AMD CPU speculation... and expert conjecture

[palladin9479]

Anyone mining cryptocurrency these days using GPUs has mostly missed the boat. FPGAs and ASICs are where the big money in that is, and it is one of the few ways to still turn a profit doing it. However, as I understand it, there are other less in demand currencies than bitcoin, so that may be what people are using those radeons for at this point. However, I know little of the newer currency and bitcoin is pretty tapped out at this point.

There is no more money to be made in mining BTC. Speculation has already took over and hyper-inflated the price. The moment people started making cheap custom ASIC's it was over with as nothing anyone else can do will equate to how fast those things can pump out SHA-256 hash's. LTC, IMO, has a much better hashing algorithm that is resistant to such mass mining methods.

But wouldn't the Scrypt in Litecoin make it even less profitable in the future given that difficulty will increase, and the difficulty increase will outpace the release of stronger mining GPUs. ASICs will have to exist for the sake of longevity of the currency.

Honestly, I think that AMD can make big $$ if they make cards specialized for cryptocoin mining, which will also make the prices of high-end AMD Cards go back to what AMD intended them to be (lower than Nvidia, which will give AMD more customers), and also get cryptocoin miners to buy their ASICs, which equates to more $$

That's not how that works.

The difficulty of the equation is changed only if the supply is increasing faster then originally planned. For BTC's it's one coin per 10mim on average. So if in a given week there is one coin per 5min created the difficulty will be doubled to bring it back down to one per 10min. Your competing against every other miner in the world for that coin. The faster you produce them the harder they will become to produce. In 2013 the difficulty of BTC increased by more then 10x due to all the ASIC's being created. Since all the values for a SHA-256 has operation can fit inside register memory, it's trivial to create a custom chip that only performs that operation. It'll be faster then anything else on the planet, doing that singular operation. Scrypt was created specifically to battle that vulnerability. Scrypt is memory intensive and all the values can't fit inside register memory so any custom chip would also have to include a high speed memory subsystem and cache, that would inflate the cost and make them non-profitable in most cases.

BTC was fine until the difficulty exploded suddenly. Less then 10% of the mined BTC's are on the market right now, meaning most people are sitting on them waiting for the price to go up further. It's caused the supply to dwindle to nothing and the price to skyrocket. As long as people are running those custom ASIC systems and hoarding the BTC's the entire system will have serious issues.

 

[juggernautxtr]

"Those APUs are planned for 2018-2020."

what are you going to do till then jaun? 4-5 years away.

i'd say that's how long till AMD's apus' are even capable of doing what the 8000 series is doing now.
the A10 is barely capable of doing rendering in the time of dgpu and high end processor.

and as many high core count cpu's on both sides from AMD and Intel, still means we are in need of the power of hp cpu.

you should read the bottom of those charts more closely juan, as they change constantly, and you should know better to rely on it.

 

[8350rocks]

There's another theory Blackstar: That, for CPU's at least, AMD abandons desktop PC's, escept for cheap APU options. Looking from that perspective, AMD's moves make a LOT of sense.

That is opposite of what I am hearing...honestly...from what I can gather, the environment just is not right for them at this time to do a large die many core CPU.

There in lies the issue...fab! The same place the issue has been for some time...anyone surprised? I am not...

 

[juggernautxtr]

I would have to agree with 8350rocks, first a massive production for consoles fab space took up, then kaveri fab space took up, plus whoever else is using fab space, there is no room to make a high core count cpu. and global foundries floor space along with tsmc are probably all used up. unless one of the 2 get some room AMD is stuck with what they can get done.
which mostly make this bet and win that it was an opt for APU which would be a bigger income.

 

[noob2222]

Juan is apparently allowed to go on spouting marketing hype as 100% facts, but anyone who thinks he is full of it gets their posts deleted. Marketing is there to promote what is available the best way they can.

For apus that means not showing what they cant compete with, an actual CPU + GPU. AMD isnt going to tell the world that GF screwed them royally for CPU nodes. That leads to slander lawsuits, that doesnt mean its not the real reason.

If one pays attention to intel, they have been releasing APUs first and 2 years later the CPU. Some people believe that AMD cant possibly start doing the same thing. Instead they would rather spread their personal opinion that AMD is done trying to be high end, going to an even lower end processor and relying on software and marketing hype to sell hardware.

If that is actually AMDs new tactic, to sell software, then I bought my last AMD cpu. The world doesnt need downgrades from the 83xx for the next 5 years. Trying to pretend that APUs are the best thing ever made is only marketing.

As others have said. Its impossible to have an 8 core APU with 2048 shader cores and 32gb of gddr5 memory.

 

[ShadowofGhosts]

If AMD is giving up on Hard-Core Gamers and focusing on mainstream users, then they are loosing a chance to be more competitive with Intel. They will only get the office worker to buy their CPUs, while PC Gaming is still there and arguably expanding with the Steam Machines generating considerable hype. Yes, devices are becoming smaller (From room-sized computers to smartphones), but that doesn't mean that the old devices are going to be obsolete. Have you heard about someone doing a huge social calculation with his iPhone or his gaming PC? The same people (like universities, labratories) still use room sized computers and they are far from being obsolete. The Standalone GPU will still exist in 2030, the room sized computer will still exist in 2030, and I doubt these technologies will be disappearing any time soon. Think about it this way, graphics are improving as fast as hardware is improving. Look at the Witcher 3 and compare it to the Witcher 2. And then look at the GPUs when the Witcher 2 came out (GTX 500 Series, Radeon 6000 Series) and look at the GPUs when the Witcher 3 is coming out (Radeon Rx 200 Series, GTX 700 Series). The rate is about the same, and it can be argued that the graphics are actually overtaking the pace of hardware releases. We are always innovating in software just as much as we are innovating in hardware. Yeah, we can get an R9 290X equivalent APU.....in 2016 or 2017, and by then, it'll be running Battlefield 5 or 6 at Medium Settings. There is no evidence that the development of the GPUs of APUs is faster than the development of graphics. In fact, Battlefield 4 bogged down my Radeon 8650g (30 FPS low settings) as much as BF3 bogged down my Intel HD 3000 2 years ago. Yes, the Kaveri APU can run BF4 Medium Settings 1080p 30FPS, like a DDR3 7750, but this is merely an integration of an entry level card unto a CPU. Expect the same thing to happen for a long time. Likewise, the 5000 Series APUs's GPUs really had (about) the same framerates on BF3 Medium Settings as we have the Kaveri APU's GPUs have on Medium Settings 1080p.

 

[de5_Roy]

whatever happens with amd's x86 after excavator's launch in carrizo is beyond this thread's scope. those information won't be coming out until later this year. if everything goes amd's way, carrizo will launch next year on glofo 20nm bulk (or soi, if glofo can do it). that's when amd will outline their next roadmap.

there's a very high chance amd might allocate majority of it's resources to developing (or co-develop with arm inc.) arm and sideline x86 (cpus-socs) for desktops and laptops. right now amd has negligible presence in tablets and phones. amd could develop (if they haven't already) arm(64bit)+gcn soc with hsa support. adding optimizations like mantle on portable, "closed" devices can go a really long way. i say this because reed seems keen on developing their semicustom division and microservers while quickly rolling out high performance x86 apus.

this is just one possibility. plenty of other things can happen. speculating beyond 2015 isn't worth the effort, imo.

 

[juanrga]

"Those APUs are planned for 2018-2020."

what are you going to do till then jaun? 4-5 years away.

i'd say that's how long till AMD's apus' are even capable of doing what the 8000 series is doing now.
the A10 is barely capable of doing rendering in the time of dgpu and high end processor.

and as many high core count cpu's on both sides from AMD and Intel, still means we are in need of the power of hp cpu.

you should read the bottom of those charts more closely juan, as they change constantly, and you should know better to rely on it.

I believe that Intel will force AMD and Nvidia to accelerate their respective plans, but you are right that there is a span. Nvidia will continue improving and selling dGPUs and Tegras before has ready its designs and prepares to abandon its traditional market. AMD will continue selling FX brand up to 2015 (see roadmaps) whereas continues developing APUs. I believe that AMD will unify big and small cores after Excavator on an APU design and it will continue developing dGPUs as Nvidia up to then.

Intel will continue increasing its emphasis on APUs (Intel continues calling them CPUs). Broadwell introduces a huge leap in iGPU (the desktop K CPUs will include Iris Pro graphics 5000 gen with L4) and Skylake will introduce a mayor revolution on the iGPU side. According to Intel (see link given before), Skylake will kill dGPUs. Intel will continue selling CPUs (aka six/eight cores and no iGPU) for desktop enthusiasts up to then.

Note that Intel will be replacing its own discrete cards with APUs in the next gen of Phi. See slide given above. Therefore it is not some weird strategy to only kill AMD and Nvidia dGPU divisions, but that the new Intel APU will be more powerful than the current Intel discrete card.

 

[juanrga]

Lol juan do you even know the law of physics? There's no way a APU can get the same type of power that a 6 core Intel and Nvidia 780 TI can. Why because of the law of physics guess what in year 2020 when those APU's come out a separate GPU and CPU will still be faster and if anything in the future its only going to get harder to put more transistors in a single package and not have to worry about temps. I might see it if we go back to the old day's and have a slot for the CPU(APU now) then it could be possible because we can have massive heatsinks prebuilt in them

We will always need the fastest hardware for rendering and encoding, as well as big servers.....Always!
Gamers will always want the best bang for buck and right now a A10 7850K makes little sense even with mantle because i can still build a x4 760K athlon and a 7770HD for about the same price and i can better manage heat when overclocking those parts.

Juan is apparently allowed to go on spouting marketing hype as 100% facts, but anyone who thinks he is full of it gets their posts deleted. Marketing is there to promote what is available the best way they can.

For apus that means not showing what they cant compete with, an actual CPU + GPU. AMD isnt going to tell the world that GF screwed them royally for CPU nodes. That leads to slander lawsuits, that doesnt mean its not the real reason.

If one pays attention to intel, they have been releasing APUs first and 2 years later the CPU. Some people believe that AMD cant possibly start doing the same thing. Instead they would rather spread their personal opinion that AMD is done trying to be high end, going to an even lower end processor and relying on software and marketing hype to sell hardware.

If that is actually AMDs new tactic, to sell software, then I bought my last AMD cpu. The world doesnt need downgrades from the 83xx for the next 5 years. Trying to pretend that APUs are the best thing ever made is only marketing.

As others have said. Its impossible to have an 8 core APU with 2048 shader cores and 32gb of gddr5 memory.

Both you have been posting personal attacks during months. Both of you were warned but insist on doing it personal. Again I will ignore the personal attacks and will focus on the points raised.

There is consensus among the entire research community and also among the engineers of Nvidia, Intel, and AMD on that APUs scale up beyond a CPU+dGPU. There is even agreement on the magnitude of the scale: the future APUs will be 10x faster than the CPU+dGPU. I already explained why. The principle of locality was first identified in a DARPA report mentioned before.

If you believe that every scientist/engineer is wrong please raise your criticism to them. You can contact AMD or Nvidia engineers and ask them if "even know the law of physics" Of course, they will continue with their designs because they know very well the laws...

Intel will be the first chip maker to replace its discrete card with an APU (Intel calls CPUs to its APUs). I already gave this slide above, but it seems it was ignored on purpose

The Intel APU will be more powerful that the discrete card version. Ask Intel if you don't believe it. The Intel APU will have 16GB of MCDRAM with a sustained bandwidth of 500GB/s. This APU will not compete against a cheap R9-290X but is aimed to compete with fastest and expensive discrete cards from Nvidia like the K40 with 12GB of GDDR5 and peak bandwidth of 288GB/s. The Intel APU will offer about 15GFLOP DP per Watt, which is about 4x what offers the best current supercomputers with dGPUs.

According to official doc from AMD that I have, their APU for ~2018 has 8 cores with 256-bit FMAC units (Piledriver/Steamroller has 128-bit), ~10TFLOP DP, shared L3 cache between CPU and iGPU, quad-memory channel, 32--64 GB stacked RAM, and 200--250W TDP.

According to official doc from Nvidia that I have, their APU for ~2018, has 8 cores, ~20TFLOP DP, 256MB shared L2 cache, octo-memory channel, 256 GB stacked RAM with 1.4TB/s bandwidth, and 300W TDP. I van give even die sizes, working voltages, max. frequencies, electronic diagrams for the SM units, code samples of CUDA... Nvidia is developing this in partnership with universities and companies. Cray is one of the members of the design team. The same Cray that designs some of the more powerful supercomputers in the world.

Neither Nvidia nor AMD have plans for CPU+dGPUs at this level, because engineers know that no CPU+dGPU could achieve that kind of performance. The Nvidia doc has graphics with scaling of performance that prove that a hypothetical 500W CPU + 500W GPU couldn't achieve that kind of performance of their 300W APU. That is why they are designing APUs. Now check the Intel slide of above.

I repeat, if both of you believe that all the engineers are ignorant of basic laws of physics, you could contact them and explain your ideas, but stop from attacking me by merely posting here which are their plans and designs for the future.

Also if you disagree with AMD plans of selling Piledriver FX up to 2015 whereas developing APUs you can sign the petition mentioned before.

EDIT: I found an article with AMD plans

The AMD plan, says Moore, is to get a 10 teraflops Fusion APU into the field that only consumes 150 watts, and to use this as the basis of an exascale machine. "You start to think that maybe we can get there," said Moore, saying that he would put a stake in the ground and predict an exascale system could be built by 2019 or 2020.

http://www.theregister.co.uk/2011/11/02/amd_exascale_supercomputing_apu/

 

[Reynod]

Juan is apparently allowed to go on spouting marketing hype as 100% facts, but anyone who thinks he is full of it gets their posts deleted. Marketing is there to promote what is available the best way they can.

For apus that means not showing what they cant compete with, an actual CPU + GPU. AMD isnt going to tell the world that GF screwed them royally for CPU nodes. That leads to slander lawsuits, that doesnt mean its not the real reason.

If one pays attention to intel, they have been releasing APUs first and 2 years later the CPU. Some people believe that AMD cant possibly start doing the same thing. Instead they would rather spread their personal opinion that AMD is done trying to be high end, going to an even lower end processor and relying on software and marketing hype to sell hardware.

If that is actually AMDs new tactic, to sell software, then I bought my last AMD cpu. The world doesnt need downgrades from the 83xx for the next 5 years. Trying to pretend that APUs are the best thing ever made is only marketing.

As others have said. Its impossible to have an 8 core APU with 2048 shader cores and 32gb of gddr5 memory.

Enjoy your two week ban.

jdwii ... your personal attacks had better stop or your joining noob.

 

[jdwii]

Juan is apparently allowed to go on spouting marketing hype as 100% facts, but anyone who thinks he is full of it gets their posts deleted. Marketing is there to promote what is available the best way they can.

For apus that means not showing what they cant compete with, an actual CPU + GPU. AMD isnt going to tell the world that GF screwed them royally for CPU nodes. That leads to slander lawsuits, that doesnt mean its not the real reason.

If one pays attention to intel, they have been releasing APUs first and 2 years later the CPU. Some people believe that AMD cant possibly start doing the same thing. Instead they would rather spread their personal opinion that AMD is done trying to be high end, going to an even lower end processor and relying on software and marketing hype to sell hardware.

If that is actually AMDs new tactic, to sell software, then I bought my last AMD cpu. The world doesnt need downgrades from the 83xx for the next 5 years. Trying to pretend that APUs are the best thing ever made is only marketing.

As others have said. Its impossible to have an 8 core APU with 2048 shader cores and 32gb of gddr5 memory.

Juan is apparently allowed to go on spouting marketing hype as 100% facts, but anyone who thinks he is full of it gets their posts deleted. Marketing is there to promote what is available the best way they can.

For apus that means not showing what they cant compete with, an actual CPU + GPU. AMD isnt going to tell the world that GF screwed them royally for CPU nodes. That leads to slander lawsuits, that doesnt mean its not the real reason.

If one pays attention to intel, they have been releasing APUs first and 2 years later the CPU. Some people believe that AMD cant possibly start doing the same thing. Instead they would rather spread their personal opinion that AMD is done trying to be high end, going to an even lower end processor and relying on software and marketing hype to sell hardware.

If that is actually AMDs new tactic, to sell software, then I bought my last AMD cpu. The world doesnt need downgrades from the 83xx for the next 5 years. Trying to pretend that APUs are the best thing ever made is only marketing.

As others have said. Its impossible to have an 8 core APU with 2048 shader cores and 32gb of gddr5 memory.

Enjoy your two week ban.

jdwii ... your personal attacks had better stop or your joining noob.

"Lol juan do you even know the law of physics?"
"Both you have been posting personal attacks during months. Both of you were warned but insist on doing it personal. Again I will ignore the personal attacks and will focus on the points raised."

Would you like me to find the personal attacks from juan from the previous pages? Not to mention a personal attack is calling names which is something i did not do, i insisted on asking if he understands the law of physics that is not hinting towards insulting his intelligence, i just wanted to see if he was educated in that area.

Also personal documents juan where are those coming from and why should we just believe you or take your word for it?


An ad hominem, short for argumentum ad hominem, is a general category of fallacies in which a claim or argument is rejected on the basis of some irrelevant fact about the author of or the person presenting the claim or argument. ...

 

[blackkstar]

@juan, my challenge to you is to prove that Phi going from AIB to LGA is an APU.

I don't know how to break it to you, LGA Phi is just a Phi chip without an additional CPU. Meaning it's an aborted GPU (Larrabee) and only a GPU in a socket. It's connected via QPI.

This is exactly what I have been saying will happen. APU is just a stop-gap measure. The real solutions will be dGPU (or aborted dGPU like Larrabee successors) with dCPUs or even APUs.

You're not going to see a Xeon Phi as a standalone product. It's a coprocessor. If it were to be an APU, it would not be a coprocessor but the main processor.

This is exactly what I predicted would happen and it just seems like Intel announced it happening sooner than AMD. I said that there are situations where a simple APU wouldn't have enough power. APUs are limited to mid-range solutions because of TDP and die size restraints. A big dCPU and a big dGPU will never, ever fit on an APU. But please don't misunderstand what I'm getting at. Yes at some point you can fit Tahiti class GPU on an APU, but at that time it'd be possible to make another > 350mm^2 GPU, which would destroy the APU.

I also said that there were needs of people who exceed the few CUs on high end AMD APU. Yet you seemed content in thinking that that market simply doesn't exist.

My question to everyone who thinks big AMD x86 CPUs are done, is how do you expect HSA to compete against something like Xeon Phi in their own sockets (theoretically multiple Phis on a single board should be possible) when the best HSA has to offer is mid-range APUs? If AMD does not come up with a serious contender to Intel Xeon + Xeon Phi(s) platform, HSA will forever be relegated to the "it will never have the hardware to scale to that Intel has" department. Meaning that the software platform designed for massive improvements will be limited by hardware while the competition will not be.

Do you see what I am getting at? Intel is already prepared to do what I have been proposing for quite some time. AMD is no doubt ready to follow. But now is not a good time for AMD to follow. Why introduce a platform that is heavily dependent on memory performance when DDR4 is "around the corner" and while fabs are a disaster.

Speaking of which, by the way, IBM is selling their fabrication division. You know, the one that includes 22nm SOI. I wonder who would buy it and I wonder if AMD would be able to make chips there.

@8350rocks, you have been wondering why AMD is saying "it's not a good time to introduce a big die chip" and here you go. They probably knew IBM was selling 22nm SOI (which they presumably didn't have access to as 22nm SOI was supposed to come to GloFo as well as it never did) and were waiting for ATIC to grab it.

IBM and AMD are BFFs. Just saying.

EDIT: and also, AMD taking these interconnects to an ARM server filled with big dGPUs with HSA enabled would be absolutely terrifyingly amazing. Both from a raw performance and a performance per watt scenario.

AMD has gone to great lengths to create chips that can be used as building blocks interchangeably. What is to say that they won't be taking this to the next level where they allow components to be added? Ergo ARM CPU + HSA dGPUs, x86 CPU with HSA dGPUs, ARM CPU with x86 APUs, etc. Isn't that sort of the point of HSAIL (and coincidentally Mantle shares similar goals)

 

[juggernautxtr]

juan is right and wrong at the same time, yes APU will become the more powerful processing unit. and more cost efficient both in power and production.

yes the apu's will be taking over, AMD will shift all production to an apu, as they get smaller they will soon add cores to both cpu and gpu. and as on kaveri they will work more closely together on same processes, which will increase processing done. fx will not last more than 3-4 more years before they have the apu's just annihilating the separated components.

the only thing i see them continuing is dgpu for gaming, where you will need the dedicated processing power.
the GPU/PPU in kaveri alone is capable of i think i read 3 Tbytes of processing power and nowhere near being utilized fully for what it can do. yeah it's not the greatest for gaming but is a much more powerful processor than being let on at the moment. the software capable of being used on them hasn't hit.

the change is coming guys, and it's going to be a whole new world.

 

[Cazalan]

It's a good summation Juanrga but I think people here are more concerned (and rightfully so) with what they can buy today or tomorrow for a desktop PC. Those future ultra high end APUs are going to continue to cost a fortune for the next 3-4 years. Those Xeon Phi CPUs are in the $3500-$5000 range. They will be cool but geared towards HPC not gaming.

People want to know what they'll be buying for BF5/BF6/BF7. If they're running a 290x now they'll still likely be buying a discrete for the next 2-4 years.

There's no magic here when it comes to APUs. It's all about transistor counts, density, yields and process nodes. TSMC/GF are just now getting their teeth wet with FinFET. Intel stumbled a bit at the start on their FinFET roll out. They're just now getting the yields high enough to make their massive 15 core server parts with gobs of cache.

We know what NVidia can do with a 7B transistor (Titan), and we know what AMD can do with a 2.4B transistor APU (Kaveri). Neither is going to change drastically without a major die shrink or much higher yields. The 16nm/14nm nodes should make for some interesting parts but how soon those will make it into consumer grade high end APU parts depends on the yields. I've heard rates as low as 10-20% for GF and 20-30% for TSMC for their FinFETs.

 

[jdwii]

It's a good summation Juanrga but I think people here are more concerned (and rightfully so) with what they can buy today or tomorrow for a desktop PC. Those future ultra high end APUs are going to continue to cost a fortune for the next 3-4 years. Those Xeon Phi CPUs are in the $3500-$5000 range. They will be cool but geared towards HPC not gaming.

People want to know what they'll be buying for BF5/BF6/BF7. If they're running a 290x now they'll still likely be buying a discrete for the next 2-4 years.

There's no magic here when it comes to APUs. It's all about transistor counts, density, yields and process nodes. TSMC/GF are just now getting their teeth wet with FinFET. Intel stumbled a bit at the start on their FinFET roll out. They're just now getting the yields high enough to make their massive 15 core server parts with gobs of cache.

We know what NVidia can do with a 7B transistor (Titan), and we know what AMD can do with a 2.4B transistor APU (Kaveri). Neither is going to change drastically without a major die shrink or much higher yields. The 16nm/14nm nodes should make for some interesting parts but how soon those will make it into consumer grade high end APU parts depends on the yields. I've heard rates as low as 10-20% for GF and 20-30% for TSMC for their FinFETs.

Thank you that was my main point

 

[juggernautxtr]

the question now is will GF/TSMC get floor space, for a possible steamroller? maybe see by/before end of the yr type thing or do you guys think they will just hold off and dump excavator on us in 2015 w/ddr4.
with the kaveri scores looking pretty good, not spectacular, what would an 8 core steamroller look like score wise?
maybe we will see a low power steam roller 4 core on fm2+(tagged as an athlon part)?

 

[Cazalan]

You're not going to see a Xeon Phi as a standalone product. It's a coprocessor. If it were to be an APU, it would not be a coprocessor but the main processor.

KL is supposedly Silvermont cores with AVX 3.1 (512bit) extensions. So it's not an APU. It doesn't have a graphical element like the shaders of AMD/NVidia GCN/CUDA cores.

Intel currently ships 2/4/8 core silvermont based products so it will be a stand alone CPU. As far as HPC deployments go I'm sure there will still be primary racks with traditional Xeon CPUs controlling the show. The compute nodes will just be more autonomous than they are today.

 

[de5_Roy]

KL is supposedly Silvermont cores with AVX 3.1 (512bit) extensions. So it's not an APU. It doesn't have a graphical element like the shaders of AMD/NVidia GCN/CUDA cores.

Intel currently ships 2/4/8 core silvermont based products so it will be a stand alone CPU. As far as HPC deployments go I'm sure there will still be primary racks with traditional Xeon CPUs controlling the show. The compute nodes will just be more autonomous than they are today.

i'd say that hawaii is quite close to being a standalone system on a board (not strictly x86). if it isn't... then it's just a few steps short due to amd being fab-bound. imagine small cores (cat or arm) by the a.c.e.s' side, connected through a crossbar like the ps4 soc, or fused with the aces.
nvidia's project debacle...ahem!... i mean project denver was supposed to be this way, but it's not clear when nvidia will launch that. it's a speculation for somewhere else.

 

[8350rocks]

A few things to add to this discussion:

1.) GCN 1.1 can already run x86 SSE4 instruction sets autonomously for GPGPU applications. Those are still not sold as stand alone Scalar processors...they make great vector processors, and can perform scalar operations that are massively parallel (see cryptocoin mining), though they are not good at micro scale scalar operations like a CPU.

2.) APUs will never be able to beat the dCPU + dGPU combo. While the world goes more integrated, the dCPU + dGPU is not going anywhere. As blackkstar pointed out, for the resources you spend to make both on one die, you could make the same size die for each and it would be several orders of magnitude more powerful in combination.

3.) The argument about localization of resources holds some merit, until you consider extremely high end networking technologies that are not currently available in mainstream systems...but will be in the next few years. When an internal system bus can transfer at 1 Tb/s rates and faster, localization is only a novelty. Those days are already coming.

4.) We are also not considering advances in materials outside of simple process node advancements. Anyone following superconductor research these days? They are at the point now where they can produce superconductors that only require LN2 to operate at the correct temp. A few years ago, you had to go to much colder temps to get superconductors to even be more than a pipe dream. Within the next 10 years I bet we see room temp superconductors, at which point everything within the entire computing world will be turned on it's head. The race will be to see who brings it to market as a technology first once they are close.

5.) With IBM selling what is honestly the most advanced fab in the world at this point (Intel may have smaller nodes with bulk, but 22nm SOI is a game changing technology above and beyond what any of us dare to imagine). That will open doors to technologies for other groups to have access to that simply do not have the ability to dump the resources IBM did for so many years into a fab. The possibility of the back biasing in 22nm FD-SOI is alone is mind bending as far as the capability to tune a specific architecture to do what you want with fewer sacrifices. If Intel had 22nm FD-SOI for their FinFETs, then Haswell might have had a 5.0 GHz stock frequency 4670k that consumed the same power as a 3570k with all their technology integrated. This technology is that good.

-my 2 cents

 

[sora]

This is honestly one of the biggest debate threads lol

 

[wh3resmycar]

we're probably 20 years away from the achieving real-time ray-tracing for video games, if the world would go arm/apu completely, then we'll be pushed back 100 years..

 

[juggernautxtr]

but who is going to grab that fab global foundries? TSMC? SONY? who? whoever gets that fab if gf gets it it will be in the major benefit of AMD. what if intel is up for the grab?

 

[Reynod]

Juan is apparently allowed to go on spouting marketing hype as 100% facts, but anyone who thinks he is full of it gets their posts deleted. Marketing is there to promote what is available the best way they can.

For apus that means not showing what they cant compete with, an actual CPU + GPU. AMD isnt going to tell the world that GF screwed them royally for CPU nodes. That leads to slander lawsuits, that doesnt mean its not the real reason.

If one pays attention to intel, they have been releasing APUs first and 2 years later the CPU. Some people believe that AMD cant possibly start doing the same thing. Instead they would rather spread their personal opinion that AMD is done trying to be high end, going to an even lower end processor and relying on software and marketing hype to sell hardware.

If that is actually AMDs new tactic, to sell software, then I bought my last AMD cpu. The world doesnt need downgrades from the 83xx for the next 5 years. Trying to pretend that APUs are the best thing ever made is only marketing.

As others have said. Its impossible to have an 8 core APU with 2048 shader cores and 32gb of gddr5 memory.

Juan is apparently allowed to go on spouting marketing hype as 100% facts, but anyone who thinks he is full of it gets their posts deleted. Marketing is there to promote what is available the best way they can.

For apus that means not showing what they cant compete with, an actual CPU + GPU. AMD isnt going to tell the world that GF screwed them royally for CPU nodes. That leads to slander lawsuits, that doesnt mean its not the real reason.

If one pays attention to intel, they have been releasing APUs first and 2 years later the CPU. Some people believe that AMD cant possibly start doing the same thing. Instead they would rather spread their personal opinion that AMD is done trying to be high end, going to an even lower end processor and relying on software and marketing hype to sell hardware.

If that is actually AMDs new tactic, to sell software, then I bought my last AMD cpu. The world doesnt need downgrades from the 83xx for the next 5 years. Trying to pretend that APUs are the best thing ever made is only marketing.

As others have said. Its impossible to have an 8 core APU with 2048 shader cores and 32gb of gddr5 memory.

Enjoy your two week ban.

jdwii ... your personal attacks had better stop or your joining noob.

"Lol juan do you even know the law of physics?"
"Both you have been posting personal attacks during months. Both of you were warned but insist on doing it personal. Again I will ignore the personal attacks and will focus on the points raised."

Would you like me to find the personal attacks from juan from the previous pages? Not to mention a personal attack is calling names which is something i did not do, i insisted on asking if he understands the law of physics that is not hinting towards insulting his intelligence, i just wanted to see if he was educated in that area.

Also personal documents juan where are those coming from and why should we just believe you or take your word for it?


An ad hominem, short for argumentum ad hominem, is a general category of fallacies in which a claim or argument is rejected on the basis of some irrelevant fact about the author of or the person presenting the claim or argument. ...

Listen buddy I warned all of you previously and he stopped ... yout two didn't.

Keep going and your off for a holiday ... this isn't an intellectual debate ... its a promise.

No more personal attacks.

 

[Antaredja Kartasasmita]

Making an undefeatable CPU that obliterates every render and calculations isn't cost-effective, for AMD. People usually bought cheap CPU and cheap GPU for just everyday uses/light gaming, Mid gamers don't tend to push CPU that far other than GPU, and High enders to Supreme gamer tends to almost maxed out their CPU, and bought even much more to GPU. IF we're talking about calculations and shit, AMD's sure gonna be tired first in a few minutes.


If we're talking about gaming, CPU performance is kinda irrelevant, exception in CPU heavy games like Skyrim or Metro last light, but going Kaveri is cost effective, and buying at least 280x, is a nice and smooth gaming all round.


PC users hierarchy:
Casual: low cpu, no gpu
Very light gaming: low cpu, low gpu
Light gaming: low cpu, low-mid gpu
Minimum requirement gaming: low cpu, mid-low gpu (higher tier)
Reccomended requirement gaming: low-mid cpu, mid gpu
High settings gaming, casual: low-mid cpu, mid-high gpu
Maxed out gaming, casual: mid-low cpu, high-mid gpu
Maxed out gaming, supreme gaming: mid cpu, high cpu
Supreme gaming, enthusiast: mid-high cpu, extreme gpu


and the amount of comsumers lowered each users, and AMD won until mid-high level

 

[de5_Roy]

Making an undefeatable CPU that obliterates every render and calculations isn't cost-effective, for AMD.

not just for amd, it simply doesn't exist outside paper/design. physics will always set the limits.

If we're talking about gaming, CPU performance is kinda irrelevant, exception in CPU heavy games like Skyrim or Metro last light, but going Kaveri is cost effective, and buying at least 280x, is a nice and smooth gaming all round.

280x... not for present prices. the 7900 cards were at really good perf/price few months ago, not anymore. currently amd's perf/price cards are r9 270, 260x and 7770. all those will fit nicely with kaveri. at higher end, geforce gtx 760 and 770 would be more suitable for kaveri.