Text Size

K10: Barcelona, Shanghai, Quad-Core Opteron, Phenom

Bobcat in Servers?

AMD's latest line-up of CPUs including Shanghai Opteron, Phenom and Phenom II X4 & X3, Athlon X2.

Bobcat in Servers?

Postby JF-AMD » Mon Nov 29, 2010 4:45 pm

Because the question has been asked here:

http://bit.ly/hHwL8t
While I work for AMD, my posts are my own opinions.

http://blogs.amd.com/work/author/jfruehe/

Follow AMD Opteron on Twitter: @JF_AMD
User avatar
JF-AMD
XIP
 
Posts: 1832
Joined: Thu Apr 23, 2009 7:27 am

Re: Bobcat in Servers?

Postby Pietro sk » Mon Nov 29, 2010 4:59 pm

i saw more than 5 news about ARM / atom servers recently
bobcat makes sense in servers too.. (if atom , arm does - why not ?)
Image
The famous intel lawsuit Image Moron is playing VIDEO-game
Not biased Cinebench ? Think again ... They say it´s "finetuned" for intel CPU´s .. Image
User avatar
Pietro sk
K8 Athlon 64 X2 (Windsor) Elite Boarder
K8 Athlon 64 X2 (Windsor) Elite Boarder
 
Posts: 3933
Joined: Fri Jan 08, 2010 3:59 pm
Location: Le sarcasm..

Re: Bobcat in Servers?

Postby BaronMatrix » Mon Nov 29, 2010 6:08 pm

JF-AMD wrote:Because the question has been asked here:

http://bit.ly/hHwL8t



At first I thought you guys had lost your minds, but you clearly delineated the advantages of Opteron ONLY for server. Not that I would consider Bobcat for a sever.
MSI 790GX X4 940 - @ 3GHz - AC64 HSF - HD 4870 512MB - 8GB DDR2 - 1TB HDD - BluRay - TX650 W - Antec 300 - 2X ACER 23"
HP DV7-3065 - M600 - 17" LCD - BD-Rom - HDMI\eSATA - .5TB HDD - HD4200
User avatar
BaronMatrix
K8 Athlon 64 (San Diego) Expert Boarder
K8 Athlon 64 (San Diego) Expert Boarder
 
Posts: 1949
Joined: Mon Jan 29, 2007 3:13 am

Re: Bobcat in Servers?

Postby JF-AMD » Mon Nov 29, 2010 6:42 pm

Pietro sk wrote:i saw more than 5 news about ARM / atom servers recently
bobcat makes sense in servers too.. (if atom , arm does - why not ?)


It all you have is a hammer, all your problems look like nails. Who is pushing ARM on servers? ARM. The reality is that the market is big enough yet to merit the work.

There are also people developing Itanium servers. Just because it is in the news does not mean that a.) there is enough of a market yet and b.) they will be successful.
While I work for AMD, my posts are my own opinions.

http://blogs.amd.com/work/author/jfruehe/

Follow AMD Opteron on Twitter: @JF_AMD
User avatar
JF-AMD
XIP
 
Posts: 1832
Joined: Thu Apr 23, 2009 7:27 am

Re: Bobcat in Servers?

Postby abinstein » Mon Nov 29, 2010 6:49 pm

JF-AMD wrote:
Pietro sk wrote:i saw more than 5 news about ARM / atom servers recently
bobcat makes sense in servers too.. (if atom , arm does - why not ?)


It all you have is a hammer, all your problems look like nails. Who is pushing ARM on servers? ARM.

Right on! :lol: :lol:

Those tiny power-efficient cores do not scale out and their aggregated memory and I/O bandwidth are usually too low. In contrast, a 2-socket Opteron 6000 server has 8 channels of memory and two 16-bit HT3 links for I/O; the 16 or 24 (or 32 next year) cores work in single system image sharing the system resources, bringing the benefits of consolidation. To get the same number of cores one needs 8 to 16 dual-core ARM (or Bobcat) systems working independently. My guess is the tiny core solution will end up consuming more power and deliver less performance.
abinstein
K8 Opteron (SledgeHammer) Moderator
K8 Opteron (SledgeHammer) Moderator
 
Posts: 7176
Joined: Sat Oct 30, 2004 9:49 pm

Re: Bobcat in Servers?

Postby MU_Engineer » Mon Nov 29, 2010 7:38 pm

abinstein wrote:Right on! :lol: :lol:

Those tiny power-efficient cores do not scale out and their aggregated memory and I/O bandwidth are usually too low. In contrast, a 2-socket Opteron 6000 server has 8 channels of memory and two 16-bit HT3 links for I/O; the 16 or 24 (or 32 next year) cores work in single system image sharing the system resources, bringing the benefits of consolidation. To get the same number of cores one needs 8 to 16 dual-core ARM (or Bobcat) systems working independently. My guess is the tiny core solution will end up consuming more power and deliver less performance.


There are some uses where very low-powered CPUs that don't scale up or out very well can succeed. Think of a home or small business file server or NAS, print server, or gateway. You're probably only going to have one of those machines, it does not need to be all that powerful, and it likely won't need to scale out to need multiple many-cored CPUs. Those would be decent candidates for low-power, not-very-expandable CPUs. There is apparently a market for them as quite a few vendors have debuted products using said chips, such as HP's Micro Server, which uses a 1.3 GHz Athlon II X2 and some ECC RAM.
Desktop: Opteron 6234, SM H8QGL, 4x4 GB PC3-12800
HTPC: 2x 2 GHz Xeon LV Sossaman, TYAN S5365, 6 GB PC2-3200R
File Server: 2x Xeon X5260, MSi MS-91A1, 12 GB PC2-5300F
Router: Athlon 5150, Biostar AM1ML, 4 GB PC3-12800
MU_Engineer
K8 Athlon 64 (Orleans) Expert Boarder
K8 Athlon 64 (Orleans) Expert Boarder
 
Posts: 2103
Joined: Sat Apr 08, 2006 5:17 am

Re: Bobcat in Servers?

Postby Pietro sk » Mon Nov 29, 2010 9:21 pm

funny or not, in server shops atom things are present several months ago.. its question of time, if arm will be visible in such shops

on the other note, specs of those "servers" are quite funny, now everybody can call "chinese scientifical calc" a server too... :lol:
------
:arrow: note how many atom boards are on sites...
example http://www.supermicro.com/products/motherboard/Atom/ (new atoms too)
----------
edit
google has published report about their ram errors, stability..
after reading that article, im happy that i use ECC 8)
Last edited by Pietro sk on Tue Nov 30, 2010 1:53 pm, edited 1 time in total.
Image
The famous intel lawsuit Image Moron is playing VIDEO-game
Not biased Cinebench ? Think again ... They say it´s "finetuned" for intel CPU´s .. Image
User avatar
Pietro sk
K8 Athlon 64 X2 (Windsor) Elite Boarder
K8 Athlon 64 X2 (Windsor) Elite Boarder
 
Posts: 3933
Joined: Fri Jan 08, 2010 3:59 pm
Location: Le sarcasm..

Re: Bobcat in Servers?

Postby Athlotron » Tue Nov 30, 2010 10:49 am

JF-AMD wrote:Because the question has been asked here:

http://bit.ly/hHwL8t

I found it interesting, your hinting at AMD making some SOC's, and seeing which of their core products would be suitable for it. This is exciting, and all of this for Cloud Computing.

AMD has it's head in the clouds now? :P
What is a cynic? A man who knows the price of everything and the value of nothing. - Oscar Wilde
User avatar
Athlotron
K7 Athlon XP (Palomino) Junior Boarder
K7 Athlon XP (Palomino) Junior Boarder
 
Posts: 534
Joined: Sat Jun 26, 2010 7:48 am
Location: Western Australia, Australia

Re: Bobcat in Servers?

Postby JF-AMD » Tue Nov 30, 2010 1:04 pm

AMD has sold over 2M Opterons into the cloud so far, I'd say we are there.
While I work for AMD, my posts are my own opinions.

http://blogs.amd.com/work/author/jfruehe/

Follow AMD Opteron on Twitter: @JF_AMD
User avatar
JF-AMD
XIP
 
Posts: 1832
Joined: Thu Apr 23, 2009 7:27 am

Re: Bobcat in Servers?

Postby BaronMatrix » Tue Nov 30, 2010 4:35 pm

Pietro sk wrote:funny or not, in server shops atom things are present several months ago.. its question of time, if arm will be visible in such shops

on the other note, specs of those "servers" are quite funny, now everybody can call "chinese scientifical calc" a server too... :lol:
------
:arrow: note how many atom boards are on sites...
example http://www.supermicro.com/products/motherboard/Atom/ (new atoms too)
----------
edit
google has published report about their ram errors, stability..
after reading that article, im happy that i use ECC 8)



There is no doubt that simple file serving can use less beefy chip with better IO but that's a very small amount. The best use for those would be as "late night" replacements. But due to the fact that it's called the world wide web, the Internet would eat them up. SMB could mabye use them but Adelaide will cover the low power arena nicely.
MSI 790GX X4 940 - @ 3GHz - AC64 HSF - HD 4870 512MB - 8GB DDR2 - 1TB HDD - BluRay - TX650 W - Antec 300 - 2X ACER 23"
HP DV7-3065 - M600 - 17" LCD - BD-Rom - HDMI\eSATA - .5TB HDD - HD4200
User avatar
BaronMatrix
K8 Athlon 64 (San Diego) Expert Boarder
K8 Athlon 64 (San Diego) Expert Boarder
 
Posts: 1949
Joined: Mon Jan 29, 2007 3:13 am

Re: Bobcat in Servers?

Postby dm7000s » Wed Mar 16, 2011 12:06 am

Reviving this older thread a little, because although AMD might or might not have a bobcat style server CPU in the works, Intel has:

Intel, Facebook Talk Up New “Micro” Servers:

"Intel expects most makers of micro servers will prefer new energy-efficient members of its Xeon line of server chips, including two new models now in production and a new version expected in the second half of the year based on a design called Sandy Bridge. Intel next year also plans to offer a server-style version of its Atom design used in portable computers that is expected to draw less than 10 watts–compared with a range of 15 watts to 65 watts for comparable Xeons."

So in case of "AMD might not have": "AMD", please read the above bold sentence, and don't be late to the party :wink:

Edit: More on this -> Intel Plans on Bringing Atom to Servers in 2012, 20W SNB Xeons in 2011
Image
User avatar
dm7000s
K6-2 Fresh Boarder
K6-2 Fresh Boarder
 
Posts: 218
Joined: Wed May 07, 2008 11:24 pm

Re: Bobcat in Servers?

Postby AussieFX » Sun Mar 20, 2011 6:10 am

I think it's time for intel and AMD to put their fighting to bed - particularly intel.

ARM is a threat and if intel/AMD get together to fight this they might have some chance. I would prefer to see that happen than have AMD take an ARM license which a lot of people are suggesting AMD should do.
Sitting back and saying we have the right solution when the market appears to be asking for more will only end in trouble.

You have to cover the oppositions moves, and in this case intel aren't the opposition.
Sent from my flippy phone thingy using TAPATALK HD_2016.1


Image
Nikon D7000 / Nikon D5000
User avatar
AussieFX
K8 Opteron (SledgeHammer) Moderator
K8 Opteron (SledgeHammer) Moderator
 
Posts: 7678
Joined: Fri May 11, 2007 1:50 pm
Location: I wish I knew...

Re: Bobcat in Servers?

Postby maduroutmb » Wed Mar 30, 2011 7:46 pm

Extremely low power server chips are in direct opposition to the current trends toward hardware utilization, consolidation, and efficiency. You take a small number of high-performance chips which are close to dense and efficient power and cooling systems and you queue up workloads in such a way that they have very high utilization rates (e.g. with virtualization). The ARM/Atom/Bobcat "server" chips are meant to shine in isolated hardware where low power usage is most valuable because the utilization will never be very high. Print and low-utilization file servers: in other words, the sort of systems that are supposed to be eliminated by consolidation.In most commercial cases, ARM servers are the equivalent of trying to drive in a screw with a hammer (as JF pointed out).

Home NAS is one area where such systems could do well, except that I believe that consolidation will replace them even here. The idea of a home server, which at once acts as a file server, media and gaming center (with the ability to stream content to thin clients around the house), and a traditional desktop. The power to perform several jobs at once (i.e. sending files to dad on a business trip via Tonido, streaming a movie to the kitchen for mom, streaming audio to one child's room while the other kids to play a video game in the den) means that the total power consumption number aren't terribly different. PCI-E wired router cards(you can functionally do this with a master-mode wireless card)and the return of PCI cable/DSL/fiber modems would allow a user to literally have one box for their home with thin clients (a good opportunity for Bobcat or Atom/ION devices but likely not ARM) located by other displays in the kitchen or bedrooms.
Phenom 9950 BE @2.52 GHz/1.2125V (Nirvana NV120), 790GX, 8 GB DDR2-800 @4-4-4-12/2.2V, 2*4850 1GB @0.625 GHz
Fedora 14 KDE 64/Windows 7 64
User avatar
maduroutmb
K7 Athlon (Argon) Junior Boarder
K7 Athlon (Argon) Junior Boarder
 
Posts: 361
Joined: Mon May 05, 2008 3:03 am
Location: Galveston, TX

Re: Bobcat in Servers?

Postby scorpiDragon » Mon Apr 04, 2011 5:55 am

I see where Bobcat and for that mater Trinity could come in real handy, is in the workstation market turn the G34 socket into an FM3 socket (with 1974+ pins = 3x PCI-e 2.0 links; 1x HT3 link to bridge; 1x HT link to HTX socket or second socket).
scorpiDragon
K5 Fresh Boarder
K5 Fresh Boarder
 
Posts: 135
Joined: Fri May 02, 2008 1:42 am

Re: Bobcat in Servers?

Postby JF-AMD » Wed Apr 06, 2011 9:23 pm

AussieFX wrote:ARM is a threat and if intel/AMD get together to fight this they might have some chance. I would prefer to see that happen than have AMD take an ARM license which a lot of people are suggesting AMD should do.


ARM has really low power. But, once you add:

ECC
64-bit memory addressing
Coherent links to allow for more than 1 socket
More than 2 cores
Higher performance to match current low power processor choices.

...you have something that is actually now drawing the same power as an x86 processor. Except that it can't run x86 code natively. There just isn't viable today.
While I work for AMD, my posts are my own opinions.

http://blogs.amd.com/work/author/jfruehe/

Follow AMD Opteron on Twitter: @JF_AMD
User avatar
JF-AMD
XIP
 
Posts: 1832
Joined: Thu Apr 23, 2009 7:27 am

Re: Bobcat in Servers?

Postby hyc » Thu Apr 07, 2011 5:59 pm

JF-AMD wrote:
AussieFX wrote:ARM is a threat and if intel/AMD get together to fight this they might have some chance. I would prefer to see that happen than have AMD take an ARM license which a lot of people are suggesting AMD should do.


ARM has really low power. But, once you add:

ECC
64-bit memory addressing
Coherent links to allow for more than 1 socket
More than 2 cores
Higher performance to match current low power processor choices.

...you have something that is actually now drawing the same power as an x86 processor. Except that it can't run x86 code natively. There just isn't viable today.


Mmm. I'd say you have something drawing 10-15% less power than x86 worst case, because you don't have the godawful instruction decoder.
hyc
K8 Athlon 64 (Winchester) Expert Boarder
K8 Athlon 64 (Winchester) Expert Boarder
 
Posts: 1349
Joined: Mon Jan 16, 2006 4:38 pm

Re: Bobcat in Servers?

Postby maduroutmb » Sat Apr 09, 2011 4:06 am

the godawful instruction decoder.


It's interesting though that the special case that CISC originally meant to solve, limited RAM, is now alive and well in the form of limited L3. That's not a justification for the massive silicon overhead that the decoder requires, but it is food for thought when dismissing CISC architectures out of hand.
Phenom 9950 BE @2.52 GHz/1.2125V (Nirvana NV120), 790GX, 8 GB DDR2-800 @4-4-4-12/2.2V, 2*4850 1GB @0.625 GHz
Fedora 14 KDE 64/Windows 7 64
User avatar
maduroutmb
K7 Athlon (Argon) Junior Boarder
K7 Athlon (Argon) Junior Boarder
 
Posts: 361
Joined: Mon May 05, 2008 3:03 am
Location: Galveston, TX

Re: Bobcat in Servers?

Postby abinstein » Sat Apr 09, 2011 7:15 am

maduroutmb wrote:
the godawful instruction decoder.


It's interesting though that the special case that CISC originally meant to solve, limited RAM, is now alive and well in the form of limited L3. That's not a justification for the massive silicon overhead that the decoder requires, but it is food for thought when dismissing CISC architectures out of hand.

The more interesting thing is what gives the "CISC" x86-64 "performance" today is terrible memory inefficient.

Take a look at assembled of any SSE program segments. Not only the average instruction size is 6~8 bytes, but also lots of time the compiler purposely align instructions to 16-byte or 32-byte boundaries. Wasting memory space is what x86-64 compiler does for getting good performance.

So what's the benefit of CISC, or in particular x86?
abinstein
K8 Opteron (SledgeHammer) Moderator
K8 Opteron (SledgeHammer) Moderator
 
Posts: 7176
Joined: Sat Oct 30, 2004 9:49 pm

Re: Bobcat in Servers?

Postby Pietro sk » Sat Apr 09, 2011 2:15 pm

how ironic, apple went from POWER to x86 :lol:
Image
The famous intel lawsuit Image Moron is playing VIDEO-game
Not biased Cinebench ? Think again ... They say it´s "finetuned" for intel CPU´s .. Image
User avatar
Pietro sk
K8 Athlon 64 X2 (Windsor) Elite Boarder
K8 Athlon 64 X2 (Windsor) Elite Boarder
 
Posts: 3933
Joined: Fri Jan 08, 2010 3:59 pm
Location: Le sarcasm..

Re: Bobcat in Servers?

Postby Am486DX100 » Sat Apr 09, 2011 5:03 pm

abinstein wrote:
maduroutmb wrote:
the godawful instruction decoder.


It's interesting though that the special case that CISC originally meant to solve, limited RAM, is now alive and well in the form of limited L3. That's not a justification for the massive silicon overhead that the decoder requires, but it is food for thought when dismissing CISC architectures out of hand.

The more interesting thing is what gives the "CISC" x86-64 "performance" today is terrible memory inefficient.

Take a look at assembled of any SSE program segments. Not only the average instruction size is 6~8 bytes, but also lots of time the compiler purposely align instructions to 16-byte or 32-byte boundaries. Wasting memory space is what x86-64 compiler does for getting good performance.

So what's the benefit of CISC, or in particular x86?


The only way to get beyond a boundary is to be inefficient by design. If the said "design" can be rid of/or can get rid of the earlier baggage, then it can survive. If the said "design" is needed, then it has to be worked out in a less efficient manner. Trade-offs are trade-offs.

...
Ah, the old days before there was this fashion for having a solo career.

Squeak?

Yes, once there were five horsemen. But you know;there's always a row. Creative disagreements, rooms being trashed, and things said that perhaps should not have been said.
User avatar
Am486DX100
K8 Athlon 64 (Venice) Expert Boarder
K8 Athlon 64 (Venice) Expert Boarder
 
Posts: 1675
Joined: Fri May 26, 2006 7:17 pm
Location: Ahhhh..... I'm @rting around!!!

Re: Bobcat in Servers?

Postby maduroutmb » Sun Apr 10, 2011 2:58 am

Take a look at assembled of any SSE program segments. Not only the average instruction size is 6~8 bytes, but also lots of time the compiler purposely align instructions to 16-byte or 32-byte boundaries. Wasting memory space is what x86-64 compiler does for getting good performance.


Very interesting. But I think that the argument has been made that x86 is so bloated that, at this point, it's a terrible representative for the idea of CISC in general. I can enable AMDLive! on my Phenom 9950 if I want. I would assume that the amount of memory needed to designated an operation is proportional to the total number of operations that must be distinguished. Would you say that a major overhaul of x86 (or a completely new CISC ISA) could avoid a significant portion of the waste that you brought up?
Phenom 9950 BE @2.52 GHz/1.2125V (Nirvana NV120), 790GX, 8 GB DDR2-800 @4-4-4-12/2.2V, 2*4850 1GB @0.625 GHz
Fedora 14 KDE 64/Windows 7 64
User avatar
maduroutmb
K7 Athlon (Argon) Junior Boarder
K7 Athlon (Argon) Junior Boarder
 
Posts: 361
Joined: Mon May 05, 2008 3:03 am
Location: Galveston, TX

Re: Bobcat in Servers?

Postby abinstein » Sun Apr 10, 2011 3:21 am

maduroutmb wrote:
Take a look at assembled of any SSE program segments. Not only the average instruction size is 6~8 bytes, but also lots of time the compiler purposely align instructions to 16-byte or 32-byte boundaries. Wasting memory space is what x86-64 compiler does for getting good performance.


Very interesting. But I think that the argument has been made that x86 is so bloated that, at this point, it's a terrible representative for the idea of CISC in general. I can enable AMDLive! on my Phenom 9950 if I want. I would assume that the amount of memory needed to designated an operation is proportional to the total number of operations that must be distinguished. Would you say that a major overhaul of x86 (or a completely new CISC ISA) could avoid a significant portion of the waste that you brought up?

The point you raised here is really not CISC vs RISC, but x86 vs non-x86. So we should look at them separately.

On x86 vs non-x86, the PC market definitely favors the former. It is for the same reason that the OS market favors Windows (or to some degree Linux) and the handheld market favors ARM. These market-favored specs all have baggage because they wouldn't have survived without supporting those baggage. So it's natural that x86 or Win32 won't be optimal for any next generation workload. IMHO, there is not much we can do either.

On CISC vs RISC, the two use completely different philosophy, and RISC is the way to get better performance or performance per watt. (There is not a single popular CPU with good performance per watt that does not actually implement RISC.)

Does that mean we can make a major overhaul of x86? It depends on the struggle between the two forces above. A danger of relying too much on baggage, however, is that it makes it more difficult for the lines of products venture into new territory. Both x86 and Windows are facing such problem. I don't expect x86/Windows to compete as-is with ARM/whatever in the handheld or ultra-low power markets. A major overhaul OTOH will break backward compatibility and make x86/Windows irrelevant.
abinstein
K8 Opteron (SledgeHammer) Moderator
K8 Opteron (SledgeHammer) Moderator
 
Posts: 7176
Joined: Sat Oct 30, 2004 9:49 pm


Return to K10: Barcelona, Shanghai, Quad-Core Opteron, Phenom

Who is online

Users browsing this forum: No registered users and 0 guests

cron