AMD CPU speculation... and expert conjecture

[8350rocks]

@8350rocks: gotta say, nicely placed info. just enough gap for "but that's not what i said before." later. ;D
btw, can you tell me if a10 7850k's pricing is also under n.d.a.? i wanted to know why amd priced the apu slightly south of $200 (which retailers brought closer to $200 at launch) until the most recent price drop. i want to match your findings with my hypotheses. i am certain that even amd knew that 7850k isn't worth paying that much premium over richland 6800k without the established hsa ecosystem, in 7850k's target market and price range. at least some of these are gonna come out anyway in their next earnings call.

as for the new info, my newbiespeculationizer (nspz) machine says the new x86 core could be a heterogenous multiprocessor uarch with 4-8 big cores with 4-8 small cores like puma+ or cheetah (i can't see a57/k12 in that without massive, incredibly reworked iommu). those take up really little place. or gcn 2.0 cores instead of gcn. 8350rocks never said which gcn iteration.

i think carrizo's (excavator) promised upgrades require at least some sort of cache-reworking. but the real improvement might be in the igpu as usual.

amd should take calxeda and nvidia as cautionary tales because arm's big casino isn't all fun and games. while arm will always win, calxeda closed down and nvidia almost got squeezed out of consumer smartphone/tablet market. nvidia's project debacle.. err... i mean denver is still m.i.a. we know jetson has 4x a15 cores! :lol:

edit:

Steamboy is a handheld Steam Machine
http://www.nextpowerup.com/news/9692/steamboy-is-a-handheld-steam-machine.html
mullins? i don't think intel has a gaming i/gpu.

For the next x86 core, forget small cores, and dream bigger....(think 8-12 for a flagship SKU).

Alright, no more hints...for now...

 

[etayorius]

I never understood that Slide a year ago and a half about the FX line would go all through 2015, that means new FX or just the current FX will be sold until 2015? I wish they could just get rid of Piledriver already... it`s been like 2 years since Piledriver arrived and we barely got small bumps in the Speed of the CPUs instead of improving the arch or replace with a newer one.

 

[8350rocks]

Oh, keep your eyes peeled for AMD Gamer Series branded SSD from an AMD partner...

 

[8350rocks]

The x86 core will most definitely be in an APU. K12 will be an APU too. Nothing announced so far will feature both ARM and x86 cores. Any HEDT space AMD will enter will be an HSA accelerated one.

It might fit into an APU board, and use an APU die...though there is no reason that the die has to have GPU cores...

Just saying...

 

[de5_Roy]

Oh, keep your eyes peeled for AMD Gamer Series branded SSD from an AMD partner...

so it'll be called "amd gamer series". i thought it'd be called "radeon ssd" from an article i pat.. i mean parroted.

 

[juanrga]

Back from Austin...LOTS of things I cannot talk about, but some things that I can...

1.) K12 is intended specifically for servers, and the single platform boards are intended for servers as well.

2.) AMD is not done in the BIG server space.

3.) Next gen HEDT is being reworked from scratch, it will be x86, and they are saying some pretty interesting things. It will not be an APU, it has an internal code name that has not yet been released so I cannot tell you what it is called, though the comments I got were along the lines of..."Well, who is to say the same die we use for an APU has to have GCN cores at all...?" (Obviously cannot say more...)

4.) Carrizo is going to be a NICE performance increase, in more ways than one...I cannot give details beyond that...though expect large gains in GPU and large gains in CPU performance for multiple reasons.

1) This is untrue.

2) Everyone knew that. The same that everyone knew that AMD is not done in HEDT.

3) It was already said to you that new x86 arch starts from zero and abandons CMT architecture of the Bulldozer family. I already gave before some features of the future FX line, such as FPU, cache tech. number of cores... Contrary to your claims the new x86 architecture will appear in APUs.

4) Shhhh! don't say to collinp, he still believes that Carrizo was cancelled.

Arm in high-end no only in juan's mind i agree with 8350fxrocks findings!

LOL

K12 is ARM cores ONLY. Internal codename for the x86 project is actually a different name. AMD does not anticipate ARM to enter consumer market space for DT PCs anytime soon.

@juanrga:

Because I know you are going to roll out the slide...that is a reference to ULP SERVERS.

Everyone knows that K12 is the name of the new ARM core. Everyone knows that the x86 sister is not still labeled. Everyone knows that since months ago.

As mentioned before ARM will enter first in servers/HPC, then laptops, and desktop will be the last platffform to migrate. Thus you are repeating stuff that I have been saygin here for months.

And no, the slide is not a reference to ULP SERVERS, never was ;-)

Also by the time HEDT dCPU hits, fab issues will have been resolved...there is another company they are looking at possibly partnering with if things get rough with GloFo and/or TSMC. They will have more options at that node.

As mentioned before, this is just one of the advantages from abandoning SOI. By selecting FINFET for the 14/16 node, AMD is open to almost any foundry that uses FinFET for he node, which is the 90% of the foundries.

Again you are not saying anything new.

as for the new info, my newbiespeculationizer (nspz) machine says the new x86 core could be a heterogenous multiprocessor uarch with 4-8 big cores with 4-8 small cores like puma+ or cheetah (i can't see a57/k12 in that without massive, incredibly reworked iommu).

Sorry, but this makes the same sense (i.e. none) that your new-Kaveri-VRAM 'speculation'.

 

[8350rocks]

Back from Austin...LOTS of things I cannot talk about, but some things that I can...

1.) K12 is intended specifically for servers, and the single platform boards are intended for servers as well.

2.) AMD is not done in the BIG server space.

3.) Next gen HEDT is being reworked from scratch, it will be x86, and they are saying some pretty interesting things. It will not be an APU, it has an internal code name that has not yet been released so I cannot tell you what it is called, though the comments I got were along the lines of..."Well, who is to say the same die we use for an APU has to have GCN cores at all...?" (Obviously cannot say more...)

4.) Carrizo is going to be a NICE performance increase, in more ways than one...I cannot give details beyond that...though expect large gains in GPU and large gains in CPU performance for multiple reasons.

1) This is untrue.

2) Everyone knew that. The same that everyone knew that AMD is not done in HEDT.

3) It was already said to you that new x86 arch starts from zero and abandons CMT architecture of the Bulldozer family. I already gave before some features of the future FX line, such as FPU, cache tech. number of cores... Contrary to your claims the new x86 architecture will appear in APUs.

4) Shhhh! don't say to collinp, he still believes that Carrizo was cancelled.

Arm in high-end no only in juan's mind i agree with 8350fxrocks findings!

LOL

K12 is ARM cores ONLY. Internal codename for the x86 project is actually a different name. AMD does not anticipate ARM to enter consumer market space for DT PCs anytime soon.

@juanrga:

Because I know you are going to roll out the slide...that is a reference to ULP SERVERS.

Everyone knows that K12 is the name of the new ARM core. Everyone knows that the x86 sister is not still labeled. Everyone knows that since months ago.

As mentioned before ARM will enter first in servers/HPC, then laptops, and desktop will be the last platffform to migrate. Thus you are repeating stuff that I have been saygin here for months.

And no, the slide is not a reference to ULP SERVERS, never was ;-)

Also by the time HEDT dCPU hits, fab issues will have been resolved...there is another company they are looking at possibly partnering with if things get rough with GloFo and/or TSMC. They will have more options at that node.

As mentioned before, this is just one of the advantages from abandoning SOI. By selecting FINFET for the 14/16 node, AMD is open to almost any foundry that uses FinFET for he node, which is the 90% of the foundries.

Again you are not saying anything new.

as for the new info, my newbiespeculationizer (nspz) machine says the new x86 core could be a heterogenous multiprocessor uarch with 4-8 big cores with 4-8 small cores like puma+ or cheetah (i can't see a57/k12 in that without massive, incredibly reworked iommu).

Sorry, but this makes the same sense (i.e. none) that your new-Kaveri-VRAM 'speculation'.

1. So you claim to know more than John Byrne...the CSO of AMD? (if you were following me on twitter you would already know who my source is by now, so I can tell you that here...) I think that says everything I need to say...

EDIT: You are right about bulk FinFET @ 14nm, though past that...

 

[-Fran-]

I got a little confused... If K12 is going to be a server class uArch, does that mean that AMD plans to scale up Puma/Cheetah for the next gen APUs or do they plan to use K12 in APUs as well? Since K12 might land a little after Carrizo, I'm not sure about the timings, haha.

I mean, having 3 uArches is kind of too much, isn't it? At least X86 uArches.

Cheers!

 

[8350rocks]

K12 = ARM

Next x86 uarch has an internal name, but they are not announcing it yet.

Additionally...cat cores will stay LP/ULP, K12 will scale up/down for ARM, and next x86 uarch is HEDT/Server

Just FYI: Profit on x86 desktop parts has the highest margins of any product AMD offers next to dGPUs. Server margins are high, though they are rethinking their strategy for large scale servers, and will have ARM and x86 solutions for that. AMD has decided to go after HEDT by storm, and it will be coming...but not yet...

Additionally, AMD does not think ARM will ever take hold in desktop (in any meaningful scale). The software ecosystem is too divergent, the vendor buy in for x86 applications conversion is nil, and the consumer does not care about power consumption in desktop PCs because there is no battery involved.

Also, announcements regarding MANTLE drivers and Linux are coming soon...

 

[jdwii]

K12 = ARM

Next x86 uarch has an internal name, but they are not announcing it yet.

Additionally...cat cores will stay LP/ULP, K12 will scale up/down for ARM, and next x86 uarch is HEDT/Server

Just FYI: Profit on x86 desktop parts has the highest margins of any product AMD offers next to dGPUs. Server margins are high, though they are rethinking their strategy for large scale servers, and will have ARM and x86 solutions for that. AMD has decided to go after HEDT by storm, and it will be coming...but not yet...

Additionally, AMD does not think ARM will ever take hold in desktop (in any meaningful scale). The software ecosystem is too divergent, the vendor buy in for x86 applications conversion is nil, and the consumer does not care about power consumption in desktop PCs because there is no battery involved.

Also, announcements regarding MANTLE drivers and Linux are coming soon...

That sounds about right 8350rocks wish i could of went to far away from anything like that, i'll probably upgrade to it when it comes out until then this 8350fx is fun to play with and it performs very well in all my games with this 770(basically can max out everything i own with smooth gameplay) when i do anything else it uses all my 8 cores and well as fast as an I7, Might get a better heat-sink though to OC this thing more just for fun. I was a little sad that my chip wasn't a very good overclocker.

I was never told much of anything when i asked a while ago all i was told was Arm is not going to replace X86 its for newer markets ones that are now starting that Arm is better for compared to X86. X86 will still be used in high-end servers as well as Gaming machines. It’s important for everyone here to remember what "high end" really is it’s not I5's or 8 core FX’s its much higher than that those 2 parts are mid-range CPU's.

A lot of things I hear on this site seem a bit off such as Arm beating X86 in performance or APU’s with 1 piece of silicon beating 2 pieces of silicon in performance. Or even W8 being a very heavy OS when Ubuntu uses more ram under my testing and we have like 3 OS’s on the Xbox one 8 based and that is not limiting its performance at all its inferior GPU is, not to mention both have very low-end CPU’s. I also still get a laugh at the A10 equals a I5 thing wonder if anyone in their right mind thinks that true when most benchmarks show it below the A10 6800K in terms of CPU performance.

 

[cdrkf]

K12 = ARM

Next x86 uarch has an internal name, but they are not announcing it yet.

Additionally...cat cores will stay LP/ULP, K12 will scale up/down for ARM, and next x86 uarch is HEDT/Server

Just FYI: Profit on x86 desktop parts has the highest margins of any product AMD offers next to dGPUs. Server margins are high, though they are rethinking their strategy for large scale servers, and will have ARM and x86 solutions for that. AMD has decided to go after HEDT by storm, and it will be coming...but not yet...

Additionally, AMD does not think ARM will ever take hold in desktop (in any meaningful scale). The software ecosystem is too divergent, the vendor buy in for x86 applications conversion is nil, and the consumer does not care about power consumption in desktop PCs because there is no battery involved.

Also, announcements regarding MANTLE drivers and Linux are coming soon...

That sounds about right 8350rocks wish i could of went to far away from anything like that, i'll probably upgrade to it when it comes out until then this 8350fx is fun to play with and it performs very well in all my games with this 770(basically can max out everything i own with smooth gameplay) when i do anything else it uses all my 8 cores and well as fast as an I7, Might get a better heat-sink though to OC this thing more just for fun. I was a little sad that my chip wasn't a very good overclocker.

I was never told much of anything when i asked a while ago all i was told was Arm is not going to replace X86 its for newer markets ones that are now starting that Arm is better for compared to X86. X86 will still be used in high-end servers as well as Gaming machines. It’s important for everyone here to remember what "high end" really is it’s not I5's or 8 core FX’s its much higher than that those 2 parts are mid-range CPU's.

A lot of things I hear on this site seem a bit off such as Arm beating X86 in performance or APU’s with 1 piece of silicon beating 2 pieces of silicon in performance. Or even W8 being a very heavy OS when Ubuntu uses more ram under my testing and we have like 3 OS’s on the Xbox one 8 based and that is not limiting its performance at all its inferior GPU is, not to mention both have very low-end CPU’s. I also still get a laugh at the A10 equals a I5 thing wonder if anyone in their right mind thinks that true when most benchmarks show it below the A10 6800K in terms of CPU performance.

The ARM > X86 argument is a complex one. It kinda is and kinda isn't. As I understand things at the moment, ARM has an advantage in perf / w, however the outright maximum performance of any particular ARM core is lower than the high performance X86 cores by quite a margin. If you have the perf / w advantage though, that does allow you to squeeze more total performance out of a machine by using lots of cores / lots of processors which is important for servers- provided your software can scale adequately and has sufficient bandwidth between processors to actually make use of it. This is where the 'freedom fabric' comes into play with AMD's SeaMicro servers- they address the bandwidth and infrastructure concerns and will allow lots and lots of efficient ARM cores to work together to (in theory at least) provide the maximum total performance at a given level of power consumption.

Also the idea being a single APU performing better than CPU + dGPU does have some basis in reality. The problem with offloading compute to a dGPU is the fact that (as things currently stand) there are a number of operations like memory copy and such that have to take place *before* the dGPU can do anything. The result is that GPU acceleration of large tasks can yield big performance gains, however for smaller tasks (which account for the bulk of work a computer does) the benefits are wiped out by the other operations. With AMD's new HSA capable APUs, the 2 components are not only on the same silicon but also share coherent memory, so it eliminates most of the delays for executing things on the GPU portion of the chip. This means *in theory* that the gpu can be used much more frequently improving overall performance.

This issue with both of the above is both cases rely on specific workloads (in the case of the Seamicro + ARM servers) or significant changes to the software ecosystem (HSA) to be fully realised. As with allot of things AMD have done, they have lots of potential however it might take a while before it's fully realised (for example 64 bit machines are finally gaining traction, only took 11 years...)

 

[juanrga]

1. So you claim to know more than John Byrne...the CSO of AMD? (if you were following me on twitter you would already know who my source is by now, so I can tell you that here...) I think that says everything I need to say...

EDIT: You are right about bulk FinFET @ 14nm, though past that...

I don't claim what you pretend. I don't know if you are lying or if he just said you what you want to hear or if he said one thing and you understood another or what.

What I know is that your claim that "K12 is intended specifically for servers" is untrue. K12 will be used in a range of products beyond servers. You can post so many irrelevant pics in your twitter account as you desire, my comment stands.

Additionally, AMD does not think ARM will ever take hold in desktop (in any meaningful scale). The software ecosystem is too divergent, the vendor buy in for x86 applications conversion is nil, and the consumer does not care about power consumption in desktop PCs because there is no battery involved.

The reason why ARM will be used in desktops has nothing to do with power consumption and batteries. I use this answer to link to some last news about it:

While it will still be a while before consumers are able to see 64-bit ARM hardware on their desk, Mozilla's Firefox web-browser on AArch64 (64-bit ARM) is working.

http://www.phoronix.com/scan.php?page=news_item&px=MTcxODU

A lot of things I hear on this site seem a bit off such as Arm beating X86 in performance or APU’s with 1 piece of silicon beating 2 pieces of silicon in performance. Or even W8 being a very heavy OS when Ubuntu uses more ram under my testing and we have like 3 OS’s on the Xbox one 8 based and that is not limiting its performance at all its inferior GPU is, not to mention both have very low-end CPU’s. I also still get a laugh at the A10 equals a I5 thing wonder if anyone in their right mind thinks that true when most benchmarks show it below the A10 6800K in terms of CPU performance.

All of this was carefully explained and demonstrated to you before, including citation to academic works and rigorous benchmarks. Your misconceptions/beliefs were corrected more than one time.

About your claims about Ubuntu... well Ubuntu is less bloated than W8, however, Ubuntu is still a bloated system. It is typical of only-windows users to believe that Ubuntu is representative of the whole linux world. Just because Ubuntu has something so bloated and odd as Unity doesn't mean that rest of distros have; at contrary, Unity is a Canonical LTD only stuff.

This misconception of only-windows users is understandable. Windows is about giving the same mediocre tech to everyone, thus when only-windows users test some linux distro they think than any linux distro will be the same than that they tested, when this is very far from reality.

Ubuntu is linux but linux is not Ubuntu.

 

[juanrga]

The ARM > X86 argument is a complex one. It kinda is and kinda isn't. As I understand things at the moment, ARM has an advantage in perf / w, however the outright maximum performance of any particular ARM core is lower than the high performance X86 cores by quite a margin. If you have the perf / w advantage though, that does allow you to squeeze more total performance out of a machine by using lots of cores / lots of processors which is important for servers- provided your software can scale adequately and has sufficient bandwidth between processors to actually make use of it. This is where the 'freedom fabric' comes into play with AMD's SeaMicro servers- they address the bandwidth and infrastructure concerns and will allow lots and lots of efficient ARM cores to work together to (in theory at least) provide the maximum total performance at a given level of power consumption.

Performance = efficiency * power

ARM superior efficiency is the reason why it beats any other ISA at same power consumption

At ~2W, ARM beats any other phone chip, including Intel.

At ~25W, ARM beats small x86 core chips, including AMD jaguar, this is why AMD is replacing its jaguar server by Seattle servers.

At ~90W, ARM beats x86 high-performance design. This is why some companies are already preparing to beat high-performance 140W Xeon chips from Intel.

 

[Embra]

@8350rocks: gotta say, nicely placed info. just enough gap for "but that's not what i said before." later. ;D
btw, can you tell me if a10 7850k's pricing is also under n.d.a.? i wanted to know why amd priced the apu slightly south of $200 (which retailers brought closer to $200 at launch) until the most recent price drop. i want to match your findings with my hypotheses. i am certain that even amd knew that 7850k isn't worth paying that much premium over richland 6800k without the established hsa ecosystem, in 7850k's target market and price range. at least some of these are gonna come out anyway in their next earnings call.

as for the new info, my newbiespeculationizer (nspz) machine says the new x86 core could be a heterogenous multiprocessor uarch with 4-8 big cores with 4-8 small cores like puma+ or cheetah (i can't see a57/k12 in that without massive, incredibly reworked iommu). those take up really little place. or gcn 2.0 cores instead of gcn. 8350rocks never said which gcn iteration.

i think carrizo's (excavator) promised upgrades require at least some sort of cache-reworking. but the real improvement might be in the igpu as usual.

amd should take calxeda and nvidia as cautionary tales because arm's big casino isn't all fun and games. while arm will always win, calxeda closed down and nvidia almost got squeezed out of consumer smartphone/tablet market. nvidia's project debacle.. err... i mean denver is still m.i.a. we know jetson has 4x a15 cores! :lol:

edit:

Steamboy is a handheld Steam Machine
http://www.nextpowerup.com/news/9692/steamboy-is-a-handheld-steam-machine.html
mullins? i don't think intel has a gaming i/gpu.

For the next x86 core, forget small cores, and dream bigger....(think 8-12 for a flagship SKU).

Alright, no more hints...for now...

3 ALU's per core?

 

[cdrkf]

The ARM > X86 argument is a complex one. It kinda is and kinda isn't. As I understand things at the moment, ARM has an advantage in perf / w, however the outright maximum performance of any particular ARM core is lower than the high performance X86 cores by quite a margin. If you have the perf / w advantage though, that does allow you to squeeze more total performance out of a machine by using lots of cores / lots of processors which is important for servers- provided your software can scale adequately and has sufficient bandwidth between processors to actually make use of it. This is where the 'freedom fabric' comes into play with AMD's SeaMicro servers- they address the bandwidth and infrastructure concerns and will allow lots and lots of efficient ARM cores to work together to (in theory at least) provide the maximum total performance at a given level of power consumption.

Performance = efficiency * power

ARM superior efficiency is the reason why it beats any other ISA at same power consumption

At ~2W, ARM beats any other phone chip, including Intel.

At ~25W, ARM beats small x86 core chips, including AMD jaguar, this is why AMD is replacing its jaguar server by Seattle servers.

At ~90W, ARM beats x86 high-performance design. This is why some companies are already preparing to beat high-performance 140W Xeon chips from Intel.

Yeah I agree- I wasn't disputing any of that. The thing that is likely to confuse people though is that to my knowledge there currently aren't any ARM cores competing in the ~90W space, so x86 appears (in pure performance terms) faster core for core, but it isn't as efficient.

What I was describing is related to the AMD A57 server parts that we will see first. They're not a particularly high performance core (as ARM designs are all geared towards low power devices) so 1 A57 core isn't going to outrun an Intel or AMD 'big' core in absolute terms. It should be much more efficient though so a server with lots of them makes allot of sense.

Based on the titbits of information released on K12, that should be an ARM based 'big' core which can be more readily compared to x86 'big' cores. It will be interesting to see how well ARM scales up to much higher TDP, and I think AMD are the ideal company to do it.

 

[Elenadavid]

Still saying today, in the latest article. His 'articles' read more like an evil old man chuckling and rubbing his hands together, rather than a news article .

 

[cdrkf]

@8350rocks: gotta say, nicely placed info. just enough gap for "but that's not what i said before." later. ;D
btw, can you tell me if a10 7850k's pricing is also under n.d.a.? i wanted to know why amd priced the apu slightly south of $200 (which retailers brought closer to $200 at launch) until the most recent price drop. i want to match your findings with my hypotheses. i am certain that even amd knew that 7850k isn't worth paying that much premium over richland 6800k without the established hsa ecosystem, in 7850k's target market and price range. at least some of these are gonna come out anyway in their next earnings call.

as for the new info, my newbiespeculationizer (nspz) machine says the new x86 core could be a heterogenous multiprocessor uarch with 4-8 big cores with 4-8 small cores like puma+ or cheetah (i can't see a57/k12 in that without massive, incredibly reworked iommu). those take up really little place. or gcn 2.0 cores instead of gcn. 8350rocks never said which gcn iteration.

i think carrizo's (excavator) promised upgrades require at least some sort of cache-reworking. but the real improvement might be in the igpu as usual.

amd should take calxeda and nvidia as cautionary tales because arm's big casino isn't all fun and games. while arm will always win, calxeda closed down and nvidia almost got squeezed out of consumer smartphone/tablet market. nvidia's project debacle.. err... i mean denver is still m.i.a. we know jetson has 4x a15 cores! :lol:

edit:

Steamboy is a handheld Steam Machine
http://www.nextpowerup.com/news/9692/steamboy-is-a-handheld-steam-machine.html
mullins? i don't think intel has a gaming i/gpu.

For the next x86 core, forget small cores, and dream bigger....(think 8-12 for a flagship SKU).

Alright, no more hints...for now...

3 ALU's per core?

That would bring it in-line with Phenom II at least... hopefully they're shooting a bit further though.

 

[gamerk316]

IHS pointed out exactly why Intel is moving aggressively into the Car market:

http://spectrum.ieee.org/cars-that-think/transportation/self-driving/driverless-cars-a-bonanza-for-chipmakers

According to a new report from analysis firm IHS Technology, microcontroller and processor units for autonomous vehicles will be a half-billion-dollar market by 2020, up 625 percent from $69 million in 2013.

Intel's following the growth. Makes sense, though MIPS has a VERY strong presence here.

 

[gamerk316]

Also the idea being a single APU performing better than CPU + dGPU does have some basis in reality. The problem with offloading compute to a dGPU is the fact that (as things currently stand) there are a number of operations like memory copy and such that have to take place *before* the dGPU can do anything. The result is that GPU acceleration of large tasks can yield big performance gains, however for smaller tasks (which account for the bulk of work a computer does) the benefits are wiped out by the other operations. With AMD's new HSA capable APUs, the 2 components are not only on the same silicon but also share coherent memory, so it eliminates most of the delays for executing things on the GPU portion of the chip. This means *in theory* that the gpu can be used much more frequently improving overall performance.

The memory point is worth discussing here. Yes, HSA allows the CPU and GPU to exist in the same memory context. The downside to this is the fact that VRAM has to go away, leaving the GPU at the mercy of main memory bandwidth. We see the results when large datasets are processed: The GPU gets starved. Contrast that to non-HSA dGPU's, with, what, 4GB of VRAM now? You have a high initial latency, but after that, the higher bandwidth of the VRAM makes up the initial performance disadvantage.

Hence the downside of HSA: Everytime you can't fit a dataset into the cache, you have to access main memory again, which is slow (from a computational standpoint). dGPUs just get around this with massive amounts of VRAM. If you want a dedicated gaming APU, I really can't see how to do it without greatly expanding the size of the CPU cache, upward to the hundreds of MB, at a minimum, simply due to the relative slowness of main memory.

 

[cdrkf]

Also the idea being a single APU performing better than CPU + dGPU does have some basis in reality. The problem with offloading compute to a dGPU is the fact that (as things currently stand) there are a number of operations like memory copy and such that have to take place *before* the dGPU can do anything. The result is that GPU acceleration of large tasks can yield big performance gains, however for smaller tasks (which account for the bulk of work a computer does) the benefits are wiped out by the other operations. With AMD's new HSA capable APUs, the 2 components are not only on the same silicon but also share coherent memory, so it eliminates most of the delays for executing things on the GPU portion of the chip. This means *in theory* that the gpu can be used much more frequently improving overall performance.

The memory point is worth discussing here. Yes, HSA allows the CPU and GPU to exist in the same memory context. The downside to this is the fact that VRAM has to go away, leaving the GPU at the mercy of main memory bandwidth. We see the results when large datasets are processed: The GPU gets starved. Contrast that to non-HSA dGPU's, with, what, 4GB of VRAM now? You have a high initial latency, but after that, the higher bandwidth of the VRAM makes up the initial performance disadvantage.

Hence the downside of HSA: Everytime you can't fit a dataset into the cache, you have to access main memory again, which is slow (from a computational standpoint). dGPUs just get around this with massive amounts of VRAM. If you want a dedicated gaming APU, I really can't see how to do it without greatly expanding the size of the CPU cache, upward to the hundreds of MB, at a minimum, simply due to the relative slowness of main memory.

True, or simply link the APU up to a 256 bit DDR5 memory pool (PS4).

 

[gamerk316]

Also the idea being a single APU performing better than CPU + dGPU does have some basis in reality. The problem with offloading compute to a dGPU is the fact that (as things currently stand) there are a number of operations like memory copy and such that have to take place *before* the dGPU can do anything. The result is that GPU acceleration of large tasks can yield big performance gains, however for smaller tasks (which account for the bulk of work a computer does) the benefits are wiped out by the other operations. With AMD's new HSA capable APUs, the 2 components are not only on the same silicon but also share coherent memory, so it eliminates most of the delays for executing things on the GPU portion of the chip. This means *in theory* that the gpu can be used much more frequently improving overall performance.

The memory point is worth discussing here. Yes, HSA allows the CPU and GPU to exist in the same memory context. The downside to this is the fact that VRAM has to go away, leaving the GPU at the mercy of main memory bandwidth. We see the results when large datasets are processed: The GPU gets starved. Contrast that to non-HSA dGPU's, with, what, 4GB of VRAM now? You have a high initial latency, but after that, the higher bandwidth of the VRAM makes up the initial performance disadvantage.

Hence the downside of HSA: Everytime you can't fit a dataset into the cache, you have to access main memory again, which is slow (from a computational standpoint). dGPUs just get around this with massive amounts of VRAM. If you want a dedicated gaming APU, I really can't see how to do it without greatly expanding the size of the CPU cache, upward to the hundreds of MB, at a minimum, simply due to the relative slowness of main memory.

True, or simply link the APU up to a 256 bit DDR5 memory pool (PS4).

GDDR5. But then platform cost comes into play, given the higher cost of GDDR. So you're trading off cost for performance, which while attractive at the high end, is a no-go for 90% of the market.

 

[8350rocks]

1. So you claim to know more than John Byrne...the CSO of AMD? (if you were following me on twitter you would already know who my source is by now, so I can tell you that here...) I think that says everything I need to say...

EDIT: You are right about bulk FinFET @ 14nm, though past that...

I don't claim what you pretend. I don't know if you are lying or if he just said you what you want to hear or if he said one thing and you understood another or what.

What I know is that your claim that "K12 is intended specifically for servers" is untrue. K12 will be used in a range of products beyond servers. You can post so many irrelevant pics in your twitter account as you desire, my comment stands.

Additionally, AMD does not think ARM will ever take hold in desktop (in any meaningful scale). The software ecosystem is too divergent, the vendor buy in for x86 applications conversion is nil, and the consumer does not care about power consumption in desktop PCs because there is no battery involved.

The reason why ARM will be used in desktops has nothing to do with power consumption and batteries. I use this answer to link to some last news about it:

While it will still be a while before consumers are able to see 64-bit ARM hardware on their desk, Mozilla's Firefox web-browser on AArch64 (64-bit ARM) is working.

http://www.phoronix.com/scan.php?page=news_item&px=MTcxODU

A lot of things I hear on this site seem a bit off such as Arm beating X86 in performance or APU’s with 1 piece of silicon beating 2 pieces of silicon in performance. Or even W8 being a very heavy OS when Ubuntu uses more ram under my testing and we have like 3 OS’s on the Xbox one 8 based and that is not limiting its performance at all its inferior GPU is, not to mention both have very low-end CPU’s. I also still get a laugh at the A10 equals a I5 thing wonder if anyone in their right mind thinks that true when most benchmarks show it below the A10 6800K in terms of CPU performance.

All of this was carefully explained and demonstrated to you before, including citation to academic works and rigorous benchmarks. Your misconceptions/beliefs were corrected more than one time.

About your claims about Ubuntu... well Ubuntu is less bloated than W8, however, Ubuntu is still a bloated system. It is typical of only-windows users to believe that Ubuntu is representative of the whole linux world. Just because Ubuntu has something so bloated and odd as Unity doesn't mean that rest of distros have; at contrary, Unity is a Canonical LTD only stuff.

This misconception of only-windows users is understandable. Windows is about giving the same mediocre tech to everyone, thus when only-windows users test some linux distro they think than any linux distro will be the same than that they tested, when this is very far from reality.

Ubuntu is linux but linux is not Ubuntu.

The pic I was referring to was the group I was there with standing with John Byrne...call that irrelevant all you want.

Also, I will take his word over yours any day as to what they are doing. I have already said this before.

As for upcoming architecture, you have guessed right where I gave you credit. The rest is all speculation, and inaccurate at best, laughably off the mark at worst.

We will see when it comes...how does that sound? In the meantime, drop the ARM BS and let the thread discuss what it is actually here to discuss.

 

[8350rocks]

Oh, keep your eyes peeled for AMD Gamer Series branded SSD from an AMD partner...

so it'll be called "amd gamer series". i thought it'd be called "radeon ssd" from an article i pat.. i mean parroted.

It will likely be branded "AMD Radeon Gamer Series"...just like the RAM.

 

[kviksand81]

Maybe this will help AMD's R&D a bit:

Intel loses 1.06bn euro fine appeal [To AMD]:

http://www.bbc.com/news/business-27809112

 

[8350rocks]

Maybe this will help AMD's R&D a bit:

Intel loses 1.06bn euro fine appeal [To AMD]:

http://www.bbc.com/news/business-27809112

Ahh, some justice in the world...