AMD CPU speculation... and expert conjecture

[Cazalan]

As for 20nm, if AMD puts their GPUs on GloFo's 20nm process and Nvidia is stuck on 28nm TSMC for the entirety of 2014, Nvidia is dead. It will basically be $300 AMD 300mm^2 GPU competing with $650 500mm^2 Nvidia GPU and the AMD would be faster. And it'd trickle down the entire product range as AMD's die size to performance advantage would be massive.

It takes 18+ months to switch processes. If AMD started today we wouldn't see graphics cards on GF process until 2015 at the earliest.

Perhaps they could have anticipated this move by Apple, but do you think they would have paid 1.2billion to break their GF contracts if they knew they were going to have to switch back? Unlikely. They could have negotiated to spend that money on new product at GF instead of it just going into the void for zero return.

This TSMC/Apple deal is going to have a pretty big impact and not just for AMD. The market has barely reacted yet but it spells problems for even bigger players like Qualcomm too.

 

[de5_Roy]

As for 20nm, if AMD puts their GPUs on GloFo's 20nm process and Nvidia is stuck on 28nm TSMC for the entirety of 2014, Nvidia is dead. It will basically be $300 AMD 300mm^2 GPU competing with $650 500mm^2 Nvidia GPU and the AMD would be faster. And it'd trickle down the entire product range as AMD's die size to performance advantage would be massive.

It takes 18+ months to switch processes. If AMD started today we wouldn't see graphics cards on GF process until 2015 at the earliest.

Perhaps they could have anticipated this move by Apple, but do you think they would have paid 1.2billion to break their GF contracts if they knew they were going to have to switch back? Unlikely. They could have negotiated to spend that money on new product at GF instead of it just going into the void for zero return.

This TSMC/Apple deal is going to have a pretty big impact and not just for AMD. The market has barely reacted yet but it spells problems for even bigger players like Qualcomm too.

er... glofo has a proven track record of fumbling with anything that has a gpu in it (eg. apus). according to amd's ye olde gpu roadmap, glofo was supposed to make cape verde and bali gpus (entry level hd7k) on their 28nm node.
here's my theory: amd wanted to have glofo make 32nm entry level gpus along with 32nm apus and wanted to do the same with 28nm. glofo failed from the begining, cape verde went to tsmc who delivered the gpus.
that's why i am not holding my breath over glofo delivering on time with 20nm, that node is optimized for various arm vendors anyway.

 

[juanrga]

Some new stuff on Steamroller

http://wccftech.com/amd-steamroller-apu-engineering-sample-bionic-research-database/

 

[esrever]

XBone receives the kiss of death:

http://www.tomshardware.com/news/Xbox-One-Steve-Ballmer-Don-Mattrick-Interactive-Entertainment-Business-Zynga,23356.html

to be honest, if ballmer was in control of it, he would have probably done a better job with the PR for the online and such. Mattrick really eff'ed up the messages they were delivering.

 

[GOM3RPLY3R]

I would just like to say, Cheers to AMD for making an APU. However, this subject is way out of hand with many people. AMD has much experience with GPUs and CPUs (like anyone doesn't know). Intel, really only has experience with CPUs mostly.

When logic comes to logic its this. It is not in the least fair to compare AMDs APUs to Intels CPUs and say that AMD wins. Yes they did win, they made a APU that runs off of low power compared to a CPU and GPU setup, and has great performance, however people need to realize that Intel isn't really out for that. They are more focused on making their CPUs better.

I also know that Intel did start to include Iris to their CPUs (making them somewhat APUs, but not quite there).

When the Gamer's perspective comes into hand, I might say, "Will I spend $1000 on a i7-3770k and a GTX 780 that will get 100 FPS on game X, and 50 FPS on game Y, or the A10-6800k for $200 and get 60 FPS on game X, and maybe 30 FPS on game Y? (Same settings)"

Price to performance really comes hand in hand when making these decisions, and also "Will it be future proof?"

In the end, my main point is that Intel isn't really out there to advance in certain criteria other than, the best performance possible for a reasonable price. Thus that being said, AMD's new, "neck of the woods," cannot be compared to with the other competitor.

However, I would like to mention. Wouldn't it be nice to have a computer for a long time, and when the GPU either dies or gets outdated and the CPU can still live on, you can just replace that part for a much better part, rather than buy a whole new thing when the CPU is still good?

UPDATE: Last mention. I can't believe that anyone hasn't come to mind that: maybe Intel made Iris so it can be more compatible with new software graphics? i.e. Microsoft Word, Excel with their new and fancy designs, the Windows 8 desktop, free to play games that are highly advancing with the average consumer and in graphical design.

Keep that in mind as food for thought.

 

[jdwii]

"however people need to realize that Intel isn't really out for that. They are more focused on making their CPUs better"

Have you ever seen haswell?

 

[esrever]

Intel has been trying to make their GPU better and trying to take on ARM

 

[jdwii]

"However, I would like to mention. Wouldn't it be nice to have a computer for a long time, and when the GPU either dies or gets outdated and the CPU can still live on, you can just replace that part for a much better part, rather than buy a whole new thing when the CPU is still good?"

This is no different than Intel CPU's(or APU's just switch the processor analogy to a GPU one)

 

[Cazalan]

If AMD shipped a desktop PS4 APU then they'd have the top APU for sure. That's like a $200 video card embedded.

 

[gamerk316]

I think that Dual Graphics is just a feature that is left unrefined, notably the software level as it does work and can potentially add a lot of value to lower cost systems. I can't confirm yet but it seems like GDDR5 models are a little problematic but won't know for sure until i run a GDDR3 version.

The reality is the APU will become AMD's processor for the future and it will eventually end up in FX line, that said they will then become more APU than they are FX, right now the DT APU is still very much evolving.

Dual Graphics is a neat idea but the designs are quite asymmetrical. I can see why they have such problems balancing the load. You got dedicated GPU ram on the discrete and the APU is competing with the CPU for it's RAM. The latency for the APU can vary quite a bit.

As games advance they could just use the APU graphics for compute and let the discrete render full speed.

As I've been saying: We're eventually going to get some form of PPU. Agiea had the right idea, just a bad implementation. That being said, as long as GPU's gain 20% each generation, the need to separate compute/rendering doesn't exist. But I think thats the direction things are headed.

 

[de5_Roy]

If AMD shipped a desktop PS4 APU then they'd have the top APU for sure. That's like a $200 video card embedded.

imo that'd be worth buying. amd could even segment them into quad and tri module (8-6 cores) with different shader counts for the gpu. since the cpu part is cheaper than the high performance ones, overall gaming performance would be more in the same price range. however, the cpu cores and memory subsystems are the two possible areas for bottleneck, as long as those two work fine, it'd be good.

 

[juanrga]

Someone has more info about Steamroller SMT? German version of Tomshardware seems to confirm that a Steamroller core can run two threads at once.

 

[8350rocks]

I think that Dual Graphics is just a feature that is left unrefined, notably the software level as it does work and can potentially add a lot of value to lower cost systems. I can't confirm yet but it seems like GDDR5 models are a little problematic but won't know for sure until i run a GDDR3 version.

The reality is the APU will become AMD's processor for the future and it will eventually end up in FX line, that said they will then become more APU than they are FX, right now the DT APU is still very much evolving.

Raja Kudori has driver's in Beta for Catalyst set to release with HD 8XXX series GPUs that will supposedly remedy CF micro-stutter entirely.

I think if the FX line begins to gain market share with steamroller, they may entertain the idea of an AM4 socket and bring out excavator in FX too, though my concern is that their plans may already be so integrated that they could be too far down that path already.

 

[jdwii]

Someone has more info about Steamroller SMT? German version of Tomshardware seems to confirm that a Steamroller core can run two threads at once.

Seems impossible to me maybe they meant one module not core and if it could that would most likely lower performance or single threaded performance which is something steamroller isn't suppose to do Amd already has great multitasking performance. And like everyone knows you improve single core performance you improve ALL performance while adding more cores does nothing half the time.

 

[juanrga]

Someone has more info about Steamroller SMT? German version of Tomshardware seems to confirm that a Steamroller core can run two threads at once.

Seems impossible to me maybe they meant one module not core and if it could that would most likely lower performance or single threaded performance which is something steamroller isn't suppose to do Amd already has great multitasking performance. And like everyone knows you improve single core performance you improve ALL performance while adding more cores does nothing half the time.

They did mean:
core --> 2 threads
module --> 4 threads

AMD is substituting 8C Piledriver CPUs by 4C Steamroller APUs:

http://www.xbitlabs.com/picture/?src=/images/news/2013-06/amd_opteron_roadmap.png

This makes a lot of sense if it is a 8 treads --> 8 threads substitution and if one SR core has about twice the performance of one PD core. I have read in several forums that a SR core will have 2x GFLOPs. E.g. here:

http://www.overclock.net/t/1396591/beyond3d-amd-steamroller-die-shot/60#post_20251035

 

[MU_Engineer]

Guys, I like a political discussion as much as the next guy but let's keep that in the News and Leisure forum.

 

[Cazalan]

They did mean:
core --> 2 threads
module --> 4 threads

AMD is substituting 8C Piledriver CPUs by 4C Steamroller APUs:

This makes a lot of sense if it is a 8 treads --> 8 threads substitution and if one SR core has about twice the performance of one PD core. I have read in several forums that a SR core will have 2x GFLOPs. E.g. here:

Those die shots were shown to be faked. AMD is focusing on the 4C APU but that's because it's outselling the FX by like 8 to 1. There's just less interest in the FX series. Steamroller module will be 2C/2T as it's been talked about many times by AMD.

By boosting 4C APU (with SR cores) they can try to gain market share in i3/i5 territory where the bulk of sales are anyway.

 

[8350rocks]

I would just like to say, Cheers to AMD for making an APU. However, this subject is way out of hand with many people. AMD has much experience with GPUs and CPUs (like anyone doesn't know). Intel, really only has experience with CPUs mostly.

When logic comes to logic its this. It is not in the least fair to compare AMDs APUs to Intels CPUs and say that AMD wins. Yes they did win, they made a APU that runs off of low power compared to a CPU and GPU setup, and has great performance, however people need to realize that Intel isn't really out for that. They are more focused on making their CPUs better.

I also know that Intel did start to include Iris to their CPUs (making them somewhat APUs, but not quite there).

When the Gamer's perspective comes into hand, I might say, "Will I spend $1000 on a i7-3770k and a GTX 780 that will get 100 FPS on game X, and 50 FPS on game Y, or the A10-6800k for $200 and get 60 FPS on game X, and maybe 30 FPS on game Y? (Same settings)"

Price to performance really comes hand in hand when making these decisions, and also "Will it be future proof?"

In the end, my main point is that Intel isn't really out there to advance in certain criteria other than, the best performance possible for a reasonable price. Thus that being said, AMD's new, "neck of the woods," cannot be compared to with the other competitor.

However, I would like to mention. Wouldn't it be nice to have a computer for a long time, and when the GPU either dies or gets outdated and the CPU can still live on, you can just replace that part for a much better part, rather than buy a whole new thing when the CPU is still good?

UPDATE: Last mention. I can't believe that anyone hasn't come to mind that: maybe Intel made Iris so it can be more compatible with new software graphics? i.e. Microsoft Word, Excel with their new and fancy designs, the Windows 8 desktop, free to play games that are highly advancing with the average consumer and in graphical design.

Keep that in mind as food for thought.

Whatever you've been smoking before you post...I, for one, would certainly like to know what it is; that way if my kid ever gets a hold of that stuff, I can make sure to flush it down the toilet. It's clearly rotting your mind, whatever it is...

EDIT: As a side note...Thunderbolt 2.0 has released...and it's still a joke:

http://semiaccurate.com/2013/07/02/thunderbolt-is-sparse-at-computex-2013/

45W TDP Richland SKUs leaked by MSI...high end laptop competition from AMD?

http://www.thinkcomputers.org/amd-a10-6700t-and-a8-6500t-richland-apus-leaked-by-msi/

 

[juanrga]

They did mean:
core --> 2 threads
module --> 4 threads

AMD is substituting 8C Piledriver CPUs by 4C Steamroller APUs:

This makes a lot of sense if it is a 8 treads --> 8 threads substitution and if one SR core has about twice the performance of one PD core. I have read in several forums that a SR core will have 2x GFLOPs. E.g. here:

Those die shots were shown to be faked. AMD is focusing on the 4C APU but that's because it's outselling the FX by like 8 to 1. There's just less interest in the FX series. Steamroller module will be 2C/2T as it's been talked about many times by AMD.

By boosting 4C APU (with SR cores) they can try to gain market share in i3/i5 territory where the bulk of sales are anyway.

Where were shown to be faked?

The FX series is usually derived from the server one, but the server roadmap only shows up to 4 SR cores. This means that even if a FX Steamroller series is finally released the next year (which I doubt now) the chances are that would be only 2C and 4C. There is no announcement of any Steamroller chip with more than 4C. Even the displayed 6C Kaveri on previous desktop roadmap has disappeared from the new roadmap.

I don't understand the development of 8-core consoles with weak cores (which will force games to go wide), clearly favouring AMD multicore FX chips 6xxx and specially 8xxx, if next CPUs/APUs will be only 2C/2T and 4C/4T.

It would be really ironic that game developers have chosen the FX-8350 as best CPU for future gaming, if that 8-threads CPU is now to be considered something of the past. It would be also ironic that developers are choosing an i7 (8 threads) in the PC-console demos if now AMD is going to compete only with i3/i5 (4 threads).

It makes little sense that in the server roadmap AMD is substituting 8T chips by 4T chips, except in the high end, where AMD makes the contrary move and drops the 2/4/8T line and maintain only the 12/16T line.

At the same time is AMD going to less cores and threads route whereas intel makes the contrary move and announces its first 8-core Haswell?

Nothing of this makes sense for me sorry.

 

[8350rocks]

Here's my take on this, for anyone who wants to hear it:

Steamroller is heavily integrating iGPU and CPU on APUs. I think separating the 2 is going to be a bit of an adjustment that will likely take some time to do. I wouldn't be surprised if Steamroller Opterons without iGPU aren't seen until sometime end of Q1 2014. It will take them some time to get everything worked out. This may be why we're not hearing a lot about desktop CPUs at this point.

Also, I think that the 12/16 core Opterons could be a way for AMD to continue FX on an AM4 CPU platform. Now, if they did that, it wouldn't be the first time...but it would be a bit crazy. Frankly, I wouldn't mind an adapted "Vishera 3.0" if that's what it had to be as a single 8 core CPU, but the steamroller front end and IPC improvements are what we are after for dedicated CPUs. HSA is great in APUs, but the more I think about it, you'd have to have some serious GPU integration, and AMD couldn't guarantee that every AMD CPU buyer would purchase an AMD GPU, or that they didn't already have an incompatible NVidia GPU already.

We'll see...but I expect something from AMD in 2014 on this front.

 

[Cazalan]

Where were shown to be faked?

I don't understand the development of 8-core consoles with weak cores (which will force games to go wide), clearly favouring AMD multicore FX chips 6xxx and specially 8xxx, if next CPUs/APUs will be only 2C/2T and 4C/4T.

In the SA threads where they first showed up a while ago.
http://semiaccurate.com/forums/showthread.php?t=6145&page=110

The reason for 8 core Jaguar is simple. It comes down to power. Sony/MS had a specific power target for the console (~150-200W) for a number of reasons (Power supply requirements and cooling requirements). They're slapping a huge (relatively speaking, 2 billion transistor) GPU on the APU. That has a TPD cost (100W). To balance it they had to go with slower clocked CPU cores. A 4 core would be woefully underpowered so they doubled that hoping advances in programming APIs can take advantage of it.

With the CPU+GPU+PCH you're looking at a 3+ billion transistor chip. That's no slouch when they're looking to yield maybe 30 million of them a year.

 

[juanrga]

Here's my take on this, for anyone who wants to hear it:

Steamroller is heavily integrating iGPU and CPU on APUs. I think separating the 2 is going to be a bit of an adjustment that will likely take some time to do. I wouldn't be surprised if Steamroller Opterons without iGPU aren't seen until sometime end of Q1 2014. It will take them some time to get everything worked out. This may be why we're not hearing a lot about desktop CPUs at this point.

The Opteron 3000 series with 8 PD cores is substituted at the end of 2013 by a Berlin CPU with 4 SR cores. Nothing with more than 4 SR cores is announced for 2014. It is difficult to believe that FX 6/8 SR cores will be released for 2014.

Also, I think that the 12/16 core Opterons could be a way for AMD to continue FX on an AM4 CPU platform. Now, if they did that, it wouldn't be the first time...but it would be a bit crazy.

Crazy is not the word I would use. Also it makes no sense for me that AMD is dropping the 4/6/8 threads Opteron and retain the 12/16 threads, whereas introduce new 2/4 threads CPUs.

Why nothing between 4 and 12 threads for the whole 2014? Again AMD roadmap only make sense for me if Steamroller is a 1C/2T design. Then the substitution of Opterons by Berlin plus Warsaw makes sense as (square brackets denote PD cores and curly brackets denote Steamroller cores):

2013: [4/6/8/12/16 threads]
2014: {4/8 threads}[12/16 threads]

And in 2015 AMD could introduce 6C/8C steamroller for substituting the Warsaw chips

2014: {4/8 threads}[12/16 threads]
2015: {4/8/12/16 threads}

 

[juanrga]

Where were shown to be faked?

I don't understand the development of 8-core consoles with weak cores (which will force games to go wide), clearly favouring AMD multicore FX chips 6xxx and specially 8xxx, if next CPUs/APUs will be only 2C/2T and 4C/4T.

In the SA threads where they first showed up a while ago.
http://semiaccurate.com/forums/showthread.php?t=6145&page=110

The reason for 8 core Jaguar is simple. It comes down to power. Sony/MS had a specific power target for the console (~150-200W) for a number of reasons (Power supply requirements and cooling requirements). They're slapping a huge (relatively speaking, 2 billion transistor) GPU on the APU. That has a TPD cost (100W). To balance it they had to go with slower clocked CPU cores. A 4 core would be woefully underpowered so they doubled that hoping advances in programming APIs can take advantage of it.

With the CPU+GPU+PCH you're looking at a 3+ billion transistor chip. That's no slouch when they're looking to yield maybe 30 million of them a year.

I read your link and only can find people discussing if it is fake or not, but without arriving to a solid conclusion. In fact, one page after the one that you linked I found a poster with very similar ideas to the ones that I posted above:

If that die shot is real and each module could execute 4 threads (2 per "core" or 4 per shared FP unit), it makes sense we see a lack of >2M parts on mainstream desktop and even server segments. If true, this SR has much greater parallelism within the module than BD/PD have. Per thread we still will have comparable to PD execution unit count ( each thread should be doing work on one ALU/AGU and/or one 128bit execution port in new FlexFP unit).

Assuming this is the case, one 2M Kaveri could effectively have throughput of 3M or even 4M PD,depending on clock speeds and effectiveness of multithreading approach AMD picked for new SR module (within cores this time).

4 cores at twice the speed wouldn't increase the average power consumption of the console, but would favour Intel chips when porting games to PC and even would go against AMD, because a FX-8xxx would not be much faster than a Fx-4xxx when running four-threaded games.

But even if the true reason was power consumption, the fact is that the final 8-thread design in consoles is going to provide AMD 8-core FX chips a clear advantage over the Intel i5 and even the i7. Again it makes no sense that AMD is going to ignore this advantage and release only 2 and 4 thread chips for the entire 2014.

Moreover, as said above it makes no sense that the server space for 2014 contains nothing in the middle between 4 and 12 threads, except a huge gap. Why would AMD drops the existent 6/8 threads chips, but leave 12/16. Again it makes no sense for me... unless Steamroller is 1C/2T. Note that this part of my reasoning is independent of if the leaked shot is fake or not.

 

[Intel God]

Any news if amd is going to grow a pair and give us a 10 core or 12 core steamroller?

 

[8350rocks]

Here's my take on this, for anyone who wants to hear it:

Steamroller is heavily integrating iGPU and CPU on APUs. I think separating the 2 is going to be a bit of an adjustment that will likely take some time to do. I wouldn't be surprised if Steamroller Opterons without iGPU aren't seen until sometime end of Q1 2014. It will take them some time to get everything worked out. This may be why we're not hearing a lot about desktop CPUs at this point.

The Opteron 3000 series with 8 PD cores is substituted at the end of 2013 by a Berlin CPU with 4 SR cores. Nothing with more than 4 SR cores is announced for 2014. It is difficult to believe that FX 6/8 SR cores will be released for 2014.

Also, I think that the 12/16 core Opterons could be a way for AMD to continue FX on an AM4 CPU platform. Now, if they did that, it wouldn't be the first time...but it would be a bit crazy.

Crazy is not the word I would use. Also it makes no sense for me that AMD is dropping the 4/6/8 threads Opteron and retain the 12/16 threads, whereas introduce new 2/4 threads CPUs.

Why nothing between 4 and 12 threads for the whole 2014? Again AMD roadmap only make sense for me if Steamroller is a 1C/2T design. Then the substitution of Opterons by Berlin plus Warsaw makes sense as (square brackets denote PD cores and curly brackets denote Steamroller cores):

2013: [4/6/8/12/16 threads]
2014: {4/8 threads}[12/16 threads]

And in 2015 AMD could introduce 6C/8C steamroller for substituting the Warsaw chips

2014: {4/8 threads}[12/16 threads]
2015: {4/8/12/16 threads}

Well, as I said, AMD have ported server parts over before, however, the biggest thing I see is that they might basically have to redesign Vishera 2.0 with a die shrink, and SR front end if they do steamroller FX cores, and the HSA features would require an AMD GPU to function.

However, I also think that they may have done the following to speed up getting product to market:

APUs have notoriously received the newest tech first...especially considering the market for OEM systems that it caters to, as well as HTPCs and other entry systems.

If they designed the integrated systems first focusing all efforts on that, it would make sense for SR cores on FX to come about 1-2 Quarters behind. Considering that there are already Kaveri LP server ES's leaked, I firmly believe we will see Kaveri early Q3, which gives AMD through end of Q1 to get SR out for FX and Opterons. I think they may be using APUs to gauge the performance and may adjust their market scheme accordingly.

While I dislike the thought that they may be "flying by the seat of their pants" on this one, it somewhat makes sense. SR is likely the last hoorah for FX series, and if Kaveri takes off poorly, then the FX series may be something they forego releasing. However, if the early indications of Kaveri's capability are anything near what it comes out to be...then we will end up with something special, and I think AMD would warrant producing the FX line.

My other thoughts run to this...

Why hire Jim Keller back, if you don't intend to compete for the high performance desktop segment? Additionally, why would Jim Keller go on record saying "AMD are on track to catch up in high performance cores"?

Adding those 2 statements into the fold makes no sense if they're not already waiting with SR FX designs that they're tweaking for retail, and Opteron server parts for the high end market.

Additionally, they have received a ton of server orders from large financial institutions running 1U configurations, and they've ordered the high end stuff to run their databases and encryption programs to protect their assets.

I think the roadmap released in Q1 2014 will show that SR FX is coming, it may not be as quickly as we like. Though, if you consider that this may be their last attempt to compete for DT market share...if I were Jim Keller, and this was my "masterpiece" for DTPCs, then I would make 100% certain that it was going to be everything I promised and perhaps a bit more.

I am ultimately not concerned.