AMD CPUs, SoC Rumors and Speculations Temp. thread 2

[jimmysmitty]

But i dont think as some people said here, that flagships will be 3,5 GHz (turbo) so AMD always granted nice GHz and bet for 64bit architectures. We can see that intel and AMD, they are doing a pulse of force on this way, giving nice hardware at (more or less) nice price. Both will stuck in 4,4 - 4,5 max GHz this year, maybe next 2017 will bring us 5GHz on stable (without need of liquid) taking a bit more TPD and PC (thermal and power consumption).

AMD didn't always granted GHz and it is not the case with Zen. AMD must be doing a pulse force with Intel. Intel is not. AMD is not in Intel radar. Intel is worried by real competition coming from Apple, IBM, Sun, and ARM server guys.

AMD Zen will not hit 4GHz. And Intel will not hit 5GHz next year. The GHz race finished time ago.

If we base our reasons on which is not possible to put in-processor modules nearest than 14nm, the war will go on optimize software and bring more caché quantity (i'll not surprised to see 1GB cache in next future) and using better caching technollogies (software again after all, based on low level architecture).

1GB cache how? L2 is not the same than L4, adding private caches is not the same than one giant shared cache...

(2 examples: AMD created 64bit atchitecture, intel said that it was no important and laugh. 2 years later inter ran to bring 64 bit architecture by needings, then intel made good multi-threading architecture. AMD start doing the same to be on the way, and bring competition in the market.
More recent: Intel brings low TDP and power consumption to get (quite less, as i tested) performance, but high end cpus at all (intel apus are not valuable with <mod edit> intel hd graphics, as for gaming or more usability, Radeon wins 10 times) Then AMD is running to bring lower TDP and PC as intel did (amd is getting really late on this point) then AMD throws HBM, now intel will bring HBM too).

HP and Intel invented a revolutionary 64-bit architecture first. Threy failed, because was too experimental. Then AMD did the easy thing and developed AMD64, just a 64bit extension of the old x86-32. The AMD64 architecture is considered one of the worst and ugliest 64bit architecture in use.

Funny you mentioned this because if I remember correctly the entire mantra for AMD during the Athlon XP/64/X2 days was "equal to XX00MHz". I.e. a AMD Athlon 64 3200+ was saying that it was clocked at 2GHz but was able to process at the same level as a 3.2GHz Intel CPU.

And that is true, the GHz race has been dead for quite some time. Now they only increase if they can fit it in a certain TDP.

And I noticed he mentioned HBM but Intel wont use HBM. We all know that Intel is heavily invested in the HMC with Micron so they would use that.

I doubt we will see a HBM or HMC CPU in recent times as it is still going to be very expensive to jump to for most people.

 

[juanrga]

But i dont think as some people said here, that flagships will be 3,5 GHz (turbo) so AMD always granted nice GHz and bet for 64bit architectures. We can see that intel and AMD, they are doing a pulse of force on this way, giving nice hardware at (more or less) nice price. Both will stuck in 4,4 - 4,5 max GHz this year, maybe next 2017 will bring us 5GHz on stable (without need of liquid) taking a bit more TPD and PC (thermal and power consumption).

AMD didn't always granted GHz and it is not the case with Zen. AMD must be doing a pulse force with Intel. Intel is not. AMD is not in Intel radar. Intel is worried by real competition coming from Apple, IBM, Sun, and ARM server guys.

AMD Zen will not hit 4GHz. And Intel will not hit 5GHz next year. The GHz race finished time ago.

If we base our reasons on which is not possible to put in-processor modules nearest than 14nm, the war will go on optimize software and bring more caché quantity (i'll not surprised to see 1GB cache in next future) and using better caching technollogies (software again after all, based on low level architecture).

1GB cache how? L2 is not the same than L4, adding private caches is not the same than one giant shared cache...

(2 examples: AMD created 64bit atchitecture, intel said that it was no important and laugh. 2 years later inter ran to bring 64 bit architecture by needings, then intel made good multi-threading architecture. AMD start doing the same to be on the way, and bring competition in the market.
More recent: Intel brings low TDP and power consumption to get (quite less, as i tested) performance, but high end cpus at all (intel apus are not valuable with <mod edit> intel hd graphics, as for gaming or more usability, Radeon wins 10 times) Then AMD is running to bring lower TDP and PC as intel did (amd is getting really late on this point) then AMD throws HBM, now intel will bring HBM too).

HP and Intel invented a revolutionary 64-bit architecture first. Threy failed, because was too experimental. Then AMD did the easy thing and developed AMD64, just a 64bit extension of the old x86-32. The AMD64 architecture is considered one of the worst and ugliest 64bit architecture in use.

Funny you mentioned this because if I remember correctly the entire mantra for AMD during the Athlon XP/64/X2 days was "equal to XX00MHz". I.e. a AMD Athlon 64 3200+ was saying that it was clocked at 2GHz but was able to process at the same level as a 3.2GHz Intel CPU.

And that is true, the GHz race has been dead for quite some time. Now they only increase if they can fit it in a certain TDP.

And I noticed he mentioned HBM but Intel wont use HBM. We all know that Intel is heavily invested in the HMC with Micron so they would use that.

I doubt we will see a HBM or HMC CPU in recent times as it is still going to be very expensive to jump to for most people.

There are two CPUs with HMC, one from Intel (Knight Landing) and one from Fujitsu (Xifx). Both are CPUs for HPC and both are expensive. The Intel design is hybrid but the Fujitsu CPU is a pure implementation.

Cost is not the reason why there is no professional Zen CPU with HBM. HBM is not used in $7000 server CPUs because HBM is a replacement for GDDR5, whereas HMC was designed to replace both GDDR5 and DDR3. This is why the expensive Zen server CPUs only use DDR4.

Cost is the reason why no Zen APU consumer with HBM is expected for 2017. The only APU with HBM is the 16-core APU for HPC, but this is scheduled for 2018. It is interesting that even the future 6TFLOPS console announced by Microsoft doesn't use HBM either.

 

[-Fran-]

I would imagine it's just a cost thing. HBM is great and all, but it's still expensive to package and I bet there has to be a yield issue with them still.

Plus, I don't think it justifies the "edge" it would give either console over GDDR5(X), since the GPUs inside of them wouldn't be that powerful to push the bandwidth of HBM to the limit. If it would give them an edge, I'm pretty sure MS or Sony would try and use HBM. I mean, MS *is* using that ESRAM approach, which is not *that* cheap either. Probably the next gen of Consoles would be using either GDDR5X or HBM.

I still think and APU would be amazing with HBM, but the cost for the consumer would put them in a place where they wouldn't sell. Not even die hard fanbois would be able to buy/justify them, haha. I agree with Jimmy and you on that.

Cheers!

 

[juanrga]

I would imagine it's just a cost thing. HBM is great and all, but it's still expensive to package and I bet there has to be a yield issue with them still.

Plus, I don't think it justifies the "edge" it would give either console over GDDR5(X), since the GPUs inside of them wouldn't be that powerful to push the bandwidth of HBM to the limit. If it would give them an edge, I'm pretty sure MS or Sony would try and use HBM. I mean, MS *is* using that ESRAM approach, which is not *that* cheap either. Probably the next gen of Consoles would be using either GDDR5X or HBM.

I still think and APU would be amazing with HBM, but the cost for the consumer would put them in a place where they wouldn't sell. Not even die hard fanbois would be able to buy/justify them, haha. I agree with Jimmy and you on that.

If it was a question of costs, then HBM would be avoided only in cheapest consumer parts and used in everything else. That is not the case. The fact HBM1 is used in Fury cards for gaming but is not used in 5x more expensive professional cards and CPUs clearly points that cost is not the reason. HBM1 has a capacity issue. That is why it is not used in professional GPUs neither by AMD nor by Nvidia. HBM2 will bring higher capacity, but again it is not used in any CPU I know, not even in expensive $7000 CPUs; therefore capacity and cost aren't the reason for the absence of HBM. The likely reason is that HBM technology has been designed as replacement for GDDR5 for graphics applications and similar. HBM is nice for GPUs both discrete and inside APUs, but HBM is not a replacement for DDR4 and cannot be used in CPUs; that is the reason why AMD uses eight-channel DDR4 in the expensive 32-core Zen CPU.

Cost apart I am not sure if we will see general purpose consumer-grade APUs with HBM, by the same reason we don't see general purpose APUs with ESRAM. ESRAM requires explicit knowledge by the programmer.

 

[cdrkf]

I would imagine it's just a cost thing. HBM is great and all, but it's still expensive to package and I bet there has to be a yield issue with them still.

Plus, I don't think it justifies the "edge" it would give either console over GDDR5(X), since the GPUs inside of them wouldn't be that powerful to push the bandwidth of HBM to the limit. If it would give them an edge, I'm pretty sure MS or Sony would try and use HBM. I mean, MS *is* using that ESRAM approach, which is not *that* cheap either. Probably the next gen of Consoles would be using either GDDR5X or HBM.

I still think and APU would be amazing with HBM, but the cost for the consumer would put them in a place where they wouldn't sell. Not even die hard fanbois would be able to buy/justify them, haha. I agree with Jimmy and you on that.

If it was a question of costs, then HBM would be avoided only in cheapest consumer parts and used in everything else. That is not the case. The fact HBM1 is used in Fury cards for gaming but is not used in 5x more expensive professional cards and CPUs clearly points that cost is not the reason. HBM1 has a capacity issue. That is why it is not used in professional GPUs neither by AMD nor by Nvidia. HBM2 will bring higher capacity, but again it is not used in any CPU I know, not even in expensive $7000 CPUs; therefore capacity and cost aren't the reason for the absence of HBM. The likely reason is that HBM technology has been designed as replacement for GDDR5 for graphics applications and similar. HBM is nice for GPUs both discrete and inside APUs, but HBM is not a replacement for DDR4 and cannot be used in CPUs; that is the reason why AMD uses eight-channel DDR4 in the expensive 32-core Zen CPU.

Cost apart I am not sure if we will see general purpose consumer-grade APUs with HBM, by the same reason we don't see general purpose APUs with ESRAM. ESRAM requires explicit knowledge by the programmer.

I think the argument for a consumer APU with HBM would be it would potentially represent an integrated package that could compete with discrete class components. Probably not much need for such a processor on the desktop, but for laptops, all in one and other smaller form factor devices a HBM equipped APU could offer serious performance that other integrated solutions would be unlikely to match.

I personally think it's a question of *when* such a processor is released rather than *if*. That said the specific makeup of it might not necessarily by HBM, as you say HMC is an alternative (which would arguably remove the need for an external memory pool entirely). I think as a performance enhancer for igpu though HBM 2 is plausible.

 

[-Fran-]

I would imagine it's just a cost thing. HBM is great and all, but it's still expensive to package and I bet there has to be a yield issue with them still.

Plus, I don't think it justifies the "edge" it would give either console over GDDR5(X), since the GPUs inside of them wouldn't be that powerful to push the bandwidth of HBM to the limit. If it would give them an edge, I'm pretty sure MS or Sony would try and use HBM. I mean, MS *is* using that ESRAM approach, which is not *that* cheap either. Probably the next gen of Consoles would be using either GDDR5X or HBM.

I still think and APU would be amazing with HBM, but the cost for the consumer would put them in a place where they wouldn't sell. Not even die hard fanbois would be able to buy/justify them, haha. I agree with Jimmy and you on that.

If it was a question of costs, then HBM would be avoided only in cheapest consumer parts and used in everything else. That is not the case. The fact HBM1 is used in Fury cards for gaming but is not used in 5x more expensive professional cards and CPUs clearly points that cost is not the reason. HBM1 has a capacity issue. That is why it is not used in professional GPUs neither by AMD nor by Nvidia. HBM2 will bring higher capacity, but again it is not used in any CPU I know, not even in expensive $7000 CPUs; therefore capacity and cost aren't the reason for the absence of HBM. The likely reason is that HBM technology has been designed as replacement for GDDR5 for graphics applications and similar. HBM is nice for GPUs both discrete and inside APUs, but HBM is not a replacement for DDR4 and cannot be used in CPUs; that is the reason why AMD uses eight-channel DDR4 in the expensive 32-core Zen CPU.

Cost apart I am not sure if we will see general purpose consumer-grade APUs with HBM, by the same reason we don't see general purpose APUs with ESRAM. ESRAM requires explicit knowledge by the programmer.

I don't know what you answered with your first paragraph, but whatever. I never mixed nor implied HBM was a replacement for regular RAM.

In any case, HBM does make sense inside a Console, and as cdrkf says, mobile/portable. Whether it's a main memory replacement or not, it's up for the maker to choose.

Cheers!

 

[juanrga]

I would imagine it's just a cost thing. HBM is great and all, but it's still expensive to package and I bet there has to be a yield issue with them still.

Plus, I don't think it justifies the "edge" it would give either console over GDDR5(X), since the GPUs inside of them wouldn't be that powerful to push the bandwidth of HBM to the limit. If it would give them an edge, I'm pretty sure MS or Sony would try and use HBM. I mean, MS *is* using that ESRAM approach, which is not *that* cheap either. Probably the next gen of Consoles would be using either GDDR5X or HBM.

I still think and APU would be amazing with HBM, but the cost for the consumer would put them in a place where they wouldn't sell. Not even die hard fanbois would be able to buy/justify them, haha. I agree with Jimmy and you on that.

If it was a question of costs, then HBM would be avoided only in cheapest consumer parts and used in everything else. That is not the case. The fact HBM1 is used in Fury cards for gaming but is not used in 5x more expensive professional cards and CPUs clearly points that cost is not the reason. HBM1 has a capacity issue. That is why it is not used in professional GPUs neither by AMD nor by Nvidia. HBM2 will bring higher capacity, but again it is not used in any CPU I know, not even in expensive $7000 CPUs; therefore capacity and cost aren't the reason for the absence of HBM. The likely reason is that HBM technology has been designed as replacement for GDDR5 for graphics applications and similar. HBM is nice for GPUs both discrete and inside APUs, but HBM is not a replacement for DDR4 and cannot be used in CPUs; that is the reason why AMD uses eight-channel DDR4 in the expensive 32-core Zen CPU.

Cost apart I am not sure if we will see general purpose consumer-grade APUs with HBM, by the same reason we don't see general purpose APUs with ESRAM. ESRAM requires explicit knowledge by the programmer.

I think the argument for a consumer APU with HBM would be it would potentially represent an integrated package that could compete with discrete class components. Probably not much need for such a processor on the desktop, but for laptops, all in one and other smaller form factor devices a HBM equipped APU could offer serious performance that other integrated solutions would be unlikely to match.

I personally think it's a question of *when* such a processor is released rather than *if*. That said the specific makeup of it might not necessarily by HBM, as you say HMC is an alternative (which would arguably remove the need for an external memory pool entirely). I think as a performance enhancer for igpu though HBM 2 is plausible.

My doubt is on how HBM would work in such consumer APUs. The only APU with HBM that I know (i.e. the HPC APU) will be used in a 'custom' design and the whole software stack will be aware of the details of the hardware (just as Microsoft Xbox1 tools are fully aware of the ESRAM pool). I don't see how AMD will introduce HBM in a general purpose APU for laptops.

 

[juanrga]

I would imagine it's just a cost thing. HBM is great and all, but it's still expensive to package and I bet there has to be a yield issue with them still.

Plus, I don't think it justifies the "edge" it would give either console over GDDR5(X), since the GPUs inside of them wouldn't be that powerful to push the bandwidth of HBM to the limit. If it would give them an edge, I'm pretty sure MS or Sony would try and use HBM. I mean, MS *is* using that ESRAM approach, which is not *that* cheap either. Probably the next gen of Consoles would be using either GDDR5X or HBM.

I still think and APU would be amazing with HBM, but the cost for the consumer would put them in a place where they wouldn't sell. Not even die hard fanbois would be able to buy/justify them, haha. I agree with Jimmy and you on that.

If it was a question of costs, then HBM would be avoided only in cheapest consumer parts and used in everything else. That is not the case. The fact HBM1 is used in Fury cards for gaming but is not used in 5x more expensive professional cards and CPUs clearly points that cost is not the reason. HBM1 has a capacity issue. That is why it is not used in professional GPUs neither by AMD nor by Nvidia. HBM2 will bring higher capacity, but again it is not used in any CPU I know, not even in expensive $7000 CPUs; therefore capacity and cost aren't the reason for the absence of HBM. The likely reason is that HBM technology has been designed as replacement for GDDR5 for graphics applications and similar. HBM is nice for GPUs both discrete and inside APUs, but HBM is not a replacement for DDR4 and cannot be used in CPUs; that is the reason why AMD uses eight-channel DDR4 in the expensive 32-core Zen CPU.

Cost apart I am not sure if we will see general purpose consumer-grade APUs with HBM, by the same reason we don't see general purpose APUs with ESRAM. ESRAM requires explicit knowledge by the programmer.

I don't know what you answered with your first paragraph, but whatever. I never mixed nor implied HBM was a replacement for regular RAM.

In any case, HBM does make sense inside a Console, and as cdrkf says, mobile/portable. Whether it's a main memory replacement or not, it's up for the maker to choose.

Cheers!

The maker doesn't have full freedom of choice; they are bound by the limits of the technology. A laptop maker can chose between a solution with DDR4 integrated graphics only or add a mobile discrete card with GDDR5 to increase performance, but no laptop maker has made a laptop with GDDR5 as system memory. By that reason I don't expect a maker to release a laptop with APU and HBM as system memory. Here the decision to avoid HBM is not based in costs, but latency and upgradeability.

Consoles? I can see HBM replacing the GDDR5 used in PS4, but the fastest console announced and coming in 2017 doesn't use HBM. Here the decision to skip HBM seems to be based in costs.

 

[gamerk316]

As I noted over a year ago: HBM is unsuitable as main system memory for a general purpose PC. Consoles probably will go that route, but PCs won't.

 

[cdrkf]

As I noted over a year ago: HBM is unsuitable as main system memory for a general purpose PC. Consoles probably will go that route, but PCs won't.

Yes, but an APU with a HBM pool for the iGPU would still use separate system memory. That's why I think it's possibly something AMD would release in future. It would only be suitable for high end of course, essentially though I still think there is an advantage to a CPU + iGPU + HBM combo processor as it removes the need for a discrete graphics board, even if it doesn't replace the system memory.

 

[simon12]

When do you think iGPUs will be powerful enough to take much advantage from HBM?

 

[-Fran-]

When do you think iGPUs will be powerful enough to take much advantage from HBM?

Since you can configure HBM to have specific bandwidth by stacking them, you can adjust the capacity and bandwidth to match that of what the iGPU can do.

In particular, since the Llano era, the iGPUs are not that weak. Current iGPUs can push a lot of pixels. The comparison might fall short, but look at the PS4. Even if you get 2/3rds of the performance in games for the same/equivalent settings using Windows, then it's a very nice baseline for performance using a single chip. Another way to look at it, is by comparing Iris Pro parts from Intel with their non-EDRAM equivalents.

Cheers!

 

[cdrkf]

When do you think iGPUs will be powerful enough to take much advantage from HBM?

Since you can configure HBM to have specific bandwidth by stacking them, you can adjust the capacity and bandwidth to match that of what the iGPU can do.

In particular, since the Llano era, the iGPUs are not that weak. Current iGPUs can push a lot of pixels. The comparison might fall short, but look at the PS4. Even if you get 2/3rds of the performance in games for the same/equivalent settings using Windows, then it's a very nice baseline for performance using a single chip. Another way to look at it, is by comparing Iris Pro parts from Intel with their non-EDRAM equivalents.

Cheers!

Yeah AMD's current flagship APU has about a third the available memory bandwidth that it could potentially exploit- for a good comparison, compare the latest A10 (with 512 gcn shaders) against the performance of the HD 7750 (a discreet 512 shader part with a 128 bit GDDR 5 memory interface).

Future APU's will have more shaders than this as well- so will be even more bandwidth starved. That said a single HBM stack should offer anywhere from 128gb/s upwards- which would be ideal for 500 - 1000 shaders imo (and about 3 to 4x what is possible over a dual channel DDR4 interface)...

 

[simon12]

Sorry still trying to understand this, are you suggesting HBM on the same chip as the APU? If so I think I get it, if not and the HBM would be on the motherboard or in a slot on the motherboard would it not make sense just to use DDR5 until the graphics are like 1070+ performance? Is the any reason an APU can't have a seperate slot for a different type of memory to the system memory.

 

[-Fran-]

Sorry still trying to understand this, are you suggesting HBM on the same chip as the APU?

Yes.

If so I think I get it, if not and the HBM would be on the motherboard or in a slot on the motherboard would it not make sense just to use DDR5 until the graphics are like 1070+ performance?

Since it is scalable, like I said, the maker can choose whatever suit him best.

Is the any reason an APU can't have a separate slot for a different type of memory to the system memory.

Complexity vs cost I'd say is the correct answer to this question. You always need to strike a balance between simplifying the MoBo circuitry so it is cheaper to manufacture (chipset logic included) and how much stuff you can slap into the APU (on-core and in same package).

Putting it to a more concrete scenario, and one that I'd love to see:
APU: Zen + GCN v5 (+ 2GB HMB ~ 256GB/s in-package).
MoBO: ITX / mATX + DDR4 (2 slots) + bazillion USB3.

That would be quite the potent little machine for a living room IMO.

Cheers!

EDIT: Formatting.

 

[gamerk316]

As I noted over a year ago: HBM is unsuitable as main system memory for a general purpose PC. Consoles probably will go that route, but PCs won't.

Yes, but an APU with a HBM pool for the iGPU would still use separate system memory. That's why I think it's possibly something AMD would release in future. It would only be suitable for high end of course, essentially though I still think there is an advantage to a CPU + iGPU + HBM combo processor as it removes the need for a discrete graphics board, even if it doesn't replace the system memory.

If you stack HBM on the iGPU, you're basically putting a consumer GPU on the CPU die at that point. From a cost perspective, that's a no-go. Nevermind power/heat requirements. At most, you'll get a small 32MB buffer, similar to how EDRAM is used by Intel, just more expensive and power hungry.

 

[cdrkf]

As I noted over a year ago: HBM is unsuitable as main system memory for a general purpose PC. Consoles probably will go that route, but PCs won't.

Yes, but an APU with a HBM pool for the iGPU would still use separate system memory. That's why I think it's possibly something AMD would release in future. It would only be suitable for high end of course, essentially though I still think there is an advantage to a CPU + iGPU + HBM combo processor as it removes the need for a discrete graphics board, even if it doesn't replace the system memory.

If you stack HBM on the iGPU, you're basically putting a consumer GPU on the CPU die at that point. From a cost perspective, that's a no-go. Nevermind power/heat requirements. At most, you'll get a small 32MB buffer, similar to how EDRAM is used by Intel, just more expensive and power hungry.

Intel already offer a 128mb edram. HBM is a much smaller package in comparison- and a cpu + (entry level) dGpu is where this is going to go imo. People want small form factor, no compromise machines- an integrated chip like this would allow that. No one ever said AMD would sell it cheap- they would probably charge a premium in fact, compared to the comparable discrete components. Apple is the sort of buyer who'd be interested.

 

[juanrga]

As I noted over a year ago: HBM is unsuitable as main system memory for a general purpose PC. Consoles probably will go that route, but PCs won't.

Yes, but an APU with a HBM pool for the iGPU would still use separate system memory. That's why I think it's possibly something AMD would release in future. It would only be suitable for high end of course, essentially though I still think there is an advantage to a CPU + iGPU + HBM combo processor as it removes the need for a discrete graphics board, even if it doesn't replace the system memory.

My doubt is on how that HBM pool would be used. According to different leaks/patents and so, HBM seems to be used more like a private memory poll for the iGPU than as a LLC. I wonder if that configuration will require specific software tunning and only used for custom designs (like consoles) or if will be fully used in commercial general workloads.

 

[juanrga]

When do you think iGPUs will be powerful enough to take much advantage from HBM?

AMD already announced a powerful APU with HBM. It is aimed at the HPC market and expected somewhat in 2018. According to leaks the GPU provides ~4 TFLOPS DP.

 

[juanrga]

Future APU's will have more shaders than this as well- so will be even more bandwidth starved. That said a single HBM stack should offer anywhere from 128gb/s upwards- which would be ideal for 500 - 1000 shaders imo (and about 3 to 4x what is possible over a dual channel DDR4 interface)...

Zen-based APUs for AM4 will use higher-clocked DDR4 @ 3200MHz to provide extra bandwidth to the larger iGPU, which is expected to be sized around 12CU.

 

[juanrga]

If you stack HBM on the iGPU, you're basically putting a consumer GPU on the CPU die at that point. From a cost perspective, that's a no-go. Nevermind power/heat requirements. At most, you'll get a small 32MB buffer, similar to how EDRAM is used by Intel, just more expensive and power hungry.

The announced HPC APU is expected to have 500x more memory than that small buffer: 16GB of HBM.

 

[Krnt]

How much of truth is in here?
http://www.guru3d.com/news-story/ams-zen-engineering-sample-specs-leaked.html

2.8GHz with 3.2 turbo seems to be as expected, but its quite high for engineering samples.

 

[juanrga]

How much of truth is in here?
http://www.guru3d.com/news-story/ams-zen-engineering-sample-specs-leaked.html

2.8GHz with 3.2 turbo seems to be as expected, but its quite high for engineering samples.

That article is based in a leak posted in the Anandtech forums and already posted in the page 32 of this thread.

The frequencies aren't rather high, because those aren't the first ES samples,¹ which means that the reported clocks are very close to the clocks that we can expect for final silicon. If you check the original Anandtech source (link given in page 32) you can see some people in that forum claiming that AMD could just release final silicon based in A0 stepping, whereas other are expecting final Zen silicon A1-stepping and 200--400MHz higher clocks.

If AMD releases final silicon with 200--400MHz higher clocks then that agrees very closely with my prediction about Zen clocks 3.0/3.5 GHz I posted in this thread (check page 32) before the leak.

2.8 + 0.2 = 3.0 GHz
3.2 + 0.4 = 3.6 GHz

If AMD releases A0-level 2.8/3.2 silicon then clocks are poor than I expected for GF14LPP.

¹ 8350rocks mentioned some time ago that former ES models were clocked at 2.4GHz, which was already a low frequency (lower than Bulldozer ES clocks) and raised some redflags about Zen.

 

[cdrkf]

How much of truth is in here?
http://www.guru3d.com/news-story/ams-zen-engineering-sample-specs-leaked.html

2.8GHz with 3.2 turbo seems to be as expected, but its quite high for engineering samples.

That article is based in a leak posted in the Anandtech forums and already posted in the page 32 of this thread.

The frequencies aren't rather high, because those aren't the first ES samples,¹ which means that the reported clocks are very close to the clocks that we can expect for final silicon. If you check the original Anandtech source (link given in page 32) you can see some people in that forum claiming that AMD could just release final silicon based in A0 stepping, whereas other are expecting final Zen silicon A1-stepping and 200--400MHz higher clocks.

If AMD releases final silicon with 200--400MHz higher clocks then that agrees very closely with my prediction about Zen clocks 3.0/3.5 GHz I posted in this thread (check page 32) before the leak.

2.8 + 0.2 = 3.0 GHz
3.2 + 0.4 = 3.6 GHz

If AMD releases A0-level 2.8/3.2 silicon then clocks are poor than I expected for GF14LPP.

¹ 8350rocks mentioned some time ago that former ES models were clocked at 2.4GHz, which was already a low frequency (lower than Bulldozer ES clocks) and raised some redflags about Zen.

3.0ghz with boost speed of 3.6 wouldn't be a disaster, although I'd hope to see a higher max boost speed for a single core than that. Base speed of 3ghz on an 8 core part wouldn't be a bad thing.

 

[gamerk316]

Putting aside boost clocks for a second, 40% IPC gains minus 25% base clocks gives a per-core improvement of 15% over PD (FX-8350).

Still expecting 20% average gains here.