AMD CPU speculation... and expert conjecture

[juanrga]

Have anyone noted that the recently announced Hierofalcom is the first AMD chip that comes with DDR4 support?

 

[palladin9479]

ARM really isn't designed to sale "up" in performance. To do that they would have to dedicate more hardware to things like cache / predictors / multiple ALU's / SIMD units / more robust OoO engine and so forth. Those all consume power and cost die space which goes against the original concept of ARM, low cost low power (as in energy usage) designs. You end up trying to reinvent the wheel and there already exists an ISA that was designed to have high scaling and has 0 licensing costs.

http://en.wikipedia.org/wiki/SPARC
http://www.oracle.com/technetwork/systems/opensparc/index.html

Not trying to be a shill, only pointing out that there is already material out there to do RISC processing on the cheap.

 

[palladin9479]

http://www.eweek.com/pc-hardware/amd-unveils-arm-based-soc-for-embedded-systems.html

The announcement comes a year after AMD officials said the company was partnering with ARM to create SoC's based on ARM's architecture for dense, low-power servers for the growing hyperscale data center market, part of AMD's heterogeneous strategy of offering systems makers and end users a choice of platforms. - See more at: http://www.eweek.com/pc-hardware/amd-unveils-arm-based-soc-for-embedded-systems.html#sthash.H1oDEZlO.dpuf

Emphasis mine.

Hierofalcon will include up to eight ARM Cortex-A57 CPUs running at up to 2GHz. It also will offer two 64-bit DDR3/4 memory channels with error correction code (ECC), 10G-bps KR Ethernet, PCI-Express Gen 3 network connectivity and support for ARM's TrustZone security technology. The TrustZone technology will be housed on a separate chip. - See more at: http://www.eweek.com/pc-hardware/amd-unveils-arm-based-soc-for-embedded-systems.html#sthash.H1oDEZlO.dpuf

It does both which is something we already knew AMD is putting on their chips. Just like how their older chips could do DDR2/3.

Those systems do not directly compete with x86 systems, different tool for a different problem. Stop thinking "one CPU to rule them all", IT is about problems vs solutions. Describe / define the problem, select the appropriate tools then engineer the solution.

 

[juanrga]

ARM Aiming for 5%-10% Share in Global Server Chip Market by 2016

Intel's Server Market Share To Decline As Competition Heats Up

AMD's Feldman: ARM Will Quickly Gain Market Share in Servers

Interesting that Feldman mentions: SPARC --> x86 --> ARM

Nvidia Launches CUDA Support for ARM Server Chips

Nvidia sharing its vision about high-performance ARM dominating everything from mobile to supercomputers

AMD 'secret' slide

Apple engineers exploring a switch to high-performance ARM chips from Intel chips

I agree with AMD, Nvidia, Apple... and applaud them.

 

[hcl123]

In the fourth place, do you still pretend that anyone can do a CPU+dGPU with UMA/hUMA?

I don't pretend nothing, its on the road map since the very beginning in 2011

http://t1.gstatic.com/images?q=tbn:ANd9GcRvGiOpiosFpbI_dQ0Fhf-2w6RanklsgXec8ZHRWDd0vWt-JvHlSQ

And this slide says its perfectly possible (its not mine its AMD) hUMA/HSA

http://mygaming.co.za/news/wp-content/uploads/2013/05/AMD-Kaveri-hUMA-benefits-600x297.jpg

Just read the slides... and stop wasting time, they say "extend to discrete GPU"... and "10 reason for full coherent hardware" with the GPU without recording for the absence of coherence (mix coherent +non-coherent) and probe filter and directories (coherency directories ala ccNUMA ) (edt)

 

[hcl123]

About ARM vs x86, it's no contest here. x86 is the more powerful uArch, ARM is the more power efficient one. ARM doesn't sale vertically very well, it's very difficulty to make a "powerful" ARM computer. Instead it scales horizontally, you can make "many" low powered ARM computers on the cheap. This lends them to being used in radically different computing then x86, different tools for different problems. Everything in computing is on a "problem vs solution" mindset. You pick out the right tool for the job and engineer your solution around that tool.

The 64 bit changed that completely... its a very powerful ISA in the same relax memory model (half way to transactional memory) of previous, but more clean less quirks, and with 32 GPR (general purpose registers) and 32 logical extension vector registers. And NO its not an extension of ARMv7 ISA, its a completely new one tough compatible.

http://www.realworldtech.com/arm64/

everything points it could be quite more powerful than x86_64 with its 16GPR and load to store model... i think they were some how thinking of servers and HPC when they did this ARMv8 ( the GPR and the vector registers don't mix, so in reality compared with x86_64 it has 64 fast access registers)...

Of course this depends on the implementer, the ISA has potential, but not many have experience with real high performance CPUs... in ARM armada AMD is the only one i think...

 

[hcl123]

Emphasis mine.

Hierofalcon will include up to eight ARM Cortex-A57 CPUs running at up to 2GHz. It also will offer two 64-bit DDR3/4 memory channels with error correction code (ECC), 10G-bps KR Ethernet, PCI-Express Gen 3 network connectivity and support for ARM's TrustZone security technology. The TrustZone technology will be housed on a separate chip. - See more at: http://www.eweek.com/pc-hardware/amd-unveils-arm-based-soc-for-embedded-systems.html#sthash.H1oDEZlO.dpuf

It does both which is something we already knew AMD is putting on their chips. Just like how their older chips could do DDR2/3.

Those systems do not directly compete with x86 systems, different tool for a different problem. Stop thinking "one CPU to rule them all", IT is about problems vs solutions. Describe / define the problem, select the appropriate tools then engineer the solution.

That "falcon" is clusters ... and is only 8 cores because of process... transition below 28nm and it can have 16 CPU cores (meaning its already on a different interconnect and cache hierarchy than ARM reference designs, that can go only to 8...also DDR4 support is not yet on ARM reference designs... )

https://semiaccurate.com/assets/uploads/2013/06/AMD-Seattle-Slide.png

An PCIe on-board is for Freedom Fabric on-board to...

 

[palladin9479]

*sigh*

I explained it above, before people started cutting my posts apart to portray my statements dishonestly.

ARM is designed to be power and cost efficient. Other modern design's achieve high performance via brute force mechanics, scheduling magic, large caching mechanisms, front end predictors and other non computing components. Those components take up both die space and power and are effective when scaling to higher clock speeds and trying to stuff more instructions per clock (proper use of the term) through the same CPU. ARM could become "high performance" but only if it does these things which just turns it into another x86 / SPARC / POWER CPU while discarding it's strengths.

Basically "ARM" can not be "high performance" while also maintaining it's higher power / space efficiency. There is a reason we use ARM CPU's in phones and mobile devices, x86 CPU's in desktops and SPARC/POWER in HPC / Enterprise processing. Attempting to discard those targeted engineered solutions without first addressing the entire reason for existence is foolish.

 

[esrever]

ARM Aiming for 5%-10% Share in Global Server Chip Market by 2016

Intel's Server Market Share To Decline As Competition Heats Up

AMD's Feldman: ARM Will Quickly Gain Market Share in Servers

Interesting that Feldman mentions: SPARC --> x86 --> ARM

Nvidia Launches CUDA Support for ARM Server Chips

Nvidia sharing its vision about high-performance ARM dominating everything from mobile to supercomputers

AMD 'secret' slide

Apple engineers exploring a switch to high-performance ARM chips from Intel chips

I agree with AMD, Nvidia, Apple... and applaud them.

so why didn't mips win yet?

 

[palladin9479]

so why didn't mips win yet?

Right tool for the job. People keep getting fixated on the mentality of

One CPU to rule them all, One CPU to find them; One CPU to bring them all and in the darkness bind them.

There is no one best metal, there is no one best food, there is no one best power system, there is no one best car. What we have is a tool bag full of tools and a whole host of problems that need solutions. Over time our problems evolve and change which in turn requires our solutions to evolve and change with them. In order to facilitate solving these problems our tools must also evolve and change. Often a tool does not evolve and finds itself part of a solution without a problem. When that happens we discard the tool in favor of ones more capable of solving our newer problems.

So when people talk about technology and things changing they need to stop with the bandwagon fanboyism of whatever is "new" or "shiny" and instead look at everything as tools to be part of solutions that solve problems. So start with the problems, identify the solutions then find the tools required to best engineer those solutions.

 

[noob2222]

ARM Aiming for 5%-10% Share in Global Server Chip Market by 2016

Intel's Server Market Share To Decline As Competition Heats Up

AMD's Feldman: ARM Will Quickly Gain Market Share in Servers

Interesting that Feldman mentions: SPARC --> x86 --> ARM

Nvidia Launches CUDA Support for ARM Server Chips

Nvidia sharing its vision about high-performance ARM dominating everything from mobile to supercomputers

AMD 'secret' slide

Apple engineers exploring a switch to high-performance ARM chips from Intel chips

I agree with AMD, Nvidia, Apple... and applaud them.

The first one, ARM themselves are being realistic. 5-10% by 2016, I can see that possibly happening. Here is the thing. that 5-10% is going to be split between the 8 companies that have already announced making microserver ARM cpus. AMD is pushing hard to get as much of that share as possible. 8 or more companies fighting over 5-10% overall marketshare isn't a whole lot for AMD to drop the ball on current products and go for table scrapping with the likes of nvidia, qualcomm, and samsung.

2nd article, of course Intel's shares are going to go down. As per the article "Intel has managed to increase its share from 86.6% in 2008 to 95.6% in 2012, as per our estimate." How much higher can you go above 95%? Can't read page 2 of that article, some stupid "register to read the rest of the article"

3rd article is nothing but marketing hype. 6 times they talk about small work loads, small cores, smaller computers ... but your claims are that ARM is this HIGH PERFORMANCE part. Unrealistic : by 2016 ARM is going to have 20% or more server market share. Of course, this could be construed to mean per core sales as it will take what 32-64 cores to equal one xeon E5/E7. ARM breaking x86 stranglehold isn't going to just magically happen. Intel has billions of dollars at stake to make sure its not an overnight takeover.

4th article .. duh ... if NVidia doens't support cuda on arm, then they don't even support their own tegra cpu within itself.

The last article pretty much sums it up right here.

And if Apple can’t design ARM-based chips that are far more powerful than current models, the company would probably need to stay with Intel to satisfy Mac power users, who need lots of computing performance for tasks such as developing software or doing high-end graphics.

ya .. arm is faster than x86 right now yet ^^ High End ARM is just that and only that, High End for ARM processors, not high end over all cpus ever made. This seems to be a mis-thought process that just because something has the words "high end", it means over all other products that are remotely close.

ARM isn't just going to magically displace x86. Aside from being slower, you will have compatiblity problems (such as windows RT can only run ARM compatible programs) That means that anyone wanting to migrate from X86 to ARM, some of them are likely going to have to keep the x86 system just for the program compatibility.

All the ARM takeover right now is wishful thinking. Can it happen? sure Will it happen next year? not a chance. Should AMD abandon current x86 designs and jump to the ARM bandwagon? wouldn't be the smartest business decision. Is intel going to put up a fight to deny ARM takover? you bet your ass they will.

 

[gamerk316]

Today the main game engines scale well. I already provided you a copy of a cpu profiling of a future PS4 game. The game uses six-cores for running six game threads. And the OS and the dev tools (including the profiler) run in the remaining two cores. A total of 8 cores working fine.

For the last bloddy time: Embedded systems != General Purpose PC.

The types of scheduling tricks you can use on a system with a single hardware spec are not feasible on a general purpose PC, as you can not guarantee at any point in time if your application is even running. You HAVE to go through the OS scheduler, you HAVE to share time with thousands of other threads, you HAVE to assume other applications can come in and kick you out, and you HAVE to synchronize your thread logic if two threads touch the same object. You can NOT make any assumptions like "X takes 1ms to occur, and Y will need the data from X after 1.2 ms, both X and Y start at the same time, so I don't have to synchronize my thread logic" that you can make on an embedded system.

I've worked on embedded systems for ages. You see a LOT of low level code optimizations. For example, you may know some command will throw a switch in the hardware that, by spec, takes a maximum of 2ns to flip, so you know you can do some other processing for 2ns before you send data across said switch. You see a LOT of really low level optimizations like this. On a PC, you can't, because you don't even know at what point your thread even runs the command that throws said switch. Abstraction is great for getting SW to work on thousands of different pieces of hardware, not so much good for performance.

Finally, the PS3 had 6 CPU cores (SPE's) for development, and the 360 had three CPU cores with 2-way SMT, giving both current gen consoles 6 CPU contexts to work with. And the same PC games that ran on 6 cores on the consoles use 2 on the PC. There's a reason for that. You will never see the low-level, fine grain control you see on consoles make their way over to the PC. I can guarantee ever CoD game this generation uses 6 cores on the consoles, likely at 100%. On the PC, 2. Same for every other title out there. I can not WAIT for Novemeber, because I will be doing nothing but posted "I TOLD YOU SO" when people find that graphical quality didn't jump, and we're still stuck on two main threads and maybe a worker thread or two.

To me, people are fooling themselves just like they did before the BD launch.

 

[de5_Roy]

^^ might as well be the same guys with different forum ids from before. i've seen this happen many times.

 

[sarinaide]

Don't you just hate NDA's What I will say is that people should be saving up for the HD9000 series cards because they are really good.

 

[esrever]

Today the main game engines scale well. I already provided you a copy of a cpu profiling of a future PS4 game. The game uses six-cores for running six game threads. And the OS and the dev tools (including the profiler) run in the remaining two cores. A total of 8 cores working fine.

For the last bloddy time: Embedded systems != General Purpose PC.

The types of scheduling tricks you can use on a system with a single hardware spec are not feasible on a general purpose PC, as you can not guarantee at any point in time if your application is even running. You HAVE to go through the OS scheduler, you HAVE to share time with thousands of other threads, you HAVE to assume other applications can come in and kick you out, and you HAVE to synchronize your thread logic if two threads touch the same object. You can NOT make any assumptions like "X takes 1ms to occur, and Y will need the data from X after 1.2 ms, both X and Y start at the same time, so I don't have to synchronize my thread logic" that you can make on an embedded system.

I've worked on embedded systems for ages. You see a LOT of low level code optimizations. For example, you may know some command will throw a switch in the hardware that, by spec, takes a maximum of 2ns to flip, so you know you can do some other processing for 2ns before you send data across said switch. You see a LOT of really low level optimizations like this. On a PC, you can't, because you don't even know at what point your thread even runs the command that throws said switch. Abstraction is great for getting SW to work on thousands of different pieces of hardware, not so much good for performance.

Finally, the PS3 had 6 CPU cores (SPE's) for development, and the 360 had three CPU cores with 2-way SMT, giving both current gen consoles 6 CPU contexts to work with. And the same PC games that ran on 6 cores on the consoles use 2 on the PC. There's a reason for that. You will never see the low-level, fine grain control you see on consoles make their way over to the PC. I can guarantee ever CoD game this generation uses 6 cores on the consoles, likely at 100%. On the PC, 2. Same for every other title out there. I can not WAIT for Novemeber, because I will be doing nothing but posted "I TOLD YOU SO" when people find that graphical quality didn't jump, and we're still stuck on two main threads and maybe a worker thread or two.

To me, people are fooling themselves just like they did before the BD launch.

Your original stance is 2 cores are all you will ever need. Now its 2 cores + worker threads. Lets see where it goes in 6 months. for your argument of the ps3 and 360, both did not use many thread because they weren't required, most parallel threads were light because: 1) ps3 could not run all the SPEs loaded because memory constraints so you would get 2-3 decently threaded things to run in parallel. The platform just didn't work out. They got more tricks later on for the sony exclusives but those are never ported to PC and likely won't be good ports anyways 2) 360 supported 6 threads but nobody wants to multithread when they don't need to, thus most games goes down to 3 threads because why use SMT when you can optimize the single thread for the cpu resources, but 360's cpu also required audio processing which could take up an entire CPU on its own. Thus we were left with 2+ bit of large threads running the game on there. 3)Developers are lazy and when porting to PC, they want to target the lowest common denominator, which is dual cores. throwing all the threads into 2 main cpu cores ends up with a port that will run on any system without the need to optimize for the high end.

We will see what developers can do with 6 symmetrical cores that are completely used for the game. You can keep slowly changing your stance all you want. Of course 2 cores can run the threads, that why minimal requirements of a lot of games are still 1-2 cores. But to get proper performance for the game, its likely 6 cores will destroy 2, 3 and 4 cores.

In the coming years, I predict you will say 4 cores is all you will ever need. Then 6 cores is all you will ever need. Maybe some day you will see what the whole semiconductor industry is moving towards.

 

[daerohn]

I am a bit confused in all those intel fans said. I have a few question that I need some real good answers.

1. The software today we use are mainly developped and optimized for intel, is it not?
2. The AMD architecture is alot more different than an Intel and thus it is emulating Intel, as I know AMD has a RISC core inside, is it right?
3. As the software are not optimized for AMD, does they not use all the abilities an AMD chip has?
4. Next year PS4 and XBOX will be using AMD APU's, so all of the games that will be developed in the next years will be optimized for AMD, is it not?
5. As XBOX has an MS operating system inside, MS will also have full optimized OS for AMD next year and can easly port it to PC's.
6. So next year this time games and OS will be optimized for AMD.

As a result we can assume that fully optimized software will be running a lot more efficiently on AMD than today. Am I wrong?
Even now an AMD can match an i7 chip in multi tasking I really would like to see how optimized software would perform.

I also know that Intel chips has some special threads inside that can recognize benchmark programs and fake them, is this wrong? If this is true why does intel need to cheat when their chips are far more superior than AMD?

I also know that most of the tests out there favors Intel, again same question why?

Can someone enlighten me I really would be so glad. I am a bit confused here.

 

[daerohn]

I am a bit confused in all those intel fans said. I have a few question that I need some real good answers.

1. The software today we use are mainly developped and optimized for intel, are not they?
2. The AMD architecture is alot more different than an Intel and thus it is emulating Intel, as I know AMD has a RISC core inside, is it right?
3. As the software are not optimized for AMD, does they use all the abilities an AMD chip has?
4. Next year PS4 and XBOX will be using AMD APU's, so all of the games that will be developed in the next years will be optimized for AMD, is it not?
5. As XBOX has an MS operating system inside, MS will also have full optimized OS for AMD next year and can easly port it to PC's.
6. So next year this time games and OS will be optimized for AMD.

As a result we can assume that fully optimized software will be running a lot more efficiently on AMD than today. Am I wrong?
Even now an AMD can match an i7 chip in multi tasking I really would like to see how optimized software would perform.

I also know that Intel chips has some special threads inside that can recognize benchmark programs and fake them, is this wrong? If this is true why does intel need to cheat when their chips are far more superior than AMD?

I also know that most of the tests out there favors Intel, again same question why?

Can someone enlighten me I really would be so glad. I am a bit confused here.

 

[gamerk316]

Your original stance is 2 cores are all you will ever need. Now its 2 cores + worker threads. Lets see where it goes in 6 months.

My stance was that you won't get nearly as much scaling beyond two cores. DX11 does change the equation a little due to multithreaded rendering, but most implementations I've seen only use 1-2 worker threads for significant amounts of work. And the two primary threads still do about 75% of the total workload, so the two cores they get assigned to are the primary performance drivers.

Physics is the X-factor though; I'm assuming that process intensive physics implementations will be pushed to the GPU, as the GPU is a more efficent processor for such computations. If those calculations are done on the CPU, I could see scaling up to maybe 4-6 cores. But the trend has been toward GPU physics.

for your argument of the ps3 and 360, both did not use many thread because they weren't required, most parallel threads were light because: 1) ps3 could not run all the SPEs loaded because memory constraints so you would get 2-3 decently threaded things to run in parallel. The platform just didn't work out. They got more tricks later on for the sony exclusives but those are never ported to PC and likely won't be good ports anyways 2) 360 supported 6 threads but nobody wants to multithread when they don't need to, thus most games goes down to 3 threads because why use SMT when you can optimize the single thread for the cpu resources, but 360's cpu also required audio processing which could take up an entire CPU on its own. Thus we were left with 2+ bit of large threads running the game on there. 3)Developers are lazy and when porting to PC, they want to target the lowest common denominator, which is dual cores. throwing all the threads into 2 main cpu cores ends up with a port that will run on any system without the need to optimize for the high end.

No, I argue the opposite. The PS360 almost always used 6 threads, which you could do because everything is always finely controlled by the developer. You can NOT do this on PCs, because of the OS. That's the penalty of abstraction, and why the PS360 can still pump out very good looking visuals at times.

And your technical analysis is WAY off. The 6 SPE's could be loaded quite easily, on the provision the data they needed was ready in the systems ESRAM. That was the primary coding challenge of the PS3: Getting the data into ESRAM to keep the SPE's fed (and why Sony's approach on the PS4 is superior overall). Secondly, audio processing takes like 1% CPU time on a single core; its trivial to do. Even a game that handles 128 voices in SW [CoH is the only title I know that does this, most use 64 to stay in EAX2 compliance] doesn't add much to processing.

So yeah, I guess your always right when you make up facts.

We will see what developers can do with 6 symmetrical cores that are completely used for the game. You can keep slowly changing your stance all you want. Of course 2 cores can run the threads, that why minimal requirements of a lot of games are still 1-2 cores. But to get proper performance for the game, its likely 6 cores will destroy 2, 3 and 4 cores.

1: Adding more cores doesn't add performance, even if they are used, if fewer cores already gets the work finished before the GPU finishes its processing.

2: As I've explained: On a PC, you have to worry a lot about data consistency because YOU CAN NOT GUARANTEE WHEN YOUR THREADS RUN, WHAT DATA IS IN MEMORY, OR WHAT RESOURCES YOU HAVE ACCESS TO. Threads that touch the same object MUST be synchronized, meaning THEY CAN NOT RUN AT THE SAME TIME. And in the worst case, your threads may end up locking eachother out for extended periods of time, sending performance off the cliff. Since you have different parts of the game engine touching the same structures, you end up tanking performance as you try and evenly split the workload up between different threads.

Games already use dozens of threads; up to the 80's in some cases. But most do very little processing, operating on chunks of data that the developers know will not be locked at the duration these threads are run. You end up with two heavy threads that handle almost all the processing (Main/Render threads), and the rest handle processing that can be offloaded, but finish so quickly they don't even show up in Task manager.

At the end of the day, you can't thread if the threads need to touch the same data at the same time. Its that simple.

In the coming years, I predict you will say 4 cores is all you will ever need. Then 6 cores is all you will ever need. Maybe some day you will see what the whole semiconductor industry is moving towards.

You apparently weren't here in 2008, when I boldly proclaimed (to JDJ I think) the "Duo's are dead".

I know better now. Quads will offer advantages, as I don't think you can squeeze too many more IPC improvements out of X86 in general, but Duo's will remain more then sufficient for a very long time.

 

[gamerk316]

I am a bit confused in all those intel fans said. I have a few question that I need some real good answers.

1. The software today we use are mainly developped and optimized for intel, are not they?

Not typically. Its not like developers go out of their way to manually plop down CPU opcodes. More often, its algorithmic processing driving performance. And guess what? Anything that doesn't naturally scale favors stronger cores, hence Intel.

2. The AMD architecture is alot more different than an Intel and thus it is emulating Intel, as I know AMD has a RISC core inside, is it right?

The approach is different: Fewer, stronger cores versus many weaker cores. Intel wins at anything until about 6 threads capable of driving each FX core to 100% full load. VERY few applications could do this if they tried.

AMD has a small ARM core in some of their chips now, but it doesn't do much processing wise. Just some extra security stuff.

3. As the software are not optimized for AMD, does they use all the abilities an AMD chip has?

See first two points.

[quote[4. Next year PS4 and XBOX will be using AMD APU's, so all of the games that will be developed in the next years will be optimized for AMD, is it not?[/quote]

Nope. See my previous post on threading, and point #1.

5. As XBOX has an MS operating system inside, MS will also have full optimized OS for AMD next year and can easly port it to PC's.

Easier. You still need to tear apart any of the hardware dependent coding, which likely includes significant amounts of the threading and rendering engines. That being said, the actual PROCESSING itself is unchanged between versions, its just the API's you need to go through, and how you get the data to them, that changes.

6. So next year this time games and OS will be optimized for AMD.

Nope. See previous points.

 

[CooLWoLF]

This thread is absolutely worthless anymore. Seriously. Its become nothing but BLIND BRAND LOYALTY at this point, even in the face of simple logic.

Now I am reading posts saying games will not run anymore optimized on AMD cpus despite the fact that AMD cpus now lie in BOTH major consoles? How hard is it to concede that AMD might get a boost now because of this development? Are people so blind to their love of a tech company that they can't admit something when its staring at them right in the face? Are you for real?

I love technology and everything about it, but then again sometimes I can't really can't stand it because of stupid stuff like this.

 

[esrever]

This thread is absolutely worthless anymore. Seriously. Its become nothing but BLIND BRAND LOYALTY at this point, even in the face of simple logic.

Now I am reading posts saying games will not run anymore optimized on AMD cpus despite the fact that AMD cpus now lie in BOTH major consoles? How hard is it to concede that AMD might get a boost now because of this development? Are people so blind to their love of a tech company that they can't admit something when its staring at them right in the face? Are you for real?

I love technology and everything about it, but then again sometimes I can't really can't stand it because of stupid stuff like this.

consoles are a very small segment of the computing market. AMD is estimating 20% of AMD's revenue, so like 2% total x86 revenue lies in consoles. MS isn't going to change their stance any time soon.

 

[juanrga]

In the fourth place, do you still pretend that anyone can do a CPU+dGPU with UMA/hUMA?

I don't pretend nothing, its on the road map since the very beginning in 2011

http://t1.gstatic.com/images?q=tbn:ANd9GcRvGiOpiosFpbI_dQ0Fhf-2w6RanklsgXec8ZHRWDd0vWt-JvHlSQ

And this slide says its perfectly possible (its not mine its AMD) hUMA/HSA

http://mygaming.co.za/news/wp-content/uploads/2013/05/AMD-Kaveri-hUMA-benefits-600x297.jpg

Just read the slides... and stop wasting time, they say "extend to discrete GPU"... and "10 reason for full coherent hardware" with the GPU without recording for the absence of coherence (mix coherent +non-coherent) and probe filter and directories (coherency directories ala ccNUMA ) (edt)

I already said you that AMD is confused/confusing at this point. The HSA specification says clearly that the HSA memory model is unified (not uniform) and Nvidia makes also clear that its next gen dGPU will include a unified memory model (not uniform). AMD Kaveri memory model will be uniform, but there is no way that the future HSA dGPUs by AMD can have an uniform memory model. Some slides from AMD are contradictory and/or confusing.

Moreover, I already mentioned you the PS4/XboX1 controversy generated by AMD. One day AMD claims that PS4 is hUMA and another day refute the claims and a third day AMD claims otherwise!!!!

 

[christoffe9311]

I honestly can see cpu's ever being what people expect, i have seen people going on about x86 and ARM debating which is better, fact is they are both better for their of reasons, there is no such thing as the best of both worlds it seems to me this is what people are looking for, its the same with cars a petrol is better for accelerating (BHP) and diesel is better for torque (pulling power) at the end of the day its personal preference you will never get the best of both. When you see a true 8 cores cpu clocked at 4ghz or more which has better single core performance than a i5 and efficient as a i7 while staying cool under load you let me know.

 

[esrever]

one thing good about the cpu in consoles is now there won't be much floatpoint math for the CPU which is where AMD lags intel. I don't expect the CPU to be very important in gaming in the future. So long as you are able to run the threads, the single core performance won't matter as much.

 

[juanrga]

ARM Aiming for 5%-10% Share in Global Server Chip Market by 2016

Intel's Server Market Share To Decline As Competition Heats Up

AMD's Feldman: ARM Will Quickly Gain Market Share in Servers

Interesting that Feldman mentions: SPARC --> x86 --> ARM

Nvidia Launches CUDA Support for ARM Server Chips

Nvidia sharing its vision about high-performance ARM dominating everything from mobile to supercomputers

AMD 'secret' slide

Apple engineers exploring a switch to high-performance ARM chips from Intel chips

I agree with AMD, Nvidia, Apple... and applaud them.

so why didn't mips win yet?

This is a very good question. Why AMD, Nvidia, Dell, HP, BSC... all them have selected ARM over MIPS? Why is Apple considering ARM instead MIPS?

Because efficiency, scalability, clean design, cost, available tools...

ARM Aiming for 5%-10% Share in Global Server Chip Market by 2016

Intel's Server Market Share To Decline As Competition Heats Up

AMD's Feldman: ARM Will Quickly Gain Market Share in Servers

Interesting that Feldman mentions: SPARC --> x86 --> ARM

Nvidia Launches CUDA Support for ARM Server Chips

Nvidia sharing its vision about high-performance ARM dominating everything from mobile to supercomputers

AMD 'secret' slide

Apple engineers exploring a switch to high-performance ARM chips from Intel chips

I agree with AMD, Nvidia, Apple... and applaud them.

The first one, ARM themselves are being realistic. 5-10% by 2016, I can see that possibly happening. Here is the thing. that 5-10% is going to be split between the 8 companies that have already announced making microserver ARM cpus. AMD is pushing hard to get as much of that share as possible. 8 or more companies fighting over 5-10% overall marketshare isn't a whole lot for AMD to drop the ball on current products and go for table scrapping with the likes of nvidia, qualcomm, and samsung.

2nd article, of course Intel's shares are going to go down. As per the article "Intel has managed to increase its share from 86.6% in 2008 to 95.6% in 2012, as per our estimate." How much higher can you go above 95%? Can't read page 2 of that article, some stupid "register to read the rest of the article"

3rd article is nothing but marketing hype. 6 times they talk about small work loads, small cores, smaller computers ... but your claims are that ARM is this HIGH PERFORMANCE part. Unrealistic : by 2016 ARM is going to have 20% or more server market share. Of course, this could be construed to mean per core sales as it will take what 32-64 cores to equal one xeon E5/E7. ARM breaking x86 stranglehold isn't going to just magically happen. Intel has billions of dollars at stake to make sure its not an overnight takeover.

4th article .. duh ... if NVidia doens't support cuda on arm, then they don't even support their own tegra cpu within itself.

The last article pretty much sums it up right here.

And if Apple can’t design ARM-based chips that are far more powerful than current models, the company would probably need to stay with Intel to satisfy Mac power users, who need lots of computing performance for tasks such as developing software or doing high-end graphics.

ya .. arm is faster than x86 right now yet ^^ High End ARM is just that and only that, High End for ARM processors, not high end over all cpus ever made. This seems to be a mis-thought process that just because something has the words "high end", it means over all other products that are remotely close.

ARM isn't just going to magically displace x86. Aside from being slower, you will have compatiblity problems (such as windows RT can only run ARM compatible programs) That means that anyone wanting to migrate from X86 to ARM, some of them are likely going to have to keep the x86 system just for the program compatibility.

All the ARM takeover right now is wishful thinking. Can it happen? sure Will it happen next year? not a chance. Should AMD abandon current x86 designs and jump to the ARM bandwagon? wouldn't be the smartest business decision. Is intel going to put up a fight to deny ARM takover? you bet your ass they will.

That 5-10% is only a prediction for 2016. Companies are not moving to ARM to obtain a 5-10% of the market. Their plans go beyond 2016 and towards taking most of the market. Moreover, I applaud competition. I prefer to see half dozen of companies in the server space before seeing only one (Intel) and almost no innovation.

I already explained you that AMD is pushing hard for ARM, but will continue supporting the traditional market. The new Warsaw chips are aimed to big corporations and other customers that will be switching to ARM slowly.

The 2nd article was not saying that Intel share would increase. In fact, the share could remain almost constant. What the article says, and this is the important thing, is that Intel share will drop. It will drop because competitors will be offering better products.

3rd article gives several relevant data. It is evident that Seattle is a High-Performance chip. Look to the embedded roadmap recently presented. AMD includes ARM in the "High-Performance" tie, together Steamroller, but puts Jaguar outside the tie.

The "32-64 cores to equal Xeon" looks as a typical handipur nonsense. Seattle comes as 8-16 cores. I already said you which will be the performance of the 16 cores version. Seattle will have about the same performance than one of the most fastest Xeons. I gave you the specific Xeon model.

You seem to interpret this as a traditional AMD vs Intel war, but there are more players in the game and some of them are much more powerful and BIG than Intel.

The relevant part of the 4th article is that NVidia introduces CUDA support for ARM __servers__.

You quote a part of the last article but there is another part which is much more interesting:

Apple engineers have grown confident that the chip designs used for its mobile devices will one day be powerful enough to run its desktops and laptops, said three people with knowledge of the work, who asked to remain anonymous because the plans are confidential.

Windows RT fiasco doesn't count. Or do you believe that companies are moving to ARM servers/laptops/supercomputers to run RT on them?

Windows is currently a minority OS in global numbers. Linux/Android has bigger market share. The entire industry is moving away from Wintel.

Nobody said you that all this will happen next year. Look tot he slide again. It says long run. Moreover, I already did an estimation and it was not 2014.