AMD's market-share will shoot up in no time. :rolleyes:
I'm sure they'll do that...as soon as nVidia makes CUDA/Phys-X work on CPUs properly. :p
Posted on Reply
#4
W1zzard
SteevoI f they just created a app for trans coding applications into OpenCL on all those beautiful unused cores.........
won't do much good. gpus/opencl are really really bad at branching (if .. then .. else ..)
Posted on Reply
#5
suraswami
AMD always thinks ahead of time and dreams for it too, but the problem is do they release the product ahead of time? And if they do, is it the way it should be?

Hopefully PC industry don't die with all these stupid hand held devices (they are nice, but gaming on PC is better).
Posted on Reply
#6
TheoneandonlyMrK
could end up as something very usefull to amd to counter intels ssmall business advantage in chips ie an arm core as a hypervisor checking for viria in the background completely unobtrucively on thin clients (apus) or for low power advantages ,could be a good move reversed too, with a low tpd chip with an arm core for main use and a steamy core to push heavy loads on a handheld ,all tied to world leading gfx ip mmm
Posted on Reply
#7
Steevo
W1zzardwon't do much good. gpus/opencl are really really bad at branching (if .. then .. else ..)
What can we do to fix them? :laugh:

I'm not certain but I believe Super PI isn't really too heavily branching. However single threaded as it is you would tend to think that a CPU core running at 4Ghz would be faster at processing division than it is. Unless of course you have cache issues still.
Posted on Reply
#8
Sasqui
suraswamiHopefully PC industry don't die with all these stupid hand held devices (they are nice, but gaming on PC is better).
Yes, hearing "Post PC era" makes me sadfaced :ohwell:
Posted on Reply
#9
Nordic
This arm thing might have some merit even if I don't see it. Why do I feel like I am going to hear something even more rediculous like amd making an amd nvidea hybrid gpu...
Posted on Reply
#10
Cortex
3

Three types of processors on single chip GCN/ARM/x86 Faildozer. Why?







------------------------------------------------------------------
Which company makes the iPhone?

Is that a trick question?
Apple or Foxconn?:rolleyes:
Posted on Reply
#11
Xzibit
This was hinted at way back when AMD bought SeaMicro at the start of the year and has been making the rounds when talking about ARM servers.

Applied Micro which is leading the ARM server charge
Now that it owns SeaMicro, AMD has hinted that it will integrate the company's technology onto its own Opteron server chips. "But that's a couple of years away,"
This could be a good medium. Having both bases covered while the industry decides which or if it moves or stays with one or the other.
Posted on Reply
#12
Ikaruga
W1zzardgpus/opencl are really really bad at branching (if .. then .. else ..)
I don't want to argue with you about this since I'm not a CUDA developer, but why is it that tech documents what I read from Nvidia used to state that this is incorrect. They say that you just have to code your stuff differently and the performance loss will be minimal. There is a huge body of research going on to "cure" branching and branch divergence problems, it's not like how it was at the beginning.

Is that nothing but meaningless PR talk to programmers?
Posted on Reply
#13
Assimilator
Eliminate bureaucracy? They'd have to fire some of their CxOs to do that, not engineers.

As for ARM, I'm really not seeing how this partnership is going to help AMD. AMD is so heavily invested in x86, does it really have the capability/time to become skilled at producing good ARM chips as well? If this is an attempt for AMD to compete with the upcoming Intel Clovertrail (Atom CPUs for tablets) and Haswell, I'm not sure what benefit there is in having heterogeneous instruction sets in a single CPU package.
Posted on Reply
#14
Nordic
I got it. Amd will then be able to say it runs Android apps natively.
Posted on Reply
#15
W1zzard
IkarugaI don't want to argue with you about this since I'm not a CUDA developer, but why is it that tech documents what I read from Nvidia used to state that this is incorrect. They say that you just have to code your stuff differently and the performance loss will be minimal.
www.gamedev.net/topic/413859-need-explanation-of-efficiency-issues-with-dynamic-branching/
post 2 and 6 in that thread have good explanations on why gpus suck at branching

wikipedia "branch prediction" should have some good info on the complexities and what cpu makers have been doing to improve it
Posted on Reply
#16
Fourstaff
So Intel is forcing x86 down, while AMD is sidestepping the problem. This is going to be interesting, especially when Intel has the firepower to change route while AMD is do or die.
Posted on Reply
#17
Ikaruga
W1zzardwww.gamedev.net/topic/413859-need-explanation-of-efficiency-issues-with-dynamic-branching/
post 2 and 6 in that thread have good explanations on why gpus suck at branching

wikipedia "branch prediction" should have some good info on the complexities and what cpu makers have been doing to improve it
Thanks for the link, it was interesting to read, but it's from 2006, and a lot of progress could have been made in the last 6 years. I'm not saying that it's any close CPU solutions, but iirc it's not that hopeless anymore. I will try to dig up some docs tomorrow, and get back to you on this one.
Posted on Reply
#18
Sasqui
Fourstaffwhile AMD is sidestepping the problem
I think it's more hedging than sidestepping... though there's a lot more to the story. Since most mobile solutions have sub-chips that perform discreet tasks that have specific host protocols already built in, they are looking to take off-the shelf parts and create a hybrid, or it certainly sounds like it.
Posted on Reply
#19
Fourstaff
SasquiI think it's more hedging than sidestepping... though there's a lot more to the story. Since most mobile solutions have sub-chips that perform discreet tasks that have specific host protocols already built in, they are looking to take off-the shelf parts and create a hybrid, or it certainly sounds like it.
Hedging is probably a better word, but either way silicon real estate is not cheap.
Posted on Reply
#20
suraswami
FourstaffSo Intel is forcing x86 down, while AMD is sidestepping the problem. This is going to be interesting, especially when Intel has the firepower to change route while AMD is do or die.
Don't worry, Intel will be forced to follow AMD's innovation (if not already). AMD innovates and Intel follows :rockout:
Posted on Reply
#21
TheoneandonlyMrK
FourstaffHedging is probably a better word, but either way silicon real estate is not cheap.
they are not hedgeing in anyway, the whole starting point and ethos of their current Ic design plan is modularity and pop in Ip ,they have made all of their chips this way over the last few years, an evolution on one core is then easily transfered to other asic's, and differentiated Ip can be added into there chip designs with minimal fuss, Pop in Ip is instrumental to there push in the embedded segment ;) an area Amd wish to gain further revinue from and i suspect its there Pop in Ip flexability that added weight to the next gen console decisions.
Posted on Reply
#22
horizon2600
theoneandonlymrkcould end up as something very usefull to amd to counter intels ssmall business advantage in chips ie an arm core as a hypervisor checking for viria in the background completely unobtrucively on thin clients (apus) or for low power advantages ,could be a good move reversed too, with a low tpd chip with an arm core for main use and a steamy core to push heavy loads on a handheld ,all tied to world leading gfx ip mmm
The lobotomized machine gun. Although excusable if you are 9 or 10 years old xD

Still sig worthy.
Posted on Reply
#23
turtile
AMD is using the ARM core for ARM's TrustZone security. It was announced months ago...

Intel bought out McAfee to develop its own solution.
Posted on Reply
#24
Thefumigator
There will always be PC, it will never die.
Its just about adding extensions to it.
Posted on Reply
#25
TRWOV
Didn't we already know this? The ARM chip is supposed to be an hypervisor. I recall reading about this on TPU about two months ago... maybe at Tom's?
Posted on Reply
Add your own comment
Apr 19th, 2024 16:30 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts