Discussion: AMD Ryzen

[TechyInAZ]

Welcome to the AMD Ryzen Discussion Thread!

This will be the home of Ryzen discussions until an offical AMD MegaThread is created. I will also do my best to update this thead with the latest news on Ryzen. If anyone has more recent news of Zen that I have not posted here, please feel free to post it down below and I will do my best to update it here.

*************************************************************************************
Updates:

4/5/2016: AMD ‘pre-announces’ 7th-generation Bristol Ridge, plans for Computex launch

6/2/2016: AMD reveals Zen Summit Ridge CPU at Computex

5/25/2016: AMD Zen Delivers Double The Performance Of The FX 8350 – Zen 8 Core CPU Die Shot Revealedhttp://wccftech.com/amd-zen-cpu-performance-double-fx-8350/

8/10/2016: AMD Zen Engineering Sample Benchmarks Leak Out

8/18/2016: AMD Zen Microarchitecture: Dual Schedulers, Micro-Op Cache and Memory Hierarchy Revealed

9/16/2016: AMD Zen CPU & AM4 Socket Pictured – PGA Design With 1331 Pins Confirmed

*************************************************************************************
My Original Review of Zen A Few Months Ago:

Spoiler

After years of being outperformed in the mainstream market by Intel, Amd says they finally have a solution to get back into the competition against Intel. It’s in their new CPU aritecture called Zen.

MOAR Cores!!!

On paper, AMD’s new Zen aritecture will be almost 50% faster than the previous Vishera series of Piledriver enthusiast CPUs. This makes it one of the largest performance bumps in AMD’s history.

Just like the time years back when the world was amazed at AMD’s ability to put 8 cores on the desktop, AMD is going to do this again with their new Zen CPUs pushing an amazing 16 cores!
However, I would’t be “dazzled” by the 16 core hype. Intel has already taught us that more cores does NOT always = better performance.

But, I believe AMD did learn their lesson with bulldozer. According to many articles I’ve read, single threaded performance on these new Zen cores should be close to Haswell speeds. Plus, add the total amount of 16 cores, we should have a pretty nice competitor for not only Skylake, but also Haswell-E.

Enthusiast CPUs:

The unfortunate fact with rumors is that they don't give exact details as to which CPUs will be specifically oriented to server or desktop or both. So far, the 32 core CPUs are indeed going to the server market only, but the 16 core CPUs are still in the air.

However, other articles are saying that Zen’s enthusiast level CPUs will still only have 8 cores. Which makes me wonder if either A, 16 core and 32 core CPUs will be server only, or B, the 16 core CPUs will be a separate high performance lineup like the Intel Extreme Edition CPUs. Only time will tell.


Memory:

It is rumored that Zen will come with 8 channel memory, 2x the bandwidth of current quad channel systems use by Intel. However I’m very sure this is only for the server market, since 8 channel RAM is insanely overkill for any desktop CPU. My best guess is that we’ll still be seeing dual and quad channel CPUs on the mainstream market.


AM4, A New Platform:

Similar to Intel, AMD will be changing the relationship between CPU and motherboard into a more simplified platform. Instead of having multiple platforms for both the mainstream CPU lineup and the budget APU lineup, AMD will be using only one socket for all lineups. This will be a nice change to customers as this allows future upgradability, flexibility, and customizability previously impossible to do with AMD's older platforms.


APUs With A Mix of Greenland:

AMD hasn’t left their APUs alone either, AMD is currently working on the newest update to their Kaveri and Godaveri APUs and should be released sometime during 2016.

It is confirmed that these new APUs will feature 16 cores overall (CPU cores AND GPU cores) which is a step up from the 12 cores found in the 7850k & 7870k.

HBM will also be added, which should make things very interesting. If both the GPU and CPU can use HBM as their cache source, were looking at the fastest cache systems ever created on a CPU. Far surpassing what L4 cache did with Broadwell. However, we don’t have a lot of info on these new APUs (there is still no codename), but it still seems like they will be using L2 and L3 cache for the majority of their caching system.

Not much has been heard on the graphics side of things, however these APUs will have what's currently codenamed as "Greenland" GPU cores in their dies. They will use the same aritecture as AMD's future Polaris GPUs.


Conclusion:
While not much is being said about Zen at the moment, it does appear that Zen will be a very very competitive CPU against Skylake and most likely Kaby Lake aswell.

There is no official time stamp, but it is set for late 2016 to early 2017.


As usual, feel free to discuss this topic to your hearts desire. But remember to not start or get involved in flame wars. Thank you!

 

[Karsten75]

OK, what was Intel's highest performing single-core processor core back in 2012? At Intel's glacial pace of performance improvement, add 8-10% cumulatively for 4 years and I'm wondering if Zen will even have parity with Skylake?

That's leaving aside the monstrous gaffes such as the Bulldozer/Piledriver architecture missteps.

Then we need to look at TDP parity.

Then reconcile it with AMD's stated intent to not go after HEDT environments anymore.

Then look at the unit cost of component manufacture. Will AMD chipset motherboards achieve price parity with Intel chipset motherboards?

So now that we have maybe achieved that, shall we look at the Intel processor line coming out after Kaby Lake, since I think Kaby Lake is due before Zen?

I'm sick of this Intel vs AMD thing. I'm pretty much getting reconciled to the fact that if I want to continue running Windows I have to bend over for Intel, Nvidia, and Microsoft. That I'm now a cash cow to be milked for every dollar I have.

We should look out, towards ARM, ChromeOS, or some other disruptor will introduce a platform that will become the next defacto environment where competition still drives innovation and price wars.

 

[turkey3_scratch]

It's like waiting for my bananas to ripen... So long!

 

[TechyInAZ]

It's like waiting for my bananas to ripen... So long!

I feel the same way. But, it's better to be safe than sorry. I'd much rather have Zen than a bunch of failed Excavator FX CPUs.

 

[TechyInAZ]

Alright! Finally we actually get to see Zen in person: http://wccftech.com/amd-zen-summit-ridge-fx-processor/

 

[turkey3_scratch]

Yeah I watched AMD's conference yesterday. Lot of new info on Zen we got. They even showed a detailed video that went "I Am Zen" and was done. :lol:

I think they said something about Zen being released in a few weeks? Unless my ears be cheated by some spell, I could have sworn I heard that, though Lisa said it'll be more mass in Q3.

 

[TechyInAZ]

Yeah I watched AMD's conference yesterday. Lot of new info on Zen we got. They even showed a detailed video that went "I Am Zen" and was done. :lol:

I think they said something about Zen being released in a few weeks? Unless my ears be cheated by some spell, I could have sworn I heard that, though Lisa said it'll be more mass in Q3.

Q4 2016 according to that recent article I just posted.

Yeah it's really exciting!!

 

[turkey3_scratch]

Yeah I watched AMD's conference yesterday. Lot of new info on Zen we got. They even showed a detailed video that went "I Am Zen" and was done. :lol:

I think they said something about Zen being released in a few weeks? Unless my ears be cheated by some spell, I could have sworn I heard that, though Lisa said it'll be more mass in Q3.

Q4 2016 according to that recent article I just posted.

Yeah it's really exciting!!

Hmm that's not what Lisa said.

 

[TechyInAZ]

Oh whoops. I forgot what she said about zen, which is why I refered to the article.

 

[James Mason]

So Zen has hyperthreading now?
Or are the AMDboys gonna call it something different?

People need to stop getting mad at Intel for making CPUs that match the software usage market, and get mad at software developers for not making software that knows how to distribute the load across all available CPU threads.

 

[Spectre694]

So Zen has hyperthreading now?
Or are the AMDboys gonna call it something different?

People need to stop getting mad at Intel for making CPUs that match the software usage market, and get mad at software developers for not making software that knows how to distribute the load across all available CPU threads.

Ya that's my understanding. Instead of the modules (two integer one FPU) each core will be one integer one FPU with two hardware threads just like Intel. Probably call it something else though I think hyperthread is a trademark.

 

[TechyInAZ]

So Zen has hyperthreading now?
Or are the AMDboys gonna call it something different?

People need to stop getting mad at Intel for making CPUs that match the software usage market, and get mad at software developers for not making software that knows how to distribute the load across all available CPU threads.

I think hyperthreading should be the standard in CPUs from now on. Just like when CPUs started going from single CPUs to dual CPU cores and higher, hyperthreading should be a normal standard that every CPU is equiped with.

 

[gamerk316]

So Zen has hyperthreading now?
Or are the AMDboys gonna call it something different?

People need to stop getting mad at Intel for making CPUs that match the software usage market, and get mad at software developers for not making software that knows how to distribute the load across all available CPU threads.

I think hyperthreading should be the standard in CPUs from now on. Just like when CPUs started going from single CPUs to dual CPU cores and higher, hyperthreading should be a normal standard that every CPU is equiped with.

In regards to Hyperthreading, most non-low power CPU architectures have had some form of SMT for years. POWER based chips feature four way SMT, or four threads per core, for example.

In regards to software, I could go into a dissertation about how the Windows scheduler works, and repeat the arguments I've been making here since 2010, but I won't. Just dig up the old Bulldozer thread for reference.

Still predicting Haswell level performance @ 3 Ghz or so.

 

[juanrga]

So Zen has hyperthreading now?
Or are the AMDboys gonna call it something different?

Hypertreading is just Intel marketing name for a technology developed by other people and named SMT. SMT has been in use by other companies before: MIPS, Sun, IBM,...

For instance, IBM last Power chips use 8-way SMT, which means each core can run up to 8 threads simultaneously.

It is worth mentioning that AMD engineers are not completely alien to SMT. The K11 family (Buldozer/Piledriver/Steamroller/Excavator) already uses a partial implementation of SMT in the FPU, because the FPU in a CMT module must be able to execute two thread each one coming from an integer core.

 

[juanrga]

I think hyperthreading should be the standard in CPUs from now on. Just like when CPUs started going from single CPUs to dual CPU cores and higher, hyperthreading should be a normal standard that every CPU is equiped with.

No. Hyperthreading (i,e, Intel marketing name for Simultaneous Multithreading or SMT) is a technique valid for transforming ILP into TLP under determined conditions: the first is that one must have a wide core. The second that compiler cannot extract enough ILP from applications. The third the existence of enough threads. Those conditions are common for server-like CPUs running server-like software. Intel consumer chips have hyperthreading because the chips are derived from the server chips. E.g. Haswell-DT shares the same microarchitecture designed for Haswell-EP and EX.

SMT is of little to no utility for mobile. That is why mobile chips dont' have SMT. Intel introduced SMT in the original Atom chip and latter eliminated it. SMT can be a problem for hardware running critical tasks because performance can be difficult to predict. SMT is also virtually useless for a narrow core, for instance a 2-wide OOO core doesn't need SMT because ILP in code is typically high enough to fill all the execution units.

It can be said that most CPU don't use SMT and don't need it.

It is worth to mention that SMT is a solution to a problem. If your code doesn't have enough ILP or if your hardware cannot extract it, then SMT helps to reuse unused hardware resources to run extra threads. But if your code has lots of ILP and the hardware can extract it then SMT is of no use.

 

[juanrga]

After years of being outperformed in the mainstream market by Intel, Amd says they finally have a solution to get back into the competition against Intel. It’s in their new CPU aritecture called Zen.

MOAR Cores!!!
On paper, AMD’s new Zen aritecture will be almost 50% faster than the previous Vishera series of Piledriver enthusiast CPUs. This makes it one of the largest performance bumps in AMD’s history.

Zen is not 50% faster than Vishera. According to AMD Zen is expected to provide about 50% higher IPC than Piledriver. But performance depends of clocks as well. And Zen will not hit the same frequencies than Piledriver.

Many expect Zen to be clocked under 3GHz, but assume it is finally 3.5GHz. That is 34% lower clocks than the FX-9590. Therefore performance will be only 12% faster, not 50%.

Just like the time years back when the world was amazed at AMD’s ability to put 8 cores on the desktop, AMD is going to do this again with their new Zen CPUs pushing an amazing 16 cores!
However, I would’t be “dazzled” by the 16 core hype. Intel has already taught us that more cores does NOT always = better performance.

But, I believe AMD did learn their lesson with bulldozer. According to many articles I’ve read, single threaded performance on these new Zen cores should be close to Haswell speeds. Plus, add the total amount of 16 cores, we should have a pretty nice competitor for not only Skylake, but also Haswell-E.

Enthusiast CPUs:

The unfortunate fact with rumors is that they don't give exact details as to which CPUs will be specifically oriented to server or desktop or both. So far, the 32 core CPUs are indeed going to the server market only, but the 16 core CPUs are still in the air.

However, other articles are saying that Zen’s enthusiast level CPUs will still only have 8 cores. Which makes me wonder if either A, 16 core and 32 core CPUs will be server only, or B, the 16 core CPUs will be a separate high performance lineup like the Intel Extreme Edition CPUs. Only time will tell.

The world wasn't amazed at "AMD’s ability to put 8 cores on the desktop". They used conjoined cores not full cores to fit moar simpler cores in the same die area than a complex core. If you have smaller cores you can fit more in the same area. A Bulldozer module (two conjoined cores) was roughly the same size than one Sandy Bridge core. It is also fair to mention that neither AMD did invent conjoined cores nor was the only company to use them.

AMD is not doing it again. Now they are using full cores for Zen, and the top desktop CPU will have 8 cores. It is Intel which has increased the core count. Broadwell-E comes with a 10-core variant.

The 16-core and 32-core variants of Zen are reserved for server and HPC.

Memory:

It is rumored that Zen will come with 8 channel memory, 2x the bandwidth of current quad channel systems use by Intel. However I’m very sure this is only for the server market, since 8 channel RAM is insanely overkill for any desktop CPU. My best guess is that we’ll still be seeing dual and quad channel CPUs on the mainstream market.

Dual-channel for customers including enthusiast desktop.
Quad-channel for the top HPC/server APU.
Octo-channel for the top server CPU.

AM4, A New Platform:

Similar to Intel, AMD will be changing the relationship between CPU and motherboard into a more simplified platform. Instead of having multiple platforms for both the mainstream CPU lineup and the budget APU lineup, AMD will be using only one socket for all lineups. This will be a nice change to customers as this allows future upgradability, flexibility, and customizability previously impossible to do with AMD's older platforms.

The reason why Intel uses different sockets/platforms is because different users have different needs. The requirements for an enthusiast platform aren't the same that for a entry mainstream plattform. The main reason why AMD is designing a common platform is because lacks the money to desing different platfforms. A common platform has the disadvantage that have to take compromises due to the contradictory requirements from different users. E.g. an enthusiast user want lots of PCIe, memory channels and higher power. An entry mainstream user want a cheap and efficient platfform. Those are contradictory requirements. Thus AMD engineers are doing what is known as a compromises desing by taking a middle point. That is for instance the reason why AM4 is only 95W and has only two memory channels.

The dangers of a compromise plattform were observed in the recent fiasco of the Carrizo/Carrizo-L common platform.

APUs With A Mix of Greenland:

AMD hasn’t left their APUs alone either, AMD is currently working on the newest update to their Kaveri and Godaveri APUs and should be released sometime during 2016.

It is confirmed that these new APUs will feature 16 cores overall (CPU cores AND GPU cores) which is a step up from the 12 cores found in the 7850k & 7870k.

HBM will also be added, which should make things very interesting. If both the GPU and CPU can use HBM as their cache source, were looking at the fastest cache systems ever created on a CPU. Far surpassing what L4 cache did with Broadwell. However, we don’t have a lot of info on these new APUs (there is still no codename), but it still seems like they will be using L2 and L3 cache for the majority of their caching system.

Not much has been heard on the graphics side of things, however these APUs will have what's currently codenamed as "Greenland" GPU cores in their dies. They will use the same aritecture as AMD's future Polaris GPUs.

The 2016 APUs don't use Zen, but Excavator, and there is a rumor that the desktop version have been cancelled. At Computex only the mobile APUs were mentioned. All them have a maximum of 12 compute cores (as Kaveri, Godavari and Carrizo), not 16.

These APUs don't use HBM. The APUs with HBM are expected somewhat in 2017.

HBM is not a replacement for a cache system. HBM is a replacement for GDDR5. HBM in GPUs is used as main graphics memory, not as cache. It is neither known which will be the configuration of the future APUs with HBM. Some diagrams and patents suggest that the HBM is only a local memory pool for the iGPU, whereas the iCPU relies on an ordinary L3 cache.

On the other hand the crystalwell Technology used by Intel is a proper cache L4 level that can be accessed by both CPU and GPU. It is also worth to mention that HBM is an inferior technology to HMC, the thecnology used by Intel in the KNL Xeon and by Fujitsu in the Xifx CPU.

Conclusion:
While not much is being said about Zen at the moment, it does appear that Zen will be a very very competitive CPU against Skylake and most likely Kaby Lake aswell.

There is no official time stamp, but it is set for late 2016 to early 2017.

Zen is a masive improvement over Piledriver or Excavator. But it is not enough for AMD. Zen will be inferior to Broadwell, Skylake, and sucessors. Not to mention desings from IBM, Sun, Broadcom, Qualcomm,...

About release date almost everyone expect a Q4 2016 launch. I expect a paper launch in late 2016 and volume availability only in early 2017

 

[gonf]

wow you are good. with only rumor out there. you explained like pro. nailed. LOL

 

[TechyInAZ]

@juanrga, hey man thanks for the critque! Much appriciated.

But, keep in mind that I made this thread almost 5 months ago. There is far far more info now on Zen now that it's closer to release. Again though, thanks!

 

[juanrga]

Let me be a bit proud and say that I started predicting K12 stuff about two years ago,

http://semiaccurate.com/forums/showpost.php?p=219627&postcount=642

I posted a resume of my expectations for K12 and Zen the past year here at Toms, in an old thread that was fully deleted,

http://www.tomshardware.co.uk/forum/id-2585271/amd-cpus-socs-rumors-speculations-temp-thread/page-3.html#15594766

fortunately a copy of what I said is stored in another part

http://semiaccurate.com/forums/showpost.php?p=232252&postcount=38

A here my last update with extra info (in the deleted thread)

http://www.tomshardware.co.uk/forum/id-2585271/amd-cpus-socs-rumors-speculations-temp-thread/page-9.html#15745175

and the copy

http://semiaccurate.com/forums/showpost.php?p=233677&postcount=422

My 3ALU+3AGU prediction was plain wrong. Zen is 4ALU+2AGU. Also I predicted ~3.5GHz for the octo-core Zen CPU, but last info from the thermal/electrical properties of the process node at Globalfoundries seem to indicate I was also plain wrong on that. Other people is expecting 2.5--2.8GHz for Zen.

I said "~40% IPC over Piledriver" and AMD initially claimed "40% IPC over Excavator", but people like the legendary "The Stilt" says that AMD has lately changed the claim to "up to 40% IPC over Excavator". In any case Zen IPC is higher than I expected but not much more.

There are also doubts about the multidie configuration for the HPC APU and server SoC. I considered that the 32-core SoC for server would be made of four 8-core dies (see diagram in above link) on interposser, with the HPC APU being made of two 8-core dies plus two Greenland dies. However Fudzilla and others claim that the HPC APU uses one 16-core die in an MCM configuration, which means that the 32-core SoC would use two 16-core dies. The correct answer is unknown at this time.

 

[TehPenguin]

Juangra, you make me feel like waiting for Zen is not worth it D:

 

[-Fran-]

Juangra, you make me feel like waiting for Zen is not worth it D:

It is worth it. It's just keeping expectations in check. Plus, at the end of the day, it's just conjecture based on information. A well educated guess.

In any case, it's good to deepen the notion that Zen won't be the second coming of the Athlon and it won't beat top tier Intel offerings (given information available).

Cheers!

 

[juanrga]

Juangra, you make me feel like waiting for Zen is not worth it D:

I suppose it depends of personals expectations. Pessimists were expecting a new Bulldozer fiasco and optimists were expecting a new K8. I was somewhat in the middle.

Let us do some math.

We know that the X4 845 (Excavator with 3.8 GHz max turbo speed) does about 95 points in single-thread Cinebench R15. The Sandy Bridge i7-2600K also have a 3.8 GHz max turbo and does 135 points. AMD claims (see slide in the OP) that Zen core has 40% higher IPC than Excavator core. Thus 95 x 1.4 = 133, so that would bring Zen IPC up to Sandy Bridge levels.

On the other hand AMD has recently shared this slide

The right hand side is not labeled. but many of us believe AMD is saying that Zen CPU for desktop is about 2x faster than Orochi on Cinebench 15 multithreaded. The million dollar question is what AMD means by Orochi?

Orochi is the codename for the server-die used in the BD-based FX and Opteron CPUs. Thus some of us assume that AMD is saying that 8-core Zen is ~2x faster than the FX-8150. The FX-8150 scores about 552 points in CBR15 multithread. Multiply by 2x and we obtain about 1104 points. The 6-core Ivy-Bridge i7 4960X does about 1097.

Others claim that Orochi means FX-8350 because Vishera chips are based in the Orochi rev. C die. Let us then take FX-8350 as baseline. The FX-8350 does about 640 points, so double that would be 1280 points. That puts us right around the E5-2687W v2. This is 8-core Ivy Bridge CPU that does about 1297 points.

 

[TerryLaze]

We know that the X4 845 (Excavator with 3.8 GHz max turbo speed) does about 95 points in single-thread Cinebench R15. The Sandy Bridge i7-2600K also have a 3.8 GHz max turbo and does 135 points. AMD claims (see slide in the OP) that Zen core has 40% higher IPC than Excavator core. Thus 95 x 1.4 = 133, so that would bring Zen IPC up to Sandy Bridge levels.

Yes and they also say that the core will have SMT,so what if you will only be able to achieve the +40% together with SMT?
Same design as FX now,part of the core is separated and needs a second thread to be utilized.
SMT is a umbrella term and can mean a lot of things,just because everybody is hoping that it will be like intel's hyperthreading doesn't make it so.

 

[juanrga]

We know that the X4 845 (Excavator with 3.8 GHz max turbo speed) does about 95 points in single-thread Cinebench R15. The Sandy Bridge i7-2600K also have a 3.8 GHz max turbo and does 135 points. AMD claims (see slide in the OP) that Zen core has 40% higher IPC than Excavator core. Thus 95 x 1.4 = 133, so that would bring Zen IPC up to Sandy Bridge levels.

Yes and they also say that the core will have SMT,so what if you will only be able to achieve the +40% together with SMT?
Same design as FX now,part of the core is separated and needs a second thread to be utilized.
SMT is a umbrella term and can mean a lot of things,just because everybody is hoping that it will be like intel's hyperthreading doesn't make it so.

I doubt that they are including SMT in the 40%. First, because execution resources on Zen (4 ALU + 2 AGU + 2x128bit FMAC) seem to be between Sandy and Haswell; therefore; a similar single-thread IPC level is expected. Second, because if they are including SMT in the percentage count, then the above Orochi/Zen slide doesn't make sense to me.

SMT is a standard acronym for Simultaneous MultiThreading, but don't take my word for it, just check the second point in the first Zen slide in the OP. And as mentioned above hypertreading is just Intel marketing name for Simultaneous MultiThreading.

 

[TehPenguin]

So... Zen won't beat Skylake? Am I getting this right?