AMD CPUs, SoC Rumors and Speculations Temp. thread 2

[jimmysmitty]

IBM Research successfully produces 7-nm test chips
http://techreport.com/news/28603/ibm-research-successfully-produces-7-nm-test-chips
how long till this trickles downs to other cpa members... if it pans out, tessie will fall behind sammy 😗
/edit2

All I'm going to say: Do you REALLY trust the developer to handle manual memory and thread management correctly the first several months (years?) after the API hits? There's a reason APIs have been going the other way for the past two decades.

i wouldn't write off every dev. and it's well understood that new paradigm will take from months to years to get going. the end results will be worth it.

@jimmysmitty: the cpu was rumored to be configured that way from the beyond 3d leaks which turned out to be fakes. those fakes have a real root though (which is what made those seem believable). it's amd's seattle soc. the seattle's cpu part is configured like a real desktop cpu. i assume the prankstar took that idea and projected on possible zen based cpu designs. even if the high level configuration as well as SMT looks the same, the underlying design architecture will likely be radically different.

I guess we will have to wait and see. Producing the chip and how well it works are vastly different. I know Intel was talking about lower nodes having issues such as the CPUs wearing out and the possibility of a CPU having backup cores.

Guess we will see how that happens to play out.

 

[Cazalan]

With the node delays we'll be lucky to see 7nm in production by 2020.

 

[Cazalan]

I guess we will have to wait and see. Producing the chip and how well it works are vastly different. I know Intel was talking about lower nodes having issues such as the CPUs wearing out and the possibility of a CPU having backup cores.

Basically taking more enterprise grade compute solutions into the consumer space. IBM has what they call hot swap cores. Available cores that can be switched on when needed to replace a core that has been generating too many errors. Once the systems are installed they don't really want to rely on physical maintenance to keep the performance up on a monthly basis.

 

[jdwii]

Just to keep things in perspective first if Microsoft bought amd they couldn't produce x86 parts anymore second zen isn't completely new Jim Keller said he took the best parts of several designs. Last bulldozers benchmarks became more reliable the last 3 months before the release with 3-6 months before it motherboard companies thought something was up due to low performance and a low clock speed.

 

[juanrga]

zen benches won't come out for a while. may be... later this year, around october or later. or may be around intel's skylake lineup launches.

There is no silicon to test still. The ES are expected for April 2016. But considering AMD finances, I would expect delays.

What is interesting me is the way Zen is being described. It is being described like I would see a Intel CPU. I do wonder if it is based on an older Intel design maybe Sandy Bridge?

Back when Intel and AMD settled, over the whatever AMD was up in arms about, they resigned a new agreement for cross patent agreements. Basically Intel and AMD had access to each others patents without having to pay or what not.

So I am wondering if AMD is using Intel ideas to create Zen. One thing that makes me thing this is SMT of course, that is an Intel design based on something DEC Alpha had (much like the IMC and the module approach for BD) but also the "new" high speed cache. Intel has always had a much faster cache design than AMD, even when their CPUs sucked their cache was faster.

It wouldn't surprise me. If you had access to ideas that worked, wouldn't be sued/fined for using them and your current ideas failed would you not utilize them to your advantage instead of investing billions of dollars you didn't have into R&D?

Just a thought.

Zen is AMD's child. Zen will have the same basic elements than any other high-performance server-like core from Intel, IBM, Broadcomm... Zen core will be a superscalar, OOOE, SMT,... because that is the optimal point in the design space of server-like cores.

The cache subsystem is entirely new. AMD has been filling cache patents since Keller return. If patents filled are representative (not all patented work finally goes into commercial designs), Zen will have a stacked cache susbsystem. This is a cache structured to favor certain prefetching techniques that exploit temporal locality of data. Prefetching techniques are used to increase IPC and, thus, performance.

The reason for cancellation of several project, the delay of K12, the rebrands of 300-series,... is because AMD is concentrating all the efforts and money on Zen design.

 

[etayorius]

zen benches won't come out for a while. may be... later this year, around october or later. or may be around intel's skylake lineup launches.

There is no silicon to test still. The ES are expected for April 2016. But considering AMD finances, I would expect delays.

What is interesting me is the way Zen is being described. It is being described like I would see a Intel CPU. I do wonder if it is based on an older Intel design maybe Sandy Bridge?

Back when Intel and AMD settled, over the whatever AMD was up in arms about, they resigned a new agreement for cross patent agreements. Basically Intel and AMD had access to each others patents without having to pay or what not.

So I am wondering if AMD is using Intel ideas to create Zen. One thing that makes me thing this is SMT of course, that is an Intel design based on something DEC Alpha had (much like the IMC and the module approach for BD) but also the "new" high speed cache. Intel has always had a much faster cache design than AMD, even when their CPUs sucked their cache was faster.

It wouldn't surprise me. If you had access to ideas that worked, wouldn't be sued/fined for using them and your current ideas failed would you not utilize them to your advantage instead of investing billions of dollars you didn't have into R&D?

Just a thought.

Zen is AMD's child. Zen will have the same basic elements than any other high-performance server-like core from Intel, IBM, Broadcomm... Zen core will be a superscalar, OOOE, SMT,... because that is the optimal point in the design space of server-like cores.

The cache subsystem is entirely new. AMD has been filling cache patents since Keller return. If patents filled are representative (not all patented work finally goes into commercial designs), Zen will have a stacked cache susbsystem. This is a cache structured to favor certain prefetching techniques that exploit temporal locality of data. Prefetching techniques are used to increase IPC and, thus, performance.

The reason for cancellation of several project, the delay of K12, the rebrands of 300-series,... is because AMD is concentrating all the efforts and money on Zen design.

Omg i hope Zen will be good, or at least good enough... i may even give AMD a chance and wait till they release Zen, if it´s good i may even build my next Desktop with it, if not ill just go with Skylake as planned, i got a ton of cash been saving for like 2 years and it´s itching for Skylake, but if AMD is ramping up on Zen it may actually be good.

 

[CooLWoLF]

Zen has to succeed. Or its bad news for all of us $$$ wise.

 

[jimmysmitty]

Zen has to succeed. Or its bad news for all of us $$$ wise.

Not really because someone will buy out AMD and then Intel will be forced to allow the transfer of the x86 license to keep the markets competitive.

 

[cdrkf]

Zen has to succeed. Or its bad news for all of us $$$ wise.

Not really because someone will buy out AMD and then Intel will be forced to allow the transfer of the x86 license to keep the markets competitive.

In theory.... although I personally wouldn't count AMD out just yet. They are putting everything they got into Zen and we know they can produce some superb kit (as they have done many times in the past). Everything appears to be lining up for them as well- they will for example be on an equal process footing with Intel for the first time in ages.

 

[jimmysmitty]

Zen has to succeed. Or its bad news for all of us $$$ wise.

Not really because someone will buy out AMD and then Intel will be forced to allow the transfer of the x86 license to keep the markets competitive.

In theory.... although I personally wouldn't count AMD out just yet. They are putting everything they got into Zen and we know they can produce some superb kit (as they have done many times in the past). Everything appears to be lining up for them as well- they will for example be on an equal process footing with Intel for the first time in ages.

Not quite equal. Cannonlake, the 10nm shrink of Skylake, is set to be out in 2016, same year as Zen. The delay for Broadwell didn't stop the other parts from continuing.

 

[cdrkf]

The Intel chip your referencing is due out imminently so of course design wins are being announced.

No. That Intel chip will be released by about 2017 or so. And the supercomputers based in it and the other supercomputers based in future IBM Power 9 chips are expected to be working in 2018.

None of customers chose Zen, K12 or GCN2, which gives a good idea of how uncompetitive those designs are.

Given that there is no silicon out yet for any of those I don't really see how it proves anything beyond the level of confidence said HPC companies have in performance predictions. AMD have ground to make up on that score after the dozer' cores failed to meet targets.

Edit: Also I really don't understand why you make such an effort to comment here if your so *sure* AMD can't produce anything worthwhile any more. I like participating in this thread as I'm interested in the tech (all tech, advances by Intel and nVidia are also good), and I personally still think there are some nice things to come from AMD in the future. Perhaps I'm wrong, but please can we try and avoid getting into an argument to prove how hopeless everything is all the time? I'd understand if I was making statements like 'AMD will dominate Intel' but lets be honest, most people here are fairly well clued up and no one is saying this.

All we're saying is we think AMD can release another decent processor (and decent isn't the same thing as saying 'Better than Intel'). If that happens then hopefully there will be a worthwhile argument for purchasing AMD cpu's again, as currently I can't really advise people to go that way given the advantages a similarly prices Intel rig provides. At least where the graphics cards are concerned we have real choice up and down the product stack. As for what happens with the next gen, that's quite a ways off and we have no real details of either architecture so lets discuss that when there is actually something to discus.

 

[jimmysmitty]

Zen has to succeed. Or its bad news for all of us $$$ wise.

Not really because someone will buy out AMD and then Intel will be forced to allow the transfer of the x86 license to keep the markets competitive.

In theory.... although I personally wouldn't count AMD out just yet. They are putting everything they got into Zen and we know they can produce some superb kit (as they have done many times in the past). Everything appears to be lining up for them as well- they will for example be on an equal process footing with Intel for the first time in ages.

Many times? Equal process footing? I guess no.

They look set to both be on 14nm (given how long Intel delayed 14nm I don't expect to see 10nm any time soon).

AMD have also made many great products, they have only struggled very recently:

- AMD 80386 faster than Intel
- AMD 80486, faster than Intel
- Pentium and K5 were equivalent.
- K6 and K62 were slower than P2, but also much cheaper thanks to keeping Super Socket 7 so offered significant advantage in perf / $ and thanks to AMD's then new '3D now' extensions were great for gaming.
- Athlon 1 was performance parity with Pentium 3
- AMD were first to 1ghz with Athlon T-Bird
- The Athlon XP processors were quite a bit faster per clock than the first gen P4
- Athlon 64 was the first 64 bit processor and made Intel look bad.
- Athlon 64 X2 was first actual dual core single die processor and was great. It was only with the introduction of Core 2 that Intel got ahead.
- Phenom I was first true monolithic quad core, as well as offering a tri level cache structure and HT instead of FSB (ok it had issues but Phenom II didn't and was a superb chip)- note Intel basically copied the top level design of Phenom with their first gen i7.

So yes, on that basis, I think it's fair to say over their history AMD have repeatedly produced great kit.

1. 386 and 386 were just clones of Intels designs and while AMD was putting them out, Intel was on the next step. For example, while AMD was releasing their first 486 in 1993 Intel was releasing the first Pentium.

2. If K8 and Conroe have taught us anything it is that clock speed is irrelevant if the IPC is low.

3. Athlon XPs were ok against first gen P4s yes but after that they were a PITA to use. Your board might have booted with it but it might not have supported the proper clock speed. Had plenty of people have that issue where they bought a Athlon XP 2600+ and it would only urn at best as an 1800+.

4. Athlon 64 ws not the first 64bit chip. There were plenty before it including Itanium, which was started in development in the mid 90s.

5. And Intel proved that a MCM design can beat a monolithic design. Core 2 Quad was superior to Phenom and was still competitive, first gen, to Phenom II. Second gen Core 2 Quad beat Phenom easily. And tell me how Nehalem could have copied Phenom when its development was started many years before even the K10 uArch was announced:

http://www.geek.com/chips/pentium-8-codenamed-nehalem-under-way-at-intel-550648/

2002 was the first sighting of it.

And AMD was not the first with L3. The Pentium 4 EE had a L3 cache. L3 was just pointless until more recently.

 

[jimmysmitty]

Zen has to succeed. Or its bad news for all of us $$$ wise.

Not really because someone will buy out AMD and then Intel will be forced to allow the transfer of the x86 license to keep the markets competitive.

In theory.... although I personally wouldn't count AMD out just yet. They are putting everything they got into Zen and we know they can produce some superb kit (as they have done many times in the past). Everything appears to be lining up for them as well- they will for example be on an equal process footing with Intel for the first time in ages.

Many times? Equal process footing? I guess no.

They look set to both be on 14nm (given how long Intel delayed 14nm I don't expect to see 10nm any time soon).

AMD have also made many great products, they have only struggled very recently:

- AMD 80386 faster than Intel
- AMD 80486, faster than Intel
- Pentium and K5 were equivalent.
- K6 and K62 were slower than P2, but also much cheaper thanks to keeping Super Socket 7 so offered significant advantage in perf / $ and thanks to AMD's then new '3D now' extensions were great for gaming.
- Athlon 1 was performance parity with Pentium 3
- AMD were first to 1ghz with Athlon T-Bird
- The Athlon XP processors were quite a bit faster per clock than the first gen P4
- Athlon 64 was the first 64 bit processor and made Intel look bad.
- Athlon 64 X2 was first actual dual core single die processor and was great. It was only with the introduction of Core 2 that Intel got ahead.
- Phenom I was first true monolithic quad core, as well as offering a tri level cache structure and HT instead of FSB (ok it had issues but Phenom II didn't and was a superb chip)- note Intel basically copied the top level design of Phenom with their first gen i7.

So yes, on that basis, I think it's fair to say over their history AMD have repeatedly produced great kit.

1. 386 and 386 were just clones of Intels designs and while AMD was putting them out, Intel was on the next step. For example, while AMD was releasing their first 486 in 1993 Intel was releasing the first Pentium.

2. If K8 and Conroe have taught us anything it is that clock speed is irrelevant if the IPC is low.

3. Athlon XPs were ok against first gen P4s yes but after that they were a PITA to use. Your board might have booted with it but it might not have supported the proper clock speed. Had plenty of people have that issue where they bought a Athlon XP 2600+ and it would only urn at best as an 1800+.

4. Athlon 64 ws not the first 64bit chip. There were plenty before it including Itanium, which was started in development in the mid 90s.

5. And Intel proved that a MCM design can beat a monolithic design. Core 2 Quad was superior to Phenom and was still competitive, first gen, to Phenom II. Second gen Core 2 Quad beat Phenom easily. And tell me how Nehalem could have copied Phenom when its development was started many years before even the K10 uArch was announced:

http://www.geek.com/chips/pentium-8-codenamed-nehalem-under-way-at-intel-550648/

2002 was the first sighting of it.

And AMD was not the first with L3. The Pentium 4 EE had a L3 cache. L3 was just pointless until more recently.

Oh dear I really didn't think suggesting in an AMD thread that AMD actually made some good kit would cause such a controversy (irrespective of all the above, all the processors I mentioned were good in their day respectively- the 'issues' with Athlon XP were down to outdated motherboards, hardly AMD's problem?). My point wasn't that "at each point AMD beat Intel" but rather all those processors were good one way or another (the first Pentiums for example was astronomically expensive so Intel and AMD based 486 machines were the mainstay for most users).

My first machine was an IBM 286, I remember all through the 90s and repeatedly AMD offered far better value for money for the low to mid range, whilst giving Intel a real run for their money in the high end when we get into the end of the 90's and early 2000s.

I'll also amend my statement to say the Athlon 64 was the first 64 bit x86 based cpu (Itanium was a totally new uarch and never really gained traction, there's a lot to be said for maintaining compatibility).

All this comes back to my original point, I've not said at *any point* to *anyone* that AMD are about to become the next Apple and 'defeat' Intel (an AMD monopoly would be just as bad for the industry as an Intel one imo). All I've dared to say is that I don't think Zen is a guaranteed failure.

One thing I think we all agree on- there is still going to be a delta between Intel and AMD in terms of raw, per core performance (dependent on application of course). However, what little we've heard of Zen also points to a design that corrects many of the biggest issues in the modular design. It's also looks likely AMD are going to tackle the performance delta by offering more cores for the money. What that suggests to me is that:

- Single thread performance will still lag, but the situation should be better than now.
- Multi thread performance is potentially very good given we are looking at full cores rather than modules (FPU performance should be notably better than dozer as an 8 core zen cpu would have 8 fpu units rather than 4).

The above doesn't sound like a bad proposition to me, provided it's priced well (note it will probably be expensive when it's first released however prices have a tendency to adjust themselves quite quickly once an post launch hype has died down).

There is information and correct information. I never said AMD was bad nor that they were not good value for the money. Just certain things, such as Nehalem copying K10 (Phenom) when it was in development longer.

Zen itself wont be modular. It was a bad design choice and AMD knows it. It will be a standard CPU just like an i7 with SMT which is the smarter choice to get a boost in performance for little to no real estate or increase in power consumption.

The current hope is that Zen will be competitive. I wish they were as Intel is at a stand still when they probably could have done much more. But the numbers we have been hearing are not something that is showing a CPU that will blow our socks off.

We could be wrong but we wont know until we get closer to launch.

AMD has offered value for the money because that was the only way they could sell their chips. When K8 came out and Intel flubbed with Netburst they gained a massive upper hand, although with a few issues such as the ability to meet demand and supply shortages. And when that happened they priced accordingly as they could demand a higher price. They were equal to or higher than what Intel had.

Back when Core 2 Quad came out you could get a quad core AMD setup. It was the Quadfather or Quad FX, two names it was known by. But the performance was not there, the power use was insane and it was $1400 dollars for two CPUs alone that couldn't even beat, at 3GHz, the 2.4GHz Core 2 Quad Q6600. Of course after that pricing went back to the way it was but AMD has shown a few more examples of them pricing where they think it could be. The FX 8150 came out at the same price as a 2600K and the FX 9590, well that was just insanely over priced much like most "extreme" CPUs.

AMD is a company and their survival comes first.

 

[juanrga]

I expect Zen to be 20% or more behind Skylake in regards to IPC, but if they sell 8 cores CPUs then i would get one even if the IPC is not as good, the thing is Intel still milking people with 4 Cores even on the i7.

But i just can`t wait till end of 2016 or early 2017 till Zen comes to Desktop, i will just have to build my next PC with what ever the Top i5 is by October-November.

Zen will appear in an enthusiast line of CPU (no integrated graphics). Those will not be cheap. Mainstream users will be directed towards the Excavator-based APUs.

 

[cdrkf]

Zen has to succeed. Or its bad news for all of us $$$ wise.

Not really because someone will buy out AMD and then Intel will be forced to allow the transfer of the x86 license to keep the markets competitive.

In theory.... although I personally wouldn't count AMD out just yet. They are putting everything they got into Zen and we know they can produce some superb kit (as they have done many times in the past). Everything appears to be lining up for them as well- they will for example be on an equal process footing with Intel for the first time in ages.

Many times? Equal process footing? I guess no.

They look set to both be on 14nm (given how long Intel delayed 14nm I don't expect to see 10nm any time soon).

AMD have also made many great products, they have only struggled very recently:

- AMD 80386 faster than Intel
- AMD 80486, faster than Intel
- Pentium and K5 were equivalent.
- K6 and K62 were slower than P2, but also much cheaper thanks to keeping Super Socket 7 so offered significant advantage in perf / $ and thanks to AMD's then new '3D now' extensions were great for gaming.
- Athlon 1 was performance parity with Pentium 3
- AMD were first to 1ghz with Athlon T-Bird
- The Athlon XP processors were quite a bit faster per clock than the first gen P4
- Athlon 64 was the first 64 bit processor and made Intel look bad.
- Athlon 64 X2 was first actual dual core single die processor and was great. It was only with the introduction of Core 2 that Intel got ahead.
- Phenom I was first true monolithic quad core, as well as offering a tri level cache structure and HT instead of FSB (ok it had issues but Phenom II didn't and was a superb chip)- note Intel basically copied the top level design of Phenom with their first gen i7.

So yes, on that basis, I think it's fair to say over their history AMD have repeatedly produced great kit.

1. 386 and 386 were just clones of Intels designs and while AMD was putting them out, Intel was on the next step. For example, while AMD was releasing their first 486 in 1993 Intel was releasing the first Pentium.

2. If K8 and Conroe have taught us anything it is that clock speed is irrelevant if the IPC is low.

3. Athlon XPs were ok against first gen P4s yes but after that they were a PITA to use. Your board might have booted with it but it might not have supported the proper clock speed. Had plenty of people have that issue where they bought a Athlon XP 2600+ and it would only urn at best as an 1800+.

4. Athlon 64 ws not the first 64bit chip. There were plenty before it including Itanium, which was started in development in the mid 90s.

5. And Intel proved that a MCM design can beat a monolithic design. Core 2 Quad was superior to Phenom and was still competitive, first gen, to Phenom II. Second gen Core 2 Quad beat Phenom easily. And tell me how Nehalem could have copied Phenom when its development was started many years before even the K10 uArch was announced:

http://www.geek.com/chips/pentium-8-codenamed-nehalem-under-way-at-intel-550648/

2002 was the first sighting of it.

And AMD was not the first with L3. The Pentium 4 EE had a L3 cache. L3 was just pointless until more recently.

Oh dear I really didn't think suggesting in an AMD thread that AMD actually made some good kit would cause such a controversy (irrespective of all the above, all the processors I mentioned were good in their day respectively- the 'issues' with Athlon XP were down to outdated motherboards, hardly AMD's problem?). My point wasn't that "at each point AMD beat Intel" but rather all those processors were good one way or another (the first Pentiums for example was astronomically expensive so Intel and AMD based 486 machines were the mainstay for most users).

My first machine was an IBM 286, I remember all through the 90s and repeatedly AMD offered far better value for money for the low to mid range, whilst giving Intel a real run for their money in the high end when we get into the end of the 90's and early 2000s.

I'll also amend my statement to say the Athlon 64 was the first 64 bit x86 based cpu (Itanium was a totally new uarch and never really gained traction, there's a lot to be said for maintaining compatibility).

All this comes back to my original point, I've not said at *any point* to *anyone* that AMD are about to become the next Apple and 'defeat' Intel (an AMD monopoly would be just as bad for the industry as an Intel one imo). All I've dared to say is that I don't think Zen is a guaranteed failure.

One thing I think we all agree on- there is still going to be a delta between Intel and AMD in terms of raw, per core performance (dependent on application of course). However, what little we've heard of Zen also points to a design that corrects many of the biggest issues in the modular design. It's also looks likely AMD are going to tackle the performance delta by offering more cores for the money. What that suggests to me is that:

- Single thread performance will still lag, but the situation should be better than now.
- Multi thread performance is potentially very good given we are looking at full cores rather than modules (FPU performance should be notably better than dozer as an 8 core zen cpu would have 8 fpu units rather than 4).

The above doesn't sound like a bad proposition to me, provided it's priced well (note it will probably be expensive when it's first released however prices have a tendency to adjust themselves quite quickly once an post launch hype has died down).

There is information and correct information. I never said AMD was bad nor that they were not good value for the money. Just certain things, such as Nehalem copying K10 (Phenom) when it was in development longer.

Zen itself wont be modular. It was a bad design choice and AMD knows it. It will be a standard CPU just like an i7 with SMT which is the smarter choice to get a boost in performance for little to no real estate or increase in power consumption.

The current hope is that Zen will be competitive. I wish they were as Intel is at a stand still when they probably could have done much more. But the numbers we have been hearing are not something that is showing a CPU that will blow our socks off.

We could be wrong but we wont know until we get closer to launch.

AMD has offered value for the money because that was the only way they could sell their chips. When K8 came out and Intel flubbed with Netburst they gained a massive upper hand, although with a few issues such as the ability to meet demand and supply shortages. And when that happened they priced accordingly as they could demand a higher price. They were equal to or higher than what Intel had.

Back when Core 2 Quad came out you could get a quad core AMD setup. It was the Quadfather or Quad FX, two names it was known by. But the performance was not there, the power use was insane and it was $1400 dollars for two CPUs alone that couldn't even beat, at 3GHz, the 2.4GHz Core 2 Quad Q6600. Of course after that pricing went back to the way it was but AMD has shown a few more examples of them pricing where they think it could be. The FX 8150 came out at the same price as a 2600K and the FX 9590, well that was just insanely over priced much like most "extreme" CPUs.

AMD is a company and their survival comes first.

Oh I agree, if Zen is a monster CPU, it will come with a commensurate price tag that's for sure 😛

With respect to Phenom vs i7, I think if anything the first gen i7 rather validates AMD's design decision with Phenom. Out of interest I remember when Phenom I came out there were some interesting deep dives onto the multi core performance and they did identify that Core 2 Quad had an issue when using all 4 cores on a single application.

At the time there was so little software that used even 2 threads, the main advantage of quad core was running multiple discreet processes in parallel (which was fine on C2Q), however I think these days the bottleneck would be more apparent. As for Phenom II, from what I remember the 'battle' between C2Q and Phenom II was never all that black and white, as AMD were able to offer very high clock speeds on Phenom II and it was always a good performer in games. Even after Nahlem, Phenom II X6 was a great CPU, I actually still run an X6 at work for cpu bound rendering duties (at the time it was unbeatable value for that, obviously there are far faster processors available now although it's fast enough for the light duties I use it for).

 

[8350rocks]

I expect Zen to be 20% or more behind Skylake in regards to IPC, but if they sell 8 cores CPUs then i would get one even if the IPC is not as good, the thing is Intel still milking people with 4 Cores even on the i7.

But i just can`t wait till end of 2016 or early 2017 till Zen comes to Desktop, i will just have to build my next PC with what ever the Top i5 is by October-November.

Zen will appear in an enthusiast line of CPU (no integrated graphics). Those will not be cheap. Mainstream users will be directed towards the Excavator-based APUs.

Can you please post the source?

I have no information regarding anything beyond something similar to the FX branding on that line of CPUs. Which does not command a premium price.

 

[juanrga]

Given that there is no silicon out yet for any of those I don't really see how it proves anything beyond the level of confidence said HPC companies have in performance predictions. AMD have ground to make up on that score after the dozer' cores failed to meet targets.

Edit: Also I really don't understand why you make such an effort to comment here if your so *sure* AMD can't produce anything worthwhile any more. I like participating in this thread as I'm interested in the tech (all tech, advances by Intel and nVidia are also good), and I personally still think there are some nice things to come from AMD in the future. Perhaps I'm wrong, but please can we try and avoid getting into an argument to prove how hopeless everything is all the time? I'd understand if I was making statements like 'AMD will dominate Intel' but lets be honest, most people here are fairly well clued up and no one is saying this.

All we're saying is we think AMD can release another decent processor (and decent isn't the same thing as saying 'Better than Intel'). If that happens then hopefully there will be a worthwhile argument for purchasing AMD cpu's again, as currently I can't really advise people to go that way given the advantages a similarly prices Intel rig provides. At least where the graphics cards are concerned we have real choice up and down the product stack. As for what happens with the next gen, that's quite a ways off and we have no real details of either architecture so lets discuss that when there is actually something to discus.

It is pretty simple. Potential customers did launch open calls for proposals. Companies submitted proposals. Customers reviewed proposals and selected winners. Multimillionary contracts were signed and some systems are already being build. Systems will be operative by 2018 or so.

All customers selected future hardware from IBM, Nvidia, Intel, Cavium... IBM will provide future Power 9 CPU. Nvidia will provide future Pascal-based GPGPUs, Intel will provide future Knight Hill Xeon, which will arrive on 2017...

No known customer will use Zen, K12 or GCN2. Do you recall that recent leak about a future HPC APU of 250W that used 16 Zen cores? That HPC APU appears in the official AMD roadmaps for 2017. No known system will use it. This is similar to the overhyped Berlin APU. I am still waiting to find a server/HPC system using it. The Berlin APU apparently only can be found on paper.

The explanation for the lack of any design win with future AMD hardware is that future AMD hardware is not competitive even in the market for which it was designed. Recall Zen main market is not PCs but servers/HPC.

This fiasco is something was expected from an understanding of basic tech and economy. Some time ago I posted in the old thread a graph showing the decline on R&D and I said this was going to hurt future products: Carrizo, Fury, 300-series, projects cancellation (Skybridge, Mantle), K12 delay,... are just confirming.

As news reported about one month ago, AMD is preparing itself for a split, and I continue to maintain that AMD will likely abandon the PC market. Numbers I have suggests just that (next AMD report will show loses on all the PC markets: CPU, APU, GPU,...). There is no objective reason to believe that AMD will do a comeback to PCs and mentioning Am386 or K5 chips is going to mislead readers not familiar with history or details.

 

[jimmysmitty]

1. 386 and 386 were just clones of Intels designs and while AMD was putting them out, Intel was on the next step. For example, while AMD was releasing their first 486 in 1993 Intel was releasing the first Pentium.

The delay was being caused by the time required to reverse engineering the Intel processor and then clone it. However, Intel sued AMD for cloning its chips and AMD had to start developing its 'own' x86 chips, but curiously all AMD in-house x86 chips were copied or derived from other chips:

- K5 derived from 29k
- K6 copied from NetxGen Nx686.
- K7 largely based on the Alpha 21264.
- AMD K8 was essentially the K7.
- AMD K9 was canceled.
- AMD K10 based in K8.
- AMD K11 (Bulldozer) was based in a design rejected by Intel. Yes even 'revolutionary' Bulldozer was copied from Intel!

I hear some rumors on forums that K12 is based in Apple Cyclone. But that makes no sense.

I thought that the Bulldozer design was based on a design that Intel had acquired when they bought DECAlpha. It was one of the two SMT methods they developed and Intel allowed AMD rights to some of the designs from Alpha (including the IMC for K8 and the module design for BD) as Intel didn't feel those designs were necessary yet (IMC) or efficient (module).

 

[jimmysmitty]

We already have a good idea of how good Zen is or isn't. No known design win uses Zen. Customers/partners chose Cavium, Intel, and IBM CPUs.

There are not even ES performance leaks at the moment. The fact that there are no design wins right now when chips will not ship for roughly a year still...not concerning.

Exactly, I think we're jumping the gun declaring an unfinished design with no physical silicon a dud before we know anything about it.

Irrelevant arguments, because none of the companies has ES performance leaks. The chips that won have not even been tape-out: Nvidia Volta is not taped-out, Power 9 is not taped-out, KNH Xeon is not taped-out... they are all in the design office still. Nvidia Volta and KNH Xeon will ship on 2017.

Maybe we would start speculating why none of those important customers chose AMD Zen and GCN2. And why AMD only has a small design win for 2016--2018.

I still think it is possible for Zen to win some more designs possibly. But GCN 2 over KNH? Not for the uses those companies want. While Larrabee might have failed to ever become a desktop GPU it turned into a killer HPC card and has been doing very well. It is solely designed for that market while GCN is designed for multiple, gaming/design/HPC, which gives it a distinct disadvantage in the HPC market.

 

[8350rocks]

We already have a good idea of how good Zen is or isn't. No known design win uses Zen. Customers/partners chose Cavium, Intel, and IBM CPUs.

There are not even ES performance leaks at the moment. The fact that there are no design wins right now when chips will not ship for roughly a year still...not concerning.

Exactly, I think we're jumping the gun declaring an unfinished design with no physical silicon a dud before we know anything about it.

Irrelevant arguments, because none of the companies has ES performance leaks. The chips that won have not even been tape-out: Nvidia Volta is not taped-out, Power 9 is not taped-out, KNH Xeon is not taped-out... they are all in the design office still. Nvidia Volta and KNH Xeon will ship on 2017.

Maybe we would start speculating why none of those important customers chose AMD Zen and GCN2. And why AMD only has a small design win for 2016--2018.

I still think it is possible for Zen to win some more designs possibly. But GCN 2 over KNH? Not for the uses those companies want. While Larrabee might have failed to ever become a desktop GPU it turned into a killer HPC card and has been doing very well. It is solely designed for that market while GCN is designed for multiple, gaming/design/HPC, which gives it a distinct disadvantage in the HPC market.

This is a major point here...

The comparison is for HPC designs...and POWER9, KNH and VOLTA are all going to be aimed squarely at HPC performance. They are going to do nothing in consumer HEDT, and while POWER9 will make a huge splash in commercial servers, etc. too...I doubt the other stuff has near the impact of POWER9 because of the established clientele IBM has in server market shares. NVidia's Volta will have zero impact on anything AMD does, outside of HPC designs, and KNH will likely make a small splash in the x86 server market as a co-processor...but it will be a very small %, of an already small %, of the world's server markets (x86 server is relatively small comparatively).

 

[jimmysmitty]

We already have a good idea of how good Zen is or isn't. No known design win uses Zen. Customers/partners chose Cavium, Intel, and IBM CPUs.

There are not even ES performance leaks at the moment. The fact that there are no design wins right now when chips will not ship for roughly a year still...not concerning.

Exactly, I think we're jumping the gun declaring an unfinished design with no physical silicon a dud before we know anything about it.

Irrelevant arguments, because none of the companies has ES performance leaks. The chips that won have not even been tape-out: Nvidia Volta is not taped-out, Power 9 is not taped-out, KNH Xeon is not taped-out... they are all in the design office still. Nvidia Volta and KNH Xeon will ship on 2017.

Maybe we would start speculating why none of those important customers chose AMD Zen and GCN2. And why AMD only has a small design win for 2016--2018.

I still think it is possible for Zen to win some more designs possibly. But GCN 2 over KNH? Not for the uses those companies want. While Larrabee might have failed to ever become a desktop GPU it turned into a killer HPC card and has been doing very well. It is solely designed for that market while GCN is designed for multiple, gaming/design/HPC, which gives it a distinct disadvantage in the HPC market.

This is a major point here...

The comparison is for HPC designs...and POWER9, KNH and VOLTA are all going to be aimed squarely at HPC performance. They are going to do nothing in consumer HEDT, and while POWER9 will make a huge splash in commercial servers, etc. too...I doubt the other stuff has near the impact of POWER9 because of the established clientele IBM has in server market shares. NVidia's Volta will have zero impact on anything AMD does, outside of HPC designs, and KNH will likely make a small splash in the x86 server market as a co-processor...but it will be a very small %, of an already small %, of the world's server markets (x86 server is relatively small comparatively).

The biggest benefit of KNH to Intel is how it can translate eventually to their CPU designs. That is about it but it is a HPC part.

Zen needs to focus more on the server market, much like AMD used to do with their designs where they focused on server then trickled down to consumer. They used to have a pretty good chunk of the server market and because it costs the same to produce a CPU of the same specs for server and consumer yet server can demand a much higher price making the margins much better.

That is where AMD made most of their money during the K8 days, which is another reason why the consumer OEMs not selling them didn't matter as their margins there were very small in comparison. I mean do you want to sell a CPU at $500 or $3000?

That is where Intel focuses. Most of the new technology we are getting is really with servers in mind. A high bandwidth interconnect (DMI/QPI), fast lanes for add in cards (PCIe), M.2/SATAe. ALl designed to remove bottlenecks in servers then trickled down to consumers.

If AMD focuses Zen on that, to take back some market share in the server world, then they will profit way more than trying to focus on the consumer market, especially considering how it has slowed down due to tablets and hybrid laptops (Surface, Asus Transformer etc).

 

[cdrkf]

We already have a good idea of how good Zen is or isn't. No known design win uses Zen. Customers/partners chose Cavium, Intel, and IBM CPUs.

There are not even ES performance leaks at the moment. The fact that there are no design wins right now when chips will not ship for roughly a year still...not concerning.

Exactly, I think we're jumping the gun declaring an unfinished design with no physical silicon a dud before we know anything about it.

Irrelevant arguments, because none of the companies has ES performance leaks. The chips that won have not even been tape-out: Nvidia Volta is not taped-out, Power 9 is not taped-out, KNH Xeon is not taped-out... they are all in the design office still. Nvidia Volta and KNH Xeon will ship on 2017.

Maybe we would start speculating why none of those important customers chose AMD Zen and GCN2. And why AMD only has a small design win for 2016--2018.

I still think it is possible for Zen to win some more designs possibly. But GCN 2 over KNH? Not for the uses those companies want. While Larrabee might have failed to ever become a desktop GPU it turned into a killer HPC card and has been doing very well. It is solely designed for that market while GCN is designed for multiple, gaming/design/HPC, which gives it a distinct disadvantage in the HPC market.

This is a major point here...

The comparison is for HPC designs...and POWER9, KNH and VOLTA are all going to be aimed squarely at HPC performance. They are going to do nothing in consumer HEDT, and while POWER9 will make a huge splash in commercial servers, etc. too...I doubt the other stuff has near the impact of POWER9 because of the established clientele IBM has in server market shares. NVidia's Volta will have zero impact on anything AMD does, outside of HPC designs, and KNH will likely make a small splash in the x86 server market as a co-processor...but it will be a very small %, of an already small %, of the world's server markets (x86 server is relatively small comparatively).

The biggest benefit of KNH to Intel is how it can translate eventually to their CPU designs. That is about it but it is a HPC part.

Zen needs to focus more on the server market, much like AMD used to do with their designs where they focused on server then trickled down to consumer. They used to have a pretty good chunk of the server market and because it costs the same to produce a CPU of the same specs for server and consumer yet server can demand a much higher price making the margins much better.

That is where AMD made most of their money during the K8 days, which is another reason why the consumer OEMs not selling them didn't matter as their margins there were very small in comparison. I mean do you want to sell a CPU at $500 or $3000?

That is where Intel focuses. Most of the new technology we are getting is really with servers in mind. A high bandwidth interconnect (DMI/QPI), fast lanes for add in cards (PCIe), M.2/SATAe. ALl designed to remove bottlenecks in servers then trickled down to consumers.

If AMD focuses Zen on that, to take back some market share in the server world, then they will profit way more than trying to focus on the consumer market, especially considering how it has slowed down due to tablets and hybrid laptops (Surface, Asus Transformer etc).

Well said, also I'll add I'm betting Zen will be focused quite a bit on power efficiency- as that will tie in well with their semi custom designs. I could see a new console coming out with GCN2 and Zen as it's basis. They have got a new as yet undisclosed semi custom win which is rumoured to be related to Nintendo. I guess we'll find out sooner or later.

Getting wins in the HPC market is certainly a nice 'feather in the cap' as they say, although in reality whilst 1 or 2 big wins are big individual orders, it's unlikely to count for much overall.

 

[8350rocks]

Spoiler

We already have a good idea of how good Zen is or isn't. No known design win uses Zen. Customers/partners chose Cavium, Intel, and IBM CPUs.

There are not even ES performance leaks at the moment. The fact that there are no design wins right now when chips will not ship for roughly a year still...not concerning.

Exactly, I think we're jumping the gun declaring an unfinished design with no physical silicon a dud before we know anything about it.

Irrelevant arguments, because none of the companies has ES performance leaks. The chips that won have not even been tape-out: Nvidia Volta is not taped-out, Power 9 is not taped-out, KNH Xeon is not taped-out... they are all in the design office still. Nvidia Volta and KNH Xeon will ship on 2017.

Maybe we would start speculating why none of those important customers chose AMD Zen and GCN2. And why AMD only has a small design win for 2016--2018.

I still think it is possible for Zen to win some more designs possibly. But GCN 2 over KNH? Not for the uses those companies want. While Larrabee might have failed to ever become a desktop GPU it turned into a killer HPC card and has been doing very well. It is solely designed for that market while GCN is designed for multiple, gaming/design/HPC, which gives it a distinct disadvantage in the HPC market.

This is a major point here...

The comparison is for HPC designs...and POWER9, KNH and VOLTA are all going to be aimed squarely at HPC performance. They are going to do nothing in consumer HEDT, and while POWER9 will make a huge splash in commercial servers, etc. too...I doubt the other stuff has near the impact of POWER9 because of the established clientele IBM has in server market shares. NVidia's Volta will have zero impact on anything AMD does, outside of HPC designs, and KNH will likely make a small splash in the x86 server market as a co-processor...but it will be a very small %, of an already small %, of the world's server markets (x86 server is relatively small comparatively).

The biggest benefit of KNH to Intel is how it can translate eventually to their CPU designs. That is about it but it is a HPC part.

Zen needs to focus more on the server market, much like AMD used to do with their designs where they focused on server then trickled down to consumer. They used to have a pretty good chunk of the server market and because it costs the same to produce a CPU of the same specs for server and consumer yet server can demand a much higher price making the margins much better.

That is where AMD made most of their money during the K8 days, which is another reason why the consumer OEMs not selling them didn't matter as their margins there were very small in comparison. I mean do you want to sell a CPU at $500 or $3000?

That is where Intel focuses. Most of the new technology we are getting is really with servers in mind. A high bandwidth interconnect (DMI/QPI), fast lanes for add in cards (PCIe), M.2/SATAe. ALl designed to remove bottlenecks in servers then trickled down to consumers.

If AMD focuses Zen on that, to take back some market share in the server world, then they will profit way more than trying to focus on the consumer market, especially considering how it has slowed down due to tablets and hybrid laptops (Surface, Asus Transformer etc).

Exactly, they need to pickup shares in server space, and none of those design wins referenced are really going to impact anything beyond HPC except POWER9, even then...POWER9 and x86 are not going to have much overlap in terms of market segments.

 

[jimmysmitty]

^Yea Fury X is a bit high in price for what it is but it is selling I guess, if you consider a "out of stock" selling as I have yet to see it in stock at Newegg. Hell Fury is out of stock. That or production is just low.

And I can attest to nVidia being the work station GPU of choice. Every workstation I have built for customers or we have bought for our employees has has nVidia in them. CUDA just has a good grip on the market.

 

[bharathbigb]

Guys,Will we see a Carrizo or Excavator based 28nm Next-gen Athlon X4 succeeding the 860K ?