[Techpowerup] AMD "Zen" CPU Prototypes Tested, "Meet all Expectations"

[AtenRa]

Above, Purley systems in H1-2017. Still waiting for your H2-2017 imaginary roadmap. Don't make me laugh with that May roadmap divided by years.

Are you trolling or what ??? On the roadmap above Purley is not H1 2017.

E5-2600 V4 (Broadwell) was scheduled for Q1 2016, now it is pushed to Q2 or even Q3 depending on the production start, lets say Q2 2016.

E7 8800/4800 V4 (Broadwell) was scheduled according to this roadmap for Q3 2016 and would be replaced 12 months (4 Quarters) later by Purley.

E5-4600 V4 (Broadwell) was scheduled according to this roadmap for Q4 2016 and would be replaced by Purley 3 Quarters later.

 

[Sweepr]

From where have you heard that Intel would develop a new Skylake core for Skylake-E that significantly differs from the one in 1151 desktop Skylake? They've not done that before on the E series.

And do you really think it would have substantially higher IPC than 1151 Skylake anyway? Intel have been struggling to just get 5% annual performance increase between CPU generations, and that's when going from one generation to the next.

There are scarce details about core count, clockspeeds, etc. For all we know it might offer 10-12C at the high-end ($1000), especially if AMD delivers. Servers are getting 55% more cores from Haswell to Skylake. I'm not expecting a big (if any) IPC jump but we know it's not the same core as client Skylake. Also upgrading the enthusiast lineup would give them some room to offer hexa-cores or octo-cores with Cannonlake. Wait, you don't think Intel will stand still and watch if AMD delivers Haswell-E performance at lower prices, right? Lol.

Are you trolling or what ??? On the roadmap above Purley is not H1 2017.

Yup, I know it's hard to accept, but... it is.

E5-2600 V4 (Broadwell) was scheduled for Q1 2016, now it is pushed to Q2 or even Q3 depending on the production start, lets say Q2 2016.

E7 8800/4800 V4 (Broadwell) was scheduled according to this roadmap for Q3 2016 and would be replaced 12 months (4 Quarters) later by Purley.

Broadwell-EP is H1-2016 and will get replaced by Skylake-EP in H1-2017. See? Pretty easy to understand.
Purley is, according to Intel, the biggest platform advancement since Nehalem. They obviously want it out as soon as possible, your denial doesn't change the facts. By the way, when is server Zen coming out? I hope earlier than that.

 

[TheELF]

From where have you heard that Intel would develop a new Skylake core for Skylake-E that significantly differs from the one in 1151 desktop Skylake?

Who said anything about significantly?
They could have the refresh(ed) cores, even if they are only 100Mhz higer.

 

[Fjodor2001]

That's just a general description of Skylake. Where does it say Skylake-E will get a new improved CPU core that differs significantly from socket 1151 Skylake?

There are scarce details about core count, clockspeeds, etc. For all we know it might offer 10-12C at the high-end ($1000), especially if AMD delivers. Servers are getting 55% more cores from Haswell to Skylake.

So increased Skylake-E count over Haswell-E and Broadwell-E is just a speculation.

As for server CPUs, that's another story. But it was not what we were discussing.

Wait, you don't think Intel will stand still and watch if AMD delivers Haswell-E performance at lower prices, right? Lol.

Do you expect that Intel can pull significantly increased IPC out of the hat just like that in no time if they desire? Haven't you seen how they've struggled to increase IPC lately? Lol right back.

 

[Fjodor2001]

Who said anything about significantly?
They could have the refresh(ed) cores, even if they are only 100Mhz higer.

Sweepr implicitly did, since he objected to this comment earlier:

In that case so should Haswell-E and Broadwell-E too, because Skylake-E won't bring much IPC increase over those anyway.

 

[ShintaiDK]

The new core theory comes from the AVX512 implementation and due to speculations of the L2 etc required.

 

[Sweepr]

That's just a general description of Skylake. Where does it say Skylake-E will get a new improved CPU core that differs significantly from socket 1151 Skylake?

I never said significantly, so you might want to edit that, otherwise you're just trolling. Dedicated server and client IP configurations and AVX-512 support qualifies it as a different core.

So increased Skylake-E count over Haswell-E and Broadwell-E is just a speculation.

Absolutely. I will be the first to post the roadmaps though, don't worry. Meanwhile, can you provide benchmarks proving this, or is it only speculation?

You're just bitter because in 2016 you'll be sitting on an expensive and slow 4 core Intel CPU, while others will be using a cheaper and faster 8 core AMD CPU.

As for server CPUs, that's another story. But it was not what we were discussing.

You do know Intel's E lineup is based on server dies, right? So they're relevant to the discussion. If servers are getting 55% more cores then it's not outrageous to expect 25% or even 50% more cores for enthusiast desktops (just like Haswell-E/EP before it).

Do you expect that Intel can pull significantly increased IPC out of the hat just like that in no time if they desire? Lol right back.

You need some AMDZone lessons if that's the best you can do.
Selective quoting now too?

I'm not expecting a big (if any) IPC jump but we know it's not the same core as client Skylake.

 

[Fjodor2001]

So in the end, after all this rambling, you did in fact have nothing of substance which indicates that my original statement that you objected to was incorrect:

In that case so should Haswell-E and Broadwell-E too, because Skylake-E won't bring much IPC increase over those anyway.

Because you've admitted IPC is not likely to be much higher on Skylake-E. And any increased core count for Skylake-E is just speculation.

 

[Sweepr]

So in the end, after all this rambling, you did in fact have nothing of substance which indicates that my original statement that you objected to was incorrect:

Because you've admitted IPC is not likely to be much higher on Skylake-E. And any increased core count for Skylake-E is just speculation.

Just like Zen performance predictions, yes, pretty much all speculation. But it's funny how nervous you and the rest of the ADF get when anyone mentions the slightest possibility of an Intel reaction if Zen delivers on its performance claims.

I could also say Zen will not bring much performance increase and you would have 'nothing of substance' to prove it wrong, other than a vague 40% better IPC claim.

 

[Fjodor2001]

Just like Zen performance predictions, yes, pretty much all speculation. But it's funny how nervous you and the rest of the ADF get when anyone mentions the slightest possibility of an Intel reaction if Zen delivers on its performance claims.

Why would I be nervous about that? I would have no problem with AMD Zen forcing Intel to provide better CPUs at lower prices. In fact that's what I desire - competition.

I could also say Zen will not bring much performance increase and you would have 'nothing of substance' to prove it wrong, other than a vague 40% better IPC claim.

The public 40% IPC increase announcement by AMD is "vague"? What better source can we expect to have at this point? Are you requesting e.g. independent benchmarks 1 year ahead of release? Is that realistic?

Are you really being serious when you expect us to put your personal speculation that "Skylake-E may have 12 cores" at the same level of trust as the public announcement from AMD?

 

[Sweepr]

Why would I be nervous about that? I would have no problem with AMD Zen forcing Intel to provide better CPUs at lower prices. In fact that's what I desire - competition.

Me too. I would gladly accept a $300 6C/12T (or even 8C/16T) CPU from Intel and if Zen forces them to do that, then props to AMD.

40% IPC improvement and 8C/16T is not a significant improvement over current AMD CPUs!?

Maybe, but can you provide a source with the benchmarks they used to arrive at that number? If not, that doesn't really tell much, does it? Could be a best case scenario. Please add Zen lineup clockspeeds while at that because we can't imply performance based on IPC alone.

And you call the public 40% IPC increase announcement by AMD "vague"? What better source can we expect to have at this point? Are you requesting e.g. independent benchmarks 1 year ahead of release? Is that realistic?

They usually specify the benchmarks they use in this IPC comparisons, see Carrizo (Excavator) for example.

So are you really being serious when you expect us to put your personal speculation that "Skylake-E may have 12 cores" at the same level of trust as the announcement from AMD?

Yours is not exactly any better.

You're just bitter because in 2016 you'll be sitting on an expensive and slow 4 core Intel CPU, while others will be using a cheaper and faster 8 core AMD CPU.

If you think it won't happen then you're suggesting Intel will stand still and stick to 8-cores at the top (desktop) while servers are getting 55% more cores and AMD could be playing catch up with their enthusiast lineup. Yes, surely your personal speculation is much more likely. Lol.

 

[Haserath]

From where have you heard that Intel would develop a new Skylake core for Skylake-E that significantly differs from the one in 1151 desktop Skylake? They've not done that before on the E series.

And do you really think it would have substantially higher IPC than 1151 Skylake anyway? Intel have been struggling to just get 5% annual performance increase between CPU generations, and that's when going from one generation to the next. So how could then a tweaked Skylake core for Skylake-E really all of a sudden provide a significant jump in IPC over that?

Intel hasn't had a different core for server before. Assuming Skylake-E is a Xeon under a different name, and David Kanter's suspicion on the L2 is correct, it will have higher performance per clock. But it is only an assumption.

How much would an 8-way 512KB cache affect performance over the 256KB 4-way? 5%? 10%? 2%?

I'm more inclined to believe Intel isn't struggling for IPC improvements yet but have different design targets in mind. Making Skylake faster would make it harder to fit into core M.

 

[xpea]

I seriously doubt they will reach 4ghz at 95 watts on an 8 core part.

Maybe on a quad core, but even on 14nm intel quads are 90+ watts at 4ghz.

but without an iGPU that takes 50% of intel die space...
still, looks like a very optimistic target

 

[Fjodor2001]

Yours is not exactly any better.

Really? We should put your personal speculation about Skylake-E having 12 cores at the same level of trust as a public announcement by AMD about Zen's IPC?

 

[cytg111]

Me too. I would gladly accept a $300 6C/12T (or even 8C/16T) CPU from Intel and if Zen forces them to do that, then props to AMD....

I dont see the sex appeal in 6 much less 8 cores for my daily tasks(coding, compiling, running threaded jobs, news, gaming).. For the threaded content where I could benefit is small batches that is under 0.01% of the total workload, that time of month I can wait 30 mins rather than 15 without dying.
I want a race back to ST performance, thats what I want. Single. Threaded. Performance. What the *, we're actually on the brink on regression on that front.. wtf.

 

[Sweepr]

Really? We should put your personal speculation about Skylake-E having 12 cores at the same level of trust as a public announcement by AMD about Zen's IPC?

I see you like repetitions, probably because you have 0 facts to answer the rest of my post. I can go on.

You're just bitter because in 2016 you'll be sitting on an expensive and slow 4 core Intel CPU, while others will be using a cheaper and faster 8 core AMD CPU.

Are you suggesting Intel will stand still and stick to 8-cores at the top (desktop Skylake-E) while servers are getting 55% more cores and (at the same time) AMD could be playing catch up with their enthusiast lineup? Not that Skylake-E would probably need more than 8C/16T to stay at the top anyway, but I see a 10C/20T bump (12C/24T if we are really optimistic) happening for the halo $1000 model, while lower-end models might get 6 or 8 cores. Haswell-EP has 50% more cores than Ivy Bridge-EP and desktops got 33% more cores from the Ivy Bridge-E to Haswell-E transition so yes, this might very well happen.

From the above and this thread's comments we can see your personal speculation is a lot worse. If we were to believe your jokes Skylake wouldn't be on the market till god knows when in 2016.

I dont see the sex appeal in 6 much less 8 cores for my daily tasks(coding, compiling, running threaded jobs, news, gaming).. For the threaded content where I could benefit is small batches that is under 0.01% of the total workload, that time of month I can wait 30 mins rather than 15 without dying.
I want a race back to ST performance, thats what I want. Single. Threaded. Performance. What the *, we're actually on the brink on regression on that front.. wtf.

Me too, but I think top notch ST performance will stay with mainstream models like Core i7 6700K and its successors. Even if Kaby Lake is another 5% bump Skylake GT4e with eDRAM should provide some serious per core performance by the time it arrives, if Core i7 5775C is anything to go by.

 

[inf64]

I expect Zen core to be roughly 10% below Haswell on average. This would put Zen+ core around Skylake level which is what AMD was initially targeting (my guess). Still this is at equal clocks and I have no clue where the clocks will land on GF 14nm node. They could be mid to high 3GHz (3.4-3.7GHz) or be closer to 3GHz.

8C/16T configuration, even at lowish clocks could be a popular SKU since it could be OCed if AMD offers an unlocked version. Haswell-E level of MT and close to that in ST should be enough even with Skylake/Broadwell type of cores on intel side. It all comes down to price and perf/watt on whole platform level. AVX256/512 is a moot point as there is effectively no software that uses it on client side.

 

[Fjodor2001]

I see you like repetitions, probably because you have 0 facts to answer the rest of my post. I can go on.

<BS cut>

So instead of answering the questions, you try to steer away the topic to something completely different that was not being discussed, and then accuse me of not answering questions related to that new topic. Nice try.

Oh, and BTW, that old quote from a different thread you've kept repeating like 1000 times in 100 threads has already been explained numerous times. Yes, it is speculation, and yes it is written in the context of the expectations people had in the thread it was written. But I know you are having just another case of intentional amnesia to create some drama and derail the thread. You are hilarious!

 

[Sweepr]

So instead of answering the questions, you try to steer away the topic to something completely different that was not being discussed.

No, that's what you do. I'm still waiting for the benchmarks used in the Zen IPC claim, and final model clockspeeds, otherwise can we conclude your performance speculation are completely baseless (pure wishful thinking) and we should wait for actual tests? I provided you an example where Intel increased the core count for both desktop enthusiast and server chips, so while 10C/20T+ for Skylake-E is speculation they did something similar in the past and have even stronger reasons to do it again if your dreams about Zen come remotely close to true.

Oh, and BTW, the quote you've kept repeating like 100 times has already been explained numerous times. Yes, it is speculation, and yes it is written in the context of the expectations people had in the thread it was written.

Which is fine when it comes to AMD, but apparently not when it's related to Intel in your book. Got it.

 

[Fjodor2001]

I'm still waiting for the benchmarks used in the Zen IPC claim [...]

Which was not what was being discussed. A recap for you: The discussion between us in this thread started by me mentioning that I think Skylake-E is not likely to have significantly higher IPC and performance than Broadwell-E and Haswell-E. You objected to that, so I asked what you based that objection on. So far you've only mentioned your personal speculation that "perhaps Skylake-E will have 12 cores", which you suggest we should put at the same level of trust as public announcements about Zen's IPC from AMD. But we all know it's ridiculous to put that much faith in personal speculation.

So now that you realize your case is falling apart, you're trying to steer away the discussion to some completely new topic to avoid answering the questions regarding the original topic. Then you accuse me of not answering questions related to this new topic you invented. I rest my case.

 

[Sweepr]

Which was not what was being discussed. A recap for you: The discussion was whether Skylake-E is likely to have significantly higher performance than Broadwell-E and Haswell-E. I said that is not likely, to which you objected.

Just like we can't count out Zen right now, I wouldn't make any prediction about Skylake-E at this point. Too many unknown variables like core count, clockspeeds, IPC, etc. It needs to convincingly beat the competition if Intel wants to stay at the top.

So far you've only mentioned your personal speculation that "perhaps Skylake-E will have 12 cores", which you suggest we should put at the same level of trust as public announcements about Zen's IPC from AMD. We all know that is ridiculous.

So far you failed to provide performance numbers to support some of your ridiculous claims, like this one:

You're just bitter because in 2016 you'll be sitting on an expensive and slow 4 core Intel CPU, while others will be using a cheaper and faster 8 core AMD CPU.

 

[dark zero]

Broadwell-EP is H1-2016 and will get replaced by Skylake-EP in H1-2017. See? Pretty easy to understand.

People will strongly avoid Broadwell-EP then... and that will heavily hurt Intel unless they skip Broadwell-EP.

 

[ShintaiDK]

People will strongly avoid Broadwell-EP then... and that will heavily hurt Intel unless they skip Broadwell-EP.

By people you mean companies, no.

 

[dark zero]

By people you mean companies, no.

Don't think so.... just one skipping of generation means millions lost to Intel and they won't allow that.
Also there is the risk that Purley ends to be another Itanium.

 

[TheELF]

Companies refresh their equipment on a predetermined schedule,every 2 to 3 years and not everytime something new comes out.