What does Zen need to be in order to get you to buy it?

[jpiniero]

Maybe that's why AMD delayed Zen a bit; so they could do another spin. Even if it doesn't really change the stock clocks since they are keeping to 95 W it seems.

 

[itsmydamnation]

Bottom line is that if AMD's Zen could hit Intel-like frequencies they would have pushed their Zen ES to match Broadwell-E base clock in their Blender demo rather than nerf the Broadwell to make their point.

Stilt also says current Summit Ridge ES chips are shipping at 2.8ghz base and 3.2ghz Max single core turbo. Max single core turbo on BDW-E is 4GHz.

then explain bulldozer ES's.

Remember you have in BDW-E a already know core, an already known cache sytsem and uncore. Compared to a complete new core being brought up for the first time, a new cache system being brought up for the first time and a new uncore being brought up for the first time.

They are in no way comparable.


Also clock/power scaling and Fmax can be two very different things.

so AMD's ES history doesn't agree with you
what we know of the architecture doesn't agree with you
so how are you determining clock speeds again?

 

[Arachnotronic]

then explain bulldozer ES's.

Remember you have in BDW-E a already know core, an already known cache sytsem and uncore. Compared to a complete new core being brought up for the first time, a new cache system being brought up for the first time and a new uncore being brought up for the first time.

They are in no way comparable.


Also clock/power scaling and Fmax can be two very different things.

so AMD's ES history doesn't agree with you
what we know of the architecture doesn't agree with you
so how are you determining clock speeds again?

Can't wait until the final silicon is out and we'll see who winds up right.

 

[itsmydamnation]

Can't wait until the final silicon is out and we'll see who winds up right.

here let me quote it for you.

I wouldn't even try to guess Fmax at this point what makes you so sure other then Derp Derp AMD?

And at this point you have no ability to justify your position from a technical perspective, I have no idea what clocks will look like, what we can say is that there is nothing obvious from the information that we have that say Zen will clock lower then CON.

if your clock predictions are correct ( cant even hit 4ghz) then it's not because of some magically insight, your just like a broken clock.

 

[The Stilt]

And at this point you have no ability to justify your position from a technical perspective, I have no idea what clocks will look like, what we can say is that there is nothing obvious from the information that we have that say Zen will clock lower then CON.

The biggest obstacle for Zeppelin reaching high Fmax is IMO: The manufacturing process itself & the aggressive L2 latency.

If someone asked me to estimate the Fmax of Piledriver which would have the same L2 latency (and reduced size) as Zeppelin does, my estimation would be around 3.6GHz. And that's on the 32nm SHP SOI process, at it's peak. And personally I think that from the pure performance aspect, the 32nm SHP SOI is still better than ANYTHING else currently available.

Historically AMD has always suffered with large caches. And with Zeppelin they are trying to pull similar latency as Intel, at twice the size and on inferior (at least currently) manufacturing process.

Maybe AMD has figured out something new, but if I had to place a bet, it would be insane not to bet against them.

 

[itsmydamnation]

Remember Latency does not equal array speed. The very high latency of BD was because of the horrible L1 to L2 interconnects not becuase the arrays are slow, thats why you hit fmax for the L2 because the arrays are running very fast. So factor that in when you consider the lowering of latency, but also remember how caches work, Zen is a high associations cache that is quite small for a L2, that will help with clocking as well.

 

[TheELF]

And at this point you have no ability to justify your position from a technical perspective, I have no idea what clocks will look like, what we can say is that there is nothing obvious from the information that we have that say Zen will clock lower then CON.

The only real information that we have is that EX has higher IPC and lower top clocks then PD and that there are quite a number of locked models as well...
AMD is trying to turn things around by showing efficient CPUs at low TDP.
If you project that only piece of real information to ZEN it does not look well,even lower clocks and even more locked CPUs...

Yup on the technical front there is zero info so anything anybody says is speculation,so stop sounding like a brocken clock by attacking anybody who says anything negative about AMD/ZEN.

 

[itsmydamnation]

The only real information that we have is that EX has higher IPC and lower top clocks then PD and that there are quite a number of locked models as well...

thats complete bull,

we know what the high level architexture looks like
http://www.anandtech.com/show/10591...t-2-extracting-instructionlevel-parallelism/8
we know what instructions can be issued to what ports
http://www.planet3dnow.de/cms/wp-content/uploads/2015/10/Zen.png
we know how compilers should treat the L3
http://dresdenboy.blogspot.com.au/2016/02/amd-zeppelin-cpu-codename-confirmed-by.html

We know about things AMD have done research in that are in the core (stack cache, bridged FMA etc). We have a "good" idea of pipeline length, we also know what BD did bad and what it did "well".

AMD is trying to turn things around by showing efficient CPUs at low TDP.
If you project that only piece of real information to ZEN it does not look well,even lower clocks and even more locked CPUs...

So a Zen ES running @3ghz doing a throughput workload is now targeting low TDP and clocking "LOW". Ignoring that AMD have said released products will be higher clocked. Remember 8 core BD at this stage was clocking @2.8ghz........

Where is this locked CPU talk comming from?

Yup on the technical front there is zero info so anything anybody says is speculation,so stop sounding like a brocken clock by attacking anybody who says anything negative about AMD/ZEN.

So now your showing your bias, all i want is a technical reason for a position, when you have a technical position it can then be discussed and we can all learn something. I dont care if your negative, wouldn't i by your logic be attacking Slilt insistently, oh look im not ..... funny that.........

 

[The Stilt]

Remember 8 core BD at this stage was clocking @2.8ghz........

Are you expecting Zeppelin to release in Q3 2017?

 

[itsmydamnation]

Are you expecting Zeppelin to release in Q3 2017?

Well from what i remember it was more around 7 months from those leaks until release (march'ish to October) .

We saw Zen running @3ghz over a month ago and now the release date is supposed to be in Q1. So if it launches march its 7months , if it launches jan its 5. BD birth wasn't exactly smooth sailing, Zen appears to be much happier.

If you go back to when people like Thevenin (semiaccurate) Neilz (beyond3d) saw Zen it will be ~1 year.

 

[The Stilt]

Well from what i remember it was more around 7 months from those leaks until release (march'ish to October) .

We saw Zen running @3ghz over a month ago and now the release date is supposed to be in Q1. So if it launches march its 7months , if it launches jan its 5. BD birth wasn't exactly smooth sailing, Zen appears to be much happier.

If you go back to when people like Thevenin (semiaccurate) Neilz (beyond3d) saw Zen it will be ~1 year.

Ok.
I just asked because Bulldozer reached significantly higher (than 2.8GHz) over a year prior the launch, and the process wasn't working properly at that point as you said.

 

[TheELF]

So now your showing your bias, all i want is a technical reason for a position, when you have a technical position it can then be discussed and we can all learn something. I dont care if your negative, wouldn't i by your logic be attacking Slilt insistently, oh look im not ..... funny that.........

Yeah we have a lot of theoretical info,but ZERO info on how it's implemented ZERO real measurements,ZEN at launch might be clocked at 5Ghz or at 2.8Ghz anything we do is just speculation.
You speculate that AMD has improved their uarch so much that they will have 8 ~i7 haswell cores running at ~3Ghz (at least) within 95w TDP others speculate that ZEN will just be a slight improvement over EX,nobody can tell the truth from just theoretical stats.

 

[mohit9206]

I'm currently on sandy bridge Pentium so for me in order to choose Zen it would have to be $100-120 true quad core with performance on par with current Skylake i3.I will pair it with either rx460 or 1050 with a goal of locked 60 fps@900p in most games at medium/high settings.

 

[DrMrLordX]

The average IPC improvement over PD should be 15.5% (according to AMD) and that's quite close to the actual average. However there are also MANY cases where Excavator performs the same or even worse than Piledriver. Still far from "smashing", even if it was always 15.5% faster IMO.

If you look at cases like SuperPi or wprime, XV's lead over PD is staggering. Clearly the potential is there, whether it's potential for XV to succeed or for PD to fail (miserably). The more results I see from XV, the more I realize that it's all over the damn map. Watching XV lose to SR in gaming benchmarks is cringe-worthy though.

Jim Keller is just one man.

What is a man but a miserable pile of secrets badly written one-liners?

 

[Arachnotronic]

edit: nevermind.

 

[The Stilt]

If you look at cases like SuperPi or wprime, XV's lead over PD is staggering. Clearly the potential is there, whether it's potential for XV to succeed or for PD to fail (miserably).

Yeah due the doubled L1D and slightly faster L2, not due the IPC itself. The same way Skylake, which performs ~129% better than Piledriver in SuperPI doesn't have even remotely that much higher IPC in extremely legacy workloads (which SuperPI is).

 

[Insert_Nickname]

The more results I see from XV, the more I realize that it's all over the damn map.

It helps if you view it as a quad core for integer workloads, and a dual core for FP. Essentially the module design at work.

My 845 can actually match my old 920 in integer, but fails miserably at anything resembling FP.

 

[DrMrLordX]

Yeah due the doubled L1D and slightly faster L2, not due the IPC itself. The same way Skylake, which performs ~129% better than Piledriver in SuperPI doesn't have even remotely that much higher IPC in extremely legacy workloads (which SuperPI is).

Right, cache is the issue there. It's the same deal with all those games where XV is actually as slow as - or slower than - Kaveri: small L2 is the culprit. Cache has a lot to do with defining where XV sits wrt predecessors.

It helps if you view it as a quad core for integer workloads, and a dual core for FP. Essentially the module design at work.

My 845 can actually match my old 920 in integer, but fails miserably at anything resembling FP.

It's certainly not "as fast as Piledriver/Vishera" though.

 

[Insert_Nickname]

It's certainly not "as fast as Piledriver/Vishera" though.

With 2133MHz memory my 845 matches or exceeds my other 6800K, despite a 600MHz clock deficit. I'd call that pretty impressive, if we weren't talking about virtually the same level of performance as the 3 year old 6800K.

Since AMD can get BR to 4.2GHz@65W stock, EX doesn't look half bad for basic use. Those resent 4.8GHz OC results look quite good, though I wonder how much power it pulls at that kind of frequency.

 

[Glo.]

It would have to "amaze" me from technological perspective.

Higher IPC than Broadwell-EP, higher core clocks(even by 100 MHz), lower TDP, and 999$ price for 8 core 3.5 GHz/4.0 GHz, unlocked Black edition.

Then I would might consider it.

Other version: 15% lower IPC, higher core clocks than Broadwell-EP, lower TDP, and 599-699$ price tag for 8 core black edition. But this would be only amazing as a value proposition. Not technological one.

 

[cytg111]

Yeah due the doubled L1D and slightly faster L2, not due the IPC itself. The same way Skylake, which performs ~129% better than Piledriver in SuperPI doesn't have even remotely that much higher IPC in extremely legacy workloads (which SuperPI is).

Dude, you are showing colors.
IPC without respect to L1 or L2 or ?
Please define your labaratory standars for defining "IPC". - You know, so we can measure them.
-"Facts without data is dreams." - quote cytg

 

[TheELF]

Dude, you are showing colors.
IPC without respect to L1 or L2 or ?
Please define your labaratory standars for defining "IPC". - You know, so we can measure them.
-"Facts without data is dreams." - quote cytg

Yeah I would love for AMD to do the same,for the exact same reasons.
"Facts without data is dreams."
Yup,a lot of dreaming folks in this here forums.

 

[The Stilt]

Dude, you are showing colors.
IPC without respect to L1 or L2 or ?
Please define your labaratory standars for defining "IPC". - You know, so we can measure them.
-"Facts without data is dreams." - quote cytg

I think you either intentionally or unintentionally missed the whole point...

 

[ultimatebob]

For me to want one, it has to be a 200% performance improvement over my existing Core i7 3770 with a price under $300.

I'm not really looking for a new CPU right now. The one I have now is fast enough for my existing workload.

 

[IEC]

Great performance, with and without OC. I don't care about power consumption.