AMD CPU speculation... and expert conjecture

[juanrga]

The facts about AMD replacing x86 jaguar for ARM were given before. AMD official server roadmap and the presentation explaining why are replacing jaguar servers by arm servers. Also AMD quote, given above, claiming that they expect the sales of ARM servers to surpase the sales of x86 servers. Do you believe AMD will invest lots of resources in future x86 servers when is predicting that sales will drop?

facts given before where? i am not going back searching through walls of text. if you have official proof of amd stating they're replacing jaguar with arm, please provide. pretty please.
what server roadmap? the last one i saw showed coexistence. if there's a different one, i'd like to see that.

The facts about AMD replacing x86 jaguar for ARM were given before. AMD official server roadmap and the presentation explaining why are replacing jaguar servers by arm servers. Also AMD quote, given above, claiming that they expect the sales of ARM servers to surpase the sales of x86 servers. Do you believe AMD will invest lots of resources in future x86 servers when is predicting that sales will drop?

So these facts are your own speculation? nice ... I wonder what "facts" I can make up.

Unless the a57 is 85% faster than the a15 ... replacing jaguar would fail.

AMD official roadmap showing how the existent jaguar-based servers will be substituted by ARM-based servers next year:

It's hardly a secret that AMD has stepped out of its x86 comfort zone to develop an ARM-based server chip

In case you don't know, the existent jaguar-based servers (Opteron X) are more faster than the jaguar-based Kabini (1.5GHz) that you mention above. But:

AMD says its 64-bit ARM server chips will be up to four times faster and more power-efficient than its latest quad-core x86 Opteron X-series chips, which draw up to 11 watts of power.

AMD is also saying us what they expect about their server portfolio with ARM (Seattle) and x86 (Berlin and Warsaw) solutions:

AMD will sell the 64-bit ARM processors alongside x86 server processors, but the company expects ARM to outsell x86 in the long run.

http://www.engadget.com/2013/06/18/amd-seattle-arm-server-chip/
http://www.pcworld.com/article/2048069/amd-to-ship-64bit-arm-chips-to-server-makers-next-year.html

 

[juanrga]

Better then anything amd have, then again we all hear amd are cheap when they are to below average to sell for any more. AMD ain't stupid, they have to sell at there prices they do otherwise they would be bankrupt as everyone would buy intel instead. intel baytrail easily beats the jaguar cpus, looks like steamroller cores are only amd hope as it seems one sided at the moment in intels favour.

Besides your love by compare prices of brand new AMD chips with second hand Intel chips, your spreading of false information about "baytrail" was broadly corrected by two posters in the other thread. But, of course, you are ignoring the corrections and posting the same nonsense here. LOL

 

[juanrga]

^^ thats 2.4 ghz turbo blue boy and 1.5 ghz isn't the fastest jaguar available. Bay trail will probably show up in time to compete with puma+

Don't worry He knows that, because two posters (including mysefl) said the same in the other thread, where he posted the same nonsense. It is more, in the other thread he tried to convince us that the new atom is a 2w TDP. It was very funny.

 

[juanrga]

As I said before an uniform memory access model provides a single access time to physical memory. A non-uniform memory access provides two times: one for local memory and other for remote memory. This is standard stuff.

In the slide, AMD clearly represent a single physical memory pool. This works for kaveri APU, but will not work for dGPUs. Once again, the HSA specification clearly states that its memory model is unified. There is not a single mention to uniform or hUMA in it.

If we want now reinvent the words and give them other meanings then the confusion is guaranteed.

The confusion is yours, there isn't such ting as single access time

UMA => single access time (uniform access).
NUMA => multiple access times (non-uniform access).

http://en.wikipedia.org/wiki/Non-Uniform_Memory_Access
http://en.wikipedia.org/wiki/Uniform_Memory_Access

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).

All very correct, but is like that by the "arbitration" scheme, which is local then remote, which invariability leads to mem "hops" and different access times for NUMA. But what defines NUMA is address spaces not access time.. meaning different "physical" address spaces in an hierarchical view (kind of "single" view but not linear unified) , which invariably leads to different access times.

Even in Numa access times can vary substantially according to the interconnect of the different mem pools... but is not because of that that there must be a different nomenclature or different scheme based on the "amount of time" .

Accept it, the distinctive feature is not access times, those are consequences, not the determinant factor. NUMA and more ccNUMA are coherent physical memory in an hierarchical view... and in here virtual memory becomes orthogonal, cache coherency is not dependent on VM.

hUMA is different for not saying the opposite... the coherent unified view is "Virtual Memory" and the cache coherency is somehow dependent of this view. (first make VM an unified view, even for "devices", then attach cache coherency on top of that).

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.

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.

 

[juanrga]

I suspect the HSA foundation includes some kind of hidden clause/agreement. Traditional ARM vendors will focus in mobile, whereas AMD will focus in servers and probably HPC, but with ARM instead of x86.

In any case, the facts are that AMD is replacing x86 jaguar based servers by ARM based servers, and has said that they expect ARM sales to outperform x86 sales in the long run (2016?). This indicates, at least to me, that AMD will be not releasing high-end x86 server chips in the future. And I suspect that AMD assault to the HPC market will be made with HSA ARM designs.

In fact, I am almost sure that I will choose HSA ARM in the poll about what architecture will dominate the world. Apparently Nvidia and others think something similar.

I can't find any "hidden clause" not even slightly suggested on HSA.

HSA is not about ARM + x86 or something like that... HSA is about an Virtual Machine with its own kind of ISA, which is called HSAIL. HSA programs even have its own "binary format" called BRIG...

http://www.slideshare.net/hsafoundation/hsail-final-11junepptx

HSA should work on x86, ARM, MIPS, Power etc... only needs is the runtime definitions ported to those CPU or LU (latency units) ISAs. See it like Java... if the runtime/JVM (java virtual machine) itself is ported to those CPU ISAs, the same Java program can run "unchanged" in a x86, ARM, MIPS or Power system without issues... The same with HSA BRIG program, only HSA is more low level, it can even have JVM on top of it, or like in the "Sumatra enable JVM" the HSA runtime is embedded in the JVM itself.

HSA is a SIMT( single instruction multi-thread) mostly JIT paradigm... to go for the GPU (though those could be accessed directly by other languages/runtimes) it needs a runtime that is dependent on the OS for the Virtual Memory management... it can have "(Cacheable) coherent system memory" mixed with "(non-cacheable) non coherent system memory" in a seamless way, dependent on the MMU of those system processing elements... an in here is where IOMMU enters, is the definition of those MMUs for HSA...

The "cacheable coherent" factor is on top of that, and it MUST be a broad definition, different vendors can have different "cache coherency" schemes (as long it is memory coherent by the "virtual memory") and i'm convinced that hUMA will be only an AMD "name", other vendors can have exactly similar functionality but call it different... and it doesn't matter, the same "programs will run unchanged" in any system (x86, ARM etc + different accels), because it depends on a runtime/driver close tight with the OS virtual memory management.

http://hsafoundation.com/hsa-developer-tools/

HSA is a Low Level Virtual Machine... very flexible, allowing direct interaction with hardware (example Sumatra JVM), and with innovating features AQL (access queue language) approach with Hardware Queue Buffers (and that will be how much of hardware definition there will be, Queues, IOMMU MMUs, and *your own* cache coherency on top if you like)..

[ Matter of fact HSA as a LLVM, can even work on the most varied discrete implementation... matter of fact "Berlin" is HSA (probably those MMUs, Queues and CC on top)... and Berlin is Seamicro clusters, so we will see a Seamicro cluster with Freedom Fabric running HSA programs out of the box ...all the way from SoC and ultraportable, to big cluster machines LOL (most probably the ARM Seattle will be the same, its not an APU, but could have discrete GPGPU, DSP, FPGA etc) ... ]

Any "hidden clause" (it it exists) will be not available to general public. LOL

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.

 

[juanrga]

ARM 64bit is clearly a better ISA(uarch) than x86_64 for a CPU... now the all it needs is a definition like the chinese of Godson are doing with MIPS for vectors, a wide vector definition, and Godson will have 512bit vectors even before Intel...

AMD could "port" (a subsidy for language) XOP 256bit or even 512bt to ARM etc (at least try)... that is, have a superior Vector definition (XOP is better than AVX), and an execute pipe definition for x86_64 that doesn't depend on Intel for licensing... in ARM and others... like Godson that has similar pipe (BUT VERY LIMITED) to help/accelerate run x86 on MIPs by "on-the-fly binary translation" (like Transmeta)... AMD could even negotiate "royalties" with ARM Holdings about that, and not only with ARM but also MIPS etc to.

**IF** that happens, honestly it starts to smell like rotten for the sides of x86... x86 which no matter its dominance on the PC and traditional small server markets, seems clearly now obsolete...

So... if one is long sighted, and clever sighted... the future of Intel is on AMD hands ... that is why they want to kill it the fast as they can.

Me if i were AMD CEO i'll start doing it tomorrow... and its late... Intel is a vicious entity, enemy of every body, even its own partners and costumers( sorry, but the long track of fines all over the world and law suits, seems to confirm this)... and deserves to die.

First no more patent agreements, what is done is done... then try to license XOP and x86_64 (what possible) to ARM and MIPS and why not even OpenPower... all joined by a good push for royalty free HSA and royalty free HyperTransport 4.0... it would be a slow transition, but for one thing the x86 dominating allover factor will be dead and buried( intel vision will be dead). If Intel dies or not is a totally different issue, but i would have the upper hand, and i think in the end Intel would be forced to "license" ARM or MIPS (other) to stay relevant( they could take back to life Alpha)... it would be slow but inexorable (my view).

XOP and x86_64 on ARM (etc) is most relevant because *the* principal factor for the success of a new BETTER ISA, was and is exactly the "software ecosystem", not the technical prowess of a ISA (that is why x86 won so far)... no "software ecosystem" and you can even have the upper duper frincking best ISA/uarch in the world, that it goes nowhere...

In the end i think something in those lines is unavoidable... even for x86... it will be Intel on one side, with its very peculiar views about heterogeneous and else, and it will be everybody else on the other side.

Make no mistake, its NOT a technical approach, its a "political" approach... follow the leader and you'll always be a second choice no matter what, and no matter advertising and publicity or propaganda, so the *wise political choice* is exactly NOT follow the leader, and more and "free" technical approaches are better than alone, and if you break then you could be dead on the leader command... like happened to nvidia chipset business, and could happen to nvidia discrete adapters (1.5B has can have an ensnaring irresistible appeal, but nvidia is dancing with dead...)

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

ARM delayed the launch of ARM 64 bits, because they decided to follow the hard route. The easy route would be simply to expand the ARM 32 bits, somewhat as AMD/Intel developed x86_64 on top of x86_32. However, ARM did return to the dashboard and cleaned up the previous design. As a consequence, the ARM 64bit is a very elegant and optimized design, which is not fully backward compatible with 32 bit ARM, because some of the oddities were eliminated.

Why do I believe that ARM could dominate the world?

- ARM is the dominator in phone, tablets, modems, printers, hard disks, cars...

- ARM is starting the assault to servers. AMD is replacing jaguar servers by more powerful and efficient ARM servers.

- Many serious people is developing ARM supercomputers.

- First ARM desktops are appearing.

- Nvidia secret plans (Boulder project).

 

[noob2222]

The facts about AMD replacing x86 jaguar for ARM were given before. AMD official server roadmap and the presentation explaining why are replacing jaguar servers by arm servers. Also AMD quote, given above, claiming that they expect the sales of ARM servers to surpase the sales of x86 servers. Do you believe AMD will invest lots of resources in future x86 servers when is predicting that sales will drop?

facts given before where? i am not going back searching through walls of text. if you have official proof of amd stating they're replacing jaguar with arm, please provide. pretty please.
what server roadmap? the last one i saw showed coexistence. if there's a different one, i'd like to see that.

The facts about AMD replacing x86 jaguar for ARM were given before. AMD official server roadmap and the presentation explaining why are replacing jaguar servers by arm servers. Also AMD quote, given above, claiming that they expect the sales of ARM servers to surpase the sales of x86 servers. Do you believe AMD will invest lots of resources in future x86 servers when is predicting that sales will drop?

So these facts are your own speculation? nice ... I wonder what "facts" I can make up.

Unless the a57 is 85% faster than the a15 ... replacing jaguar would fail.

AMD official roadmap showing how the existent jaguar-based servers will be substituted by ARM-based servers next year:

Don't cut off what you don't want to answer. What is beema, mullins, and Nolan and Puma+?

Arm cores are going to go along-side not replace. You don't just jump both feet into a market thats not proven or is destined to be the slowest thing around and abandon everything else in hopes "that it might work out in the long run". They might as well rename the company to VIA if thats the case.

The difference with the updated roadmap is there is no pretty "picture" to go along with it. http://news.softpedia.com/news/AMD-Updates-2014-2015-Roadmap-Kabini-Beema-Carizo-and-Excavator-378897.shtml

 

[de5_Roy]

Therefore, you have been missing the reviews, the news, and the marketing of last years... LOL

AMD fusion plans started to be developed many years ago. Why do you believe AMD bought ATI? And those plans were made public in many occasions:

http://arstechnica.com/gadgets/2011/10/can-amd-survive-bulldozers-disappointing-debut/

riight. then this happened:
http://www.xbitlabs.com/news/other/display/20120119151031_AMD_Dumps_Fusion_System_Architecture_Brand.html
because of this:
http://www.techpowerup.com/159096/arctic-to-sue-amd-over-fusion-brand-name.html
toldja they got sued. AWH i geddit now... this was not last year's news, it happened before last year. if you're really counting, more than 12 months prior. my bad.

Spoiler

Therefore, the modern HSA idea of using the GPU to compute tasks, offloading the CPU, was invented many years ago by AMD, much before the HSA foundation was born. If you believed that AMD idea of APUs was something with cheap integrated graphics, then you were very wrong.

AMD is promoting the HSA foundation for spreading this new heterogeneous technology, because they did learn of the mistakes when they tried to push the (then new) 64bits technology alone.

blah blah blah

i'ts not like i'm avoiding this. it's not remotely related to my query.

AMD official roadmap showing how the existent jaguar-based servers will be substituted by ARM-based servers next year:

It's hardly a secret that AMD has stepped out of its x86 comfort zone to develop an ARM-based server chip

oh noes! two yellowish-orange-ish bars side-by-side means replace?!?! gasp. then the two teal(blue? in my defense, i am using an uncalibrated tn monitor. cheap!) bars in the following promo slide (that every c.a.l.f. parroted a while ago) must mean 'replacement' as well...

or the 'fact' that 16 core server cpus will be 'replaced' by 20 core(!) cpus:
http://www.xbitlabs.com/images/news/2010-11/amd_server_roadmap_nov2010.jpg
here's a version with amd seal, in case you're wondering:
http://blogs.pcmag.com/miller/assets_c/2010/11/AMD%20server%20roadmap%202011-thumb-450x257-17242.png
the bars are side by side, just like the promo slide you posted.

none of the soures you posted here even mention the word 'replace'. they say 'outsell', but not replace. i'll ask, for the 3rd(i think, could be more...) time, please provide the official source where it says amd will replace jaguar x86 with arm server socs/cpus.

even in the promo slide you posted, it never says that seattle won't compete/sell along with jaguar's successor architecture (i don't know it's name yet).

In case you don't know, the existent jaguar-based servers (Opteron X) are more faster than the jaguar-based Kabini (1.5GHz) that you mention above. But:

AMD says its 64-bit ARM server chips will be up to four times faster and more power-efficient than its latest quad-core x86 Opteron X-series chips, which draw up to 11 watts of power.

4x faster how? i'll give an example how amd(and other vendors) twists 'facts' - by comparing 4c opteron X cpu with 16c/20c arm soc in a specialized bench.

AMD is also saying us what they expect about their server portfolio with ARM (Seattle) and x86 (Berlin and Warsaw) solutions:

AMD will sell the 64-bit ARM processors alongside x86 server processors, but the company expects ARM to outsell x86 in the long run.

http://www.engadget.com/2013/06/18/amd-seattle-arm-server-chip/
http://www.pcworld.com/article/2048069/amd-to-ship-64bit-arm-chips-to-server-makers-next-year.html

see, even in your own excerpt it says 'alongside' and 'outsell', not 'replace'.

 

[Ranth]

Well this is pretty interesting: http://www.anandtech.com/portal/amd

"You've landed on the AMD Portal on AnandTech. This section is sponsored by AMD. It features a collection of all of our independent AMD content, as well as Tweets & News from AMD directly. AMD will also be running a couple of huge giveaways here so check back for those."

It is gonna be interesting if this changes something in the reviews.... AnAMDtech?

 

[Ranth]

Can't we just ignore Hafi? He's obviously trolling, so just let him be...

 

[esrever]

Well this is pretty interesting: http://www.anandtech.com/portal/amd

"You've landed on the AMD Portal on AnandTech. This section is sponsored by AMD. It features a collection of all of our independent AMD content, as well as Tweets & News from AMD directly. AMD will also be running a couple of huge giveaways here so check back for those."

It is gonna be interesting if this changes something in the reviews.... AnAMDtech?

AMD finally realized the anandtech is just a shill and is paying them eh?

 

[juanrga]

The facts about AMD replacing x86 jaguar for ARM were given before. AMD official server roadmap and the presentation explaining why are replacing jaguar servers by arm servers. Also AMD quote, given above, claiming that they expect the sales of ARM servers to surpase the sales of x86 servers. Do you believe AMD will invest lots of resources in future x86 servers when is predicting that sales will drop?

facts given before where? i am not going back searching through walls of text. if you have official proof of amd stating they're replacing jaguar with arm, please provide. pretty please.
what server roadmap? the last one i saw showed coexistence. if there's a different one, i'd like to see that.

The facts about AMD replacing x86 jaguar for ARM were given before. AMD official server roadmap and the presentation explaining why are replacing jaguar servers by arm servers. Also AMD quote, given above, claiming that they expect the sales of ARM servers to surpase the sales of x86 servers. Do you believe AMD will invest lots of resources in future x86 servers when is predicting that sales will drop?

So these facts are your own speculation? nice ... I wonder what "facts" I can make up.

Unless the a57 is 85% faster than the a15 ... replacing jaguar would fail.

AMD official roadmap showing how the existent jaguar-based servers will be substituted by ARM-based servers next year:

Don't cut off what you don't want to answer. What is beema, mullins, and Nolan and Puma+?

Arm cores are going to go along-side not replace. You don't just jump both feet into a market thats not proven or is destined to be the slowest thing around and abandon everything else in hopes "that it might work out in the long run". They might as well rename the company to VIA if thats the case.

The difference with the updated roadmap is there is no pretty "picture" to go along with it. http://news.softpedia.com/news/AMD-Updates-2014-2015-Roadmap-Kabini-Beema-Carizo-and-Excavator-378897.shtml

Pay attention to my words. Look closely to the bold face part of my posts. Go back and read my other posts. I have been talking about AMD servers not about desktops.

 

[juanrga]

Therefore, you have been missing the reviews, the news, and the marketing of last years... LOL

AMD fusion plans started to be developed many years ago. Why do you believe AMD bought ATI? And those plans were made public in many occasions:

http://arstechnica.com/gadgets/2011/10/can-amd-survive-bulldozers-disappointing-debut/

riight. then this happened:
http://www.xbitlabs.com/news/other/display/20120119151031_AMD_Dumps_Fusion_System_Architecture_Brand.html
because of this:
http://www.techpowerup.com/159096/arctic-to-sue-amd-over-fusion-brand-name.html
toldja they got sued. AWH i geddit now... this was not last year's news, it happened before last year. if you're really counting, more than 12 months prior. my bad.

LOL I have been talking about the original idea and the technology, not about the brand name. They changed the name by legal reasons. The name.

Spoiler

Therefore, the modern HSA idea of using the GPU to compute tasks, offloading the CPU, was invented many years ago by AMD, much before the HSA foundation was born. If you believed that AMD idea of APUs was something with cheap integrated graphics, then you were very wrong.

AMD is promoting the HSA foundation for spreading this new heterogeneous technology, because they did learn of the mistakes when they tried to push the (then new) 64bits technology alone.

blah blah blah

i'ts not like i'm avoiding this. it's not remotely related to my query.

AMD official roadmap showing how the existent jaguar-based servers will be substituted by ARM-based servers next year:

It's hardly a secret that AMD has stepped out of its x86 comfort zone to develop an ARM-based server chip

oh noes! two yellowish-orange-ish bars side-by-side means replace?!?! gasp. then the two teal(blue? in my defense, i am using an uncalibrated tn monitor. cheap!) bars in the following promo slide (that every c.a.l.f. parroted a while ago) must mean 'replacement' as well...

LOL Orange means 28nm, Blue means 32nm. Forget colours. At the left what is being replaced. At the right the replacement. Can you differentiate between left and right or need assistance?

AMD is also saying us what they expect about their server portfolio with ARM (Seattle) and x86 (Berlin and Warsaw) solutions:

AMD will sell the 64-bit ARM processors alongside x86 server processors, but the company expects ARM to outsell x86 in the long run.

http://www.engadget.com/2013/06/18/amd-seattle-arm-server-chip/
http://www.pcworld.com/article/2048069/amd-to-ship-64bit-arm-chips-to-server-makers-next-year.html

see, even in your own excerpt it says 'alongside' and 'outsell', not 'replace'.

Look to the AMD official roadmap again. Seatle (ARM) "alongside" Berlin (x86) and Warsaw (x86). Jaguar servers (Opteron X) will be replaced by ARM servers (Seattle) next year. And AMD wait sales of Seattle (ARM) to "outsell x86 in the long run". I suspect AMD will move to ARM in near future (2016? 2018?) with x86 being a legacy arch.

 

[juanrga]

Better then anything amd have, then again we all hear amd are cheap when they are to below average to sell for any more. AMD ain't stupid, they have to sell at there prices they do otherwise they would be bankrupt as everyone would buy intel instead. intel baytrail easily beats the jaguar cpus, looks like steamroller cores are only amd hope as it seems one sided at the moment in intels favour.

Besides your love by compare prices of brand new AMD chips with second hand Intel chips, your spreading of false information about "baytrail" was broadly corrected by two posters in the other thread. But, of course, you are ignoring the corrections and posting the same nonsense here. LOL

I bought my i5 4670k brand new in box for £150 with grid 2 and my i5 4430 is used wir box for £97. i5 4670k was around £152 a few months back on amazon anyway.

Your dishonesty of comparing brand new prices of AMD chips to prices of used Intel chips was already mentioned before. The first time that you wrote "£97" you didn't mention that it was used. You only write it now "used" after being caught again in another pile of dishonest marketing.

 

[juanrga]

Well this is pretty interesting: http://www.anandtech.com/portal/amd

"You've landed on the AMD Portal on AnandTech. This section is sponsored by AMD. It features a collection of all of our independent AMD content, as well as Tweets & News from AMD directly. AMD will also be running a couple of huge giveaways here so check back for those."

It is gonna be interesting if this changes something in the reviews.... AnAMDtech?

Anand did an interview to new AMD CEO and said he feel impressed:

http://www.anandtech.com/show/7281/understanding-amd-semi-custom-strategy

Pay attention to the slide. The one saying that AMD will be not "designing / building high performance products with emphasis in cores". Does this sound as expect 8/12/16 FX Steamroller/Excavator? I believe not.

I know this part will raise the ire of some fanboys but here goes:

Rory’s playbook for AMD is actually very similar to how he ran things at Lenovo. Lenovo was stuck in a similar position not too long ago: it had a relatively high margin enterprise business that it used to fund and grow a much lower margin consumer business. The new AMD strategy is quite similar.

Traditional margins on x86 CPUs are nothing short of tremendous. The fabless semiconductor manufacturers that compete with Intel don’t operate on anywhere near the same margins. AMD used to play in the same space Intel did, and as a result was often viewed as disappointing as their margins wouldn’t hold up. The problem was that although AMD shed its fab burden, its margins were always viewed as needing to be up at Intel levels. That perception has to change.

Rory’s strategy is to use high margin revenues from existing markets (e.g. PC client, GPUs) and use it to fund a low cost structure expansion into new markets. On paper it’s a sound approach, but the unexpected quick decline of AMD PC sales/shipments threw a wrench in the plan.

The company had to shrink in order to deal with lower than expected revenues than what Rory had planned on when he first took the job almost two years ago. Based on what Raja Koduri told me after he joined a few months ago, it seems like it’s working:

“Raja returns to a very different AMD than the one he left. I asked him what’s different and he responded by saying the AMD he left acted like a company that was 10x its size. Today, AMD is a much smaller and more agile company. Raja believes AMD is in a better position to take advantage of new opportunities vs. being in the hopeless position of never being able to catch up in mature markets.”

So you take the higher margin PC revenues, and use them to invest in lower cost products in new markets. Rory expects semi-custom silicon to see heavy use in these new markets by the way. Lower cost typically means lower margin, which is something the new AMD is ok with. Making Intel margins (or even traditional AMD margins) is tough, in these new markets AMD just needs to be making better margins than the ARM players as the company transitions from being fully PC supported, to PC + additive revenue from these new markets and finally to a position where AMD’s revenues are dominated by these new revenue sources. What are the new markets in specific? Getting anyone at AMD to answer that question today is tough, but I suspect the first place to look is among all of those other players I talked about earlier. The companies tasked with competing with these vertically integrated powerhouses could rely heavily on AMD.

 

[de5_Roy]

LOL I have been talking about the original idea and the technology, not about the brand name. They changed the name by legal reasons. The name.

by the original idea you mean just fusion? that's a marketing bait and misnomer at that. if amd was really honest, they'd say 'future is fusion through virtualization' or something that doesn't involve lying by omission. that's why they changed into hsa because that is the real technology standard.

Forget colors. At the left what is being replaced. At the right the replacement.

Look to the AMD official roadmap again. Seatle (ARM) alongside Berlin (x86) and Warsaw (x86). It is jaguar servers which will be replaced by ARM servers now. And AMD wait sales of Seattle (ARM) to "outsell x86 in the long run". I suspect AMD will move to ARM in near future (2016? 2018?).

i just proved, with evidence, that official roadmaps are fickle sandy beaches. being at the 'right side of the promo slide' does not mean replacement. it means nothing, really. thus, my question still remains unanswered. so consider this me asking again. i'll even help you out, try looking for direct quotes from amd executives or ceo, make sure you skip words that some overzealous idiot put in their mouth to make their argument by interpreting according to their convenience.
i don't doubt arm v8 will pose a considerable threat to x86 and who knows, might take over eventually.

the way you 'promo slide this','promo slide that' makes me wonder if you ever really read between the lies that red/green/blue spout.

 

[juanrga]

Forget colors. At the left what is being replaced. At the right the replacement.

Look to the AMD official roadmap again. Seatle (ARM) alongside Berlin (x86) and Warsaw (x86). It is jaguar servers which will be replaced by ARM servers now. And AMD wait sales of Seattle (ARM) to "outsell x86 in the long run". I suspect AMD will move to ARM in near future (2016? 2018?).

i just proved, with evidence, that official roadmaps are fickle sandy beaches. being at the 'right side of the promo slide' does not mean replacement. it means nothing, really. thus, my question still remains unanswered. so consider this me asking again. i'll even help you out, try looking for direct quotes from amd executives or ceo, make sure you skip words that some overzealous idiot put in their mouth to make their argument by interpreting according to their convenience.
i don't doubt arm v8 will pose a considerable threat to x86 and who knows, might take over eventually.

the way you 'promo slide this','promo slide that' makes me wonder if you ever really read between the lies that red/green/blue spout.

AMD will replace jaguar-based servers by ARM-based servers in 2014. They have explained why. You don't believe it? Ok, that is fine for me.

 

[de5_Roy]

AMD will replace jaguar-based servers by ARM-based servers in 2014. They have explained why. You don't believe it? Ok, that is fine for me.

it's not about me believing. it's about amd actually stating if they would, according to your claim: replace jaguar based servers by arm based servers in 2014. you have not been able to present any official statement or any proof of that after multiple times asking. two side-by-side yellow blocks on a promo slide means very little.
what i do understand though, is that you believe marketing promo slides a bit too much.

 

[The Q6660 Inside]

AMD should try to kill x86 with ARM ASAP but they will need to do it very slowly (4 years per say). I guarantee they will still offer but not promote x86 Jaguar servers when ARM servers are released.

 

[noob2222]

Pay attention to my words. Look closely to the bold face part of my posts. Go back and read my other posts. I have been talking about AMD servers not about desktops.

AMD will replace jaguar-based servers by ARM-based servers in 2014. They have explained why. You don't believe it? Ok, that is fine for me.

Pay attention to my words. This move makes no sense. They will go from competetive with Atom servers to noncompetetive and incompatible with software. You have no proof, just some rumor you made up that puma+ is not socket compatible.

AMD will not be alone in this ARM server market. They will be competing with Samsung, Qualcomm, Applied Micro, Cavium, Calxeda, Marvell, and even Nvidia.

Does completely abandoning the only competetive microserver chip AMD have against Intel in favor of competing with 7 other vendors make any sense at all to you?

 

[de5_Roy]

Pay attention to my words. This move makes no sense. They will go from competetive with Atom servers to noncompetetive and incompatible with software. You have no proof, just some rumor you made up that puma+ is not socket compatible.

AMD will not be alone in this ARM server market. They will be competing with Samsung, Qualcomm, Applied Micro, Cavium, Calxeda, Marvell, and even Nvidia.

Does completely abandoning the only competetive microserver chip AMD have against Intel in favor of competing with 7 other vendors make any sense at all to you?

i think you've already read his stance. to repeat:

I suspect the HSA foundation includes some kind of hidden clause/agreement. Traditional ARM vendors will focus in mobile, whereas AMD will focus in servers and probably HPC, but with ARM instead of x86.

doesn't make sense to me, from a business standpoint.
they're businesses, not secret hobby club. why would anyone in the right mind give up a lucrative sector so that amd can make money off of selling arm-based server solutions?

Because AMD is licensing the fusion technology to the HSA foundation. Do you really believe AMD would say hey guys take all this technology for almost free, which will improve your products a lot of, and now compete against me please. It seems evident that AMD is licensing to mobile players, because those will be competing with Intel in the mobile market. Whereas AMD has already shared its intention that they want take Intel servers market using ARM servers. Everyone in the HSA foundation wins each one in a market.

apparently other vendors will let amd make money because of a secret handshake in the hsa foundation. since amd invented modern hsa (his words, not mine) and other arm vendors are 'mobile players'. i think this includes nvidia who is just a mobile player and has no position in servers and hpc. however, nvidia is not hsa foundation member, afaik. i wonder how many smartphones and tablets calxeda (to my limited knowledge, a vendor that sells arm server machines) sells....
to summerize: he thinks amd included a secret non-competition clause somewhere. [strike]to summerize further: he thinks amd is intel. maybe he's from blue team after all... *notices hafijur victory-dancing around the fire* [/strike] disregard the struck-out part.
okay i'm done.
edit: sorted out the quote-system mess.

 

[GOM3RPLY3R]

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?

 

[blackkstar]

Pay attention to my words. Look closely to the bold face part of my posts. Go back and read my other posts. I have been talking about AMD servers not about desktops.

AMD will replace jaguar-based servers by ARM-based servers in 2014. They have explained why. You don't believe it? Ok, that is fine for me.

Pay attention to my words. This move makes no sense. They will go from competetive with Atom servers to noncompetetive and incompatible with software. You have no proof, just some rumor you made up that puma+ is not socket compatible.

AMD will not be alone in this ARM server market. They will be competing with Samsung, Qualcomm, Applied Micro, Cavium, Calxeda, Marvell, and even Nvidia.

Does completely abandoning the only competetive microserver chip AMD have against Intel in favor of competing with 7 other vendors make any sense at all to you?

I am with noob on this quite a bit. Everyone is basically suggesting that AMD make a massive gamble on desktop AND on servers at the same time.

AMD still needs big cores to fund their expansion into custom SoC builder. Eventually they will move out of that market, however, I don't think they're remotely close regarding how mature HSA actually is.

I just have one big question for you guys that think AMD is going to completely abandon high performance x86 cores instead of running some sort of special blend.

What will happen to the folks who are content creators who need Autodesk products, Adobe products, etc? Do you really think that every feature in these programs will be HSA accelerated? Can you look at something like the Autodesk suite of software and think that every feature in every piece of software there is going to use HSA and will be just fine in a weak two module AMD system?

You think AMD is going to walk away and give Intel a monopoly on high end x86 CPUs?

And then there's gaming... Going to give Intel a monopoly on anything beyond mid-range gaming CPUs too?

That's comedy gold right there. Completely abandon markets served by $1000+ CPUs to replace them with $50 SoCs. That's just about the exact opposite of what every semiconductor design company has done, except for Via.

None of this seems completely idiotic to you guys? To expect mounds and mounds of x86 software, much of which people actually NEED to make money to just be completely forgotten because a few HSA accelerated features in your APU are 5 times faster than a regular CPU while the CPU itself is much slower in the vast majority of workloads that don't have HSA?

amd64 worked because it didn't break backwards compatibility. Other than, say, Windows XP 64 bit drivers, there wasn't much of a reason to not go 64-bit. Everything you had before worked just the same.

Now, for AMD to abandon traditional x86 CPUs in favor of HSA making up the performance gap between low-end CPUs, you think AMD is going to pull this off when they throw away performance on old programs? FX 8350 is very popular on amazon and newegg, more so than FX 6000 and all the APUs. People still want that performance and they want those cores. They want to do work. They want to stream games.

HSA can not replace a high performance core for a large number of users. It never will.

 

[de5_Roy]

ditching jaguar based opteron x less in than 12 months in favor of arm v8 without accounting for validation, certifications, backwards compatibility check, re-training, maintenance training, customer support and so on, yet expecting such high revenue the instant arm v8 servers start shipping against existing vendors (that sell 32 bit servers)... amd must have some kind of a secret.

 

[8350rocks]

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.

2.) Gaming Evolved is premised on high powered gaming solutions. You can't have that without high powered cores. It just doesn't work that way...I suppose you think they're going to start promoting Radeon HD GPUs with Intel CPUs???

3.) AMD is wise to move on emerging markets, they can do so and boost their margins and volume increasing profits. It does make sense from a business angle, if they leverage it and keep their existing market share. AMD has the highest penetration in x86 among gamers, they'd be fools not to expand upon that. In fact, I think that's the entire premise behind Gaming Evolved.

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....)