AMD CPU speculation... and expert conjecture

[blackkstar]

Fottemberg at S|A forums, who has a good history of being right

has said about Sweclokers leaks

I don't think that these leaks are reliable. ^_^

At the time of writing my previous post, I was not aware of the sweclocker's rumor and I was not referencing them at all. I was merely referencing Fottemberg @ S|A.

Also, you under-estimate how much power the jump from 32nm to 14nm will save. We also already have 95w 4m/8c Bulldozer CPUs.

The sweclocker's leak, if true about a core being equal to a module, indicates to me that each core will be massively beefed up compared to Bulldozer. If those rumors are true, a single Zen core will have the entire transistor budget of a Bulldozer module, which includes (as of Steamroller), two decoders, two integer units, two dispatch, shared L2 cache, and a 256-bit FPU with SMT.

Seeing a single core get all of those available transistors used for those shared parts in Bulldozer module does not indicate a weak, efficient core to me. It leaves me expecting something that would leave each core much closer to a full bulldozer's module in multi-thread performance while not having the weak single thread performance that comes from sharing as much as Bulldozer modules share.

Basically, I think it should be reasonable to expect this to perform as a theoretical 8m/16c Bulldozer chip at Fx 8370-E frequencies (3.3ghz base 4.3ghz turbo) while having improved single-thread performance.

If each Zen CPU was rumored around the size of a Bulldozer CORE, I would agree with you about it being a downsizing. But an entire core consuming the transistor budget of two cores sharing things indicates to me that there will be more transistors allocated per core, pointing to a more high performance design.

I am rather confused how you think a single core using the resources and being the same size of two cores that share things is a downsizing. That seems like an upsizing to me. Unless you're letting the TDP thing overwhelm the thoughts of how transistors would be allocated between a Zen core and a Bulldozer module.

And when I say Bulldozer module, I'm talking about the entire family, not FX x100 series. So don't go strawmanning me or whatever silly thing you like to do like you did in the post I quoted.

 

[juanrga]

Original source: http://www.sweclockers.com/nyhet/19954-amd-zen-forst-ut-i-summit-ridge-for-14-nanometer

Sorry, I don't like WCCF's interpretations :>

In this case WCCFTECH simply copied and pasted about word for word.

The Sweclockers report says lots of ordinary stuff: FM3 as replacement for FM2+, DDR4 as replacement for DDR3, 14nm (which was confirmed before), SMT (mentioned at K12 presentation),...

This leaves 95W TPD and the 8-core leak as the only new.

I still have my doubts about SMT, but let us assume the rest is accurate. Max of 95W TDP is current Kaveri top level and I don't consider that HEDT, but mainstream.

8-cores on 95W APU seems to confirm that the cores will be small and efficient rather than big and powerful.

Considering the TDP of Intel's top parts like 4770k, etc...95W on a 14nm process could very well equate to 125W TDP on 32nm process...

The 95W Zen APU on 14nm will be not competing against 22nm Haswell APUs but against the 95W Skylake APUs on 14nm.

All points to a revival of the old Bulldozer vs Sandy fight:

8 Zen cores will be faster than 4 Skylake cores on highly multithreaded integer code (at least 8 threads), but 8 Zen cores will be slower than 4 Skylake cores on FP code.

Once again AMD will try to hide their core performance deficit with moar cores. Luckily for AMD, Mantle/DX12 will hide partially the core deficit, but that will not help them beyond gaming.

 

[jdwii]

SMT on Amd cores that just sounds wrong in everyway unless they increase the size of the core. Heck i can see that hurting performance like the pentium 4 days.

If juan is right and i think he said 40% better single core performance(would put that around the haswell CPU under my benchmarks) then fine.

 

[szatkus]

SMT on Amd cores that just sounds wrong in everyway unless they increase the size of the core. Heck i can see that hurting performance like the pentium 4 days.

Performance of P4 wasn't fault of HT. Proper SMT usually doesn't decrease performance (just look at Intel Core).

 

[juanrga]

Also, you under-estimate how much power the jump from 32nm to 14nm will save. We also already have 95w 4m/8c Bulldozer CPUs.

The sweclocker's leak, if true about a core being equal to a module, indicates to me that each core will be massively beefed up compared to Bulldozer. If those rumors are true, a single Zen core will have the entire transistor budget of a Bulldozer module, which includes (as of Steamroller), two decoders, two integer units, two dispatch, shared L2 cache, and a 256-bit FPU with SMT.

The sweclocker's leak doesn't say anything about the size of cores. It only says 8-cores on 95W APUs, which --as I showed-- just point in the opposite direction: smaller cores.

Fottemberg also said that Zen/K12 will be very small cores. He said "Mini-Core dall'alto IPC (Jaguar-Puma style)".

K12/Zen are being designed for embedded, semicustom, consoles, dense servers... those are traditional markets for small, efficient, cheap cores.

Seeing a single core get all of those available transistors used for those shared parts in Bulldozer module does not indicate a weak, efficient core to me. It leaves me expecting something that would leave each core much closer to a full bulldozer's module in multi-thread performance while not having the weak single thread performance that comes from sharing as much as Bulldozer modules share.

Basically, I think it should be reasonable to expect this to perform as a theoretical 8m/16c Bulldozer chip at Fx 8370-E frequencies (3.3ghz base 4.3ghz turbo) while having improved single-thread performance.

As said, the leaks don't indicate that Zen will be a core so big as a whole Bulldozer module.

In any case, a hypothetical Zen core so big as a whole Bulldozer module will produce much less multi-thread performance than the module (I am talking here about integer code since the FPU is shared on a module). There are equations in my last work explaining why.

And that hypothetical big Zen core would be still far from Skylake specially in FP performance.

Finally, I doubt very much that Zen will hit 4GHz, because Samsung/GF 14FF node is optimized for density and efficiency. I doubt that Zen will hit 3.7GHz (turbo) if you pressure me.

 

[szatkus]

Congratulations. I haven't seen such ridiculous post from you for a long time.

Mini-Core dall'alto IPC (Jaguar-Puma style) in grado di scalare in frequenza fino a 4 GHz (a seconda del processo produttivo, dai 28nm ai 14nm);

dall'alto - from on high (yep, Google Translate get it wrong)

The sweclocker's leak doesn't say anything about the size of cores. It only says 8-cores on 95W APUs, which --as I showed-- just point in the opposite direction: smaller cores.

As I said, look at the 18-cores Xeon@22nm.

In any case, a hypothetical Zen core so big as a whole Bulldozer module will produce much less multi-thread performance than the module (I am talking here about integer code since the FPU is shared on a module). There are equations in my last work explaining why.

I would agree if you were comparing similar architectures. Bulldozer is quite bad and Intel's core which more less the same size is able to deliver better ST and similar MT performance. Practice != theory.

And that hypothetical big Zen core would be still far from Skylake specially in FP performance.

Even if your speculations about Zen's FPU will be right, AVX isn't that important for mere consumers. I doubt that AMD will try to sirously attack markets where AVX512 are important.

Finally, I doubt very much that Zen will hit 4GHz, because Samsung/GF 14FF node is optimized for density and efficiency. I doubt that Zen will hit 3.7GHz (turbo) if you pressure me.

They usually have few different flavours of process. High Performance, Low Power, whatever floats your boat. Even Intel has 3 or 4 types of 14nm.

 

[juanrga]

SMT on Amd cores that just sounds wrong in everyway unless they increase the size of the core. Heck i can see that hurting performance like the pentium 4 days.

If juan is right and i think he said 40% better single core performance(would put that around the haswell CPU under my benchmarks) then fine.

SMT on mini-cores "Jaguar-Puma style" doesn't make any sense. SMT makes sense when you have BIG cores (e.g. 8-wide) because superscalar cannot extract enough ILP from most code and several execution ports are iddle. SMT just allow those idle units to execute instructions from another thread.

My 40% higher IPC is an estimation based in the very little information that I have about the microarchitecture. The percent can be accurate or not depending of the accuracy of the info. It would put Zen between Ivy Bridge and Haswell in raw performance.

With available info (assuming is accurate) 8 Zen cores would outperform 4 Skylake cores on integer code, but 4 Skylake cores would outperform 8 Zen cores on FP code.

 

[juanrga]

Mini-Core dall'alto IPC (Jaguar-Puma style) in grado di scalare in frequenza fino a 4 GHz (a seconda del processo produttivo, dai 28nm ai 14nm);

dall'alto - from on high (yep, Google Translate get it wrong)

"Mini-core" and "Jaguar-Puma style" don't need translation. "dall'alto IPC" means "of high IPC".

The sweclocker's leak doesn't say anything about the size of cores. It only says 8-cores on 95W APUs, which --as I showed-- just point in the opposite direction: smaller cores.

As I said, look at the 18-cores Xeon@22nm.

I did, now just show the relevance to the argument/calculation given.

In any case, a hypothetical Zen core so big as a whole Bulldozer module will produce much less multi-thread performance than the module (I am talking here about integer code since the FPU is shared on a module). There are equations in my last work explaining why.

I would agree if you were comparing similar architectures. Bulldozer is quite bad and Intel's core which more less the same size is able to deliver better ST and similar MT performance. Practice != theory.

There is nothing more practical than a good theory! Moreover, the above comment is rather independent of the muarch.

And that hypothetical big Zen core would be still far from Skylake specially in FP performance.

Even if your speculations about Zen's FPU will be right, AVX isn't that important for mere consumers. I doubt that AMD will try to sirously attack markets where AVX512 are important.

AVX must be more important for "mere consumers" than HSA, but my point is that there are life beyond what you call "mere consumers".

Finally, I doubt very much that Zen will hit 4GHz, because Samsung/GF 14FF node is optimized for density and efficiency. I doubt that Zen will hit 3.7GHz (turbo) if you pressure me.

They usually have few different flavours of process. High Performance, Low Power, whatever floats your boat. Even Intel has 3 or 4 types of 14nm.

Intel has several 'flavours' because has the production volume and can charge the price for the 'enthusiasts' chips.

AMD doesn't provide volume enough for Samsung to design an exclusive 'flavour' that hits 4GHz. Neither AMD has enough quality to charge a premium to users.

 

[szatkus]

Mini-Core dall'alto IPC (Jaguar-Puma style) in grado di scalare in frequenza fino a 4 GHz (a seconda del processo produttivo, dai 28nm ai 14nm);

dall'alto - from on high (yep, Google Translate get it wrong)

"Mini-core" and "Jaguar-Puma style" don't need translation. "dall'alto IPC" means "of high IPC".

Broadwell is also "mini" compared to Sandy Bridge. I don't know, ask him for clarification on SA...

The sweclocker's leak doesn't say anything about the size of cores. It only says 8-cores on 95W APUs, which --as I showed-- just point in the opposite direction: smaller cores.

As I said, look at the 18-cores Xeon@22nm.

I did, now just show the relevance to the argument/calculation given.

That shows it is not a big problem to put 8 big cores (big as Bulldozer/Haswell, of course physically they will be smaller) at 14nm.

In any case, a hypothetical Zen core so big as a whole Bulldozer module will produce much less multi-thread performance than the module (I am talking here about integer code since the FPU is shared on a module). There are equations in my last work explaining why.

I would agree if you were comparing similar architectures. Bulldozer is quite bad and Intel's core which more less the same size is able to deliver better ST and similar MT performance. Practice != theory.

There is nothing more practical than a good theory! Moreover, the above comment is rather independent of the muarch.

As long as the theory is close enough to real world to be practical.

And that hypothetical big Zen core would be still far from Skylake specially in FP performance.

Even if your speculations about Zen's FPU will be right, AVX isn't that important for mere consumers. I doubt that AMD will try to sirously attack markets where AVX512 are important.

AVX must be more important for "mere consumers" than HSA, but my point is that there are life beyond what you call "mere consumers".

There are a lot of interesting markets for different CPUs. AMD isn't at a good position to risk. Microservers market isn't stable enough to put everything at it. AMD usually aims at people who like Solitaire, Excel and Crysis.

Finally, I doubt very much that Zen will hit 4GHz, because Samsung/GF 14FF node is optimized for density and efficiency. I doubt that Zen will hit 3.7GHz (turbo) if you pressure me.

They usually have few different flavours of process. High Performance, Low Power, whatever floats your boat. Even Intel has 3 or 4 types of 14nm.

Intel has several 'flavours' because has the production volume and can charge the price for the 'enthusiasts' chips.

AMD doesn't provide volume enough for Samsung to design an exclusive 'flavour' that hits 4GHz. Neither AMD has enough quality to charge a premium to users.

[/quotemsg]
It's not that expensive. GF usually has MPU process mainly for AMD even their Kaveri isn't the bestseller.

 

[8350rocks]

@juan:

You are trying to defend your position because it is unraveling at a rate that will soon be untenable.

I can say this much: puma+ may be inspiration to a degree, In the same manner that the uss john f kennedy was inspired by earlier carrier designs...(except more modern and much larger)

 

[cdrkf]

SMT on Amd cores that just sounds wrong in everyway unless they increase the size of the core. Heck i can see that hurting performance like the pentium 4 days.

If juan is right and i think he said 40% better single core performance(would put that around the haswell CPU under my benchmarks) then fine.

SMT on mini-cores "Jaguar-Puma style" doesn't make any sense. SMT makes sense when you have BIG cores (e.g. 8-wide) because superscalar cannot extract enough ILP from most code and several execution ports are iddle. SMT just allow those idle units to execute instructions from another thread.

My 40% higher IPC is an estimation based in the very little information that I have about the microarchitecture. The percent can be accurate or not depending of the accuracy of the info. It would put Zen between Ivy Bridge and Haswell in raw performance.

With available info (assuming is accurate) 8 Zen cores would outperform 4 Skylake cores on integer code, but 4 Skylake cores would outperform 8 Zen cores on FP code.

An 8 core chip with per core performance between Ivy and Haswell would be a *big* step up from where they are atm. Currently they don't have a scenario where they can actually offer anything outright faster than intel no matter the metric. If they can get a more balanced design that has sufficient single thread performance, and can actually offer a bit better multi thread performance for a similar transistor budget then they'd have a winner. That was what bulldozer was supposed to deliver but didn't. The idea of more cores isn't outright bad, provided they scale up well.

At the end of the day, it all boils down to what you expect from AMD. I don't think for 1 minute they're going to pull another Athlon 64 out of the hat (not *yet* at any rate, they're coming from too far behind), as in release a CPU that wins in every conceivable benchmark. That said, given their current position, if they can release a CPU that is broadly competitive with a few strong areas it would be a big win in my opinion. Focusing the new chip on multi-threaded performance might give them the strong areas they need to make a good case for the design, the trick will be in avoiding creating too much of a bottleneck for single thread in the process.

 

[gamerk316]

With available info (assuming is accurate) 8 Zen cores would outperform 4 Skylake cores on integer code, but 4 Skylake cores would outperform 8 Zen cores on FP code.

Assuming perfect scaling. So for non-benchmark programs, 4 Skylake >= 8 Zen.

 

[gamerk316]

SMT on Amd cores that just sounds wrong in everyway unless they increase the size of the core. Heck i can see that hurting performance like the pentium 4 days.

If juan is right and i think he said 40% better single core performance(would put that around the haswell CPU under my benchmarks) then fine.

SMT on mini-cores "Jaguar-Puma style" doesn't make any sense. SMT makes sense when you have BIG cores (e.g. 8-wide) because superscalar cannot extract enough ILP from most code and several execution ports are iddle. SMT just allow those idle units to execute instructions from another thread.

My 40% higher IPC is an estimation based in the very little information that I have about the microarchitecture. The percent can be accurate or not depending of the accuracy of the info. It would put Zen between Ivy Bridge and Haswell in raw performance.

With available info (assuming is accurate) 8 Zen cores would outperform 4 Skylake cores on integer code, but 4 Skylake cores would outperform 8 Zen cores on FP code.

An 8 core chip with per core performance between Ivy and Haswell would be a *big* step up from where they are atm. Currently they don't have a scenario where they can actually offer anything outright faster than intel no matter the metric. If they can get a more balanced design that has sufficient single thread performance, and can actually offer a bit better multi thread performance for a similar transistor budget then they'd have a winner. That was what bulldozer was supposed to deliver but didn't. The idea of more cores isn't outright bad, provided they scale up well.

At the end of the day, it all boils down to what you expect from AMD. I don't think for 1 minute they're going to pull another Athlon 64 out of the hat (not *yet* at any rate, they're coming from too far behind), as in release a CPU that wins in every conceivable benchmark. That said, given their current position, if they can release a CPU that is broadly competitive with a few strong areas it would be a big win in my opinion. Focusing the new chip on multi-threaded performance might give them the strong areas they need to make a good case for the design, the trick will be in avoiding creating too much of a bottleneck for single thread in the process.

Which is fine, but AMD really needs to pick a sector and start focusing on it. Because right now, I have very little idea on what AMD wants to be. A server CPU vendor? Desktop APUs? Price/Performance king? Low end vendor? Mobile? Right now, AMD is trying to stay alive in ALL of these, and its bleeding them dry.

 

[gamerk316]

Zen SMT confirmed?

http://www.extremetech.com/extreme/198386-amds-next-gen-cpu-leak-14nm-simultaneous-multithreading-and-ddr4-support

 

[-Fran-]

Well, AMD is aiming to be a "jack of all trades" with their "semi custom" design business, gamerk. At least, the consoles are the biggest win they've had thanks to that approach. Like it or not, it seems it *might* work out in the mid term for AMD. I don't know if long term is enough to keep them afloat.

Cheers!

 

[gamerk316]

Well, AMD is aiming to be a "jack of all trades" with their "semi custom" design business, gamerk. At least, the consoles are the biggest win they've had thanks to that approach. Like it or not, it seems it *might* work out in the mid term for AMD. I don't know if long term is enough to keep them afloat.

Cheers!

Long term, not much impact. What are the total sales going to be for consoles when all is said and done; ~200 Million combined? At the end of the day, that pales in comparison to PC sales, which can easily ship 30 Million a QUARTER. They aren't making a lot per sale, and they aren't selling a lot of product.

Jack-of-all-Trades is nice, but with Via/ARM below, and IBM/Intel above, leaves AMD in a very uncomfortable position. Likewise, "semi-custom" requires a LOT of engineering know-how, which itself is expensive. Nevermind ARM is the more attractive semi-custom design at this stage.

I said it years ago: AMD made a colossal mistake not pushing hard in the mobile market. They made another huge mistake basically ceding the high-margin server market. They focused on the high-end desktop instead, and AMD broke its back trying to compete against Intel/NVIDIA. I don't see them going away, but I can easily envision them becoming sort of like Via/S3 in terms of what they offer.

 

[gamerk316]

Well, AMD is aiming to be a "jack of all trades" with their "semi custom" design business, gamerk. At least, the consoles are the biggest win they've had thanks to that approach. Like it or not, it seems it *might* work out in the mid term for AMD. I don't know if long term is enough to keep them afloat.

Cheers!

Long term, not much impact. What are the total sales going to be for consoles when all is said and done; ~200 Million combined? At the end of the day, that pales in comparison to PC sales, which can easily ship 30 Million a QUARTER. They aren't making a lot per sale, and they aren't selling a lot of product.

Jack-of-all-Trades is nice, but with Via/ARM below, and IBM/Intel above, leaves AMD in a very uncomfortable position. Likewise, "semi-custom" requires a LOT of engineering know-how, which itself is expensive. Nevermind ARM is the more attractive semi-custom design at this stage.

I said it years ago: AMD made a colossal mistake not pushing hard in the mobile market. They made another huge mistake basically ceding the high-margin server market. They focused on the high-end desktop instead, and AMD broke its back trying to compete against Intel/NVIDIA. I don't see them going away, but I can easily envision them becoming sort of like Via/S3 in terms of what they offer.

 

[gamerk316]

And the full explanation of the GTX 970 memory architecture:

http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-Discloses-Full-Memory-Structure-and-Limitations-GTX-970

 

[de5_Roy]

Zen SMT confirmed?

http://www.extremetech.com/extreme/198386-amds-next-gen-cpu-leak-14nm-simultaneous-multithreading-and-ddr4-support

what are the chances amd will use excavator's branch predictor in zen and go up to 4 way set assoc. for L1 inst. cache(at least)? i don't see how amd can boost per-core and multi-core performance otherwise. bulk silicon will likely hinder high clockrate. amd might add another integer alu to the cores too.

 

[szatkus]

Zen SMT confirmed?

http://www.extremetech.com/extreme/198386-amds-next-gen-cpu-leak-14nm-simultaneous-multithreading-and-ddr4-support

what are the chances amd will use excavator's branch predictor in zen and go up to 4 way set assoc. for L1 inst. cache(at least)? i don't see how amd can boost per-core and multi-core performance otherwise. bulk silicon will likely hinder high clockrate. amd might add another integer alu to the cores too.

Zen is a redesign. All of these things are possible.

 

[szatkus]

http://www.kitguru.net/components/cpu/anton-shilov/amd-preps-bristol-ridge-apus-carrizo-for-desktops/

Carrizo without HDL?

 

[blackkstar]

Original source: http://www.sweclockers.com/nyhet/19954-amd-zen-forst-ut-i-summit-ridge-for-14-nanometer

Sorry, I don't like WCCF's interpretations :>

In this case WCCFTECH simply copied and pasted about word for word.

The Sweclockers report says lots of ordinary stuff: FM3 as replacement for FM2+, DDR4 as replacement for DDR3, 14nm (which was confirmed before), SMT (mentioned at K12 presentation),...

This leaves 95W TPD and the 8-core leak as the only new.

I still have my doubts about SMT, but let us assume the rest is accurate. Max of 95W TDP is current Kaveri top level and I don't consider that HEDT, but mainstream.

8-cores on 95W APU seems to confirm that the cores will be small and efficient rather than big and powerful.

Considering the TDP of Intel's top parts like 4770k, etc...95W on a 14nm process could very well equate to 125W TDP on 32nm process...

The 95W Zen APU on 14nm will be not competing against 22nm Haswell APUs but against the 95W Skylake APUs on 14nm.

All points to a revival of the old Bulldozer vs Sandy fight:

8 Zen cores will be faster than 4 Skylake cores on highly multithreaded integer code (at least 8 threads), but 8 Zen cores will be slower than 4 Skylake cores on FP code.

Once again AMD will try to hide their core performance deficit with moar cores. Luckily for AMD, Mantle/DX12 will hide partially the core deficit, but that will not help them beyond gaming.

Do you even have any conjecture as to why the core would behave like this? It seems to me that we would be seeing the opposite happen since AMD is going to dedicate more transistors to a single core than with how things are split in Bulldozer family. Also, if Zen supports AVX-512, which I expect it to, I find it highly unlikely the FPU will be as weak as you're claiming. Specifically because the FPU will be dedicated per core if CMT is gone and will not longer have to share resources with other decoders and such.

The only way I see FPU not significantly improving over Bulldozer family is if AMD does something like only give Zen a 128-bit FPU. And if they're going to increase their transistor budget per core like I specified earlier, they have no incentive to do that.

Also, 14nm is supposedly around twice as dense as 32nm at GloFo. So, 8 core Zen, given a Zen core is the same size as a Bulldozer module, would be the same size as an 8m/16c Piledriver with L3 on 14nm. They could probably ditch the L3 entirely and use HBM or something on an interposer and reduce the die size to actually be significantly smaller than a 32nm Orochi die.

Do not underestimate the jump from 32nm to 14nm. We haven't seen a jump like this ever, it's more than halving the advertised process size.

 

[juanrga]

Mini-Core dall'alto IPC (Jaguar-Puma style) in grado di scalare in frequenza fino a 4 GHz (a seconda del processo produttivo, dai 28nm ai 14nm);

dall'alto - from on high (yep, Google Translate get it wrong)

"Mini-core" and "Jaguar-Puma style" don't need translation. "dall'alto IPC" means "of high IPC".

Broadwell is also "mini" compared to Sandy Bridge. I don't know, ask him for clarification on SA...

The sweclocker's leak doesn't say anything about the size of cores. It only says 8-cores on 95W APUs, which --as I showed-- just point in the opposite direction: smaller cores.

As I said, look at the 18-cores Xeon@22nm.

I did, now just show the relevance to the argument/calculation given.

That shows it is not a big problem to put 8 big cores (big as Bulldozer/Haswell, of course physically they will be smaller) at 14nm.

You must be the only person in the world that characterizes Broadwell as "mini-core" "Jaguar-Puma style".

I am rather sure Zen will be only a fraction of the complexity and size of Broadwell.

Finally, I note you avoided to show with numbers what I asked you.

 

[juanrga]

I think amd is spoon feeding info with little clarification to cya itself if plans change. 4ghz could be reference to Bristol Ridge, kaveri replacement as well as the 95 watt tdp
http://www.sweclockers.com/nyhet/19958-amd-bristol-ridge-eftertrader-kaveri-refresh?utm_source=sprinkle&utm_medium=widget&utm_content=www.sweclockers.com

http://www.kitguru.net/components/cpu/anton-shilov/amd-preps-bristol-ridge-apus-carrizo-for-desktops/

Carrizo without HDL?

Sweclockers claims now that Carrizo is coming to the desktop in Q3 2016 on 28nm. I see three possibilities:

(i) Zen has been delayed/canceled and AMD will release Carrizo to fill the roadmap hole.
(ii) Zen is a small core and AMD releases Carrizo with excavator cores for desktop.
(iii) Sweclockers is spreading misinformation.

Choose yourself :-D

Zen SMT confirmed?

http://www.extremetech.com/extreme/198386-amds-next-gen-cpu-leak-14nm-simultaneous-multithreading-and-ddr4-support

That is a copy of the sweclockers report about Zen, which was discussed here before.

 

[szatkus]

(iv) Bristol is low-end (28nm is cheaper than 14nm and for desktop users TDP isn't that important) and Summit (that design with Zen) is high-end.