AMD CPU speculation... and expert conjecture

[juanrga]

I understand that some people here doesn't like facts. I can live with that.

What I find difficult to accept is when this same people pretends that the facts don't exist. I have given the AMD official roadmap together with AMD reasons why they are replacing jaguar servers with ARM servers. An official roadmap is not a "promo slide" nor "personal speculation". If you don't know how to read a roadmap or are confused about the meaning of blue/orange colours, that is not my problem.

Kyoto is known as the Opteron X-series and is available now, based on Jaguar and offering GPU compute enhancements as well as increased CPU performance. The Seattle family will replace these CPUs in the near future and will represent a new era for AMD as these chips will be clusters of ARM Cortex-A57 on AMD's advanced Freedom Fabric.

This week, AMD made public their server roadmap for 2013 and 2014, that revealed some details of future microprocessors and APUs for server and microserver markets.

[...]

A few weeks ago, AMD introduced X-Series Opteron CPUs, based on recently launched "Jaguar" microarchitecture. The roadmap states that these processors will be on the market until the second half of 2014, at which time AMD is going to replace them with "Seattle" systems-on-a-chip, built on ARM Cortex-57 IP core.

AMD has been hinting at major changes to its server strategy for a while now, but with its announcements of new server chips yesterday it has firmly pivoted away from a focus on the traditional two-processor and four-processor servers that drive most enterprise servers and toward lower-power, single-processor servers typically used in Web servers and cloud applications. More interestingly, it plans to take this even further next year, replacing the x86 cores in its extreme low-power server chips with new ARM-based cores. This represents an enormous shift in direction.

[...]

In the second half of 2014, AMD will be replacing the X-series with a new ARM-based chip known as "Seattle." This will be a System-on-Chip that is purely a CPU (in other words, it won't have integrated graphics) based on ARM Cortex-A57 cores. Initially it will be offered as an eight-core server chip, but with a 16-core version due to follow. The ARM cores are expected to run at 2GHz or a bit faster and AMD says they should be in the same general performance ranges as the "Jaguar" cores.

The whole industry is applauding AMD plans and new strategy/products, even Anandtech!!! LOL This is what Anandtech said about AMD plans of replacing jaguar servers by arm servers:

The chip that has the potential to give Intel some real headaches is “Seattle”. It is a pretty revolutionary design for being an AMD CPU.

[...]

It looks like the Intel Avoton will have a very potent challenger in Q1 2014.

[...]

However, the combined AMD, ARM and Seamicro technology inside AMD’s new Seattle CPU look extremely promising: these are probably the best specs of a micro server CPU we have seen so far. And since all the right components are now in place, it looks like the micro server is ready for prime time. There is little doubt that Seamicro servers will continue to thrive in their niche market while HP's Moonshot and Dell's Viking will make the market much more popular. So there is good chance that AMD will make a big comeback in 2014 in the server market.

http://www.pcper.com/news/General-Tech/AMDs-plans-keep-their-ARMs-server-room
http://www.cpu-world.com/news_2013/2013061901_AMD_reveals_server_roadmap.html
http://forwardthinking.pcmag.com/architecture/312794-amd-pivots-to-arm-on-servers
http://www.anandtech.com/show/7079/amd-evolving-fast-to-survive-in-the-server-market-jungle-/5

I find amazing how many AMD hate has floured in this thread recently.

 

[hcl123]

I am using the following definitions (from the above links):

Uniform Memory Access (UMA) is a shared memory architecture used in parallel computers. All the processors in the UMA model share the physical memory uniformly. In a UMA architecture, access time to a memory location is independent of which processor makes the request or which memory chip contains the transferred data. Uniform Memory Access computer architectures are often contrasted with Non-Uniform Memory Access (NUMA) architectures.

What if more than one processor in a MPU (example 4 cores) makes a memory request at the same time ?

Sure happens all the time... and the answer is; "its arbitrated" ... the IMC (Integrated Memory Controller) has queues for requests

Non-Uniform Memory Access (NUMA) is a computer memory design used in multiprocessing, where the memory access time depends on the memory location relative to a processor. Under NUMA, a processor can access its own local memory faster than non-local memory (memory local to another processor or memory shared between processors).

According to those definitions the memory model used in Kaveri is UMA (hUMa), but the memory model used in future AMD HSA dGPUs will be NUMA, because access to local memory will be faster than to remote memory (through the PCIe).

HSA and Nvidia seem to agree with me and both use the term unified, but not the term uniform, because the memory model is Non-Uniform. In fact, the term "hUMA" is not mentioned in the HSA specification.

I think we can agree on that we disagree.

In NUMA the "local" cache/memory is *probed* first for the memory request, and the next in the chain is "probed" next for memory request, like in a MCM (another CPU close coupled) with 4 channels (2 per CPU die), than another, or like AMD that has direct link to all other sockets, after the MCM is probed, the all other sockets are probed...

Of course there is a clear "separation" of priorities, and so of access times... but those different access times are consequence of the arbitration priorities... access times are *dependent" on the arbitration priorities, its *NOT* the arbitration priorities that are dependent on access times, more so because ANY mem request doesn't really know where the "target is" without probing first, it may happen its not in DRAM and the request passes to disk which is quite more time (time can vary tremendously, YOU NEVER KNOW "A PRIORI", the request can go to DRAM, to disk, or to network... in case of "clusters" where BERLIN APU will be headed, and BERLIN APU is hUMA/HSA, its in the slides, its a cluster and mem request can happen over a network... and yes future AMD HSA dGPUs will be special kind of NUMA for the "physical memory", but dependent on "Virtual Memory", that is the part that is "unified"... got it ?)

Fell free to disagree how much you like ( to me is utter pointless bickering).

 

[hcl123]

You are right that HSA is agnostic about the CPU. In principle HSA should work on x86, ARM, MIPS, Power etc..., but I was talking about crude reality. Look to the HSA members: only x86 and ARM. I doubt MIPS will join to HSA, because they are now positioning themselves in direct competition with both ARM and x86 (take a look to what they say about ARM in their talks LOL). I doubt still more that IBM will join, because IBM has just founded the PowerOpen consortium, whose first project will be the integration of Power8 with CUDA GPUs for heterogeneous compute.

I am not 100% sure about this, but I believe that Seattle is not HSA. The official roadmap only mentions that Berlin is HSA. I suppose that it is ARM who has to design a HSA ARM CPU and then license it to AMD.

MIPS has already joined the HSA almost since the very beginning... if you didn't know, MIPS is propriety of "Imagination" ( of PowerVR fame) since more than 1 year ago... almost at the same time HSA debuted, and "Imagination" is one of the founding members of HSA.

IBM is another story... but what better than a royalty free standard like HSA for a "open ISA/uarch" like OpenPower ?
(If it will ever join is a different matter)

 

[hcl123]

I doubt very much that AMD will port XOP to ARM, when AMD has eliminated XOP support from jaguar.

Doesn't mean much if anything... in a steady slow but inexorable "transition", the first designs to go, to be replaced by ARM equivalents, are exactly those Jaguar type of chips.

Sure XOP could never be ported *directly* to ARM, the "prefixes" defining instruction types are completely different, but the "opcodes" could be very similar, facilitating tremendously that way recompiling a program from x86 to ARM ( this is what i meant).

A x86 like pipe like Godson has, is different... its supposed to crunch x86 instructions unchanged by several tricks having the same effects if a real x86 pipe... AMD could *expand* this kind of functionality tremendously, since is the owner of x86_64 IP.. and without having to pay royalties or licensing from Intel.

In the end, what will be more important and pertinent is that in AMD ARM designs, the same "Cluster Multi-Threding" the same "vertical multi-threading" and the same modular approach could go to a ARM design... and x86 run natively on an ARM CPU/APU by binary translation, with a OS + WINE + Binary Translation + native x86, kind of infrastructure, if not kind of dual boot with a x86 OS like windows... akin to check-mate lol

 

[juanrga]

I am using the following definitions (from the above links):

Uniform Memory Access (UMA) is a shared memory architecture used in parallel computers. All the processors in the UMA model share the physical memory uniformly. In a UMA architecture, access time to a memory location is independent of which processor makes the request or which memory chip contains the transferred data. Uniform Memory Access computer architectures are often contrasted with Non-Uniform Memory Access (NUMA) architectures.

What if more than one processor in a MPU (example 4 cores) makes a memory request at the same time ?

Sure happens all the time... and the answer is; "its arbitrated" ... the IMC (Integrated Memory Controller) has queues for requests

You continue misunderstanding the definition. Imagine that processor A makes an early request to a given address and the access time is T ns, and processor B access time is T+D ns, with D being the delay due to the shared resource being busy with the other request. The inverse case is also possible: processor B making an early request to a given address and the access time being T ns, whereas it is T+D ns for processor A.

The point is that both A and B have equivalent access to memory in equivalent situations, because there is no topological distinction in the memory model. The possible delay caused to A or B by busy resources doesn't split the physical memory into local plus remote.

In any case all of this discussion is completely useless. I am using the standard definitions for UMA and NUMA. If you don't like the definitions it is better that you discuss this with the people who did them. I will continue using both definitions.

 

[juanrga]

You are right that HSA is agnostic about the CPU. In principle HSA should work on x86, ARM, MIPS, Power etc..., but I was talking about crude reality. Look to the HSA members: only x86 and ARM. I doubt MIPS will join to HSA, because they are now positioning themselves in direct competition with both ARM and x86 (take a look to what they say about ARM in their talks LOL). I doubt still more that IBM will join, because IBM has just founded the PowerOpen consortium, whose first project will be the integration of Power8 with CUDA GPUs for heterogeneous compute.

I am not 100% sure about this, but I believe that Seattle is not HSA. The official roadmap only mentions that Berlin is HSA. I suppose that it is ARM who has to design a HSA ARM CPU and then license it to AMD.

MIPS has already joined the HSA almost since the very beginning... if you didn't know, MIPS is propriety of "Imagination" ( of PowerVR fame) since more than 1 year ago... almost at the same time HSA debuted, and "Imagination" is one of the founding members of HSA.

IBM is another story... but what better than a royalty free standard like HSA for a "open ISA/uarch" like OpenPower ?
(If it will ever join is a different matter)

You are entirely right (and I was completely wrong), not only Imagination is already member but that are developing HSA MIPS CPUs. Thanks by the correction.

In the end, what will be more important and pertinent is that in AMD ARM designs, the same "Cluster Multi-Threding" the same "vertical multi-threading" and the same modular approach could go to a ARM design... and x86 run natively on an ARM CPU/APU by binary translation, with a OS + WINE + Binary Translation + native x86, kind of infrastructure, if not kind of dual boot with a x86 OS like windows... akin to check-mate lol

This is a very interesting idea. In fact, Nvidia secret plan was to run x86 on their forthcoming custom ARM chip using transmeta code-morphing technology. They had to abandon the original x86 project, because Intel did not license them x86. However, AMD has a x86 license. I don't know the details of the license and if this would allow to AMD to do something similar to what Nvidia tried. However, AMD has a big advantage: cross-licensing AMD64 to Intel.

 

[christoffe9311]

Does anyone know when the new steam roller is ready for coming out? I'm in a right situation, i decided to build a new pc back in April, half way through i got made redundant and had to leave my project. I have a brand new motherboard still in my pc that's not been used a gigabyte 990fxa-ud3 that needs a CPU, this is my situation, i don't want to be putting money into a pc that will be a old generation motherboard and cpu as soon a the steam roller comes out. I'm amusing the steam roller wont be am3+ so that's a new motherboard as well, the motherboard i have now that's not being used cost me full price over £112 if i leave it standing around until steam roller it will loose its value and be worth less than £50, so should i cut my losses and sell my motherboard while they are still pulling their money and abandon this build and wait for the new amd steam roller and the new line of motherboards or carry on with this build and risk loosing out lots of money by paying full rrp on soon to be old generation parts?

 

[de5_Roy]

I understand that some people here doesn't like facts. I can live with that.

What I find difficult to accept is when this same people pretends that the facts don't exist. I have given the AMD official roadmap together with AMD reasons why they are replacing jaguar servers with ARM servers. An official roadmap is not a "promo slide" nor "personal speculation". If you don't know how to read a roadmap or are confused about the meaning of blue/orange colours, that is not my problem.

you see, the facts you claim to exist, do not exist in the first place. your so-called official roadmap is actually a promo slide and the facts you're claiming are based on your personal speculation as you have repeatedly failed to provide an official statement.

[quote excerpts deleted because those are the parts that seems to fit your claim]

in reality they don't fit. none of these contain any official statement from amd or amd official. they're simply interpreting the promo slide.

The whole industry is applauding AMD plans and new strategy/products, even Anandtech!!! LOL This is what Anandtech said about AMD plans of replacing jaguar servers by arm servers:
<blah blah>

yeah, anandtech said that, not amd. geddit?

http://www.pcper.com/news/General-Tech/AMDs-plans-keep-their-ARMs-server-room
http://www.cpu-world.com/news_2013/2013061901_AMD_reveals_server_roadmap.html
http://forwardthinking.pcmag.com/architecture/312794-amd-pivots-to-arm-on-servers
http://www.anandtech.com/show/7079/amd-evolving-fast-to-survive-in-the-server-market-jungle-/5
I find amazing how many AMD hate has floured in this thread recently.

interesting links, especially the pcmag ones. but, before i go into those, i will present yet another result of parroting courtesy of cpu-world, with a before-after scenario:
Before, official roadmap:
http://www.cpu-world.com/news_2011/2011080201_AMD_2012_-_2013_server_roadmap.html
six months after, official roadmap:
http://www.cpu-world.com/news_2012/2012020302_AMD_server_roadmap_for_2012_2013.html
surprise!
to put it in perspective, the products mentioned in your 'official roadmap' are 6 months away since the promo slide came out, arm ones are further away.

in the pcmag links, it seemed like amd plans to introduce some enhanced microserver-specific technologies only into the seattle socs, which may make them more viable but they never disclose if they'd do the same with jaguar's successor or not. chances are amd wants to sell microservers solely based on arm v8 and nothing else. amd pretty much sidestepped the corporations' conservative tendencies and their habit of longer hardware refresh cycles. i mention this because anyone who'd buy a jaguar-based server will get a dead end machine unless amd introduces the arm v8 socs on the same hardware platform (due to my limited knowledge i don't know if this is possible).

amd hate? where in heck did you get that idea from? the main point was asking you to provide an official source or statement since you kept claiming those were facts, not speculations.

 

[noob2222]

I understand that some people here doesn't like facts. I can live with that.

What I find difficult to accept is when this same people pretends that the facts don't exist. I have given the AMD official roadmap together with AMD reasons why they are replacing jaguar servers with ARM servers. An official roadmap is not a "promo slide" nor "personal speculation". If you don't know how to read a roadmap or are confused about the meaning of blue/orange colours, that is not my problem.

you see, the facts you claim to exist, do not exist in the first place. your so-called official roadmap is actually a promo slide and the facts you're claiming are based on your personal speculation as you have repeatedly failed to provide an official statement.

[quote excerpts deleted because those are the parts that seems to fit your claim]

in reality they don't fit. none of these contain any official statement from amd or amd official. they're simply interpreting the promo slide.

The whole industry is applauding AMD plans and new strategy/products, even Anandtech!!! LOL This is what Anandtech said about AMD plans of replacing jaguar servers by arm servers:
<blah blah>

yeah, anandtech said that, not amd. geddit?

http://www.pcper.com/news/General-Tech/AMDs-plans-keep-their-ARMs-server-room
http://www.cpu-world.com/news_2013/2013061901_AMD_reveals_server_roadmap.html
http://forwardthinking.pcmag.com/architecture/312794-amd-pivots-to-arm-on-servers
http://www.anandtech.com/show/7079/amd-evolving-fast-to-survive-in-the-server-market-jungle-/5
I find amazing how many AMD hate has floured in this thread recently.

interesting links, especially the pcmag ones. but, before i go into those, i will present yet another result of parroting courtesy of cpu-world, with a before-after scenario:
Before, official roadmap:
http://www.cpu-world.com/news_2011/2011080201_AMD_2012_-_2013_server_roadmap.html
six months after, official roadmap:
http://www.cpu-world.com/news_2012/2012020302_AMD_server_roadmap_for_2012_2013.html
surprise!
to put it in perspective, the products mentioned in your 'official roadmap' are 6 months away since the promo slide came out, arm ones are further away.

in the pcmag links, it seemed like amd plans to introduce some enhanced microserver-specific technologies only into the seattle socs, which may make them more viable but they never disclose if they'd do the same with jaguar's successor or not. chances are amd wants to sell microservers solely based on arm v8 and nothing else. amd pretty much sidestepped the corporations' conservative tendencies and their habit of longer hardware refresh cycles. i mention this because anyone who'd buy a jaguar-based server will get a dead end machine unless amd introduces the arm v8 socs on the same hardware platform (due to my limited knowledge i don't know if this is possible).

amd hate? where in heck did you get that idea from? the main point was asking you to provide an official source or statement since you kept claiming those were facts, not speculations.

Pretty much, all the speculation aka "Facts" are based on the pretty picture and not what AMD said.

“Our strategy is to differentiate ourselves by using our unique IP to build server processors that are particularly well matched to a target workload and thereby drive down the total cost of owning servers. This strategy unfolds across both the enterprise and data centers and includes leveraging our graphics processing capabilities and embracing both x86 and ARM instruction sets,” said Andrew Feldman, general manager of the Server Business Unit, AMD. “AMD led the world in the transition to multicore processors and 64-bit computing, and we intend to do it again with our next-generation AMD Opteron families.”
...
Seattle” will be the industry’s only 64-bit ARM-based server SoC from a proven server processor supplier. “Seattle” is an 8- and then 16-core CPU based on the ARM Cortex-A57 core and is expected to run at or greater than 2 GHz. The “Seattle” processor is expected to offer 2-4X the performance of AMD’s recently announced AMD Opteron X-Series processor with significant improvement in compute-per-watt. It will deliver 128GB DRAM support, extensive offload engines for better power efficiency and reduced CPU loading, server caliber encryption, and compression and legacy networking including integrated 10GbE. It will be the first processor from AMD to integrate AMD’s advanced Freedom™ Fabric for dense compute systems directly onto the chip. AMD plans to sample “Seattle” in the first quarter of 2014 with production in the second half of the year.

hmm ... x86 and arm, not one or the other. The funny part is the fact that its going to take 16 arm cores to be 2x-4x as fast as a 4-core x86 jaguar. sounds pretty un-impressive.

The only AMD hate is stemming from AMD becoming a half support cpu company according to "facts". AMD is abandoning high end completely, concentrating on low-middle end APUs, and now low end is going even lower end. What is AMD going to be able to sell anyone who wants to upgrade?

Noting according to someone's "facts".

If AMD goes fully down this speculated "fact" road, they are done. No one is going to buy their APUs in hope that 2-3 years from now software will support its functionality.

 

[juanrga]

Before, official roadmap:
http://www.cpu-world.com/news_2011/2011080201_AMD_2012_-_2013_server_roadmap.html
six months after, official roadmap:
http://www.cpu-world.com/news_2012/2012020302_AMD_server_roadmap_for_2012_2013.html
surprise!

No surprise here. Just AMD changed the official roadmaps and the strategy. This happens to all companies. But that is different from pretending that the existent official roadmap for 2014 doesn't say what it says.

 

[juanrga]

Does anyone know when the new steam roller is ready for coming out? I'm in a right situation, i decided to build a new pc back in April, half way through i got made redundant and had to leave my project. I have a brand new motherboard still in my pc that's not been used a gigabyte 990fxa-ud3 that needs a CPU, this is my situation, i don't want to be putting money into a pc that will be a old generation motherboard and cpu as soon a the steam roller comes out. I'm amusing the steam roller wont be am3+ so that's a new motherboard as well, the motherboard i have now that's not being used cost me full price over £112 if i leave it standing around until steam roller it will loose its value and be worth less than £50, so should i cut my losses and sell my motherboard while they are still pulling their money and abandon this build and wait for the new amd steam roller and the new line of motherboards or carry on with this build and risk loosing out lots of money by paying full rrp on soon to be old generation parts?

AMD says Steamroller comes in first quarter 2014 in APU form for the new FM2+ platform. AMD has not said anything about Steamroller CPU coming for AM3+.

 

[de5_Roy]

No surprise here. Just AMD changed the official roadmaps and the strategy. This happens to all companies. But that is different from pretending that the existent official roadmap for 2014 doesn't say what it says.

sigh.. even the only promo slide that you keep refering to does not mention anything about replacing. it just shows some image, leaving the text up to reader/audience's imagination.
i'll bet (with 100% certainty) that at the end of the slide show that the slide (you keep refering to) is from, there is a concluding slide that contains a clearly written official statement (and the only 'official' part of the slideshow) from amd that everything presented previously may be misinformation - leading to the very, very strong impression that amd's promo slides are just enthusiastic projections full of forward-looking speculation. not facts.

in reality, no job-loving executive will say something like this without answering about the massive transition process and everything associated with it. if someone from amd has said something - that's what i want to see.

 

[juanrga]

Pretty much, all the speculation aka "Facts" are based on the pretty picture and not what AMD said.

“Our strategy is to differentiate ourselves by using our unique IP to build server processors that are particularly well matched to a target workload and thereby drive down the total cost of owning servers. This strategy unfolds across both the enterprise and data centers and includes leveraging our graphics processing capabilities and embracing both x86 and ARM instruction sets,” said Andrew Feldman, general manager of the Server Business Unit, AMD. “AMD led the world in the transition to multicore processors and 64-bit computing, and we intend to do it again with our next-generation AMD Opteron families.”
... <---- HERE GOES THE PART YOU SNIPPED
Seattle” will be the industry’s only 64-bit ARM-based server SoC from a proven server processor supplier. “Seattle” is an 8- and then 16-core CPU based on the ARM Cortex-A57 core and is expected to run at or greater than 2 GHz. The “Seattle” processor is expected to offer 2-4X the performance of AMD’s recently announced AMD Opteron X-Series processor with significant improvement in compute-per-watt. It will deliver 128GB DRAM support, extensive offload engines for better power efficiency and reduced CPU loading, server caliber encryption, and compression and legacy networking including integrated 10GbE. It will be the first processor from AMD to integrate AMD’s advanced Freedom™ Fabric for dense compute systems directly onto the chip. AMD plans to sample “Seattle” in the first quarter of 2014 with production in the second half of the year.

hmm ... x86 and arm, not one or the other. The funny part is the fact that its going to take 16 arm cores to be 2x-4x as fast as a 4-core x86 jaguar. sounds pretty un-impressive.

You snipped the central paragraph from the official presentation found in AMD site. I reproduce below the paragraph that you gave and the one that you snipped:

“Our strategy is to differentiate ourselves by using our unique IP to build server processors that are particularly well matched to a target workload and thereby drive down the total cost of owning servers. This strategy unfolds across both the enterprise and data centers and includes leveraging our graphics processing capabilities and embracing both x86 and ARM instruction sets,” said Andrew Feldman, general manager of the Server Business Unit, AMD. “AMD led the world in the transition to multicore processors and 64-bit computing, and we intend to do it again with our next-generation AMD Opteron families.”

In 2014, AMD will set the bar in power-efficient server compute with the industry’s premier ARM server CPU. The 64-bit CPU, code named “Seattle,” is based on ARM Cortex™-A57 cores and is expected to provide category-leading throughput as well as setting the bar in performance-per-watt. AMD will also deliver a best-in-class APU, code named “Berlin.” “Berlin” is an x86 CPU and APU, based on a new generation of cores named “Steamroller.” Designed to double the performance of the recently available “Kyoto” part, “Berlin” will offer extraordinary compute-per-watt that will enable massive rack density. The third processor announced today is code named “Warsaw,” AMD’s next-generation 2P/4P offering. It is optimized to handle the heavily virtualized workloads found in enterprise environments including the more complex compute needs of data analytics, xSQL and traditional databases. “Warsaw” will provide significantly improved performance-per-watt over today’s AMD Opteron™ 6300 family.

When Andrew Feldman said "both x86 and ARM instruction sets” he did mean Berlin (x86), Warsaw (x86), and Seattle (ARM). He didn't say that ARM Seattle will be sold together with the old jaguar based Opteron X. He couldn't because the Opteron X (x86) will be replaced by the new Seattle (ARM) as the official roadmap and presentation show.

The 2x as fast as a 4-core jaguar refers to the 8-core version of Seattle (not to the 16-core as you pretend), whereas the 4x as fast refers to the 16-core Seattle. And Seattle achieves that raw performance with lower power consumption than jaguar. That is the reason why jaguar will be replaced with ARM in 2014. Moreover the new Seattle servers have other advantages over the jaguar servers.

The industry has praised AMD plans/strategy/products. Even Anandtech has praised Seattle; as shown above, they consider Seattle to be impressive and probably the best design in the market. However, your "sounds pretty un-impressive" is another example of the AMD hate that exists in this thread.

 

[8350rocks]

Pretty much, all the speculation aka "Facts" are based on the pretty picture and not what AMD said.

“Our strategy is to differentiate ourselves by using our unique IP to build server processors that are particularly well matched to a target workload and thereby drive down the total cost of owning servers. This strategy unfolds across both the enterprise and data centers and includes leveraging our graphics processing capabilities and embracing both x86 and ARM instruction sets,” said Andrew Feldman, general manager of the Server Business Unit, AMD. “AMD led the world in the transition to multicore processors and 64-bit computing, and we intend to do it again with our next-generation AMD Opteron families.”
... <---- HERE GOES THE PART YOU SNIPPED
Seattle” will be the industry’s only 64-bit ARM-based server SoC from a proven server processor supplier. “Seattle” is an 8- and then 16-core CPU based on the ARM Cortex-A57 core and is expected to run at or greater than 2 GHz. The “Seattle” processor is expected to offer 2-4X the performance of AMD’s recently announced AMD Opteron X-Series processor with significant improvement in compute-per-watt. It will deliver 128GB DRAM support, extensive offload engines for better power efficiency and reduced CPU loading, server caliber encryption, and compression and legacy networking including integrated 10GbE. It will be the first processor from AMD to integrate AMD’s advanced Freedom™ Fabric for dense compute systems directly onto the chip. AMD plans to sample “Seattle” in the first quarter of 2014 with production in the second half of the year.

hmm ... x86 and arm, not one or the other. The funny part is the fact that its going to take 16 arm cores to be 2x-4x as fast as a 4-core x86 jaguar. sounds pretty un-impressive.

And you snipped the following paragraph from AMD official claim. I reproducer both the one that yu gave and the one that you snipped:

“Our strategy is to differentiate ourselves by using our unique IP to build server processors that are particularly well matched to a target workload and thereby drive down the total cost of owning servers. This strategy unfolds across both the enterprise and data centers and includes leveraging our graphics processing capabilities and embracing both x86 and ARM instruction sets,” said Andrew Feldman, general manager of the Server Business Unit, AMD. “AMD led the world in the transition to multicore processors and 64-bit computing, and we intend to do it again with our next-generation AMD Opteron families.”

In 2014, AMD will set the bar in power-efficient server compute with the industry’s premier ARM server CPU. The 64-bit CPU, code named “Seattle,” is based on ARM Cortex™-A57 cores and is expected to provide category-leading throughput as well as setting the bar in performance-per-watt. AMD will also deliver a best-in-class APU, code named “Berlin.” “Berlin” is an x86 CPU and APU, based on a new generation of cores named “Steamroller.” Designed to double the performance of the recently available “Kyoto” part, “Berlin” will offer extraordinary compute-per-watt that will enable massive rack density. The third processor announced today is code named “Warsaw,” AMD’s next-generation 2P/4P offering. It is optimized to handle the heavily virtualized workloads found in enterprise environments including the more complex compute needs of data analytics, xSQL and traditional databases. “Warsaw” will provide significantly improved performance-per-watt over today’s AMD Opteron™ 6300 family.

When Andrew Feldman said "both x86 and ARM instruction sets” he did mean Berlin (x86) Warsaw (x86) and Seattle (ARM), not that ARM Seattle will be sold togheter with the old replaced jagiuar based Opteron X.

The funny part is that you are plain wrong again. The 2x as fast as a 4-core jaguar refers to the 8-core version of Seattle (not to 16-core as you pretend), whereas the 4x as fast refers to the 16-core Seattle. And Seattle achieves that performance with lower power consumption than jaguar. Moreover Seattle has other advantages over the jaguar servers.

Even Anandtech has praised Seattle. However your "sounds pretty un-impressive" is another example of the AMD hate mentioned before. LOL

Juan, as a game designer, and someone who will be hosting servers for a MMO soon in our office, this is my outlook:

1.) I would do a SeaMicro server with 16 core opterons for the backend of the MMO servers. ARM will not solve that issue for me in any way, shape or form without serious issues to contend with.

2.) The 8350 I currently run in my home PC for gaming/productivity applications will not be sufficiently replaced by any APU short of something 6 cores or more, and I would still run a dGPU over the iGPU because I simply need the horsepower a dGPU offers.

3.) Software will not be fully integrated for HSA for some time to come. We are still not fully transitioned to 64 bit software in Windows!!! It's been 10 years since that came around, and Intel supports that technology too...how long do you think it will take them to integrate something that AMD supports and not Intel? 15 years? 20?

4.) HSA/hUMA is an impressive step forward, ahead of it's time for sure. Though I tend to run toward thinking it may be too fully embraced, too far ahead of it's time for this to be successful without a medium/high end x86 CPU with 6+ cores.

In short, they're abandoning established markets where they have fair (and increasing) market share, especially in certain niches...all for the sake of chasing an emerging market that may take years to come to fruition. Additionally, they're abandoning a LARGE chunk of their customer base without support, if they offer no successor to the FX line.

 

[gamerk316]

Physx will not run on AMD GPUs. Any software implementations would be at cost of the efficiency of the hardware. AMD doesn't need a physics engine since its up to the devs to either implement one or license one. For a dev to choose to use physx is unimaginable in the current situation. If nvidia could have just done gpu physx in software, why wasn't it in the ps3 or 360?

Are you sure? What about those benchmarks where they used PhysX for AMD GPU's to test it vs OGL?

NVIDIA choose to implement PhysX via a CUDA, which AMD doesn't support. For AMD to support PhysX, they would have to support CUDA, which they will NEVER do. Likewise, because NVIDIA owns the license, AMD can't get access to the source to re-implement via DirectCompute.

Older drivers you could force a split AMD-NVIDIA configuration, with AMD as the main GPU, and NVIDIA doing CUDA, but NVIDIA now does a driver check to prevent that.

 

[gamerk316]

My thoughts here run to the following:

1.) 30% is a massive number to just up and abandon...especially for a company the size of AMD. I anticipate that they know this, and are likely working out a solution to keep those folks happy. Rory Read is from IBM originally, if he has half the training I know he received from IBM in market analysis...he knows they need that market share.

Vendor lock in isn't as big a problem as it used to be. In short: Going AMD once doesn't tie you to them, so focusing on Market Share, rather then margins, would be a mistake.

4.) It will take AMD a while to gain a foot hold worth mentioning in ARM SoCs. Primarily because there is no 64 bit software available at this time. Anyone recall how long the transition took for AMD64 to really take over the market...? (In case you're wondering, there are still 32 bit Windows versions only 10 years later now....)

No ARM software either. Hence the main problem: X86-64, SPARC, PPC, MIPS, and ARM. Too many vendors, and I simply don't see the performance out of ARM to be competitive in this space. I view ARM as the eventual replacement of PPC in the embedded world, but don't see it making a real impact in server-land.

 

[gamerk316]

In the end, what will be more important and pertinent is that in AMD ARM designs, the same "Cluster Multi-Threding" the same "vertical multi-threading" and the same modular approach could go to a ARM design... and x86 run natively on an ARM CPU/APU by binary translation, with a OS + WINE + Binary Translation + native x86, kind of infrastructure, if not kind of dual boot with a x86 OS like windows... akin to check-mate lol

As someone who's worked on emulators: Binary Translation = Slow as ****. Nevermind the need to be cycle accurate on every instruction/memory access to avoid screwing up the CPU state royally. And X86 goes out of its way to make this impossible to do.

For example: X86 processors since the Pentium 3 have been capable of doing multiple instructions at a time. Here's the hitch: The actual algorithm that does this is still not known outside of Intel/AMD. So, when emulating a SPECIFIC X86 CPU, even on ANOTHER X86 CPU, you are basically reduced to running one instruction at a time to ensure you run things in the right order, killing your throughput. And this would need to be emulated on an ARM Instruction Set, running a totally different scheduling algorithm.

Good luck with that. Short of adding a low-power X86 core (which defeats the purpose of going ARM in the first place!), you aren't getting X86 software running on ARM, or any other instruction set.

 

[gamerk316]

http://www.nextpowerup.com/news/2378/playstation-4-packs-weak-single-threaded-amd-apu-powerful-multi-threaded-performance.html

This basically means now game developers are going to have to code for weak cores to. I hope the quality of physics don't go down because of it. I still don't get why amd release so slow per core cpus. I think it wil be a nightmare to code games on these consoles if the cpu is at 1.6ghz as its like the equivalent of a hypothetical intel sytem with 8 cores running at 800mhz performance wise.

Not going to happen. Weak CPU? Set settings to low. That's why we have multiple graphics settings, and why C2D's @ 2.4 are still the typical minimum requirement.

 

[juanrga]

Juan, as a game designer, and someone who will be hosting servers for a MMO soon in our office, this is my outlook:

1.) I would do a SeaMicro server with 16 core opterons for the backend of the MMO servers. ARM will not solve that issue for me in any way, shape or form without serious issues to contend with.

2.) The 8350 I currently run in my home PC for gaming/productivity applications will not be sufficiently replaced by any APU short of something 6 cores or more, and I would still run a dGPU over the iGPU because I simply need the horsepower a dGPU offers.

3.) Software will not be fully integrated for HSA for some time to come. We are still not fully transitioned to 64 bit software in Windows!!! It's been 10 years since that came around, and Intel supports that technology too...how long do you think it will take them to integrate something that AMD supports and not Intel? 15 years? 20?

4.) HSA/hUMA is an impressive step forward, ahead of it's time for sure. Though I tend to run toward thinking it may be too fully embraced, too far ahead of it's time for this to be successful without a medium/high end x86 CPU with 6+ cores.

In short, they're abandoning established markets where they have fair (and increasing) market share, especially in certain niches...all for the sake of chasing an emerging market that may take years to come to fruition. Additionally, they're abandoning a LARGE chunk of their customer base without support, if they offer no successor to the FX line.

1) The 16 core Warsaw is released for legacy users. Maybe you would be choosing that.

2) Enthusiast 8350 users can update to 9000 series.

3) Software requiring more memory was scarce and, thus, the need to upgrade to 64bit almost non-existent. Moreover the 64bit was pushed by AMD alone, which is too small to force fast adoption by the market. And Microsoft controlled the market. Things are different with HSA. Enabling HSA you can see a 2x-8x boost in many applications. Companies enabling HSA in their software will be on big advantage over others. Moreover, AMD founded the HSA foundation to push the market to adopt it. And Microsoft is no more controlling the market, in fact Windows 8 is almost not used: Linux (Android) and Unix are controlling the market now. Didn't you read Lisa Su official claim that AMD has broken exclusivity with Windows?

http://gigaom.com/2013/06/05/sorry-microsoft-amd-cheats-on-windows-plans-a-fling-with-chrome-and-android/

Why do you believe AMD has ready a HSA enabled linux kernel? Why do you believe AMD has joined OpenSUSE and others to develop HSA enabled software?

4) The mainstream market is in 2/4 cores. The number of users purchasing 6/8 cores is very small.

As I already said before Steam statistics show that the number of 8350 users is of less than a 0.3% of the total. I know a little the local market here and the immense majority of buyers prefer a quad A10 over an FX-8000 series.

http://www.nextpowerup.com/news/2378/playstation-4-packs-weak-single-threaded-amd-apu-powerful-multi-threaded-performance.html

This basically means now game developers are going to have to code for weak cores to. I hope the quality of physics don't go down because of it. I still don't get why amd release so slow per core cpus. I think it wil be a nightmare to code games on these consoles if the cpu is at 1.6ghz as its like the equivalent of a hypothetical intel sytem with 8 cores running at 800mhz performance wise.

Nonsense. Those weak cores are so powerful but more efficient than anything Intel could offer at the time. Only the next generation of Intel Atoms will be competing with jaguar cores.

A jaguar core at 2GHz offers near the same performance than an SB i3 core at 2.5GHz. Evidently one jaguar core at 1.6 GHz will be not so powerful as one Piledriver core at 4GHz neither as one IB core at 3.5GHz. That is evident. But guess what? the PS4/XBox1 have 8 cores of those, not only one core.

Some developers are accustomed to coding for few cores (some are simply dumb or incompetent), but others prefer "moar cores". John Carmack has shared his desire of 16 core CPUs for gaming. Another dev that prefers "moar cores" is one of the Metro Last Night developers:

Digital Foundry: Let's talk about next-gen console. What's your take on the general design in terms of CPU and graphics processing power?

Oles Shishkovstov: We are talking PS4, right? I am very excited about both CPU and GPU. Jaguar is a pretty well-balanced out-of-order core and there are eight of them inside. I always wanted a lot of relatively-low-power cores instead of single super-high-performance one, because it's easier to simply parallelise something instead of changing core-algorithms or chasing every cycle inside critical code segment (not that we don't do that, but very often we can avoid it).

As explained to you before, the physics is expected to be run in the GPU. This is not anything revolutionary. In fact existent PC games such as Metro or Batman offer the better results when PhysX is run in the GPU (PCs with Nvidia GPUs).

 

[noob2222]

Pretty much, all the speculation aka "Facts" are based on the pretty picture and not what AMD said.

“Our strategy is to differentiate ourselves by using our unique IP to build server processors that are particularly well matched to a target workload and thereby drive down the total cost of owning servers. This strategy unfolds across both the enterprise and data centers and includes leveraging our graphics processing capabilities and embracing both x86 and ARM instruction sets,” said Andrew Feldman, general manager of the Server Business Unit, AMD. “AMD led the world in the transition to multicore processors and 64-bit computing, and we intend to do it again with our next-generation AMD Opteron families.”
... <---- HERE GOES THE PART YOU SNIPPED
Seattle” will be the industry’s only 64-bit ARM-based server SoC from a proven server processor supplier. “Seattle” is an 8- and then 16-core CPU based on the ARM Cortex-A57 core and is expected to run at or greater than 2 GHz. The “Seattle” processor is expected to offer 2-4X the performance of AMD’s recently announced AMD Opteron X-Series processor with significant improvement in compute-per-watt. It will deliver 128GB DRAM support, extensive offload engines for better power efficiency and reduced CPU loading, server caliber encryption, and compression and legacy networking including integrated 10GbE. It will be the first processor from AMD to integrate AMD’s advanced Freedom™ Fabric for dense compute systems directly onto the chip. AMD plans to sample “Seattle” in the first quarter of 2014 with production in the second half of the year.

hmm ... x86 and arm, not one or the other. The funny part is the fact that its going to take 16 arm cores to be 2x-4x as fast as a 4-core x86 jaguar. sounds pretty un-impressive.

You snipped the central paragraph from the official presentation found in AMD site. I reproduce below the paragraph that you gave and the one that you snipped:

“Our strategy is to differentiate ourselves by using our unique IP to build server processors that are particularly well matched to a target workload and thereby drive down the total cost of owning servers. This strategy unfolds across both the enterprise and data centers and includes leveraging our graphics processing capabilities and embracing both x86 and ARM instruction sets,” said Andrew Feldman, general manager of the Server Business Unit, AMD. “AMD led the world in the transition to multicore processors and 64-bit computing, and we intend to do it again with our next-generation AMD Opteron families.”

In 2014, AMD will set the bar in power-efficient server compute with the industry’s premier ARM server CPU. The 64-bit CPU, code named “Seattle,” is based on ARM Cortex™-A57 cores and is expected to provide category-leading throughput as well as setting the bar in performance-per-watt. AMD will also deliver a best-in-class APU, code named “Berlin.” “Berlin” is an x86 CPU and APU, based on a new generation of cores named “Steamroller.” Designed to double the performance of the recently available “Kyoto” part, “Berlin” will offer extraordinary compute-per-watt that will enable massive rack density. The third processor announced today is code named “Warsaw,” AMD’s next-generation 2P/4P offering. It is optimized to handle the heavily virtualized workloads found in enterprise environments including the more complex compute needs of data analytics, xSQL and traditional databases. “Warsaw” will provide significantly improved performance-per-watt over today’s AMD Opteron™ 6300 family.

When Andrew Feldman said "both x86 and ARM instruction sets” he did mean Berlin (x86), Warsaw (x86), and Seattle (ARM). He didn't say that ARM Seattle will be sold together with the old jaguar based Opteron X. He couldn't because the Opteron X (x86) will be replaced by the new Seattle (ARM) as the official roadmap and presentation show.

The 2x as fast as a 4-core jaguar refers to the 8-core version of Seattle (not to the 16-core as you pretend), whereas the 4x as fast refers to the 16-core Seattle. And Seattle achieves that raw performance with lower power consumption than jaguar. That is the reason why jaguar will be replaced with ARM in 2014. Moreover the new Seattle servers have other advantages over the jaguar servers.

The industry has praised AMD plans/strategy/products. Even Anandtech has praised Seattle; as shown above, they consider Seattle to be impressive and probably the best design in the market. However, your "sounds pretty un-impressive" is another example of the AMD hate that exists in this thread.

first off where does he say opteron x is being replaced? .... where ... Your putting words into his statement, period.

2nd, you really think arm cortex A-57 is going to be just as powerful as an x86 core? good luck, they need a boost of about 85% to achieve jaguar speeds.

As far as others praising AMD for making ARM servers, yes, thats good. Not at the cost of other faster products being canned, thats the stupid part, one you seem to pretend is a genious move. Thats the reason Intel hasn't adopted ARM. Its supposedly super efficient but extremely slow.

Im not so easily convinced that ARM is the solution to the world's problems. No one really posts what the power consumption is on ARM devices, they just try to convince you its the best thing on the planet.

http://www.tomshardware.com/reviews/atom-z2760-power-consumption-arm,3387-5.html

 

[8350rocks]

juangra are you really being serious or joking. Anyway you do realise an i3 3220 ivb is faster then the 8 core console cpu don't you. My estimation is a 2ghz jaguar core is equivalent to an intel sb cpu at like 1ghz. I think this explains why an i3 3220 is faster then an 8 core jaguar at 1.6ghz as 4 cores score 1.6 in cinebench double that to get 8 cores at 1.6ghz=3.2 score vs 3.3 for i3 3220. Oh and the next generations of atom cores will be about 2-3x better performance per watt then amd jaguar.
http://gamrconnect.vgchartz.com/thread.php?id=159069

Thanks, but the comic relief is getting old now...until you learn a new song and dance, I am afraid the crowd is just going to throw tomatoes.

Once you can have an intelligent discussion about processors and how they work...try again. Until then, you're just a troll who's fishing with flame bait.

 

[noob2222]

If you don't want an intelligent discussion then why post. I am the most knowledgeable person on here. Some of the crazy stuff I have seen on this thread that are not even close to reality to put amd in a good light is quite amusing as you amd fanboys really believe this. I mean a gem I saw once here 10-15% ipc improvement and steamroller is on par with sandy or ivy bridge ipc wise when they need 80% to get near sandy bridge letalone ivy bridge which is 100% better.

HAHAHAHAH!!!!!!

thats a good one little boy blue who cried Intel.

You have been shown over and over and over, yet refuse to accept anyting other than stupid pi as a benchmark, thinking that its a great benchmark tool. That right there shows everyone how knowledgeable you are.

 

[griptwister]

If you don't want an intelligent discussion then why post. I am the most knowledgeable person on here.

You've got to be sh***ing me.

 

[juanrga]

juangra are you really being serious or joking. Anyway you do realise an i3 3220 ivb is faster then the 8 core console cpu don't you. My estimation is a 2ghz jaguar core is equivalent to an intel sb cpu at like 1ghz. I think this explains why an i3 3220 is faster then an 8 core jaguar at 1.6ghz as 4 cores score 1.6 in cinebench double that to get 8 cores at 1.6ghz=3.2 score vs 3.3 for i3 3220. Oh and the next generations of atom cores will be about 2-3x better performance per watt then amd jaguar.
http://gamrconnect.vgchartz.com/thread.php?id=159069

But I was discussing real world performance, not incorrect estimations that are only in your imagination. From your own link benchmarks:

ST:
4x1 jaguar @ 2Ghz: 403.8
2x2 sandy @ 2.5Ghz: 495.5

MT:
4x1 jaguar @ 2Ghz: 840.7
2x2 sandy @ 2.5Ghz: 749.6

Which agrees with what I said:

A jaguar core at 2GHz offers near the same performance than an SB i3 core at 2.5GHz.

The problem of the Intel 3220 is not that it is slower than the consoles CPU. The real problem is that the 3220 is very inefficient and power hungry. Even the 3220T is still inefficient and power hungry.

Your pretensions about next atom efficiency are just another pile of pure nonsense.

first off where does he say opteron x is being replaced? .... where ... Your putting words into his statement, period.

2nd, you really think arm cortex A-57 is going to be just as powerful as an x86 core? good luck, they need a boost of about 85% to achieve jaguar speeds.

It is said in the official roadmap attached to the official presentation. Also the roadmap says that Berlin and Seattle are 28nm, whereas Warsaw is 32nm. Does Feldman say the node size used for each? No, because that info is already available in the roadmap. I don't want he repeating as a parrot what I can ready by myself, I want him giving me extra information not contained in the roadmap.

What part of the new Seattle servers will be about 2x-4x faster than the Opteron X you don't get?

EDIT: I will say you more. There exists a slide that AMD has maintained under embargo until recently, where AMD clearly states its firm prediction on that ARM will dominate the world. AMD is releasing Warsaw chips for the institutional customers that will take longer to migrate to an ARM ecosystem, not because AMD believes that big x86 cores have some future...

 

[christoffe9311]

http://www.bit-tech.net/hardware/2012/11/06/amd-fx-8350-review/4
gimp uses 1 core so it shows this advantage. Thats why in some games you see amd cpu get like 70fps at a higher clock then an intel cpu that gets like 130fps as the double ipc is in work.

edit: find realworld tasks that use 1 core that intel is not 2x faster with ivy bridge or haswell at same clock. Prove me wrong then if you think I am wrong.

Both companies have their own advantages and dis advanatges, amd's cores arent as powerful as intels core but intel cant be overclocked as well as the fx line from amd which also has 8 cores. The amd 8350 can be taken upto 5ghz before is becomes unstable maxing around 55-60c where the 3570k can be taken upto 4.2ghz on the same cooling reaching 77c and they both tie neck and neck with each other considering the fx 8350 has 8 cores running at 5ghz on the overclock the temps are very impresive comparing it to the i5 in terms of cores, voltage and temps. You dont need powerful cores when you a more weakers ones works like a nest of wasps weak and harmless one their own but when they come together they form something very impresive.